A Generic Review of Integrating Artificial

Intelligence in Cognitive Behavioral Therapy

Meng Jiang1 , Qing Zhao1 , Jianqiang Li1 , Fan Wang2 ,
Tianyu He2 , Xinyan Cheng2 , Bing Xiang Yang2 ,
Grace W.K. Ho3 , Guanghui Fu4*
1 School of Software Engineering, Beijing University of Technology,
Beijing, China.
2 School of Nursing, Wuhan University, Wuhan, China.
arXiv:2407.19422v1 [cs.AI] 28 Jul 2024

3 School of Nursing, Joint Research Centre for Primary Health Care, The

Hong Kong Polytechnic University, Hong Kong, China.

4 Sorbonne Université, ICM, CNRS, Inria, Inserm, AP-HP, Hôpital de la

Pitié-Salpêtrière, Paris, France.

*Corresponding author(s). E-mail(s): [email protected];

Abstract
Cognitive Behavioral Therapy (CBT) is a well-established intervention for mit-
igating psychological issues by modifying maladaptive cognitive and behavioral
patterns. However, delivery of CBT is often constrained by resource limitations
and barriers to access. Advancements in artificial intelligence (AI) have pro-
vided technical support for the digital transformation of CBT. Particularly, the
emergence of pre-training models (PTMs) and large language models (LLMs)
holds immense potential to support, augment, optimize and automate CBT deliv-
ery. This paper reviews the literature on integrating AI into CBT interventions.
We begin with an overview of CBT. Then, we introduce the integration of AI
into CBT across various stages: pre-treatment, therapeutic process, and post-
treatment. Next, we summarized the datasets relevant to some CBT-related
tasks. Finally, we discuss the benefits and current limitations of applying AI to
CBT. We suggest key areas for future research, highlighting the need for fur-
ther exploration and validation of the long-term efficacy and clinical utility of
AI-enhanced CBT. The transformative potential of AI in reshaping the practice
of CBT heralds a new era of more accessible, efficient, and personalized mental
health interventions.

1
 Keywords: Artificial intelligence, Mental health, Cognitive behavioral therapy, Large
language model, Machine learning, Deep learning

1 Introduction
Mental health issues have become increasingly prevalent, especially in the post-
pandemic era Penninx et al (2022). Psychological distress brought on by COVID-19
has added to the existing mental health crisis, leading to higher rates of depression,
anxiety, and suicidal ideation. Consequently, there is an urgent need for effective and
accessible mental health interventions. Cognitive Behavioral Therapy (CBT) is widely
recognized as one of the most important psychological interventions for addressing
a range of mental health issues Foreman and Pollard (2016); David et al (2018).
As the gold standard for treating depression and anxiety, CBT has been integrated
into healthcare systems worldwide. Its primary goal is to identify and restructure
patients’ maladaptive cognitive frameworks to help them develop coping skills, address
behavioral issues, and alleviate symptoms. However, the delivery of CBT, as it is tra-
ditionally conducted individually and face-to-face by a therapist, faces barriers in real
life, including social stigma and limited access to qualified therapists, particularly in
underserved areas. Only 27% of those receiving psychological therapy actually access
standardized care Bandelow et al (2017). To address these challenges, technological
advancements have led to the development of Computer-Based Cognitive Behavioral
Therapy (CCBT) and Internet-Based Cognitive Behavioral Therapy (ICBT) Webb
et al (2017). These variations use computer software and the Internet as a medium
to deliver CBT interventions based on established theories and techniques. Several
established online CBT platforms, such as MoodGYM Australian National Univer-
sity (2024) and 30-Day Self-Service Psychological Expert China Cognitive Behavioral
therapy professional organization (2024), have emerged. These online platforms have
partially alleviated some of the issues related to the shortage of mental health profes-
sionals and poor accessibility Christ et al (2020). However, they still face challenges,
such as limited interactivity and high dropout rates. In addition, they offer interven-
tions that address only general cognitive issues rather than individual specific needs.
Given this challenge, there is an urgent need for more flexible and adaptive forms of
CBT to better meet the diverse needs of different patient populations.
In recent years, rapid advancements in artificial intelligence (AI) technology have
led various industries to explore its innovative applications. AI’s exceptional capabil-
ities in data analysis, pattern recognition, and automation offer vast potential in the
delivery of mental health services, providing new tools and methods for assessment,
diagnosis, and treatment of mental health conditions Lee et al (2021); Malgaroli et al
(2023); Demszky et al (2023a). For instance, deep learning (DL) models can auto-
mate mental health assessments and diagnostic processes, alleviating the workload of
professionals while enhancing the accuracy and efficiency of diagnoses Vuyyuru et al
(2023). Furthermore, AI can also aid in detecting and understanding patients’ unique

2
emotional expressions, thereby offering recommendations for personalized psycholog-
ical interventions and support Assunção et al (2022). Given the shortage of skilled
mental health professionals, innovative AI approaches have been developed to guide
peer-to-peer mental health support Sharma et al (2023a). Additionally, AI-based psy-
chological counseling support systems, such as those utilizing large language models
(LLMs), have also been developed to assist junior counselors in providing online psy-
chological support Fu et al (2023). These studies demonstrate that AI not only expands
the boundaries of traditional mental health services, but also brings new opportunities
for improving their accessibility and utility globally Graham et al (2019); Demszky
et al (2023b). In the realm of CBT, the integration of AI has led to revolution-
ary advancements, particularly with the rise of LLMs. Notable research in this area
includes the development of CBT-specific prompts and tailored datasets, leading to
models like CBT-LLM, which are specifically designed for CBT delivery Na (2024).
Other efforts have resulted in conversational counseling agents based on LLMs, such
as utilizing GPT-2 in CBT to generate human-like textual narratives to provide psy-
chological support Rajagopal et al (2021), providing psychological counseling using
CBT techniques Lee et al (2024), and delivering dialogue modules focused on Socratic
questioning using LLMs like OsakaED and GPT-4 Izumi et al (2024). These advance-
ments have made cognitive behavioral interventions more personalized and precise,
better addressing the needs of diverse patients.
Given the abundance of research emerging in this field, several review articles have
summarized innovative developments in CBT. However, the exsiting reviews focused
mainly on CBT applications Huguet et al (2016); Denecke et al (2022) and the devel-
opment of ChatGPT in redesigning CBT for different age groups and genders Chandra
et al (2023). There is a lack of comprehensive review summarizing the application of
AI’s role across various stages of CBT delivery. To address this, we provide a detailed
literature review of AI’s role in enhancing CBT in this paper.

2 Background
Cognitive behavioral therapy (CBT) stands as a structured, time-limited psycholog-
ical treatment that focuses on the interplay between cognitive processes, emotional
responses, and behavioral patterns. This therapeutic approach operates on the premise
that thoughts, emotions, and behaviors are interconnected, and aims to enhance psy-
chological well-being by addressing maladaptive cognitive patterns and behaviors Beck
and Beck (2011). The theoretical foundation of CBT encompasses both cognitive and
behavioral aspects. Regarding the origins of CBT, there are varying views in the aca-
demic community. However, it is generally accepted that the popularization of the
“cognition” began with Ellis’s Rational Emotive Therapy (RET) in the 1950s, followed
by Beck’s Cognitive Therapy (CT) in the 1960s. Since then, CBT has continuously
integrated various behavioral therapies and theories, evolving into the widely used
psychological treatment seen today.
CBT interventions typically target three domains: cognition, behavior, and emo-
tion McGinn and Sanderson (2001). In the cognitive domain, CBT focuses on
individuals’ cognitive processes, including thoughts, beliefs, and interpretations. This

3
involves aiding clients in identifying and understanding negative or distorted thought
patterns, such as negative automatic thoughts and cognitive distortions (CD) (e.g.,
jumping to conclusions, all-or-nothing thinking and mental filtering). Techniques like
cognitive restructuring (CR) are used to adjust these patterns, fostering a more objec-
tive and positive perspective toward oneself and the world. In the behavioral domain,
CBT addresses individuals’ maladaptive behavioral patterns and habits. Therapists
collaborate with clients to explore their unhealthy behaviors, such as social with-
drawal, avoidance, and substance or alcohol misuse. Techniques such as behavioral
experiments, graded exposure, and behavioral activation (BA) are then employed. Fur-
ther, CBT acknowledges the interplay between the body and the mind. Physiological
responses such as tension, fear, and anxiety can impact cognition and behavior. Thus,
CBT also focuses on regulating physiological responses through techniques such as
deep breathing, progressive muscle relaxation, and physical activity to enhance emo-
tional regulation and psychological well-being. For the emotional domain, CBT targets
clients’ emotional awareness and ways to manage them by aiding clients to recog-
nize negative emotions as they arise, accurately identify them, and utilize appropriate
cognitive and behavioral strategies to support better emotional well-being. The three
domains are intertwined, and CBT typically tailors a range of techniques and strategies
based on the individual’s specific circumstances and needs, aiding in problem-solving,
improving mental health, and building resilience Dobson and Dozois (2021).
CBT has been employed to address a variety of mental health issues, including
anxiety disorder Olatunji et al (2010), depression Tymofiyeva et al (2019), schizophre-
nia Turkington et al (2004), attention deficit and hyperactivity disorder Pan et al
(2019), insomnia Benard and Lukandu (2018), eating disorders Linardon et al (2017),
bipolar disorder Driessen and Hollon (2010), substance use disorders McHugh et al
(2010), and obsessive-compulsive disorder Moody et al (2017). Beyond mental health,
CBT has been explored as a tool for managing chronic health conditions like low
back pain Piette et al (2016); Heapy et al (2017), asthma Parry et al (2012), and
tinnitus Parry et al (2012). Prior to initiating CBT, therapists typically conduct assess-
ments to determine symptom severity and treatment goals. These assessments may
involve dialogue between therapist and client or self-administered tools like the Beck
Depression Inventory (BDI-II) or the Beck Anxiety Inventory (BAI) Bech et al (1996);
Beck et al (1988); Foa et al (1993). Based on this assessment, therapists work with
the clients to establish treatment goals and offer personalized cognitive and behavioral
strategies tailored to their needs.

3 Artificial intelligence in CBT

Integrating AI into CBT presents a myriad of potential applications, promising to
enhance the effectiveness and utility of therapy. To provide a summary of the extant
literature on the integration of AI into CBT across the delivery process, we con-
ducted a literature review by performing searches on the ArXiv and Google Scholar
database without date restrictions. Search terms considered in this review were selected
based on two elements, CBT (e.g., cognitive behavioral therapy, cognitive distortions,

4
cognitive restructuring, and exposure therapy) and technology (e.g., artificial intelli-
gence, machine learning, deep learning, natural language processing, large language
model, chatbot, and virtual reality). Articles that enhance, support, or implement
CBT through AI technology will be selected. Results were categorized based on the
stage at which AI was integrated in the CBT delivery process and were synthesized. In
this section, we present the synthesized findings describing the role of AI in different
stages of the CBT treatment process, as illustrated in Figure 1.
To be clear, this article only covers the stages of CBT that current
AI technology has addressed. Due to the complexity of CBT treatment,
some details and specific treatment stages have not yet been covered by
AI, and thus are not discussed here. Future research and technological
advancements may further expand AI’s role in CBT delivery, encompassing
more stages of the treatment process.

‎Integration of AI in various stages of CBT treatment

‎Disease diagnosis and symptom severity assessment.

‎Identifying cognitive distortions. ‎Mental Health Assessment

‎Emotion analysis.

‎Who will benefit from CBT？

‎CBT or other psychotherapy？ ‎Personalized Treatment Selection

‎Pre-treatment ‎CBT alone or with medication？

‎Basic Concepts of CBT.

‎ ommon Patterns of Negative Emotions, Thoughts,

C
‎and Behaviors.
‎Client psychoeducation

‎ ow to evaluate automatic thoughts using various

H
‎strategies？
‎How to implement cognitive restructuring to adjust ‎The therapeutic process also involves client
‎automatic thoughts？ ‎psychoeducation.

‎ I augmentation of therapeutic strategies for CBT

A
‎delivery.
‎CBT treatment
‎ eveloping AI tools for comprehensive delivery of
D
‎CBT.

‎ eal-time monitoring by mobile APP and wearable

R
‎devices.
‎Real-Time Monitoring and Feedback
‎Collect data from patients during treatment.
‎Therapeutic process

‎ etermine specific treatment outcome predictors or

D
‎potential biomarkers.
‎Early Prediction of CBT Outcomes
‎Prediction of treatment outcome.

‎Therapist treatment quality assessment.

‎Evaluation of Therapists and Clients
‎Predictive analysis of clients treatment adherence.

‎Will the disease recur or not？

‎ ong-term Treatment Response and
L
‎Post-treatment
‎Which stage and data yield the highest accuracy？ ‎Relapse Prediction

Fig. 1: The role of AI in various stages of CBT treatment.

5
3.1 Integration of AI in the Pre-Treatment Stage
A comprehensive assessment is a foundational step in all psychological treatments,
including CBT. This initial assessment involves gathering information on the client’s
history, current issues, and therapeutic goals through structured or semi-structured
interviews, questionnaires, and standardized scales. In CBT, particular emphasis is
placed on assessing the client’s cognitive and behavioral patterns, as well as their
emotional responses. The insights from this detailed assessment guides therapists in
crafting tailored CBT treatment plans that effectively address the client’s concerns,
thereby maximizing therapeutic outcomes. Traditional CBT assessment methods,
while thorough, often rely heavily on manual processes, which can be subjective and
time-consuming. The reliance on the therapist’s clinical skills and professional acumen
can may also limit diagnostic accuracy and treatment efficiency. In contrast, AI tech-
nology, with its advanced data processing and analysis capabilities, offers a powerful
alternative. AI can rapidly analyze vast amounts of client data to identify patterns
and correlations, thereby assisting therapists in assessing patient symptoms more
accurately and swiftly, and in identifying adverse emotions and cognitive distortions.

3.1.1 Mental health assessment

Disease diagnosis and symptom severity assessment
AI is increasingly employed for the diagnosis and assessment of common mental ill-
nesses such as depression and anxiety. Wanderley Espinola et al (2022) proposed a
methodology to support the diagnosis of mental disorders in psychiatric emergency
context using vocal acoustic analysis and machine learning, targeting disorders like
major depressive disorder, schizophrenia, bipolar disorder, and generalized anxiety
disorder. Siddiqua et al (2023) employed a survey questionnaire distributed among
Bangladeshi students with a total of 684 responses were obtained, and utilized ten
machine learning models and two deep learning models to predict three levels of depres-
sion severity (normal, moderate, and extreme) with remarkable accuracy. Vuyyuru
et al (2023)introduced the Trans-CNN, which effectively integrates the strengths of
Transformer and Convolutional Neural Network (CNN) architectures. This model
demonstrates superior performance in identifying depressive disorders compared to
traditional CBT methods and standalone Transformer or CNN models by analyzing
textual data such as patient narratives, treatment records, and diagnostic reports.
Despite these advancements, the existing research is centered around AI mod-
els focused on the diagnosis and assessment of singular psychological disorders, with
limited research dedicated to the development of comprehensive models capable of
simultaneously diagnosing multiple co-occurring psychological conditions.

Identifying cognitive distortions

CBT postulates an individual’s emotions and behaviors are influenced by their percep-
tion and interpretation of events. Cognitive distortions refer to erroneous or irrational
patterns of thinking that can lead to negative emotions and maladaptive behaviors.
During the initial assessment phase of CBT, therapists collaborate with clients to
explore these cognitive patterns and reactions to specific events, aiming to assist the

6
client in identifying and understanding potential cognitive distortions, which were
categorized into ten types by Burns theory Burns and Beck (1999). They include:
all-or-nothing thinking, over generalization, mental filter, disqualifying the positive,
jumping to conclusions, magnification and minimization, emotional reasoning, should
statements, labeling and mislabeling, and blaming oneself or others.
Currently, AI technology facilitates the identification of cognitive distortions pri-
marily through text classification techniques. Researchers utilize various textual data
as inputs to build models and algorithms that automatically identify and categorize
these cognitive distortions. However, due to the lack of publicly available structured
datasets specifically designed for detecting cognitive distortions, researchers often turn
to alternative data sources such as social media data Alhaj et al (2022); Wang et al
(2023b), personal blogs Simms et al (2017), and everyday narratives Xing et al (2017);
Shickel et al (2020); Mostafa et al (2021). Shreevastava and Foltz (2021) compared
five classification algorithms for detecting cognitive distortions using a therapist Q&A
dataset obtained from Kaggle. Tauscher et al (2023) applied NLP methods to iden-
tify cognitive distortions from text messages exchanged between patients with severe
mental illnesses and their clinical therapists. Challenges in this task include dealing
with short texts that lack contextual information and imbalanced data where certain
distortion types are underrepresented, leading to poorer classification performance.
To address these issues, researchers have proposed various solutions. Alhaj et al
(2022) suggest enriching short textual representations to improve cognitive distortions’
classification of the Arabic context over Twitter. It also utilizes a transformer-based
topic modeling algorithm (BERTopic) that employs a pre-trained language model
(AraBERT). Ding et al (2022) tackled data imbalance with approaches like data
augmentation and domain-specific models, demonstrating the effectiveness of the pre-
trained language model MentalBERT. Recent advances have expanded the scope
of cognitive distortion detection to include multimodal datasets. This cross-modal
research approach provides a more holistic perspective, enabling a more comprehensive
detection of cognitive distortions. For instance, Singh et al (2023) proposed a multitask
framework that integrates text, audio, and visual data to detect cognitive distortions,
achieving significant performance improvements over existing state-of-the-art models.
Recently, LLMs have also shown promise in complex psychological tasks such as
identifying cognitive distortions Qi et al (2023); Nazarova (2023); Chen et al (2023);
Lim et al (2024). Qi et al (2023) conduct experiments to compare LLMs and super-
vised learning in cognitive distortion identification and suicide risk classification. The
experimental results indicate that LLMs struggle with accurately identifying complex
cognitive distortions in Chinese social media data, suggesting that deep learning algo-
rithms remain the preferred solution for complex psychological tasks like cognitive
distortion identification. Nazarova (2023) developed TeaBot, an AI tool fine-tuned
on GPT-3, and employs CBT techniques to aid users in identifying and challenging
distorted thoughts. Chen et al (2023) introduced the Diagnosis of Thought (DoT)
framework, which strategically prompts the LLMs to produce diagnosis rationales per-
tinent for cognitive distortions detection. Although the DoT method demonstrates its
capability in classifying cognitive distortions, it also exhibits a notable flaw, namely,

7
the model tends to over-diagnose cognitive distortions even when the user’s state-
ments are benign. Lim et al (2024) addressed this by introducing the ERD framework,
involving extraction, reasoning, and debate among multiple LLM agents to classify
cognitive distortions from user utterances. This approach significantly mitigates the
issue of overdiagnosing cognitive distortions in the DoT method.

Emotion analysis
Emotion analysis is essential for understanding individuals’ emotional states. This
understanding enables therapists or conversational agents to offer more empathetic
support and feedback in CBT interventions, thereby strengthening the therapeutic
alliance and enhancing the interactive experience Brave et al (2005); Provoost et al
(2019). Furthermore, emotional states also serve as indicators of intrinsic goals, observ-
able behaviors, and treatment efficacy. During the therapeutic process, individuals are
more likely to benefit from treatment when they can effectively manage their emo-
tions Mehta et al (2021). Therefore, during the initial assessment phase, therapists
explore the clients’ emotional responses and assist them acquire healthier emotion reg-
ulation strategies. AI-assisted emotion analysis has become increasingly prevalent in
mental health Tanana et al (2021); Assunção et al (2022); Khare et al (2023). Provoost
et al (2019) employed an emotion mining algorithm to assess the overall sentiment
and five specific emotions expressed in texts written by patients during Internet-based
cognitive behavioral therapy (ICBT), finding moderate agreement between the algo-
rithm and human judgment in evaluating the overall sentiment, while the agreement
was low for specific emotions. Patel et al (2019) proposed an intelligent social thera-
peutic chatbot. This chatbot defined several basic emotion labels, and based on these
emotion labels, three deep learning algorithms were employed to extract emotions
from user chat data. Kozlowski et al (2023) introduced Terabot, a conversational sys-
tem that enhanced sentiment and emotion recognition by integrating CBT techniques
and replacing BERT with RoBERTa in a neural language model framework. However,
Striegl et al (2023) pointed out that some emotion recognition methods Fitzpatrick
et al (2017); Provoost et al (2019) categorize emotions into discrete classes, failing to
capture the continuum and diversity of emotions accurately. To address this, they pro-
posed a deep learning-based dimensional text emotion recognition system within the
context of CBT using the ALBERT pre-trained model Lan et al (2019), fine-tuned on
emotion-annotated data for dimensional score prediction.
In recent years, ChatGPT, as a prominent example of conversational AI, shows
potential in assisting individuals with emotion understanding and management.
Rathje et al (2023) initially assessed GPT’s capability for overall sentiment analysis
(positivity, negativity, or neutrality) in English and Arabic texts. They found that
GPT demonstrates effective multilingual sentiment analysis, achieving performance
levels comparable to top-performing machine learning models from previous years.
Furthermore, they examined GPT’s ability to accurately discern more nuanced dis-
crete emotions such as anger, joy, fear, and sadness. The results demonstrated a high
level of consistency between GPT’s performance and human judgement. Elyoseph
et al (2023) also underscored GPT’s superiority over humans in emotional cognition,
as highlighted in their study.

8
 Despite AI’s capabilities in emotion detection and understanding, there remains a
perception that individuals may perceive AI-driven support as lacking genuine emo-
tional resonance compared to human interaction Yin et al (2024). Thus, future research
should focus on making AI responses more conversational and human-like to address
this perception. Additionally, to ensure rigor and utility in real clinical settings, human
cross-validation is required.

3.1.2 Personalized treatment selection: identifying CBT

beneficiaries
Within the realm of precision medicine, aligning clients with the most suitable treat-
ment is a critical and significant objective Ozomaro et al (2013). This involves
selecting an effective treatment plan before commencing therapy, thereby avoiding the
risks associated with ineffective treatments. CBT is a well-established psychotherapy
approach to address mental health issues, however, its effectiveness varies among indi-
viduals experiencing different mental health conditions. This variability highlights the
need to identify which clients are likely to respond favorably to CBT and to assess its
effectiveness in treating specific mental health conditions. Researchers are increasingly
exploring the potential of AI models to predict which clients may benefit from CBT
and to determine the most appropriate treatment for each person Ball et al (2014);
Reggente et al (2018); Vieira et al (2022). Studies suggest that AI can analyze extensive
clinical data to forecast which patients are more likely to respond positively to CBT.
This capability can assist clinicians in selecting suitable candidates for CBT, thereby
optimizing resource allocation. For example, Reggente et al (2018) used machine
learning with cross-validation to investigate the potential of functional connectivity
patterns to predict obsessive-compulsive disorder (OCD) symptom severity in individ-
uals after treatment. The findings have important clinical implications, particularly
in the development of personalized medicine approaches aimed at identifying OCD
patients who would derive the greatest benefit from intensive CBT. Similarly, Vieira
et al (2022) conducted a comprehensive review and quantitative synthesis of previous
studies utilizing machine learning to predict the outcomes of CBT treatment in var-
ious diagnostic categories. They concluded that, on an individual level, the accuracy
of predicting CBT benefit was approximately 74.0%.
Besides CBT, other types of psychological interventions, such as Psychodynamic
Therapy (PDT) and person-centered counseling for depression may be equally effec-
tive. Moreover, not all clients may require high-intensity CBT interventions. Some
individuals might benefit from low-intensity interventions. Traditionally, the trial-and-
error method in psychotherapy can be both time-consuming and costly, with the risk
of clients undergoing multiple ineffective treatments before finding the one that works
best for them. This not only delays effective care but also increases the overall cost of
treatment. Therefore, determining the most suitable treatment for specific individuals
before the delivery of therapy is crucial. Addressing this gap, Delgadillo and Gonzalez
Salas Duhne (2020) and Schwartz et al (2021) had applied machine learning algorithms
to recommend optimal treatment approaches based on patients’ pre-treatment char-
acteristics. This method can contribute to substantial cost savings in mental health
care, making high-quality psychotherapy more accessible and affordable.

9
 Additionally, there is a growing interest in whether combining CBT with medica-
tion or other treatments can enhance outcomes. Several researchers have explored this
possibility Gunlicks-Stoessel et al (2020); Pei et al (2022); Stephenson et al (2023). For
example, Gunlicks-Stoessel et al (2020) found that combining CBT with medication
is equally effective and more long-lasting than medication alone. Furthermore, they
noted that the combination of medication and CBT could increase response rates and
prolong the duration of effectiveness, especially when CBT is administered to clients
responsive to pharmacotherapy.
These research underscores the potential of AI in refining the selection of psy-
chological treatments and highlights the importance of personalized approaches in
enhancing therapeutic outcomes.

3.1.3 Client psychoeducation

In CBT, client education is a crucial step both before and throughout the therapy
process, as shown in Figure 1. Before commencing therapy, psychoeducation involves
introducing the fundamental principles and techniques of CBT, aiding them in under-
standing common patterns of negative emotions, thoughts, and behaviors and coping
strategies McGinn and Sanderson (2001). In the treatment process, psychoeduca-
tion is reflected in teaching clients how to identify and assess automatic thoughts,
how to perform cognitive restructuring, and how to find alternative explanations for
automatic thoughts, thereby adjusting these automatic thoughts. By conveying that
individuals can acquire and master skills to address psychological issues, psychoed-
ucation enhances clients’ sense of self-efficacy and can even contribute to symptom
reduction on its own Cuijpers (1998).
Psychoeducation can be delivered through various formats, including written mate-
rials, videos, audio recordings, as well as organizing specialized lectures and seminars.
The advent of mobile applications has introduced novel avenues for delivering psy-
chological education and support, capitalizing on their convenience and adaptability.
Psychological education has been seamlessly integrated into several mobile applica-
tions, some of which have demonstrated efficacy in treating various mental disorders,
including stress Rose et al (2013), anxiety and depression Harrison et al (2011), eat-
ing disordersCardi et al (2013), post-traumatic stress disorder Reger et al (2013), and
obsessive-compulsive disorder Whiteside et al (2014). For instance, Whiteside et al
(2014) developed the smartphone application “Mayo Clinic Anxiety Coach”, designed
to deliver CBT for anxiety disorders, including OCD. This application comprises a psy-
chological education module that instructs users on the utilization of the application,
conceptualization of anxiety, descriptions of various anxiety disorders, explanations of
CBT, and guidance on assessing alternative forms of therapy. Similarly, Martı́nez-
Miranda et al (2019) developed a mobile-based embodied conversational agent (ECA)
for the prevention and detection of suicidal behavior. The ECA introduces users to a
segment of psychoeducational video in which a combination of cartoon-like images in
motion are accompanied by a narrator explaining the basic elements of CBT. Users
can access the video at any time through the application’s main menu. The advantages
of delivering psychological education through mobile applications are self-evident:

10
information becomes portable, and patients can access it easily. Heng (2021) inte-
grated CBT within immersive gaming experiences to assist individuals suffering from
generalized anxiety disorder (GAD). In this approach, CBT elements are seamlessly
interwoven with the diegetic components of the game, providing psychoeducation in
an engaging and entertaining manner.
AI-powered models facilitate personalized psychoeducation by tailoring modules to
each patient’s understanding and preferences. For example, Bhaumik et al. Bhaumik
et al (2023) introduced MindWatch, a system harnessing AI-driven language mod-
els for early symptom detection and personalized psychological education. Within
the psychological education module, they utilized the foundational Llama 2 model
within the Amazon SageMaker Studio environment to deliver tailored education to
individuals experiencing mental health issues. Numerous chatbots incorporate mod-
ules for psychoeducation as well. For instance, Jang et al (2021) developed Todaki, a
chatbot for managing Attention Deficit Hyperactivity Disorder (ADHD). This chat-
bot offers tailored psychoeducation and brief CBT sessions, enabling individuals with
ADHD to acquire self-help skills for managing their condition effectively. Similarly,
Su et al (2022) created XIAO AN, an AI-assisted psychotherapy robot utilizing multi-
modal signal recognition and natural interaction technology. XIAO AN monitors
emotions and offers brief, comprehensive psychological therapy based primarily on
CBT, integrating psychological education modules.
Through these approaches, integrating CBT principles into various technological
platforms has facilitated the dissemination of psychological education, empowering
individuals to better understand and manage their mental health.

3.2 Integration of AI in CBT therapeutic process

CBT intervention encompasses the implementation of various strategies, including cog-
nitive restructuring, behavioral activation, and exposure therapy, etc. These strategies
are intrinsically connected to cognition, emotions, and behavior. During CBT treat-
ment, psychotherapists select appropriate strategies based on the patients’ symptoms
and needs to provide targeted treatment. In section 3.2.1, we will delve into how AI
technology can enhance each CBT strategy. Following this, Section 3.2.2 will focus on
how researchers are developing chatbots to integrate multiple CBT strategies, offer-
ing comprehensive mental health support to users. Next, Section 3.2.3 introduces
how wearable devices or smartphone applications equipped with AI technology can
now be used to monitor patients’ psychological states during CBT treatment. Finally,
Section 3.2.4 introduces some researchers utilize AI to make early predictions about
the outcomes of CBT treatments.

3.2.1 AI Augmentation of Therapeutic Strategies for CBT delivery

Cognitive restructuring (CR)
Cognitive restructuring (CR) is a fundamental component of CBT, characterized as
a structured, goal-oriented, and collaborative intervention. It involves the patient
and therapist working together to identify and correct irrational thoughts, evalua-
tions, and beliefs by generating more adaptive alternative cognitive patterns Clark

11
(2013). Thereby alleviating negative emotions, improving mental health, and pro-
moting healthier behaviors and coping strategies. To some extent, it is considered
analogous to Cognitive Reappraisal Shurick et al (2012); Zhan et al (2024). Recent
advancements have leveraged AI models to conduct CR effectively. The majority of
these efforts aim to provide evidence that AI language models can effectively engender
reframed thinking in response to negative emotional situations, as well as to compare
the efficacy of different models in CR. For instance, de Toledo Rodriguez et al (2021)
collected a small-scale CBT dataset via the crowd-sourcing platform Prolific. They
validated Google’s T5 transformer and BERT model for their ability to transform
initial negative cognition into more positive or realistic alternatives. Human evalu-
ation revealed that T5-generated reconstructions resembled original human-written
responses more closely, while BERT showed a relatively lower performance but higher
positive sentiment. Maddela et al (2023) introduced PATTERNREFRAME, a novel
dataset that incorporates personas and classical unhelpful thought patterns, extending
the reframing task to include the identification and generation of thoughts corre-
sponding to a given persona and unhelpful pattern. They evaluated various language
models using prompt-based and fine-tuning methods for their efficacy in identifying
and reframing these thoughts. Unlike de Toledo Rodriguez et al (2021) and Mad-
dela et al (2023), who conceptualize CR tasks as a sentence rewriting task, Xiao et al
(2024) emphasize client empowerment over reliance on therapist-driven solutions. They
propose, Helping and Empowering through Adaptive Language in Mental Enhance-
ment (HealMe), a model utilizing Large Language Models (LLMs) for CR through
three steps: distinguishing between situations and thoughts for a rational perspective,
generating alternative perspectives through brainstorming to alleviate negative think-
ing, and providing suggestions that recognize the client’s efforts and promote positive
action.
Previous research primarily focused on the perspective of holistic CR, attempting
to address how language models can be utilized to generate CR. However, Sharma
et al (2023c,b) have begun to delve deeper into exploring various dimensions of CR,
and examining people’s preferences for particular types of restructuring. To be more
specific, Sharma et al (2023c) investigated how negative thinking can be restruc-
tured, how language models can be utilized to facilitate such restructuring, and which
types of restructuring individuals experiencing negative thoughts prefer. They intro-
duced a framework comprising seven linguistic attributes for reconstructing thoughts
and trained a retrieval-enhanced in-context learning model to generate reconstructed
thoughts. Additionally, they conducted a randomized field study on the Mental Health
America (MHA) website. In another work, Sharma et al (2023b) designed and eval-
uated a CR tool based on human language model interaction. This system leverages
language models to assist individuals in various steps of CR, including identifying cog-
nitive traps within thoughts and selecting more actionable, empathetic, or personalized
reconstruction suggestions when reconstructing negative thoughts. Furthermore, they
demonstrated that language models can not only be utilized for generating reassess-
ments but also aid individuals in enhancing their own reassessment abilities. Wang
et al (2024b) argue that while Sharma et al (2023c,b) explored the enhancement
of reframed thoughts within a single attribute in one generation, their efforts were

12
limited in addressing multiple features. Consequently, they developed ReframeGPT,
leveraging GPT-3 as an inference engine to generate and iteratively refine reframed
thoughts across various features, aiming to achieve high-quality reframing. Li et al
(2024) focused on aligning GPT-4’s reconstructive thinking with human performance,
enhancing model performance in CR tasks by understanding the differences between
human and AI-generated reconstructions.
Identifying a patient’s automatic thoughts and emotions is crucial for effective
CR. In traditional face-to-face CBT sessions, therapists refine identified automatic
thoughts through methods such as Socratic questioning. However, when delivering
CR in ICBT, accurately capturing thoughts and emotions poses challenges. Hence,
Furukawa et al (2023) trained the T5 model to predict emotions associated with each
automatic thought. They then compared these predictions with judgments from rele-
vant experts, demonstrating the accuracy of T5 predictions. The application of T5 in
iCBT platforms holds promise for achieving more efficient CR. Shidara et al (2022)
and Jiang et al (2024) also have both underscored the significance of automatically
identifying patients’ cognitive distortions and emotions in CR. In response to this,
corresponding efforts have been made. Particularly, Jiang et al (2024), drawing on
the ABCD model in CBT, employed the pre-trained model ERNIE 3.0 to construct a
hierarchical text classification model for extracting automatic thoughts and emotions
from user statements. Furthermore, they also validated the efficacy of LLMs in this
task.

Behavioral activation (BA)

Behavioral activation (BA) is a essential therapeutic strategy in CBT, focusing on
addressing observable behaviors for intervention Petersen et al (2016). In BA, ther-
apists collaborate with clients to formulate specific goals and plans, encourage the
modification of negative behavior patterns, gradually increase the frequency and diver-
sity of positive activities, and help clients regain a sense of pleasure and meaningful
engagement in daily life. Lewinsohn (1974) and Jacobson et al (1996) demonstrated
that behavioral activation achieves outcomes comparable to full CBT and is com-
parable to pharmacotherapy Dimidjian et al (2006). Individuals with mental disease
may have problems with reduced daily activities, loss of interest, avoidance behavior
and social avoidance. To facilitate the application of BA, researchers have developed
innovative interventions.
Innovative interventions have been developed to facilitate BA application. Dahne
et al (2019) developed a Spanish-language behavioral activation self-help mobile appli-
cation called ¡Aptı́vate!. Within this app, the user identifies individualized values and
associated activities across five life areas, including relationships, daily responsibilities,
recreation, career and education, and health. Evaluation results also demonstrated the
feasibility and preliminary effectiveness of this app. Rohani et al (2020) developed
a mobile health system called MUBS (Mobile-based Behavioral Activation System)
to support patients with depression through behavioral activation therapy. The sys-
tem implements a unique content-based probabilistic recommendation algorithm to
motivate patients to engage in positive behaviors by providing activity directories and
personalized rewards, helping them build awareness of behavioral activation.

13
 While these studies primarily focus on evaluating the efficacy of BA or developing
applications to assist, prompt and motivate patients to actively engage in various
activities, without explicitly linking AI to BA. Against this background, Madhu et al
(2022) and Rathnayaka et al (2022) explore how AI technology can be utilized to
enhance the effectiveness of behavioral activation. Madhu et al (2022) proposed a
novel approach for activity recognition utilizing AI. They employed multi-modal data
(combining speech and text) and utilized a BERT model for emotion recognition, along
with logistic regression for sentiment detection. Subsequently, an activity classification
model was employed wherein emotion and sentiment were integrated with keywords
to accurately identify the appropriate activity for behavioral activation. This method
achieved accuracies exceeding 80% in emotion recognition, sentiment detection, and
activity recognition tasks. Rathnayaka et al (2022) designed and developed a chatbot
named Bunji using AI technology to provide emotional support, personalized behavior
activation, and remote health monitoring. Participatory evaluation in a pilot study
environment also demonstrated the practicality and effectiveness of the chatbot.

Exposure therapy (ET)

Exposure therapy, a common technique in CBT, involves gradually exposing patients
to anxiety-provoking or phobias-provoking situations to help them adapt and reduce
their fear responses Abramowitz (2013). It is used to treat various psychological disor-
ders, including specific phobias, panic disorder, post-traumatic stress disorder (PTSD),
obsessive-compulsive disorder (OCD), and social anxiety disorder. In recent years,
virtual reality (VR) technology has emerged as a potent adjunct to traditional expo-
sure therapy Jameel (2020), leading to the development of Virtual Reality Exposure
Therapy (VRET), validated by research for its effectiveness within CBT Scozzari and
Gamberini (2011); Maples-Keller et al (2017); Ørskov et al (2022).
While VRET may not yield superior results in terms of therapeutic efficacy com-
pared to traditional exposure therapy within the CBT framework, it presents unique
advantages like enhanced patient comfort, safety, real-time customization of expo-
sure content, and simulation of personalized real-life scenarios. The integration of VR
with CBT has led to the development of interactive applications and serious games to
augment therapeutic interventions. Heng (2021) developed ReWIND, an RPG-styled
serious game based on the ABCDE model of CBT. The design intention of ReWIND is
to immerse players diagnosed with symptoms of Generalized Anxiety Disorder (GAD)
in various anxiety-inducing situations within a virtual environment and provide them
with constructive psychological interventions. The game aims to deliver psychologi-
cal education to players in an engaging and interactive manner. Similarly, Giordano
et al (2022) emphasize that CBT is the most successful protocol in gambling disorder
(GD) treatment, and they developed a serious virtual game called Alter Game, based
on VRET, aimed at preventing relapse in patients with gambling disorder. Addition-
ally, Michelle et al (2014) proposed an Android application named CBT Assistant,
where patients can choose, set, and customize their own exposure fear ladders based
on a generic template for graded exposure therapy. This feature distinguishes CBT
Assistant from other CBT applications. The fusion of VRET with AI technology fur-
ther enhances its efficacy in addressing mental health challenges. Åhs et al (2020)

14
proposed and discussed the idea of utilizing Explainable Artificial Intelligence (XAI)
to enhance CBT treatment for speech anxiety in VR settings. Rahman et al (2022)
explore a range of machine learning models’ performance on the task of arousal pre-
diction using publicly available datasets. They propose a pipeline to address the model
selection issue with various parameter configurations in the context of VRET.

Homework
A CBT session typically lasts 45-60 minutes, which is often insufficient for many
patients. Homework assignments bridge the gap between sessions, allowing patients
to apply learned skills and enabling therapists to assess skill acquisition and main-
tenance Beck (1979); LeBeau et al (2013). In the professional literature, homework
assignments are frequently delineated as precise, structured therapeutic tasks upon
during sessions, intended for completion between sessions. This process entails collab-
orative delineation of therapeutic objectives for the homework, determining pertinent
activities or data collection as components of the homework, strategizing the prac-
tical execution of the homework, and subsequently reviewing the homework during
subsequent sessions Kazantzis et al (2010). Homework assignments may encompass a
range of activities within each session, such as engaging with relevant materials, doc-
umenting thoughts and emotions, practicing specific skills or behaviors, or engaging
in communication with others Tang et al (2017).
Research has found that clients who consistently complete homework derive greater
benefits from interventions compared to those who complete little or no home-
work Burns and Spangler (2000), yet traditional paper-based CBT homework pose
various impediments that can significantly undermine users’ motivation to complete
tasks as instructed. Non-compliance with homework is cited as one of the most com-
mon reasons for the failure of CBT treatments Helbig and Fehm (2004), persisting as
a prevalent issue in clinical practice. The prevalence of digital devices and the internet
has enabled the transformation of traditional homework into digital formats, thereby
enhance CBT homework compliance. However, there is a lack of guidelines for design-
ing mobile phone apps tailored for this purpose. Consequently, Tang et al (2017)
proposes six essential features of an optimal mobile app aimed at maximizing CBT
homework compliance, aiming to provide theoretical guidance for the development of
such applications.
Innovative approaches, such as the integration of traditional diary writing with
mobile technology and LLM, offer promising solutions to enhance homework compli-
ance. For example, Peretz et al (2023) developed a machine learning model capable
of recognizing the presence of homework assignments during therapy sessions based
on natural language dialogue between therapists and clients in real-world settings, as
well as determining the type of homework assigned. Such advancements hold signifi-
cant promise in bolstering therapists’ ability to assign and monitor homework tasks,
ultimately fostering enhancements in therapeutic outcomes. Nepal et al (2024) inte-
grated traditional diary writing with mobile technology and LLM to create a diary
application with contextual awareness, named MindScape. Specifically, the application
utilizes real-time analysis of behavioral data collected from smartphones and employs
LLM to provide personalized, contextually relevant writing prompts. These prompts

15
are designed to guide users in reflection and contemplation, facilitating the recording
of their thoughts within daily life contexts. This innovative approach not only fosters
a habit of regular self-reflection but also addresses the challenge of homework compli-
ance in CBT. During CBT for tinnitus alleviates the patients are typically assigned
various homework tasks, including diary writing and self-monitoring. These homework
assignments primarily consist of handwritten text data. However, analyzing this data
can be extremely time-consuming for therapists, leading to decreased treatment effi-
ciency. To address this issue, Jeong et al (2024) proposed utilizing LLMs like GPT-2
to analyze the homework data of patients undergoing CBT. Their goal is to predict
the Tinnitus Handicap Inventory (THI) scores from the homework, which can, in turn,
predict the outcomes of CBT treatment, thereby enabling the selection of more per-
sonalized and effective treatment plans. Additionally, they compared the performance
of the latest language models, particularly Google’s T5 and Flan-T5, in predicting
THI scores. Finally, they looked ahead the application of this research to monitor and
predict the effectiveness of CBT treatment in patients with depression.
In summary, AI augments various CBT strategies by leveraging natural language
processing and machine learning techniques, improving effectiveness and engagement.

3.2.2 Developing AI tools for comprehensive delivery of CBT

While CBT strategies are often discussed individually, in clinical practice, they are
typically combined to address the specific needs of clients. Recent advancements in
AI have fostered the creation of digital tools that utilize the principles and techniques
of CBT. These tools integrate multiple CBT strategies, thus providing a comprehen-
sive approach to treatment. AI-powered chatbots, also referred to as conversational
agents or relational agents exemplify such innovations, with chatbots being particu-
larly notable. According to a review by Abd et al. Abd-Alrazaq et al (2019), out of
17 chatbots offering psychotherapy, 10 were based on CBT. They are capable of mim-
icking the communication styles of human therapists by engaging users in interactive
dialogues that include both verbal and non-verbal exchanges. This allows them to
offer efficient, personalized, and readily accessible CBT treatment, available 24/7. Pio-
neering chatbots like Woebot, Wysa, Tess, and Youper are emblematic of this trend.
Their innovative approaches and effectiveness have garnered significant attention and
recognition as effective psychological support systems based on CBT. In recent years,
in response to diverse needs, researchers have developed many new AI tools based on
CBT principles Oliveira et al (2021); Rizea (2022); Rani et al (2023); Sabour et al
(2023); Su et al (2022); Nwoye et al (2024). As summarized in Table 1, we have outlined
specific AI tools currently utilized in CBT interventions

16
 Table 1: AI tools used in current CBT intervention
AI tools Description
1
Woebot Woebot is a chatbot that offers CBT-based therapy for depres-
sion and anxiety. It engages users in daily conversations, tracks
their emotions, and introduces CBT concepts through short
videos or interactive word games. Using decision trees and
natural language processing, Woebot responds empathetically
and provides helpful suggestions. It can also detect concerning
language and directs users to external resources if needed.
Wysa2 Wysa is a therapy chatbot designed to support mental health
issues like depression, anxiety, stress, and loneliness. It utilizes
CBT, mindfulness, and positive psychology techniques. Instead
of AI-generated responses, Wysa uses pre-crafted therapeutic
conversations developed by clinicians for safe and effective inter-
actions. Its adaptive AI understands complex user inputs and
offers empathetic feedback and tailored CBT-based tools.
Youper3 Youper is a chatbot that delivers CBT through three steps: a
personalized mental health assessment, instant support via con-
versations, and symptom monitoring. It uses a decision tree
to select responses, conducts real-time emotion analysis, and
provides CBT interventions based on the user’s emotional state.
Tess4 Tess is a psychological AI chatbot providing brief conversations
for mental health support, psychoeducation, and reminders. It
uses clinician-prepared statements to deliver interventions based
on user-reported moods. Tess adjusts its responses based on user
feedback, favoring CBT-based interventions for positive reac-
tions and offering alternatives for neutral or negative ones. The
platform is customizable to align with specific treatments or
user demographics.
BetterHelp5 BetterHelp is an online therapy platform that uses AI to match
patients with licensed therapists and offers various approaches
like CBT and psychodynamic therapy.
Rumi6 Oliveira
Rumi is a chatbot uses Rumination-focused Cognitive Behav-
et al (2021)
ioral Therapy (RFCBT) to explore the relationship between
thoughts, feelings, and actions, aiming to improve mental health
and reduce depressive and anxious symptoms.
Cloud
Bot Rizea Cloud Bot is a chatbot that utilized NLP technology to func-
(2022) tion as a psychologist. It focuses on applying a cognitive
restructuring CBT technique to address users’ issues.

17
 Table 1: AI tools used in current CBT intervention (continued)
AI tools Description
Saarthi Rani
Saarthi is a chatbot using NLP and AI for delivering CBT
et al (2023)
and remote health monitoring to people with mental health
issues. It offers real-time, evidence-based treatment that is
accessible, affordable, and convenient, aiming to reduce anxiety
and depression symptoms and enable long-term mental health
monitoring.
SchizoBot Nwoye
SchizoBot is a chatbot using artificial neural networks to deliver
et al (2024)
CBT for managing schizophrenia, aiding clinicians and ensuring
consistent therapy administration for patients.
XIAO AN Su
XIAO AN is a Chinese AI psychotherapy robot designed to
et al (2022)
monitor emotions and provide effective therapy, primarily using
CBT principles. It has shown effectiveness in treating anxiety
disorders in clinical trials without replacing therapists.
Emohaa Sabour
Emohaa is a Chinese conversational agent, which consists of
et al (2023)
two platforms: one is template-based (CBT-Bot) for structured
conversations and exercises based on Cognitive Behavioral Ther-
apy principles, while the other (ES-Bot) allows for open-ended
discussions on emotional issues and provides emotional support.

By offering efficient, personalized, and readily accessible CBT treatment, these AI

agents provide round-the-clock support to individuals undergoing therapy, which to
some extent helps alleviate the pressure of insufficient medical resources. Nevertheless,
it’s worth noting that given the sensitivity of mental health support, it is crucial to
assess the chatbots before presenting them to users.

3.2.3 Real-time monitoring and feedback

Traditional CBT often lacks real-time monitoring capabilities, relying solely on self-
reported data from clients. However, the advent of mobile health technologies offers
new possibilities for monitoring client health outside CBT sessions. According to
research by Yang et al (2018), CBT is well-suited for mobile platform applications,
which facilitate real-time monitoring of user states and provide valuable data to ther-
apists. Consequently, numerous mobile applications have been developed to deliver
CBT treatment and track users’ psychological states in real-time. Addepally and

1
Woebot: https://woebothealth.com/
2
Wysa: https://www.wysa.com/
3
Youper: https://www.youper.ai/
4
Tess: https://www.cass.ai/x2ai-home
5
BetterHelp: https://www.betterhelp.com/
6
Rumi: https://www.facebook.com/rumibot.bot/

18
Purkayastha (2017) noted that loneliness can be a risk factor for depression. To address
this issue, the MoodTrainer application was developed, which tracks users’ locations
and isolating behaviors in real-time and provides CBT interventions when it detects
relevant behaviors. Also, Michelle et al (2014) introduced an Android application
named CBT Assistant. This APP analyzes input data from individuals with social
anxiety disorder (SAD) to identify stressors or situations triggering their mental health
issues and assesses their severity. Additionally, some mobile applications are designed
for specific CBT treatment scenarios, such as continuously tracking the sleep pat-
terns of individuals with sleep disorders Schabus et al (2023) or providing cessation
monitoring and CBT examples to smokers Alsharif and Philip (2015), thereby enhanc-
ing their success rates. These mobile applications leverage the capabilities of mobile
devices to collect real-time data and provide personalized interventions and support
to individuals undergoing CBT treatment.
In recent years, wearable devices and smartphone applications equipped with AI
technology are emerging as a trend for monitoring patients’ psychological states.
During CBT process, AI algorithms can monitor stress levels, physical activity,
speech changes, and other indicators to detect variations in a patient’s psychologi-
cal state. These algorithms can send timely alerts to both patients and healthcare
providers. Garcia-Ceja et al (2015) embedded accelerometer sensors into smartphones
to detect stress levels using classification algorithms like naive bayes and decision
trees, achieving 71% accuracy. Additionally, the long-term collection and analysis of
large datasets can help therapists and patients gain a deeper understanding of the
patient’s mental health patterns. With the assistance of AI, therapists can access more
comprehensive information, enabling them to discern trends in mental health, iden-
tify triggers, and evaluate treatment efficacy. These insights are crucial for making
informed therapeutic decisions and designing personalized interventions. Goodwin
et al (2019) collected physiological and movement data from wrist-worn biosensors
in 20 adolescents diagnosed with ASD. They developed prediction models utiliz-
ing ridge-regularized logistic regression. These models demonstrate high accuracy
in forecasting instances of aggressive behavior towards others occurring within the
subsequent minute. Such advancements lay the groundwork for proactive behavioral
interventions and timely adaptive intervention systems in the future.
Despite the growing use of AI-equipped wearable devices and mobile applications
for monitoring psychological states and data collection, there is relatively limited
research specifically focused on their application to CBT, particularly considering the
unique demands of real-time interaction inherent in this therapeutic approach.

3.2.4 Early prediction of treatment outcomes

Even among those for whom CBT proves effective, there may be a subset with limited
benefits, potentially leading to a misallocation of both client time and limited health-
care resources. Hence, predicting treatment outcomes early in the therapeutic journey
assumes critical importance, particularly to prevent the misallocation of client time
and healthcare resources. Recent advancements in AI and large-scale data analysis
have enabled the development of models capable of forecasting individual responses to
CBT with high accuracy. Hahn et al (2015) applied a novel machine learning approach

19
to predict individual-level response to CBT using fMRI data in patients diagnosed with
panic disorder and agoraphobia (PD/AG). Similarly, Tolmeijer et al (2018) utilze the
machine learning methods to predict how people will respond when offered CBT for
psychosis. Their two-step methodology involved first identifying potentially predictive
regions and then developing a model based on these regions to make individual-level
predictions. In a large-scale study, Kaldo et al (2021) analyzed data from over 6000
patients undergoing Internet-delivered CBT (ICBT). They evaluated the accuracy of
various machine learning algorithms in predicting treatment outcomes and explored
the integration of these algorithms into an Adaptive Treatment Strategy. Isacsson et al
(2023) further examined the clinical utility of machine learning in predicting ICBT
outcomes. They investigated the optimal timing within the treatment process for the
model’s predictive accuracy to support adaptive treatment strategies, proposing an
optimal predictive model and offering specific recommendations based on their com-
prehensive analysis. Despite advancements, traditional predictive models often exhibit
limited accuracy, particularly in assessing the effectiveness of treating adolescent social
anxiety. Under this background, Zheng et al (2022) addressed this by employing deep
learning techniques to construct a predictive model for the correlation between CBT
and adolescent social anxiety, showcasing significantly improved predictive accuracy
and reduced complexity compared to traditional models. Prasad et al (2023) seeks to
develop a state-of-the-art deep-learning framework for predicting clinical outcomes in
ICBT by leveraging large-scale, high-dimensional time-series data of client-reported
mental health symptoms and platform interaction data.
Numerous researchers have also explored the use of specific treatment outcome pre-
dictors and potential biomarkers associated with particular diseases. For instance, Wei
et al (2023) utilized the Hamilton Depression Rating Scale (HDRS) score as the pri-
mary outcome measure to investigate symptom changes in subjects undergoing CBT.
Employing machine learning algorithms, they developed a support vector regression
model, ultimately identifying left dorsolateral prefrontal cortex (DLPFC) Regional
Homogeneity (ReHo) as a neuroimaging biomarker for the therapeutic effects of CBT
in depression. Many previous studies have relied on highly selective samples to predict
the outcomes of CBT. However, few have utilized routine available socio-demographic
and clinical data to accomplish this task. Therefore, Hilbert et al (2020, 2021) applied
machine learning methods to clinical and socio-demographic data to predict the men-
tal health treatment outcomes of individual patients. Their findings suggest that using
routine data alone can feasibly predict treatment outcomes for mental disorders, with
accuracy significantly surpassing chance levels.
These studies collectively highlight the promise of integrating advanced AI method-
ologies with clinical practice to enhance the early prediction of CBT treatment
outcomes, offering potential pathways for more tailored and effective therapeutic
interventions.

3.2.5 Evaluation of therapists and clients in CBT treatment

The application of AI in CBT extends beyond the diagnosis and assessment of dis-
orders, it also encompasses crucial evaluations of therapists and clients by analyzing
conversation data during therapy sessions.

20
Therapist treatment quality assessment
Given the prevalence of mental health issues, ensuring the quality of psychotherapy
is crucial to addressing the growing mental health demands and the complexities of
the social environment. Traditionally, quality assessment is performed by human eval-
uators who listen to therapy recordings and review therapy notes to assess specific
therapeutic skills. However, this approach is costly and time-consuming, resulting
in limited feasibility and hindering widespread implementation in practical settings.
To address these challenges, some researchers and technology developers have begun
exploring the use of automated techniques to monitor the quality of psychotherapy.
AI offers automated solutions for assessing therapy sessions and monitoring treatment
fidelity of CBT sessions Chen et al (2022a). For example, Ewbank et al (2020) used
a large-scale dataset containing session transcripts from more than 14000 patients
receiving internet-enabled CBT (IECBT) to train a deep learning model to automat-
ically categorize therapist utterances according to the role that they play in therapy,
generating a quantifiable measure of treatment delivered. The closer the content pro-
vided by the therapist aligns with standard CBT protocols, the more positively it
correlates with significant symptom improvement in patients. Conversely, the quan-
tity of content unrelated to therapy shows a negative association. This method allows
for the indirect assessment of the efficacy of CBT psychotherapy provided by the
therapist. However, Flemotomos et al (2021) emphasize that, for CBT, the most com-
monly utilized coding scheme is the Cognitive Therapy Rating Scale (CTRS), which
defines a set of 11 session-level codes reflecting skills and techniques specific to the
intervention. Consequently, they introduced a model for quality assessment of psy-
chotherapy sessions based on adapted BERT representations of therapy language use.
Their analysis focused on the binary classification of CBT sessions concerning the over-
all CTRS score. Chen et al (2022b) also utilized CTRS scores to assess the quality of
CBT sessions. However, unlike Flemotomos et al (2021), they proposed a hierarchical
framework for automatically evaluating the quality of transcribed CBT interactions.
Ardulov et al (2022) proposed an approach that explicitly focuses on control-affine
dynamical system models. They attempted to extract local dynamic modes from short
windows of conversation and learn to correlate the observed dynamics with CBT com-
petence. Furthermore, some studies aim to identify areas where the therapist excels
and areas where improvement is needed by analyzing recordings of therapy sessions,
comparing therapists’ language and behavior against standards of specific therapeutic
models Stirman et al (2021); Flemotomos et al (2022); Zhang et al (2023); Wang et al
(2024a). This process aids therapists in enhancing their professional skills, thereby
improving the overall quality of therapy. Particularly, Wang et al (2024a) developed
PATIENT-Ψ, a novel patient simulation framework for CBT training. Specifically,
they constructed diverse patient profiles and corresponding cognitive models based
on CBT principles, and used a large language model to act as a simulated therapy
patient. This role-playing therapeutic scenario helps mental health trainees practice
CBT skills.

21
Predictive analysis of clients treatment adherence
Treatment adherence refers to the extent to which clients actively engage in and com-
ply with the advice and instructions provided by healthcare professionals, thereby
adhering to the treatment regimen, and it directly impacts the effectiveness and out-
comes of the treatment DiMatteo et al (2002). Therefore, client adherence evaluation is
essential components of effective healthcare delivery. Particularly, against the backdrop
of ICBT, there has been a reduction in face-to-face interactions between healthcare
professionals and clients, which may lead to low client engagement and high dropout
rates. AI models can analyze client behavior and responses during therapy sessions
to evaluate their understanding, engagement, and adherence to treatment content.
Côté-Allard et al (2022) presents a minimally data-sensitive approach, based on a self-
attention deep neural network, to perform adherence forecasting of clients undergoing
G-ICBT. This study leverages between 7 to 42 days of user-interaction data (login/l-
ogout) from the eMeistring platform. This analysis can identify individuals who may
be at risk of early dropout from intervention. With this information, clinicians can
implement meaningful and targeted interventions such as reminders, scheduling direct
interactions, or modifying the treatment approach to prevent premature termination.

3.3 Integration of AI in Post-treatment Stage

3.3.1 Long-term treatment response and relapse prediction
Since current CBT interventions typically have relatively short durations, leading to
inadequate durability of effects for some patients post-treatment, with a considerable
proportion experiencing relapse after treatment cessation. Hence, predicting individ-
ual long-term clinical responses and relapse risks remains a significant challenge. AI
presents a promising avenue for predicting relapse risks and enhancing the durabil-
ity of CBT’s effects in patients undergoing CBT. Månsson et al (2015) proposed the
first study using the multivariate SVM–fMRI method to successfully predict long-term
treatment response of ICBT for social anxiety disorder, achieving a predictive accu-
racy reached 92% (95% confidence interval 73.2–97.6). Lorimer et al (2021) utilized
machine learning techniques, specifically the XGBoost algorithm, to analyze follow-
up data of patients undergoing Low-Intensity Cognitive Behavioral Therapy (LiCBT).
They developed a dynamic prediction tool capable of identifying cases with higher
relapse risk at four distinct time points during the patients’ treatment journey, with
the ability to dynamically adjust prediction accuracy. This tool’s capacity for early
identification of high-relapse cases, combined with targeted relapse prevention mea-
sures, can significantly enhance the long-term efficacy of LiCBT in situations where
psychological resources are limited.
In conclusion, the integration of AI in CBT holds promise for revolutionizing the
delivery and efficacy of psychological interventions. By leveraging AI technologies
throughout various stages of the treatment process, clinicians can enhance treat-
ment outcomes, personalize interventions, and optimize resource allocation, ultimately
improving mental health outcomes for individuals on a global scale.

22
4 Datasets
Datasets play a crucial role in driving research at the intersection of CBT and AI, pro-
viding foundational material for training, testing, and validating AI algorithms, and
furnishing practitioners and researchers with the bedrock for constructing and evaluat-
ing CBT models. This section aims to review existing publicly datasets relevant to the
application of CBT and AI. In the context of disease diagnosis and assessment tasks,
numerous datasets have already been extensively reviewed in various comprehensive
articles. Consequently, we will not elaborate on these datasets further in this paper.
Moreover, it is worth noting that for the task of selecting personalized treatment strate-
gies, there is a noticeable lack of publicly available datasets. This paper will not cover
these datasets in detail. Instead, we focus on datasets for specific CBT-related tasks
such as identifying and classifying cognitive distortions, conducting cognitive restruc-
turing, and analyzing CBT conversation data. For clarity, datasets where descriptions
are unclear or ambiguous in the literature will not be discussed in this paper.
Table 2 provides a comprehensive overview of datasets used for detecting and
classifying cognitive distortions. As can be seen from this table, most of the cogni-
tive distortion data sets are in English, and fewer are in Chinese. Moreover, these
datasets present several notable challenges: First, low reliability. The labeling crite-
ria vary significantly across different datasets, leading to inconsistencies. Moreover,
the inherently subjective nature of labeling cognitive distortions further compromises
reliability. Second, there is a pronounced issue of data imbalance, with certain classes
of cognitive distortions lacking adequate representation. This imbalance hampers the
model’s ability to generalize well across all classes. Table 3 summarizes datasets rel-
evant to cognitive restructuring. Additionally, Table 4 outlines datasets that involve
CBT conversations. For these areas, high-quality, annotated datasets are particularly
scarce for both cognitive restructuring and CBT conversation analysis.

Table 2: Cognitive Distortions Dataset. For a clean presentation, we have only shown
the simplified version of the cognitive distortion dataset, see Appendix A for more
details.
Study Description
Size of Dataset: A total of 1644 data entries across 11 types
of cognitive distortions, and 2000 entries for normal cases.
Wang et al (2023b)
Language: Not reported.
Data Modalities: Text.
Size of Dataset: A total of 34370 samples across 14 types
of cognitive distortions.
Elsharawi and
Language: English.
El Bolock (2024)
Data Modalities: Text.

23
 Table 2: Cognitive Distortions Dataset (continued)
Study Description
Size of Dataset: Dataset CrowdDist contained 7,666 texts
across all 15 distortions, with an average of 511 responses
per distortion. Dataset MH contained two subsets: MH-
C was annotated with 15 cognitive distortion labels with
Shickel et al (2020) 1164 distorted texts, and the MH-D dataset was annotated
with binary distorted/non-distorted labels, distorted for 1605
texts, not distorted for 194 texts.
Language: English.
Data Modalities: Text.
Size of Dataset: A total of 2530 samples across 10 types of
cognitive distortions.
Lim et al (2024)
Language: English.
Data Modalities: Text.
Size of Dataset: A total of 3000 samples, 39.2% were
marked as not distorted, while the remaining were identified
Shreevastava and to 10 type of distortions.
Foltz (2021) Language: English.
Data Modalities: Text.
Size of Dataset: A total of 200 samples across 14 types of
cognitive distortions.
de Toledo Rodriguez
Language: English.
et al (2021)
Data Modalities: Text.
Size of Dataset: A total of 1077 samples across 13 types of
cognitive distortions.
Sharma et al
Language: English.
(2023c)
Data Modalities: Text.
Size of Dataset: About 10k samples across 10 types of
cognitive distortions.
Maddela et al
Language: English.
(2023)
Data Modalities: Text.
Size of Dataset: 7,500 cognitive distortion thoughts across
7 types of common cognitive distortions.
Wang et al (2023a)
Language: Chinese.
Data Modalities: Text.
Size of Dataset: A total of 3407 posts across 12 types of
cognitive distortions.
Qi et al (2023)
Language: Chinese.
Data Modalities: Text.

24
 Table 2: Cognitive Distortions Dataset (continued)
Study Description
Size of Dataset: A total of 22,327 samples across 10 types
of cognitive distortions.
Na (2024)
Language: Chinese.
Data Modalities: Text.

Table 3: Cognitive Restructuring Dataset.

Study Tasks Dataset source Description
Cognitive
Thought Records Size of Dataset: A total of 300
Reframing of
Sharma Dataset and situations-thoughts pairs, reframed
Negative Thoughts
et al Mental Health thoughts per situation.
through
(2023c) America (MHA) Component: situation; thought;
Human-Language
website. reframe; thinking traps addressed.
Model Interaction.
Language: Chinese.
Data Modalities: Text.
Open Source: Yes1 .

The corpus
labeling utilizes
Detection of one specialized
Size of Dataset: A total of 1900 sen-
Cognitive opensource
Lin et al tences.
Distortion and dataset, the
(2024) Component: original text and recon-
cognitive Chinese
struction text.
restructuring. psychological Q&A
Language: Chinese.
dataset PsyQA Sun
Data Modalities: Text.
et al (2021).
Open Source: Yes2 .

Identification and
Shidara evaluation of Size of Dataset: Not reported.
Recruit
et al automatic thoughts Language: Japanese.
participants.
(2022) and cognitive Data Modalities: Text.
restructuring. Open Source: Yes. It can be obtained
by sending an email.

5 Discussion
The integration of AI into CBT has led to significant advances in pre-treatment assess-
ment, the therapeutic process, and post-treatment follow-up. First, AI has improved

1
Sharma et al (2023c): https://github.com/behavioral-data/Cognitive-Reframing
2
Lin et al (2024):
https://github.com/405200144/Dataset-of-Cognitive-Distortion-detection-and-Positive-Reconstruction
1
Lee et al (2023): https://github.com/behavioral-data/Empathy-Mental-Health

25
 Table 4: CBT session dataset.
Study Tasks Dataset source Description
Size of Dataset: Three levels of
Enhancing empathy strategies, and the num-
Empathetic Crowdsourced ber of pairs for each strategy was
Response of Large Reddit posts of as follows: “emotion reaction”=1,047,
Lee et al
Language Models mental health “exploration”=481, and “interpreta-
(2023)
Based on from Sharma et al tion”=1,436.
Psychotherapy (2020). Language: English.
Models. Data Modalities: Text.
Open Source: Yes1 .

Size of Dataset: 22,327 entries, each

Get PsyQA
comprising questions, descriptions, and
Questions Sun
CBT responses.
et al (2021), which
Enhance the Language: Chinese.
is derived from the
precision and Data Modalities: Text.
Chinese online
Na efficacy of Open Source: Yes. Follow the data
mental health
(2024) psychological copyright protocols and obtain it by
support forum
support through email.
Yixinli, then
LLMs Note: The questions in the dataset
utilizes CBT
originate from online mental health
Prompt to generate
forum, and the responses are generated
CBT answers.
by ChatGPT, not professionals.

efficiency by assisting with pre-treatment screening and diagnosis, reducing therapist

workload, and enabling real-time adjustments to treatment plans through predictive
capabilities. Second, AI has enabled more personalized CBT by analyzing rich patient
data and identifying subtle patterns, enabling customized treatment plans that go
beyond the traditional reliance on therapist expertise alone. Finally, AI-powered CBT
platforms have increased the accessibility of mental health services by providing remote
and cost-effective treatment options through online platforms and mobile applications
that offer 24/7 support. However, there are several limitations to current AI applica-
tions in CBT. In pre-treatment assessment, while AI excels in diagnosing psychological
disorders and assessing cognitive distortions and emotional states using textual data,
it has fewer applications in analyzing video, audio, and behavioral data. The potential
of multimodal data for comprehensive diagnosis remains largely untapped. During the
therapeutic process, AI enhances individual CBT strategies but struggles to cover the
complexity of interventions comprehensively. AI tools like chatbots and virtual therapy
assistants vary in quality and lack standardized development, leading to inconsisten-
cies. Moreover, there are inadequate metrics for evaluating feasibility, engagement,
and satisfaction, complicating platform comparison and improvement. Most AI-driven
wearable devices and mobile applications focus on general indicators like heart rate
and activity levels, lacking specialized tools for monitoring CBT-specific indicators.
In post-treatment, while AI aids in predicting long-term treatment responses and
relapse risks, challenges include variability in prediction data and insufficient research
on utilizing predictive results to formulate personalized intervention strategies.

26
 The future of AI in CBT holds great potential. Autonomous learning and adap-
tive therapy systems can emulate human CBT therapists, engaging in multi-round
interactions with patients and adjusting strategies based on real-time feedback. Group
intelligence support and decision-making systems can improve both therapist guidance
and patient outcomes by aggregating the experience of experienced practitioners and
facilitating intelligent social support. Cognitive augmentation and assistive systems
can develop personalized tools to enhance cognitive function, thereby increasing the
effectiveness of CBT. Customized, personalized CBT models can adapt to users’ spe-
cific data, such as social background, culture, education, and environment, to provide
tailored responses and interventions that increase therapeutic effectiveness and user
satisfaction. Despite the potential of AI, several challenges need to be addressed. Data
security and privacy are paramount, requiring compliance with privacy regulations,
anonymization of sensitive information, and advanced encryption techniques. Ethical
and algorithmic bias must be mitigated by ensuring data diversity, involving multiple
stakeholders in development, and continuously monitoring AI systems. Model explain-
ability and transparency are essential for responsible AI decisions, requiring methods
to improve interpretability and the use of rigorously tested models. Over-reliance on
AI is a risk because the success of CBT depends on the therapist-patient relationship,
which AI cannot replicate. AI should be a complementary tool, not a replacement.
Evaluation of models in clinical practice is necessary to assess real-world effectiveness,
acceptance, trust, and usability, taking into account training costs and impact on med-
ical practice. In summary, while AI offers promising enhancements to CBT, it must
be used responsibly and ethically, complementing the guidance of professional thera-
pists. The ultimate goal is to use technology to support, not replace, human-centered
mental health care.

6 Conclusion
In this paper, we have conducted a comprehensive literature review of the integration
of AI technology into CBT . We explored the application of AI throughout the CBT
process, highlighting its significant transformative impact and existing limitations.
Subsequently, We have summarized publicly available datasets relevant to various
CBT-related tasks to provide a foundation for future research. We suggested future
research directions and acknowledged the practical challenges that AI faces in clinical
settings. Overall, our review illuminates the multifaceted integration of AI in CBT,
highlighting its potential while providing a nuanced understanding of its capabilities.
We hope that the findings will guide future research, bring new perspectives to clinical
practice, and contribute to the advancement of mental health care.
Acknowledgements. This work was supported by grants from the National Natu-
ral Science Foundation of China (grant numbers:72174152, 72304212 and 82071546),
Wuhan University Innovation and Entrepreneurship Projects for College Students
(No: 202410486100), Fundamental Research Funds for the Central Universities (grant
numbers: 2042022kf1218; 2042022kf1037), and the Young Top-notch Talent Cultiva-
tion Program of Hubei Province. Guanghui Fu is supported by a Chinese Government
Scholarship provided by the China Scholarship Council (CSC).

27
Declarations
• Competing Interests: The authors have no competing interests to declare that are
relevant to the content of this article.

28
Appendix A Cognitive distortion dataset
In this section, we provide full detail of the dataset as described in Section 4.

Table A1: Cognitive Distortions Dataset.

Dataset
Study Tasks Description
source
Size of Dataset: Before expansion:
Published There were a total of 353 data entries for
research papers all cognitive distortions, and 1000 entries
on cognitive for normal cases. After expansion: There
distortions, are now a total of 1644 data entries across
examples from 11 types of cognitive distortions, and 2000
Cognitive
Wang et al articles on the entries for normal cases.
distortion
(2023b) web that Label: 12 labels, including 11 cognitive
classification.
introduce distortions and no cognitive distortions.
cognitive Language: Not reported.
distortions and Data Modalities: Text.
social media Open Source: Yes. However, the arti-
posts. cle says it will share, but no link is
given.

Size of Dataset: Dataset CrowdDist

Dataset
contained 7,666 text responses across
CrowdDist,
all 15 distortions, with an average of
comes from the
511 responses per distortion. Dataset
popular
MH contained two subsets: The MH-C
crowdsourcing
Detection dataset was annotated with 15 cogni-
platform
Shickel and tive distortion labels with 1164 distorted
Mechanical
et al Classification texts, and the MHD dataset was anno-
Turk(MTurk).
(2020) of Cognitive tated with binary distorted/non-distorted
Dataset MH,
Distortions. labels, distorted for 1605 texts, not dis-
comes from
torted for 194 texts.
TAO Connect,
Label: situation; thought; reframe; think-
an online
ing traps addressed.
mental health
Language: English.
therapy service.
Data Modalities: Text.

29
 Table A1: Cognitive Distortions Dataset (continued)
Dataset
Study Tasks Description
source

Size of Dataset: A total of 34370 sam-

Facebook ples.
Detection
Elsharawi Empathetic Label: 14 labels.
and
and Data, Twitter Language: English.
Classification
El Bolock Data and Data Modalities: Text.
of Cognitive
(2024) Crowd- Open Source: Yes. However, the arti-
Distortions.
Sourcing. cle says it will share, but no link is
given.

Dataset
Detection Therapist Q&A Size of Dataset: A total of 2530 samples.
and comes from Label: 10 labels.
Lim et al
Classification crowd-sourced Language: English.
(2024)
of Cognitive data science Data Modalities: Text.
Distortions. repository, Open Source: Yes1 .
Kaggle.

Size of Dataset: A total of 3000 sam-

ples, 39.2% were marked as not distorted,
Dataset while the remaining were identified to
Automatic
Therapist Q&A have some type of distortion.
Shree- Detection
comes from Label: 11 labels, including 10 cognitive
vastava and
crowd-sourced distortions and no cognitive distortions.
and Foltz Classification
data science Language: English.
(2021) of Cognitive
repository, Data Modalities: Text.
Distortions.
Kaggle. Open Source: Yes. However, the arti-
cle says it will share, but no link is
given.

Convert A variety of
Size of Dataset: A total of 200 samples.
negative or sources such as
de Toledo Rodriguez Label: 14 labels.
distorted CBT books,
et al Language: English.
thoughts into forums and
(2021) Data Modalities: Text.
more realistic public content
Open Source: Yes2 .
alternatives. aggregators.

30
 Table A1: Cognitive Distortions Dataset (continued)
Dataset
Study Tasks Description
source

Cognitive
Reframing of
Negative Thought Size of Dataset: A total of 1077 samples.
Sharma Thoughts Records Dataset Label: 13 labels.
et al through and Mental Language: English.
(2023c) Human- Health America Data Modalities: Text.
Language (MHA) website. Open Source: Yes3 .
Model
Interaction.

Obtain a
diversity of
Size of Dataset: About 10k examples of
contexts,
Training thoughts containing unhelpful thought.
situations and
models to Label: 10 labels.
thoughts from
Mad- generate, Language: English.
PERSONA-
dela et al recognize, Data Modalities: Text.
CHAT
(2023) and reframe Open Source: Yes4 . However, this
dataset Zhang
unhelpful article says it will be shared and
et al (2018),
thoughts. gives a github link, but it does not
and ask
contain data.
crowdworkers to
rewrite them.

Cognitive
Carefully select
Distortion
and train
Detection
volunteers to
and Size of Dataset: 7,500 cognitive distor-
observe scenes
investigate tion thoughts.
and note
Wang et al the Label: 7 common cognitive distortions.
possible
(2023a) association Language: Chinese.
cognitive
between Data Modalities: Text.
distortions, then
cognitive Open Source: Yes5 .
have experts
distortions
evaluate the
and mental
results.
health.

31
 Table A1: Cognitive Distortions Dataset (continued)
Dataset
Study Tasks Description
source

Size of Dataset: A total of 3407 posts.

Comments on a
Classification Label: 12 labels.
Qi et al Sina Weibo post
of Cognitive Language: Chinese.
(2023) by the user
Distortions. Data Modalities: Text.
”Zoufan”.
Open Source: Yes6 .

Get PsyQA
Size of Dataset: A total of 22,327 sam-
Questions Sun
ples.
et al (2021),
Label: 10 labels.
Enhance the which is derived
Language: Chinese.
precision and from the
Data Modalities: Text.
efficacy of Chinese online
Open Source: Yes. Follow the data copy-
Na (2024) psychological mental health
right protocols and obtain it by sending
support support forum
an email.
through Yixinli, then
Note: The questions in the dataset orig-
LLMs. utilizes CBT
inate from online mental health forum,
Prompt to
and the responses are generated by Chat-
generate
GPT, not professionals.
CBTanswers.

1
Lim et al (2024):
https://www.kaggle.com/datasets/sagarikashreevastava/cognitive-distortion-detetction-dataset
2
de Toledo Rodriguez et al (2021): https://github.com/itoledorodriguez/cbt-dataset
3
Sharma et al (2023c): https://github.com/behavioral-data/Cognitive-Reframing
4
Maddela et al (2023):
https://github.com/facebookresearch/ParlAI/tree/main/projects/reframe thoughts
5
Wang et al (2023a): https://github.com/bcwangavailable/C2D2-Cognitive-Distortion
6
Qi et al (2023):https://github.com/HongzhiQ/SupervisedVsLLM-EfficacyEval

32
Appendix B Abbreviation
In this section, as shown in Table B2, we summarize the full names and abbreviations
of various specialized terms mentioned throughout the text. By providing this list of
abbreviations, our aim is to assist readers in quickly referencing and understanding
the meanings of these terms, thereby enhancing comprehension of the content and its
context within the paper.

Table B2: Full name and abbreviation of terms

Abbreviation Full Name
ADHD Attention Deficit Hyperactivity Disorder
AG Agoraphobia
AI Artificial Intelligence
AR Augmented Reality
BA Behavioral Intervention
CBT Cognitive Behavioral Therapy
CCBT Computer-Based Cognitive Behavioral Therapy
CD Cognitive Distortions
CNN Convolutional Neural Network
CR Cognitive Restructuring
CTRS Cognitive Therapy Rating Scale
DLPFC Dorsolateral Prefrontal Cortex
DoT Diagnosis of Thought
ECA Embodied Conversational Agent
ET Exposure Therapy
GAD Generalized Anxiety Disorder
GD Gambling Disorder
GPT Generative Pretrained Transformer
HDRS Hamilton Depression Rating Scale
ICBT Internet-Based Cognitive Behavioral Therapy
IECBT Internet-enabled CBT
LiCBT Low-Intensity Cognitive Behavioral Therapy
LLM Large Language Model
MHA Mental Health America
MUBS Mobile-based Behavioral Activation System
NLP Natural Language Processing
OCD Obsessive-Compulsive Disorder
PD Panic disorder
PDT Psychodynamic Therapy
PTM Pre-trained Language Model
PTSD Post-Traumatic Stress Disorder
Q&A Question and Answering
ReHo Regional Homogeneity
RET Rational Emotive Therapy
SAD Social Anxiety Disorder
THI scores Tinnitus Handicap Inventory scores
VR Virtual Reality
VRET Virtual Reality Exposure Therapy
XAI Explainable Artificial Intelligence

33
References
Abd-Alrazaq AA, Alajlani M, Alalwan AA, et al (2019) An overview of the features
of chatbots in mental health: A scoping review. International journal of medical
informatics 132:103978

Abramowitz JS (2013) The practice of exposure therapy: Relevance of cognitive-

behavioral theory and extinction theory. Behavior therapy 44(4):548–558

Addepally SA, Purkayastha S (2017) Mobile-application based cognitive behavior

therapy (cbt) for identifying and managing depression and anxiety. In: Digital
Human Modeling. Applications in Health, Safety, Ergonomics, and Risk Manage-
ment: Health and Safety: 8th International Conference, DHM 2017, Held as Part
of HCI International 2017, Vancouver, BC, Canada, July 9-14, 2017, Proceedings,
Part II 8, Springer, pp 3–12

Åhs F, Mozelius P, Dobslaw F (2020) Artificial intelligence supported cognitive behav-

ioral therapy for treatment of speech anxiety in virtual reality environments. In:
ECIAIR 2020, Academic Conferences and Publishing International Limited

Alhaj F, Al-Haj A, Sharieh A, et al (2022) Improving arabic cognitive distortion

classification in twitter using bertopic. International Journal of Advanced Computer
Science and Applications 13(1):854–860

Alsharif AH, Philip N (2015) Cognitive behavioural therapy embedding smoking ces-
sation program using smart phone technologies. In: 2015 5th World Congress on
Information and Communication Technologies (WICT), IEEE, pp 134–139

Ardulov V, Creed TA, Atkins DC, et al (2022) Local dynamic mode of cognitive
behavioral therapy. arXiv preprint arXiv:220509752

Assunção G, Patrão B, Castelo-Branco M, et al (2022) An overview of emotion in

artificial intelligence. IEEE Transactions on Artificial Intelligence 3(6):867–886

Australian National University (2024) Welcome to moodgym.

https://www.moodgym.com.au/, accessed: 2024-06-19

Ball TM, Stein MB, Ramsawh HJ, et al (2014) Single-subject anxiety treat-
ment outcome prediction using functional neuroimaging. Neuropsychopharmacology
39(5):1254–1261

Bandelow B, Michaelis S, Wedekind D (2017) Treatment of anxiety disorders.

Dialogues in clinical neuroscience 19(2):93–107

Bech A, Steer R, Brown G (1996) Beck depression inventory—second edition: manual.

San Antonio: The Psychological Corporation 4:561–71

Beck AT (1979) Cognitive therapy and the emotional disorders. Penguin

34
Beck AT, Epstein N, Brown G, et al (1988) An inventory for measuring clinical anxiety:
psychometric properties. Journal of consulting and clinical psychology 56(6):893

Beck JS, Beck AT (2011) Cognitive behavior therapy. New York: Basics and beyond
Guilford Publication pp 19–20

Benard AO, Lukandu IA (2018) A q-learning model for cognitive behavioural therapy
of insomnia patients. Int J Comput Inf Technol 7(3):1–7

Bhaumik R, Srivastava V, Jalali A, et al (2023) Mindwatch: A smart cloud-based ai

solution for suicide ideation detection leveraging large language models. medRxiv
pp 2023–09

Brave S, Nass C, Hutchinson K (2005) Computers that care: investigating the effects
of orientation of emotion exhibited by an embodied computer agent. International
journal of human-computer studies 62(2):161–178

Burns DD, Beck AT (1999) Feeling good: The new mood therapy. Avon New York

Burns DD, Spangler DL (2000) Does psychotherapy homework lead to improvements

in depression in cognitive–behavioral therapy or does improvement lead to increased
homework compliance? Journal of consulting and clinical psychology 68(1):46

Cardi V, Clarke A, Treasure J (2013) The use of guided self-help incorporating a

mobile component in people with eating disorders: A pilot study. European Eating
Disorders Review 21(4):315–322

Chandra P, Joshi G, Bhagwat R (2023) Chatgpt’s evolution in reshaping cognitive

behavioral therapy. In: 2023 IEEE Engineering Informatics, IEEE, pp 1–9

Chen Z, Flemotomos N, Imel ZE, et al (2022a) Leveraging open data and task
augmentation to automated behavioral coding of psychotherapy conversations in
low-resource scenarios. arXiv preprint arXiv:221014254

Chen Z, Flemotomos N, Singla K, et al (2022b) An automated quality evaluation

framework of psychotherapy conversations with local quality estimates. Computer
speech & language 75:101380

Chen Z, Lu Y, Wang WY (2023) Empowering psychotherapy with large language mod-

els: Cognitive distortion detection through diagnosis of thought prompting. arXiv
preprint arXiv:231007146

China Cognitive Behavioral therapy professional organization (2024) China’s first ccbt
psychological self-service platform. https://www.psy.com.cn/therapy/, accessed:
2024-06-19

Christ C, Schouten MJ, Blankers M, et al (2020) Internet and computer-based cog-

nitive behavioral therapy for anxiety and depression in adolescents and young

35
 adults: systematic review and meta-analysis. Journal of medical Internet research
22(9):e17831

Clark DA (2013) Cognitive restructuring. The Wiley handbook of cognitive behavioral

therapy pp 1–22

Côté-Allard U, Pham MH, Schultz AK, et al (2022) Adherence forecasting for

guided internet-delivered cognitive behavioral therapy: A minimally data-sensitive
approach. IEEE Journal of Biomedical and Health Informatics

Cuijpers P (1998) A psychoeducational approach to the treatment of depression: a

meta-analysis of lewinsohn’s “coping with depression” course. Behavior therapy
29(3):521–533

Dahne J, Collado A, Lejuez C, et al (2019) Pilot randomized controlled trial of a

spanish-language behavioral activation mobile app (¡ aptı́vate!) for the treatment
of depressive symptoms among united states latinx adults with limited english
proficiency. Journal of Affective Disorders 250:210–217

David D, Cristea I, Hofmann SG (2018) Why cognitive behavioral therapy is the

current gold standard of psychotherapy. Frontiers in psychiatry 9:333730

Delgadillo J, Gonzalez Salas Duhne P (2020) Targeted prescription of cognitive–

behavioral therapy versus person-centered counseling for depression using a machine
learning approach. Journal of Consulting and Clinical Psychology 88(1):14

Demszky D, Yang D, Yeager DS, et al (2023a) Using large language models in

psychology. Nature Reviews Psychology 2(11):688–701

Demszky D, Yang D, Yeager DS, et al (2023b) Using large language models in

psychology. Nature Reviews Psychology 2(11):688–701

Denecke K, Schmid N, Nüssli S, et al (2022) Implementation of cognitive behavioral

therapy in e–mental health apps: Literature review. Journal of medical Internet
research 24(3):e27791

DiMatteo MR, Giordani PJ, Lepper HS, et al (2002) Patient adherence and medical
treatment outcomes: a meta-analysis. Medical care 40(9):794–811

Dimidjian S, Hollon SD, Dobson KS, et al (2006) Randomized trial of behavioral

activation, cognitive therapy, and antidepressant medication in the acute treat-
ment of adults with major depression. Journal of consulting and clinical psychology
74(4):658

Ding X, Lybarger K, Tauscher J, et al (2022) Improving classification of infrequent

cognitive distortions: domain-specific model vs. data augmentation. In: Proceedings

36
 of the 2022 conference of the North American chapter of the association for compu-
tational linguistics: human language technologies: Student Research Workshop, pp
68–75

Dobson KS, Dozois DJ (2021) Handbook of cognitive-behavioral therapies. Guilford

Publications

Driessen E, Hollon SD (2010) Cognitive behavioral therapy for mood disorders:

efficacy, moderators and mediators. Psychiatric Clinics 33(3):537–555

Elsharawi N, El Bolock A (2024) C-journal: A journaling application for detecting

and classifying cognitive distortions using deep-learning based on a crowd-sourced
dataset. In: Proceedings of the 2024 Joint International Conference on Computa-
tional Linguistics, Language Resources and Evaluation (LREC-COLING 2024), pp
3224–3234

Elyoseph Z, Hadar-Shoval D, Asraf K, et al (2023) Chatgpt outperforms humans in

emotional awareness evaluations. Frontiers in Psychology 14:1199058

Ewbank MP, Cummins R, Tablan V, et al (2020) Quantifying the association between

psychotherapy content and clinical outcomes using deep learning. JAMA psychiatry
77(1):35–43

Fitzpatrick KK, Darcy A, Vierhile M (2017) Delivering cognitive behavior therapy

to young adults with symptoms of depression and anxiety using a fully automated
conversational agent (woebot): a randomized controlled trial. JMIR mental health
4(2):e7785

Flemotomos N, Martinez VR, Chen Z, et al (2021) Automated quality assessment

of cognitive behavioral therapy sessions through highly contextualized language
representations. PloS one 16(10):e0258639

Flemotomos N, Martinez VR, Chen Z, et al (2022) Automated evaluation of psy-

chotherapy skills using speech and language technologies. Behavior Research
Methods 54(2):690–711

Foa EB, Riggs DS, Dancu CV, et al (1993) Reliability and validity of a brief instrument
for assessing post-traumatic stress disorder. Journal of traumatic stress 6(4):459–473

Foreman EI, Pollard C (2016) Cognitive Behavioural Therapy (CBT): Your Toolkit to
Modify Mood, Overcome Obstructions and Improve Your Life. Icon Books, Limited

Fu G, Zhao Q, Li J, et al (2023) Enhancing psychological counseling with large lan-

guage model: A multifaceted decision-support system for non-professionals. arXiv
preprint arXiv:230815192

37
Furukawa TA, Iwata S, Horikoshi M, et al (2023) Harnessing ai to optimize thought
records and facilitate cognitive restructuring in smartphone cbt: An exploratory
study. Cognitive Therapy and Research 47(6):887–893

Garcia-Ceja E, Osmani V, Mayora O (2015) Automatic stress detection in working

environments from smartphones’ accelerometer data: a first step. IEEE journal of
biomedical and health informatics 20(4):1053–1060

Giordano R, Donati MA, Zamboni L, et al (2022) Alter game: A study protocol on

a virtual “serious game” for relapse prevention in patients with gambling disorder.
Frontiers in Psychiatry 13:854088

Goodwin MS, Mazefsky CA, Ioannidis S, et al (2019) Predicting aggression to others

in youth with autism using a wearable biosensor. Autism research 12(8):1286–1296

Graham S, Depp C, Lee EE, et al (2019) Artificial intelligence for mental health and
mental illnesses: an overview. Current psychiatry reports 21:1–18

Gunlicks-Stoessel M, Klimes-Dougan B, VanZomeren A, et al (2020) Developing

a data-driven algorithm for guiding selection between cognitive behavioral ther-
apy, fluoxetine, and combination treatment for adolescent depression. Translational
psychiatry 10(1):321

Hahn T, Kircher T, Straube B, et al (2015) Predicting treatment response to cognitive

behavioral therapy in panic disorder with agoraphobia by integrating local neural
information. JAMA psychiatry 72(1):68–74

Harrison V, Proudfoot J, Wee PP, et al (2011) Mobile mental health: review of the
emerging field and proof of concept study. Journal of mental health 20(6):509–524

Heapy AA, Higgins DM, Goulet JL, et al (2017) Interactive voice response–based
self-management for chronic back pain: the copes noninferiority randomized trial.
JAMA Internal Medicine 177(6):765–773

Helbig S, Fehm L (2004) Problems with homework in cbt: Rare exception or rather
frequent? Behavioural and cognitive psychotherapy 32(3):291–301

Heng YK (2021) Rewind: psychoeducation game leveraging cognitive behavioral ther-

apy (cbt) to enhance emotion control for generalized anxiety disorder. In: Extended
Abstracts of the 2021 CHI Conference on Human Factors in Computing Systems,
pp 1–5

Hilbert K, Kunas SL, Lueken U, et al (2020) Predicting cognitive behavioral therapy

outcome in the outpatient sector based on clinical routine data: A machine learning
approach. Behaviour research and therapy 124:103530

38
Hilbert K, Jacobi T, Kunas SL, et al (2021) Identifying cbt non-response among ocd
outpatients: A machine-learning approach. Psychotherapy Research 31(1):52–62

Huguet A, Rao S, McGrath PJ, et al (2016) A systematic review of cognitive behavioral

therapy and behavioral activation apps for depression. PloS one 11(5):e0154248

Isacsson NH, Abdesslem FB, Forsell E, et al (2023) Machine learning predictions of

outcome in internet-based cognitive behavioral therapy: methodological choices and
clinical usefulness. Preprint posted online on April

Izumi K, Tanaka H, Shidara K, et al (2024) Response generation for cognitive

behavioral therapy with large language models: Comparative study with socratic
questioning. arXiv preprint arXiv:240115966

Jacobson NS, Dobson KS, Truax PA, et al (1996) A component analysis of cognitive-
behavioral treatment for depression. Journal of Consulting and Clinical Psychology
64(2):295

Jameel L (2020) Virtual-reality assisted cbt for social difficulties: a feasibility study
in early intervention for psychosis services. PhD thesis, King’s College London

Jang S, Kim JJ, Kim SJ, et al (2021) Mobile app-based chatbot to deliver cognitive
behavioral therapy and psychoeducation for adults with attention deficit: A devel-
opment and feasibility/usability study. International journal of medical informatics
150:104440

Jeong Y, Song JJ, Yang J, et al (2024) Advancing tinnitus therapeutics: Gpt-2 driven
clustering analysis of cognitive behavioral therapy sessions and google t5-based
predictive modeling for thi score assessment. IEEE Access

Jiang M, Yu YJ, Zhao Q, et al (2024) Ai-enhanced cognitive behavioral therapy: Deep

learning and large language models for extracting cognitive pathways from social
media texts. arXiv preprint arXiv:240411449

Kaldo V, Isacsson N, Forsell E, et al (2021) Ai-driven adaptive treatment strategies

in internet-delivered cbt. European Psychiatry 64(S1):S20–S20

Kazantzis N, Arntz AR, Borkovec T, et al (2010) Unresolved issues regarding home-

work assignments in cognitive and behavioural therapies: An expert panel discussion
at aacbt. Behaviour Change 27(3):119–129

Khare SK, Blanes-Vidal V, Nadimi ES, et al (2023) Emotion recognition and artifi-
cial intelligence: A systematic review (2014–2023) and research recommendations.
Information Fusion p 102019

39
Kozlowski M, Gabor-Siatkowska K, Stefaniak I, et al (2023) Enhanced emotion and
sentiment recognition for empathetic dialogue system using big data and deep learn-
ing methods. In: International Conference on Computational Science, Springer, pp
465–480

Lan Z, Chen M, Goodman S, et al (2019) Albert: A lite bert for self-supervised learning
of language representations. arXiv preprint arXiv:190911942

LeBeau RT, Davies CD, Culver NC, et al (2013) Homework compliance counts in
cognitive-behavioral therapy. Cognitive behaviour therapy 42(3):171–179

Lee EE, Torous J, De Choudhury M, et al (2021) Artificial intelligence for men-

tal health care: clinical applications, barriers, facilitators, and artificial wisdom.
Biological Psychiatry: Cognitive Neuroscience and Neuroimaging 6(9):856–864

Lee S, Kang J, Kim H, et al (2024) Cocoa: Cbt-based conversational counseling

agent using memory specialized in cognitive distortions and dynamic prompt. arXiv
preprint arXiv:240217546

Lee YK, Lee I, Shin M, et al (2023) Chain of empathy: Enhancing empathetic

response of large language models based on psychotherapy models. arXiv preprint
arXiv:231104915

Lewinsohn PM (1974) A behavioral approach to depression.

Li JZ, Herderich A, Goldenberg A (2024) Skill but not effort drive gpt overperformance
over humans in cognitive reframing of negative scenarios. Preprint posted online on
April

Lim S, Kim Y, Choi CH, et al (2024) Erd: A framework for improving llm reasoning
for cognitive distortion classification. arXiv preprint arXiv:240314255

Lin S, Wang Y, Dong J, et al (2024) Detection and positive reconstruction of

cognitive distortion sentences: Mandarin dataset and evaluation. arXiv preprint
arXiv:240515334

Linardon J, Wade TD, De la Piedad Garcia X, et al (2017) The efficacy of cognitive-

behavioral therapy for eating disorders: A systematic review and meta-analysis.
Journal of consulting and clinical psychology 85(11):1080

Lorimer B, Delgadillo J, Kellett S, et al (2021) Dynamic prediction and identification

of cases at risk of relapse following completion of low-intensity cognitive behavioural
therapy. Psychotherapy Research 31(1):19–32

Maddela M, Ung M, Xu J, et al (2023) Training models to generate, recognize, and

reframe unhelpful thoughts. In: Proceedings of the 61st Annual Meeting of the
Association for Computational Linguistics (Volume 1: Long Papers), pp 13641–
13660

40
Madhu SH, Kumar SS, Pal M, et al (2022) Activity recognition for behavioral acti-
vation in depression with artificial intelligence. In: 2022 IEEE 4th PhD Colloquium
on Emerging Domain Innovation and Technology for Society (PhD EDITS), IEEE,
pp 1–2

Malgaroli M, Hull TD, Zech JM, et al (2023) Natural language processing for mental
health interventions: a systematic review and research framework. Translational
Psychiatry 13(1):309

Månsson KN, Frick A, Boraxbekk CJ, et al (2015) Predicting long-term outcome of

internet-delivered cognitive behavior therapy for social anxiety disorder using fmri
and support vector machine learning. Translational psychiatry 5(3):e530–e530

Maples-Keller JL, Bunnell BE, Kim SJ, et al (2017) The use of virtual reality tech-
nology in the treatment of anxiety and other psychiatric disorders. Harvard review
of psychiatry 25(3):103–113

Martı́nez-Miranda J, Martı́nez A, Ramos R, et al (2019) Assessment of users’ accept-

ability of a mobile-based embodied conversational agent for the prevention and
detection of suicidal behaviour. Journal of medical systems 43(8):246

McGinn LK, Sanderson WC (2001) What allows cognitive behavioral therapy to be

brief: Overview, efficacy, and crucial factors facilitating brief treatment. Clinical
psychology: science and practice 8(1):23

McHugh RK, Hearon BA, Otto MW (2010) Cognitive behavioral therapy for substance
use disorders. Psychiatric Clinics 33(3):511–525

Mehta A, Niles AN, Vargas JH, et al (2021) Acceptability and effectiveness of artificial
intelligence therapy for anxiety and depression (youper): Longitudinal observational
study. Journal of medical Internet research 23(6):e26771

Michelle TQY, Jarzabek S, Wadhwa B (2014) Cbt assistant: Mhealth app for psy-
chotherapy. In: 2014 IEEE Global Humanitarian Technology Conference-South Asia
Satellite (GHTC-SAS), IEEE, pp 135–140

Moody T, Morfini F, Cheng G, et al (2017) Mechanisms of cognitive-behavioral ther-

apy for obsessive-compulsive disorder involve robust and extensive increases in brain
network connectivity. Translational psychiatry 7(9):e1230–e1230

Mostafa M, El Bolock A, Abdennadher S (2021) Automatic detection and classification

of cognitive distortions in journaling text. In: WEBIST, pp 444–452

Na H (2024) Cbt-llm: A chinese large language model for cognitive behavioral therapy-
based mental health question answering. arXiv preprint arXiv:240316008

41
Nazarova D (2023) Application of artificial intelligence in mental healthcare: Genera-
tive pre-trained transformer 3 (gpt-3) and cognitive distortions. In: Proceedings of
the Future Technologies Conference, Springer, pp 204–219

Nepal S, Pillai A, Campbell W, et al (2024) Contextual ai journaling: Integrating

llm and time series behavioral sensing technology to promote self-reflection and
well-being using the mindscape app. arXiv preprint arXiv:240400487

Nwoye EO, Muslehat AA, Umeh C, et al (2024) Schizobot: Delivering cogni-

tive behavioural therapy for augmented management of schizophrenia. Digital
Technologies Research and Applications 3(2):24–40

Olatunji BO, Cisler JM, Deacon BJ (2010) Efficacy of cognitive behavioral therapy for
anxiety disorders: a review of meta-analytic findings. Psychiatric Clinics 33(3):557–
577

Oliveira ALS, Matos LN, Junior MC, et al (2021) An initial assessment of a chatbot
for rumination-focused cognitive behavioral therapy (rfcbt) in college students. In:
Computational Science and Its Applications–ICCSA 2021: 21st International Con-
ference, Cagliari, Italy, September 13–16, 2021, Proceedings, Part VI 21, Springer,
pp 549–564

Ørskov PT, Lichtenstein MB, Ernst MT, et al (2022) Cognitive behavioral ther-
apy with adaptive virtual reality exposure vs. cognitive behavioral therapy with in
vivo exposure in the treatment of social anxiety disorder: A study protocol for a
randomized controlled trial. Frontiers in psychiatry 13:991755

Ozomaro U, Wahlestedt C, Nemeroff CB (2013) Personalized medicine in psychiatry:

problems and promises. BMC medicine 11:1–35

Pan MR, Huang F, Zhao MJ, et al (2019) A comparison of efficacy between cognitive
behavioral therapy (cbt) and cbt combined with medication in adults with attention-
deficit/hyperactivity disorder (adhd). Psychiatry research 279:23–33

Parry GD, Cooper CL, Moore JM, et al (2012) Cognitive behavioural intervention
for adults with anxiety complications of asthma: prospective randomised trial.
Respiratory Medicine 106(6):802–810

Patel F, Thakore R, Nandwani I, et al (2019) Combating depression in students using

an intelligent chatbot: a cognitive behavioral therapy. In: 2019 IEEE 16th India
Council International Conference (INDICON), IEEE, pp 1–4

Pei W, He T, Yang P, et al (2022) Acupuncture combined with cognitive–behavioural

therapy for insomnia (cbt-i) in patients with insomnia: study protocol for a
randomised controlled trial. BMJ open 12(12):e063442

42
Penninx BW, Benros ME, Klein RS, et al (2022) How covid-19 shaped mental health:
from infection to pandemic effects. Nature medicine 28(10):2027–2037

Peretz G, Taylor CB, Ruzek JI, et al (2023) Machine learning model to predict assign-
ment of therapy homework in behavioral treatments: Algorithm development and
validation. JMIR Formative Research 7:e45156

Petersen TJ, Sprich SE, Wilhelm S, et al (2016) The massachusetts general hospital
handbook of cognitive behavioral therapy. Tech. rep., Springer

Piette JD, Krein SL, Striplin D, et al (2016) Patient-centered pain care using artificial
intelligence and mobile health tools: protocol for a randomized study funded by the
us department of veterans affairs health services research and development program.
JMIR research protocols 5(2):e4995

Prasad N, Chien I, Regan T, et al (2023) Deep learning for the prediction of

clinical outcomes in internet-delivered cbt for depression and anxiety. Plos one
18(11):e0272685

Provoost S, Ruwaard J, Van Breda W, et al (2019) Validating automated sentiment

analysis of online cognitive behavioral therapy patient texts: an exploratory study.
Frontiers in psychology 10:432272

Qi H, Zhao Q, Song C, et al (2023) Evaluating the efficacy of supervised learning

vs large language models for identifying cognitive distortions and suicidal risks in
chinese social media. arXiv preprint arXiv:230903564

Rahman MA, Brown DJ, Shopland N, et al (2022) Towards machine learning driven
self-guided virtual reality exposure therapy based on arousal state detection from
multimodal data. In: International Conference on Brain Informatics, Springer, pp
195–209

Rajagopal A, Nirmala V, Andrew J, et al (2021) Novel ai to avert the mental health

crisis in covid-19: Novel application of gpt2 in cognitive behaviour therapy. Preprint
posted online on January

Rani K, Vishnoi H, Mishra M (2023) A mental health chatbot delivering cogni-

tive behavior therapy and remote health monitoring using nlp and ai. In: 2023
International Conference on Disruptive Technologies (ICDT), IEEE, pp 313–317

Rathje S, Mirea D, Sucholutsky I, et al (2023) Gpt is an effective tool for multilingual

psychological text analysis. psyarxiv. Preprint posted online on July 17

Rathnayaka P, Mills N, Burnett D, et al (2022) A mental health chatbot with cognitive

skills for personalised behavioural activation and remote health monitoring. Sensors
22(10):3653

43
Reger GM, Hoffman J, Riggs D, et al (2013) The “pe coach” smartphone applica-
tion: An innovative approach to improving implementation, fidelity, and homework
adherence during prolonged exposure. Psychological services 10(3):342

Reggente N, Moody TD, Morfini F, et al (2018) Multivariate resting-state func-

tional connectivity predicts response to cognitive behavioral therapy in obsessive–
compulsive disorder. Proceedings of the National Academy of Sciences 115(9):2222–
2227

Rizea A (2022) Deep learning-based solution for mental health issues. DATABASE
SYSTEMS p 57

Rohani DA, Quemada Lopategui A, Tuxen N, et al (2020) Mubs: A personalized

recommender system for behavioral activation in mental health. In: Proceedings of
the 2020 CHI Conference on Human Factors in Computing Systems, pp 1–13

Rose RD, Buckey Jr JC, Zbozinek TD, et al (2013) A randomized controlled trial
of a self-guided, multimedia, stress management and resilience training program.
Behaviour research and therapy 51(2):106–112

Sabour S, Zhang W, Xiao X, et al (2023) A chatbot for mental health support: explor-
ing the impact of emohaa on reducing mental distress in china. Frontiers in digital
health 5:1133987

Schabus M, Eigl ES, Topalidis P, et al (2023) Efficacy of digital cognitive behavioural

therapy for insomnia: A randomised controlled trial using a new app that tracks
sleep continuously using hrv. In: World Sleep 2023

Schwartz B, Cohen ZD, Rubel JA, et al (2021) Personalized treatment selection in

routine care: Integrating machine learning and statistical algorithms to recommend
cognitive behavioral or psychodynamic therapy. Psychotherapy Research 31(1):33–
51

Scozzari S, Gamberini L (2011) Virtual reality as a tool for cognitive behavioral

therapy: a review. Advanced computational intelligence paradigms in healthcare 6
Virtual reality in psychotherapy, rehabilitation, and assessment pp 63–108

Sharma A, Miner A, Atkins D, et al (2020) A computational approach to understand-

ing empathy expressed in text-based mental health support. In: Proceedings of the
2020 Conference on Empirical Methods in Natural Language Processing (EMNLP),
pp 5263–5276

Sharma A, Lin IW, Miner AS, et al (2023a) Human–ai collaboration enables more
empathic conversations in text-based peer-to-peer mental health support. Nature
Machine Intelligence 5(1):46–57

44
Sharma A, Rushton K, Lin IW, et al (2023b) Facilitating self-guided mental health
interventions through human-language model interaction: A case study of cognitive
restructuring. arXiv preprint arXiv:231015461

Sharma A, Rushton K, Lin IW, et al (2023c) Cognitive reframing of negative thoughts

through human-language model interaction. arXiv preprint arXiv:230502466

Shickel B, Siegel S, Heesacker M, et al (2020) Automatic detection and classification

of cognitive distortions in mental health text. In: 2020 IEEE 20th International
Conference on Bioinformatics and Bioengineering (BIBE), IEEE, pp 275–280

Shidara K, Tanaka H, Adachi H, et al (2022) Automatic thoughts and facial expres-

sions in cognitive restructuring with virtual agents. Frontiers in Computer Science
4:762424

Shreevastava S, Foltz P (2021) Detecting cognitive distortions from patient-therapist

interactions. In: Proceedings of the Seventh Workshop on Computational Linguistics
and Clinical Psychology: Improving Access, pp 151–158

Shurick AA, Hamilton JR, Harris LT, et al (2012) Durable effects of cognitive
restructuring on conditioned fear. Emotion 12(6):1393

Siddiqua R, Islam N, Bolaka JF, et al (2023) Aida: Artificial intelligence based

depression assessment applied to bangladeshi students. Array 18:100291

Simms T, Ramstedt C, Rich M, et al (2017) Detecting cognitive distortions through

machine learning text analytics. In: 2017 IEEE international conference on health-
care informatics (ICHI), IEEE, pp 508–512

Singh GV, Ghosh S, Ekbal A, et al (2023) Decode: Detection of cognitive distortion

and emotion cause extraction in clinical conversations. In: European Conference on
Information Retrieval, Springer, pp 156–171

Stephenson C, Jagayat J, Kumar A, et al (2023) Comparing clinical decision-making of

ai technology to a multi-professional care team in an electronic cognitive behavioural
therapy program for depression: protocol. Frontiers in Psychiatry 14:1220607

Stirman SW, Gutner CA, Gamarra J, et al (2021) A novel approach to the assessment
of fidelity to a cognitive behavioral therapy for ptsd using clinical worksheets: A
proof of concept with cognitive processing therapy. Behavior therapy 52(3):656–672

Striegl J, Richter JW, Grossmann L, et al (2023) Deep learning-based dimensional

emotion recognition for conversational agent-based cognitive behavioral therapy.
Preprint posted online on December

Su S, Wang Y, Jiang W, et al (2022) Efficacy of artificial intelligence-assisted psy-

chotherapy in patients with anxiety disorders: a prospective, national multicenter

45
 randomized controlled trial protocol. Frontiers in Psychiatry 12:799917

Sun H, Lin Z, Zheng C, et al (2021) Psyqa: A chinese dataset for generating long
counseling text for mental health support. In: Findings of the Association for
Computational Linguistics: ACL-IJCNLP 2021, pp 1489–1503

Tanana MJ, Soma CS, Kuo PB, et al (2021) How do you feel? using natural lan-
guage processing to automatically rate emotion in psychotherapy. Behavior research
methods pp 1–14

Tang W, Kreindler D, et al (2017) Supporting homework compliance in cogni-

tive behavioural therapy: essential features of mobile apps. JMIR mental health
4(2):e5283

Tauscher JS, Lybarger K, Ding X, et al (2023) Automated detection of cognitive

distortions in text exchanges between clinicians and people with serious mental
illness. Psychiatric services 74(4):407–410

de Toledo Rodriguez I, Salton G, Ross R (2021) Formulating automated responses to

cognitive distortions for cbt interactions. In: Proceedings of the 4th International
Conference on Natural Language and Speech Processing (ICNLSP 2021), pp 108–
116

Tolmeijer E, Kumari V, Peters E, et al (2018) Using fmri and machine learning to pre-
dict symptom improvement following cognitive behavioural therapy for psychosis.
NeuroImage: Clinical 20:1053–1061

Turkington D, Dudley R, Warman DM, et al (2004) Cognitive-behavioral therapy for

schizophrenia: a review. Journal of Psychiatric Practice® 10(1):5–16

Tymofiyeva O, Yuan JP, Huang CY, et al (2019) Application of machine learning to

structural connectome to predict symptom reduction in depressed adolescents with
cognitive behavioral therapy (cbt). NeuroImage: Clinical 23:101914

Vieira S, Liang X, Guiomar R, et al (2022) Can we predict who will benefit from
cognitive-behavioural therapy? a systematic review and meta-analysis of machine
learning studies. Clinical Psychology Review 97:102193

Vuyyuru VA, Krishna GV, Mary SSC, et al (2023) A transformer-cnn hybrid model
for cognitive behavioral therapy in psychological assessment and intervention for
enhanced diagnostic accuracy and treatment efficiency. International Journal of
Advanced Computer Science and Applications 14(7)

Wanderley Espinola C, Gomes JC, Mônica Silva Pereira J, et al (2022) Detection of

major depressive disorder, bipolar disorder, schizophrenia and generalized anxiety
disorder using vocal acoustic analysis and machine learning: an exploratory study.
Research on Biomedical Engineering 38(3):813–829

46
Wang B, Deng P, Zhao Y, et al (2023a) C2d2 dataset: A resource for the cognitive
distortion analysis and its impact on mental health. In: The 2023 Conference on
Empirical Methods in Natural Language Processing

Wang B, Zhao Y, Lu X, et al (2023b) Cognitive distortion based explainable depression

detection and analysis technologies for the adolescent internet users on social media.
Frontiers in Public Health 10:1045777

Wang R, Milani S, Chiu JC, et al (2024a) PATIENT-Ψ: Using large language mod-
els to simulate patients for training mental health professionals. arXiv preprint
arXiv:240519660

Wang X, Sharma D, Kumar D (2024b) Cognitive reframing via large language models
for enhanced linguistic attributes. In: The Second Tiny Papers Track at ICLR 2024

Webb CA, Rosso IM, Rauch SL (2017) Internet-based cognitive-behavioral therapy

for depression: current progress and future directions. Harvard review of psychiatry
25(3):114–122

Wei Y, Zhang R, Wang Y, et al (2023) Towards a neuroimaging biomarker for

predicting cognitive behavioural therapy outcomes in treatment-naive depression:
Preliminary findings. Psychiatry Research 329:115542

Whiteside SP, Ale CM, Vickers Douglas K, et al (2014) Case examples of enhanc-
ing pediatric ocd treatment with a smartphone application. Clinical Case Studies
13(1):80–94

Xiao M, Xie Q, Kuang Z, et al (2024) Healme: Harnessing cognitive reframing in large

language models for psychotherapy. arXiv preprint arXiv:240305574

Xing Z, Zhao X, Miao C (2017) Identifying cognitive distortion by convolutional neural

network based text classification. International Journal of Information Technology

Yang S, Gao B, Jiang L, et al (2018) Iot structured long-term wearable social sensing
for mental wellbeing. IEEE Internet of Things Journal 6(2):3652–3662

Yin Y, Jia N, Wakslak CJ (2024) Ai can help people feel heard, but an ai
label diminishes this impact. Proceedings of the National Academy of Sciences
121(14):e2319112121

Zhan H, Zheng A, Lee YK, et al (2024) Large language models are capable of offering
cognitive reappraisal, if guided. arXiv preprint arXiv:240401288

Zhang S, Dinan E, Urbanek J, et al (2018) Personalizing dialogue agents: I have a dog,

do you have pets too? In: Proceedings of the 56th Annual Meeting of the Association
for Computational Linguistics (Volume 1: Long Papers), pp 2204–2213

47
Zhang X, Tanana M, Weitzman L, et al (2023) You never know what you are
going to get: Large-scale assessment of therapists’ supportive counseling skill use.
Psychotherapy 60(2):149

Zheng Y, Ye Y, et al (2022) Prediction of cognitive-behavioral therapy using deep

learning for the treatment of adolescent social anxiety and mental health conditions.
Scientific Programming 2022

48