https://doi.org/10.2352/ISSN.2470-1173.2021.3.

MOBMU-035
© 2021, Society for Imaging Science and Technology

Integration of NLP and Speech-to-text Applications with

Chatbots
D. Inupakutika, M. Nadim, G. R. Gunnam, S. Kaghyan, D. Akopian; The University of Texas at San Antonio; San Antonio, P. Chalela,
A. G. Ramirez; University of Texas Health Science Center at San Antonio, TX, USA

Abstract virtual assistants to decide a response or action, entities that

With the evolving artificial intelligence technology, the illustrate important information such as date, times and locations,
chatbots are becoming smarter and faster lately. Chatbots are and contexts correspond to the context of the object the user is
typically available round the clock providing continuous support referring to [4]. The accurate response to users’ input by a chatbot
and services. A chatbot or a conversational agent is a program or requires the combination of these intents, entities and contexts.
software that can communicate using natural language with The tech giants and leading cloud service providers such as
humans. The challenge of developing an intelligent chatbot still Google, Amazon, Microsoft, Facebook, and IBM have been
exists ever since the onset of artificial intelligence. The functionality developing cloud-based platforms/ services with NLU engines, AI/
of chatbots can range from business oriented short conversations to machine learning stack, and necessary development and deployment
healthcare intervention based longer conversations. However, the tools, to enable fast prototyping of digital virtual assistants/
primary role that the chatbots have to play is in understanding conversational agents. These companies perform computation on
human utterances in order to respond appropriately. To that end, their servers, and offer the NLU applications and modules as cloud-
there is an increased emergence of Natural Language based service, on their platform. Some of these services do not
Understanding (NLU) engines by popular cloud service providers. require knowledge of NLU algorithms, and hence, are the black-box
The NLU services identify entities and intents from the user machine learning models, but expect users to input training
utterances provided as input. Thus, in order to integrate such examples. In this paper, we perform an in-depth study of the primary
understanding to a chatbot, this paper presents a study on existing NLP applications that are available as cloud-based services,
major NLU platforms. Then, we present a case study chatbot evaluate the performance of a case study chatbot integrated with two
integrated with Google DialogFlow and IBM Watson NLU services of the cloud-based NLU services Google DialogFlow [5] and IBM
and discuss their intent recognition performance. Watson [6] in a healthcare-based chatbot context. Note that the
terms NLP applications and NLU services or engines are used
Index Terms – chatbot, natural language processing, dialog interchangeably in this study.
system, apps, deep learning The rest of the paper is organized as follows: In State-of-the-Art
section, the chatbot types, typical chatbot architecture and major
NLP (NLU and Speech-to-text) applications are discussed. NLU
Introduction Core Features section presents the classification of parameters for
the comparison of these applications. In Case Study Chatbot
There has been significant development in Artificial Intelligence Section, the prototype healthcare-based chatbot created by
(AI) in recent years. While the trend of AI has been there for the past integrating with two of the NLP applications are discussed. In
several years [1], it received great attention since AlphaGo [2] won Experiments and Results Section, the testing is explained followed
at the Google’s DeepMind challenge. Lately, ever since the rise of by the intent classification results of these two platforms.
AI, specifically natural language processing (NLP) and Conclusion Section concludes our study and provides possible
computational linguistics, chatbots are becoming the modern future research directions.
technological aspect in the next generation of conversational
services. The chatbots are the human-computer dialog systems that State-of-the-Art
interact with users in a natural language [3]. Digital assistants,
intelligent interactive agents, and artificial conversational entities Conversational software systems that are designed to mimic
are also known as chatbots. The use of natural language provides communication abilities of humans involving automatic interaction
offers an intuitive, flexible and easy way in human-machine with a user are called chatbots. They are a new, advanced, and
interaction. Chatbots provide responses or perform actions by modern customer assistance form driven by AI through a chat
processing text or speech, which humans sent digitally and can interface, also known as channel. In this section, we discuss the
closely mimic human conversations through NLP techniques and taxonomy of chatbots, typical chatbot architecture and its
sentiment analysis. The natural language understanding (NLU) constituent components, and major NLU/ speech-to-text
engine is the core of a chatbot’s NLP task. It is a process of applications for integration.
implementing natural user interfaces. NLU allows identifying
underlying intents, extract context, meanings and domain-specific Chatbots Types
entities in users’ utterances. The steps that a chatbot takes when
specific intents are triggered by user utterances are the actions and The chatbot classifications are done based on the aspects such as the
these have parameters that mention detailed information about it. process of response creation, the domain and duration of
Three aspects in a sentence are typically identified by NLU, namely, conversation, and whether they include AI self-learning programs
intent that maps users’ utterance to a specific class that lets digital or not. Chatbots are generally classified into two broad groups called

IS&T International Symposium on Electronic Imaging 2021

Mobile Devices and Multimedia: Enabling Technologies, Algorithms, and Applications 2021 035-1
task-oriented and conversational non-task-oriented [7]. Task- Train
oriented chatbots assist users to accomplish a particular task and Annotate with
with learning
have shorter conversations related to that task. Alexa or Siri helping intents language
Collect Examine
Publish users’ system
users in finding nearby restaurants, playing a song, making a phone and models
input results
entities (ML/
call or sending text messages are some of such examples. Non-task- DL)
oriented chatbots are the conversational ones that engage users in a
conversation while answering their questions. These chatbots rely
on NLP to extract user information and react with highly matched
response and can be setup to either have shorter or longer Fig. 2. NLP development process
conversations. More recent works [8] on chatbot implementation
have divided chatbots into four groups namely, goal-based, A generic architecture of a chatbot consists of some mandatory and
knowledge-based, service-based, and response generated-based. some optional components. Two types of communication modes as
These categories also fall under the former classification of task- and input are possible through the presentation layer constituting of user
non-task-oriented chatbots. Goal-based chatbots are based on the interface: text or speech. A voice/ speech input needs to be
goal to achieve consist of activity-based, conversational and processed to a text understandable by machines using automatic
informative types. Knowledge-based chatbots are based on the speech recognition (ASR). Next, the language understanding
knowledge they extract from the data sources or the datasets they component called NLU parses this text or direct user text input to
are trained on include open- and closed-domain or domain-specific extract user’s intention, context and other annotations such as intent
ones. Service-based chatbots are based on the conveniences and entities. The dialog management unit performs further
provided to users including inter-, intra-personal, and inter-agent processing by evaluating with the backend system and continuously
purposes. Finally, the response generated-based type are based on asking for clarification in case it requires more context information.
the process used for response creation and these constitute template- This evaluation is done either through external application systems
based, generative, retrieval-based, and search engine models, that through API calls or by querying the data sources such as database
are further classified based on the different types of neural network consisting of user interaction history or knowledge base. The
and word embedding models [8]. chatbot performs the requested actions based on this evaluation. The
result is then converted to a natural language via natural language
Furthermore, authors in [9] also classified chatbots on an additional generation (NLG) unit, that is a response generation component. If
aspect related to the permissions provided by the development needed, the generated text can be sent as audio using text-to-speech
platform as well as the chatbot component that we prefer to adapt in component or sent as text typically.
our use-case. Such platforms fall under two categories namely,
open-source such as RASA [10], or closed or proprietary platforms,
NLU and Speech-to-text Applications
typically by aforementioned large companies and tech giants that are
In this study, we focus primarily on the NLU applications that
mostly black boxes and only expose some of the aspects of their
comprise of both text-to-text and speech-to-text based. More
implementation in order for us to integrate in our development
specifically, the NLU incorporated in the existing frameworks for
environment. One major advantages with the services by large
developers. To that end, there are several NLU cloud services, that
companies is the large amount of data that they collect for
are available for integration into an existing chatbot development
continuous improvement of their underlying models. In this study,
environment in order to be able to understand natural language.
we are interested in the NLU services by these providers or
Some of these modern and famous ones are: Google Dialogflow [5],
development frameworks. A related work in [11] presents an
Facebook Wit.ai [12], Microsoft LUIS [13], Amazon Lex [14], IBM
analysis of three additional widely used chatbot development web
Watson Conversation [6], and SAP Conversation AI [15], among
platforms called Chatfuel, ManyChat, and It’s Alive.
others [9]. One can integrate these services even without having a
Technical Chatbot Architecture complete knowledge of NLP and ML. These services map input text
or the machine-readable text converted from speech input into
A chatbot has several major components and a combination/ choice semantic slots. Fig. 2 shows a standard process involved in NLP.
of these components result in an efficient chatbot. Based on the use- Initially, the entered data is annotated with intents and entities
case and the chatbot type, the developers choose the algorithms, similar to entering data as part of cloud-based NLU service. The
models, NLP applications, development platform and tools to build language processing ML/ DL models are trained using the annotated
it. In fig. 1, we present a generic technical architecture of a chatbot. data. The trained model is then published and evaluated with test
user utterances. The NLU services discussed in this study
continuously evaluate the bot performance and collect user inputs to
improve the model. Most of the mentioned cloud NLU services
facilitate creation of intents known as user intent category (at the
user utterance level), and entities for extracting word-level
information (named entity recognition (NER) and slot filling). NLU
services also include pre-defined slots that can be used in our
implementation for different scenarios. Most of the NLU services
have multi-language support. We discuss some of these services
next.

Fig. 1. Generic Chatbot Architecture 1) Google Dialogflow

IS&T International Symposium on Electronic Imaging 2021

035-2 Mobile Devices and Multimedia: Enabling Technologies, Algorithms, and Applications 2021
Dialogflow’s NLU service is part of Google cloud platform. It is able to extract customized insights with Watson Knowledge
facilitates integration of natural language processing technologies to Studio (WKS). A custom model with the domain-specific use-case
existing chatbots and enables developers to provide users with related entities and relations, allows to identify all the occurrences
conversation interface for both voice (takes care of speech-to-text of an entity in users' utterances. WKS uses ground truth to train a
conversion) and text inputs. Intents, entities, and contexts are the model. Model from WKS can then be deployed to our NLU instance
primary elements that enable modeling of chatbot behavior. within the WKSs user interface. The integration with our chatbot is
Additionally, it provides functionality solely for keyword detection then possible by making API calls with NLU. Watson speech to text
too in case the dialog management is taken care by a separate service enables better service to users, allowing them to ask questions using
in our chatbot application. Such feature is possible with webhooks natural language, fast-tracking them to the answers, with custom
and fulfillment to customize behavior of chatbots. Intents in speech-to-text capabilities. This service is driven by ML and has an
Dialogflow link user utterances with the chatbot response or action. AI-powered speech recognition and transcription.
Contexts enable differentiation of user requests with different
meanings. Dialogflow links to the user interface via many channels 5) Amazon Lex
such as Slack, Facebook Messenger, Skype, Telegram, and Alexa,
among others. Amazon Lex enables conversational AI for chatbots, is an Amazon
web services (AWS) service developed by Amazon. This NLU
2) Microsoft LUIS
service can be integrated to any application using voice and text. It
provides advanced deep learning functionalities of ASR (speech to
Microsoft’s language understanding information service (LUIS) is a text), and NLU for recognizing the intent from users utterances.
cloud-based domain-specific NLU service and comes with the Bot Lately, the latest deep learning technologies that power Amazon
framework SDK. An intent is triggered once the user input is Alexa are available with Lex that enables integration of NLU
matched with the prior knowledge after semantic check. Similar to (comparatively more complex integration as compared to Watson)
Dialgflow, LUIS also has entities to extract information. Unlike with our chatbots for accurate natural language conversation. Lex
other NLU services, LUIS facilitates creation of composite entities has in-built integration with AWS Lambda and with other services
required for understanding complex information in user utterance. on AWS. Few example phrases need to be inserted, using which Lex
This service does not let developers control the context parameters builds a complete natural language model with which users can
and does not have diverse set of domains unlike Dialogflow. LUIS, interact with voice or text. Unlike Watson, the dataset preparation is
however requires Microsoft Azure subscription but enables complex, and the mapping between utterances and entities are
seamless integration with Azure Bot service. Voice commands as relatively critical.
input are supported in LUIS as well through the speech recognition
service. LUIS learns from its usage with active learning technology. NLU Services Core Features
Similar to Dialogflow, LUIS can be integrated with multiple
communication channels for user input such as Telegram, Twilio, A cloud-based NLU service such as the ones discussed in the
and Skype, among others. previous section provides some advantages over custom-built
software. Some of these are as follows: simple user interfaces to
3) Facebook Wit.ai create and modify conversation flow or dialog, easy way of
accessing data especially for non-technical users, privacy of user
Wit.ai is NLP API that assists developers in turning both speech and
data is protected and in a way with the service provider, the system
text into actionable data. It is developed primarily to help with
is scalable on demand and with growing needs, and the easy
voice-to-text input for Facebook Messenger. Wit.ai service
usability and reusability. The ease of use and reuse is related to the
improves Facebook’s understanding of voice’s semantic meaning.
fact that the implementation can be adaptable for other use-cases and
It has voice commands API. Wit.ai is free, even for commercial use.
can be redeployed for future needs. In this section, we adapt the
Wit.ai builds the behavior of chatbot using stories. Wit.ai learns
taxonomy discussed in [17] for evaluating the cloud-based NLU
from the examples related to real life scenarios that constitute the
applications and later integrating with our healthcare chatbot
stories. Wit.ai processes the similar examples as user input, extracts
enabling faster prototyping without sacrificing the chatbot’s intent
entities and applies logic from already learnt examples [16]. Each
recognition performance. The core features on which the discussed
story has many user intents. The flow of conversation is controlled
NLU applications are compared are as follows:
using branches and represented by these intents, which can be
expressed in graph format. Developers can define and use the
1. Usability: Relates to the ease of use of the interface provided
predefined entities. Similar to Dialogflow, Wit.ai facilitates
by the NLU application. High indicates simple and intuitive
webhook integration.
interface. Medium corresponds to relative difficulty in usage.
High relates to difficulty in usage without documentation or
4) IBM Watson Conversation user guide.
IBM Watson NLU is a rule-based AI native cloud product and a 2. Pre-build Intents: The number of pre-build or inbuilt keyword
powerful API. It uses deep learning to extract metadata such as groups that the NLU platform offers.
keywords, categories, sentiment, emotion, syntax, relations, and
entities from text. Watson NLU can be deployed on any cloud or 3. Prebuild-Entities: The number of pre-build entities that the
behind the firewall with the text enrichments available out-of-the- NLU platform provides.
box. It has the support for more than 13 languages to analyze our 4. Default Fallback Intent: The fallback mechanism is required
unstructured data. Domain customization is possible and Watson if the recognized intent is not present as existing pre-build or
can be trained to understand our use-case domain (healthcare-based defined intent. This feature indicates whether the NLU
chatbots or bots for technical sales department, among others) and

IS&T International Symposium on Electronic Imaging 2021

Mobile Devices and Multimedia: Enabling Technologies, Algorithms, and Applications 2021 035-3
 platform has the fallback mechanism for intents. Without this We created four intents on both the Dialogflow and Watson NLU
mechanism, every user utterance will correspond to a defined services: DiseaseIntensity, Span, TobaccoCraving, and
intent. TobaccoCraving-medicine. Further, we also enabled default
fallback intent in these two applications. Fig. 3 shows the list of
5. Automatic Context: Indicates whether the platform has the intents in the SymptomAssistance chatbot. For each of the NLU
functionality to manage the context of conversation services, we trained with 10-15 sentences in each intent.
automatically or developers have to explicitly define it at their
end of implementation
6. Modality: It is the conversation flow composition mode
adopted by the platform such as form-based or block-based.
7. SDK/ Webhook Availability: Indicates the existence of toolkits
associated with the NLU platform that allows developers to
integrate their chatbots with the platform or external software.
8. Channel Integration: Lists the third-party integrations
available with the service.
Thus, considering the functionalities and ease of integration from
the aforementioned NLU services discussion, in the current study,
we compare our case study chatbot with the intent classification
algorithms from Google Dialogflow and IBM Watson
Conversation. We focus on English language-based chatbot.

Experiments and Results

For the performance evaluation of the NLU applications, we chose Fig. 3. Intents in the NLU services
Google Dialogflow NLU and IBM Watson Conversation, both of
which have speech and text inputs and relevant post processing Intents are the verbs corresponding to the activities that the user
before passing to NLU for recognition of user utterances. Based on needs to do. Since we want to capture a request and perform an
our case study, which is discussed next, we chose these NLU action, intents need to be defined. Entities are the nouns or the
services primarily because of the availability of webhook/ SDK, content for the action that needs to be performed. Fig. 4 shows an
high usability and automatic context management. In this study, example NLU slot filling and intent parsing tree. Additionally, fig.
through the evaluation, we are interested in knowing how accurately 5 shows an annotation example of entities for a sample sentence in
the NLU application identifies the intent for a particular user the current use-case. For every training sentence entered for intent,
utterance/ sentence and the entities in the sentences. Furthermore, specific entities and their requirement is highlighted in the NLU
the identified intent is the keyword for the case study chatbot that in service. We also utilized pre-build intents and entities for initial
turn triggers relevant action and response to the user. conversation initiation where users enter their phone numbers.

Case Study: SymptomAssistance Chatbot

This section presents our case study healthcare chatbot for smoking
related Symptom Assistance for the evaluation of performance of
the cloud-based NLU services (Google Dialogflow and IBM
Watson Conversation) integration. The original version of the
chatbot has SMS and Facebook Messenger integration for user input
but rule-based predefined set of responses for user’s input. Users of
the chatbot were typically aware of the sequence of steps to follow.
However, with the increasing complexity and functionality
extension of this interactive messaging along with users repeated
entry of free text, the longer conversations have been becoming Fig. 4. NLU slot filling and Intent parsing tree
tedious to handle at the backend. Thus, the necessity of integrating
NLP applications arose. The readers are directed to [18] [19] for
background and details about the smoking cessation chatbot. In the
current study, we adapted a small part of the protocol from [18] [19]
interventions in order to show the feasibility of off-the-shelf NLU
services integration, and assess the performance of two of these
services that closely suit our development needs in humanizing the
chatbot for more engaging conversation with participants.

Data Description
Fig. 5. Intents and Entities Annotation

IS&T International Symposium on Electronic Imaging 2021

035-4 Mobile Devices and Multimedia: Enabling Technologies, Algorithms, and Applications 2021
 Table 1: Classified Intents with their confidence scores

Results Test User Google Dialogflow IBM Watson

Utterances
The NLU applications mostly differ in handling automatic context, I am having DiseaseIntensity – 1.0 DiseaseIntensity –
default fallback intent, pre-defined intents and entities, and the headache but I 1.0
availability of SDK/ webhook. The details of core features on which never smoke.
we analyzed the five NLU applications in the current study are as I am suffering DiseaseIntensity - 1.0, DiseaseIntensity -
follows: from chest pain Span - 0.99, 1.0, Span - 0.85,
1. Google Dialogflow has high usability. It has 34 and 60 pre- for the past 2 TobaccoCraving -1.0 TobaccoCraving -
weeks. I have 0.99
build intents and entities respectively ranging from addresses
been smoking for
to colors. It has a default fallback intent and can handle last 10 years.
automatic context. It employs form-based conversation I am looking for DiseaseIntensity - DiseaseIntensity -
composition mode. The online integration is possible with 14 a treatment for 1.0,TobaccoCraving- 1.0
channels ranging from Telegram to Alexa. Webhook and SDK heart burn. medicine - 0.92
are available for faster prototyping and integration with What causes DiseaseIntensity - DiseaseIntensity -
external software applications. chest pain? 0.99 0.88
2. Microsoft LUIS has medium usability. It has 20 and 13 pre-
build intents and entities respectively ranging from numbers to Overall, the two NLU services perform decently well for the test
geography. It has a default fallback intent but cannot handle user utterances. For the first sentence, both the applications detected
automatic context. It employs form-based conversation the correct intent successfully with highest accuracy of 1.0. For
composition mode. Webhook and SDK are available for faster complex sentences, such as second, third and fourth in the table,
prototyping and integration with external software Dialogflow detects all the associated intents more accurately as
applications. From SDK v4, it is possible to integrate the bot compared to Watson. The creation of composite entities is found to
created with Azure Bot Service to standard channels such as be more accurate in Dialogflow intention detection algorithm.
Alexa, Facebook messenger, and Slack, among others.
3. Facebook Wit.ai has high usability. It has 41 and 22 pre-build Conclusion and Future work
intents and entities respectively ranging from location to email.
It does not have a default fallback intent and cannot handle In this paper, we compared the major cloud-based general-purpose
automatic context. It employs form-based conversation NLU applications based on the eight core features for faster
composition mode. The online integration is possible with only prototyping of natural language chatbots. Based on these features,
FB messenger currently. Webhook and SDK are available for we found Google Dialogflow and IBM Watson Conversation
faster prototyping and integration. It has 5 pre-build traits platforms best suitable for integration with our case study symptom
corresponding to sentiments, greetings, etc. assistance chatbot for smoking cessation. Google Dialogflow
4. Watson has high usability. It has lately added a lot of pre-build performs better in recognizing all the associated intents for complex
intents that focus on customer care, industry, etc. and has 7 sentences. The discussed NLU applications with extensions such as
entities respectively ranging from time to person. It has a webhook and SDKs enable us in achieving required functionality.
default fallback intent and can handle automatic context. It In future work, we intend to extend our chatbots for further custom
employs both form- and block-based conversation composition and tailored functionality with deep neural network-based language
modes. The online integration is possible with channels models such as LSTM and Transformers.
ranging from Slack to Facebook Messenger. Webhook and
SDK are available for faster prototyping and integration with
References
external software applications.
[1] Artificial Intelligence. “Artificial Intelligence. (2020). [Online].
5. Amazon Lex has a relatively low usability. It has 15 and 93
Available:
pre-build intents and entities respectively through Alexa. It has https://en.wikipedia.org/wiki/History_of_artificial_intelligence
a fallback intent and can handle automatic context. It employs [Accessed Nov. 25, 2020].
form-based conversation composition mode. The online
integration is possible with channels such as Facebook [2] AlphaGo. “AlphaGo”. (2020). [Online]. Available:
Messenger, slack, and Twilio SMS. Webhook and SDK are https://en.wikipedia.org/wiki/AlphaGo_versus_Lee_Sedol [Accessed
available for faster prototyping and integration with external Dec. 1, 2020].
software applications. [3] Sameera A Abdul-Kader and JC Woods, “Survey on chatbot design
techniques in speech conversation systems,” International Journal of
Additionally, we also evaluated Google Dialogflow and IBM Advanced Computer Science and Applications, 2015, vol. 6, no. 7, pp.
Watson Conversation NLU applications with four test user 1–6.
utterances and calculated the confidence levels as a metric for
accuracy of right intent detection by the intent classification [4] Ramesh, K., Ravishankaran, S., Joshi, A., Chandrasekaran, K, “A
algorithms in the discussed NLU applications. Table 1 shows the survey of design techniques for conversational agents,” in Proc. Of
results of evaluation on the four composite sentences with the ICICCT 2017. CCIS, vol. 750, pp. 336–350. Springer, Singapore
recognized intent and confidence levels for the two NLU (2017). doi: https://doi.org/10.1007/978-981-10-6544-6_31.
applications.

IS&T International Symposium on Electronic Imaging 2021

Mobile Devices and Multimedia: Enabling Technologies, Algorithms, and Applications 2021 035-5
[5] Google Dialogflow. “Google Dialogflow”. (2020). [Online]. Author Biography
Available: https://cloud.google.com/dialogflow [Accessed June 15, Devasena Inupakutika is a Ph.D. candidate in the Department of Electrical
2020]. Engineering at the University of Texas at San Antonio (UTSA). She is
[6] IBM Watson. “IBM Watson”. (2020). [Online]. Available: pursuing her studies in the development and analysis of systems and methods
https://www.ibm.com/watson [Accessed Sept. 5, 2020]. for enhancing mobility. Her research interests include web and mobile
application development, cloud-IoT integration and deep learning based
[7] Chen, H., et al., “A Survey on Dialogue Systems: Recent Advances and wireless LAN indoor positioning systems. For correspondence. Email:
New Frontiers,” ACM SIGKDD Explorations Newsletter, 2017. 19(2): [email protected].
p. 25-35.
Mohammad Nadim is currently a Ph.D. student at the University of Texas at
[8] M. Nuruzzaman and O. K. Hussain, "A Survey on Chatbot San Antonio (UTSA). He received his B.Sc. degree from Bangladesh
Implementation in Customer Service Industry through Deep Neural University of Engineering and Technology (BUET) in 2015. His current
Networks," in Proc. of 2018 IEEE 15th International Conference on e- research includes human-machine interaction systems, mobile health
Business Engineering (ICEBE), doi: 10.1109/ICEBE.2018.00019. applications, and computer system security.
[9] E. Adamopoulou, L. Moussiades, "An Overview of Chatbot Ganesh Reddy Gunnam is a Ph. D. candidate in the Department of Electrical
Technology," in Proc. of 2020 AIAI 2020, IFIP AICT 584, pp. 373– Engineering at the University of Texas at San Antonio (UTSA). He is
383, 2020. doi: https://doi.org/10.1007/978-3-030-49186-4_31. pursuing his studies in the Smart Deep-logic Chatbot Design, Development
and Performance Assessment Methodology. His research interests include
[10] Rasa: Open source conversational AI. (2020). [Online]. Available:
Chatbot cloud deployment, virtualization and cloud computing.
https://rasa.com/ [Accessed July 5, 2020].
Sahak Kaghyan is a Postdoctoral Research Scientist at the University of
[11] S. Kaghyan, S. Sarpal, A. Zorilescu, D. Akopian, The University of
Texas at San Antonio (UTSA). He received his Ph.D. in Computer Science
Texas at San Antonio, "Review of Interactive Communication Systems
from Russian- Armenian University in 2014. His research interests include
for Business-to-Business (B2B) Services," in Proc. of IS&T
Full Stack Web Development, Mobile application development,
International Symposium on Electronic Imaging, 2018. doi:
Conversational AI design and development, Machine Learning and Software
https://doi.org/10.2352/ISSN.2470-1173.2018.06.MOBMU-117.
Engineering.
[12] Wit.ai. “Facebook Wit.ai”. (2020). [Online]. Available: https://wit.ai/
David Akopian is a Professor and Associate Dean of Research at the
[Accessed May 15, 2020].
University of Texas at San Antonio (UTSA). Prior to joining UTSA, he was
[13] LUIS (Language Understanding) – Cognitive Services – Microsoft a specialist with Nokia from 1999 to 2003. From 1993 to 1999, he was a staff
Azure. “Microsoft LUIS”. (2020). [Online]. Available: member at the Tampere University of Technology, Finland, where he
https://www.luis.ai/ [Accessed June 13, 2020]. received his Ph.D. degree in 1997. His current research interests include
signal processing algorithms for communication and navigation receivers,
[14] Amazon Lex – Build Conversation Bots. “Amazon Lex”. (2020). and implementation platforms for software defined radio, and mHealth.
[Online]. Available: https://aws.amazon.com/lex/ [Accessed Nov. 10,
2020]. Patricia Chalela is an associate professor at the Institute for Health
Promotion Research at UT Health San Antonio and the associate director
[15] SAP Conversational AI | Automate Customer Service With AI for Education and Training programs. She received her BS in community
Chatbots. “SAP Conversational AI”. (2021). [Online]. Available: health in 1981 from Universidad Industrial de Santander, MPH in Health
https://www.sap.com/products/conversational-ai.html [Accessed Jan Promotion/ Health Communications in 1995 from UT Health San Antonio
16, 2021]. Houston, TX and DPH in Health Promotion in 2006 from UT Health Science
[16] Rahman, A., Al Mamun, A., Islam, A. “Programming challenges of Center at Houston, TX. Her research interests are in chronic disease
chatbot: current and future prospective,” in Proc. of IEEE Region 10 prevention and control, particularly the role of epidemiological,
Humanitarian Technology Conference (R10-HTC), pp. 75–78. IEEE environmental and individual psychosocial factors on health and disease,
(2017). and on racial/ethnic disparities with emphasis on Latino populations.

[17] Canonico, M. and De Russis, L., “A comparison and critique of natural Amelie G. Ramirez is a chair of the Department of Population Health
language understanding tools,” in Proc. of Cloud Computing, 2018, Sciences and a director and professor of the Salud America at the Institute
p.120. for Health Promotion Research at UT Health San Antonio. She received her
BS in Psychology in 1973 from the University of Houston, TX, MPH in
[18] Chalela P, McAlister AL, Muñoz E, Despres C, Akopian D, Kaghyan Health Services Administration in 1977 and DPH in Health Promotion in
S, Fernandez A, Sukumaran P, Ramirez AG, “Reaching Latinos 1992 from the UT Health Science Center Houston, TX. She is an
through social media and SMS for smoking cessation,” in Advancing internationally recognized cancer health disparities researcher, has spent
the Science of Cancer in Latinos, 2020 (pp. 187-196). Springer, Cham. 30 years directing research on human and organizational communication to
reduce chronic disease and cancer health disparities affecting Latinos,
[19] Chalela P, McAlister AL, Munoz E, Despres C, Sukumaran P,
including cancer risk factors, clinical trial recruitment, tobacco prevention,
Kaghyan S, Akopian D, Ramirez AG, “Development of a bilingual
obesity prevention, healthy lifestyles, and more. She also trains/mentors
Facebook Messenger chat to promote smoking cessation among young
Latinos in behavioral sciences and is on the board of directors for
adults,” in ANNALS OF BEHAVIORAL MEDICINE, 2020 May 1 (Vol.
LIVESTRONG, Susan G. Komen for the Cure, and others. She is a member
54, pp. S52-S52). JOURNALS DEPT, 2001 EVANS RD, CARY, NC
of the Institute of Medicine (IOM) of the National Academies.
27513 USA: OXFORD UNIV PRESS INC.

IS&T International Symposium on Electronic Imaging 2021

035-6 Mobile Devices and Multimedia: Enabling Technologies, Algorithms, and Applications 2021
 JOIN US AT THE NEXT EI!

IS&T International Symposium on

Electronic Imaging
SCIENCE AND TECHNOLOGY

Imaging across applications . . . Where industry and academia meet!

• SHORT COURSES • EXHIBITS • DEMONSTRATION SESSION • PLENARY TALKS •

• INTERACTIVE PAPER SESSION • SPECIAL EVENTS • TECHNICAL SESSIONS •

www.electronicimaging.org
imaging.org