Synopsis

On

AI POWERED FAKE NEWS DETECTOR

AND
SUMMARIZER
Submitted in partial fulfillment of the requirement
For the award of the degree of

Bachelor Of Technology
In

Computer Science & Engineering

Submitted By
ANMOL KUMAR 2200680100059
ASIT UPADHYAY 2200680100086
MUSKAN JAIN 2200680100217
YASH GUPTA 2200680100379
Under the guidance of
Dr. Satendra Kumar

Department of Computer Science & Engineering

Meerut Institute of Engineering & Technology, Meerut
 TABLE OF CONTENTS

TOPIC PAGE NO.

1. Introduction 3
2. Literature Review/Survey 4
3. Identification of Research Problem/Objectives 5
4. Expected Impact on Academics/Industry 6
5. Research Methodology 7
6. References 9
 1. Introduction

The unchecked spread of fake news across digital platforms poses a significant threat to public
awareness, safety, and trust in media. With the increasing use of social media, fake news travels
faster than ever before, often leading to confusion, misinformation, and in some cases, real-
world harm. Manual fact-checking is slow, limited, and insufficient in handling the volume
and speed of today’s information flow.

Limitations of traditional models: Conventional fake news detection systems rely on pre-
trained machine learning models that must be retrained periodically to stay relevant. This
approach is not only time-consuming and resource-intensive but also fails to respond to newly
emerging news in real time. Even if models are retrained at fixed intervals, there remains a
significant lag in detecting fresh misinformation.

Our solution: This project introduces a dynamic, AI-powered fake news detection system that
eliminates the need for traditional model training. Instead of relying on static datasets, it gathers
real-time, factual content from trusted sources on the internet using the Tavily API. The system
then leverages a Large Language Model (LLM) like OpenAI’s GPT-4 to compare the user’s
news statement against the gathered evidence. Based on this intelligent evaluation, the system
returns a final decision.

The output includes a clear verdict (Fake or Valid), a natural language explanation, a 60-
word summary, and a list of reliable source links used in the verification process. This
enables users not only to receive accurate results but also to verify the reasoning and references
behind them.

The system uses a modular, agent-based architecture powered by UAgent, where intelligent
agents manage input handling, query validation, and response generation. The frontend is built
in React, providing an intuitive interface, while the backend is entirely agent-driven, ensuring
scalability, modularity, and ease of deployment. The complete system is deployed using
AgentVerse, a cloud platform designed to host and run autonomous agents in real-time.
2. Literature Review/Survey

1. Shu et al. (2022)

Introduced a data mining approach for fake news detection using linguistic patterns
and user engagement behaviour. Models like RoBERTa achieved significant accuracy
in identifying misinformation based on textual features.
Result: Language and context-aware patterns are crucial for reliable detection.

2. Ahmed et al. (2021)

Developed a hybrid classifier to support multilingual fake news detection. It
combined content-based and metadata-based classifiers to improve accuracy across
different languages.
Result: Highlighted the importance of language flexibility and regional context in
fake news identification.

3. Chen et al. (2023)

Demonstrated the use of large language models such as GPT for interpretability and
reasoning in classification systems. It used prompt-based queries to extract
explanation-style results from LLMs.
Result: GPT models can enhance user trust by justifying their predictions in natural
language.

4. Zhou et al. (2020)

Proposed a hierarchical attention network to detect fake news by capturing both
sentence-level and document-level semantics. The model achieved strong
performance on multi-class fake news datasets by leveraging contextual embeddings.
Result: Hierarchical language understanding improves deep classification
performance for complex narratives.

5. Monti et al. (2019)

Introduced a graph-based approach where news articles, users, and user comments
were modeled as nodes in a relational graph. GCN (Graph Convolutional Networks)
were applied to classify news authenticity based on propagation patterns.
Result: Social engagement structure and user interaction data significantly enhance
fake news detection.

6. Kaliyar et al. (2021)

Developed a fake news detection model named FNDNet using deep convolutional
networks trained solely on textual features. The model was designed for low-latency
predictions on online news content.
Result: Lightweight CNNs can deliver fast and reliable detection performance with
minimal computational resources.
3. Identification of Research Problem/Objectives

Identification of Research Problem

With increasing reliance on online platforms for news and updates, misinformation has
become a prominent issue. Fake news spreads rapidly and has the potential to mislead
the public, damage reputations, and incite social unrest. Traditional manual verification
systems are not scalable and often lack transparency.
Furthermore, most existing fake news detection models are either limited to
classification only or rely on complex backend architectures that are difficult to scale
or modularize. These systems also rarely explain their verdict or summarize the content
for better user understanding.

This project seeks to address these challenges by building a system that:

• Detects fake or valid news statements automatically

• Justifies its decision through human-readable reasoning
• Generates a short, informative summary
• Uses a fully agent-based architecture for modularity and scalability
• Runs entirely through UAgent agents

Objectives of the Study

• To understand agent-based AI architecture using the UAgent framework.

• To build a fake news detection model capable of returning verdict, reasoning, and
summary.
• To design an intuitive web interface using React JS.
• To integrate modular agents such as REST agents and logic agents.
• To deploy the entire system using Agent Verse for accessibility.
• To explore future support for multilingual inputs and voice commands.
 4. Expected Impact on Academics/Industry

Academic Impact

This project provides a deep insight into modern approaches to misinformation detection using
agent-based systems. It highlights how AI can be made modular and explainable through
intelligent agents.

The academic significance includes:

• Demonstrates the use of UAgent-based intelligent agent architecture.

• Offers hands-on application of Natural Language Processing (NLP) techniques.
• Introduces Explainable AI (XAI) through reasoning-based outputs.
• Encourages modular and scalable design thinking in AI systems.
• Enhances understanding of agent communication and coordination.
• Reinforces frontend-backend integration skills using modern frameworks.
• Serves as a real-world example for misinformation detection research.
• Promotes interdisciplinary learning across AI, UX, and social computing.
• Demonstrates integration of real-time web search and LLM validation in agent-based
systems.

This project can also be extended into advanced academic research areas such as deep
learning-based misinformation detection, context-aware content analysis, and
deployment of intelligent agents in mobile or low-resource environments for real-time fake
news filtering.

Industry Impact

From an industry perspective, this project contributes to the growing demand for content
moderation tools. Media houses, government fact-checkers, and tech platforms can benefit
from:

• Automated content verification at scale

• Transparent decisions with explainable reasoning
• Modular deployment without maintaining full-scale server backends
• Easy extensibility into real-time systems like messaging apps and browser plugins

By eliminating the dependency on heavy infrastructure, this agent-based solution opens doors
to lightweight deployment, especially in resource-constrained environments.
 5. Research Methodology

1. Frontend Design (React JS)

A clean and responsive web interface is developed using React JS. It includes components for
user input, a “Check News” button, and sections to display the verdict (Fake/Valid), reasoning,
and a 60-word summary. The UI is designed to be user-friendly and accessible.

2. Input Handling via REST Agent

A REST Agent is implemented using the UAgent framework. It acts as the communication
bridge between the frontend and internal logic. It receives the user’s input from the interface
and passes it to the processing agent for analysis.

3. Fake News Detection Agent

The core of the project lies in the FakeNewsDetector agent, which manages all core processing
tasks. Upon receiving the user’s query, it first uses the Tavily API to perform a live web search
and fetch trusted, relevant content. Then, using the OpenAI LLM (GPT-4), it validates the
news statement against the retrieved information. Based on this comparison, the agent
generates:
• A classification verdict (Fake or Valid)
• A short reasoning explaining the decision
• A 60-word summary of the input
• A list of the sources used during verification

4. Agent-Based Architecture
The system is designed using UAgent’s modular, asynchronous agent-based architecture. Each
agent functions independently and communicates using defined protocols. This structure
ensures scalability and flexibility for future enhancements like multilingual support or plugin
integration.

5. Response Handling and Output Display

The output from the FakeNewsDetector agent (verdict, reasoning, summary) is passed back to
the REST agent, which forwards it to the frontend. The results are displayed in 4 structured
blocks on the frontend: verdict, reasoning, summary, and source URLs. The system also
presents relevant source URLs used during verification, allowing users to cross-check the
information with trusted references.

6. Tools & Technologies Used

• Frontend: React JS, JavaScript, HTML, CSS
• Backend/AI Logic: UAgent Framework, REST Agent, FakeNewsDetector Agent
• Communication: JSON over HTTP (REST architecture)
• Deployment Platform: AgentVerse for scalable agent hosting
• Version Control: Git and GitHub for team collaboration
• Tavily API – for contextual search
• OpenAI GPT-4 API – for LLM-based reasoning and summarization
• Testing – Postman for local API testing and debugging.
Agent-Based Processing Flow of the Proposed System
 6. References
[1]. Kai Shu, Amy Sliva, Suhang Wang, Jiliang Tang, and Huan Liu, “Fake News
Detection on Social Media: A Data Mining Perspective,” ACM SIGKDD Explorations
Newsletter, Vol. 19, Issue 1, 2017.
[Link]/10.1145/3137597.3137600
(Keywords: “Fake news, Social media, Text classification, Data mining, NLP,
Disinformation, Trustworthiness.”)

[2]. Hossain Ahmed, Mohammad Tahmid Hossain, and Md. Samiul Haque, “Detecting
Fake News Using Machine Learning Techniques,” Proceedings of the 2017 International
Conference on Innovations in Science, Engineering and Technology (ICISET),
Chittagong, Bangladesh, 2017.
[Link]/10.1109/ICISET.2017.8264682
(Keywords: “Fake news detection, Machine learning, News classification, Natural language
processing, TF-IDF, Naive Bayes.”)

[3]. Yizhong Chen, Jian Guan, Zihan Liu, and Minlie Huang, “Generating Reasonable
and Diverse Explanations for Fake News Detection,” 2020 Conference on Empirical
Methods in Natural Language Processing (EMNLP), Online, 2020.
[Link]/10.18653/v1/[Link]-main.584
(Keywords: “Explainable AI, Reason generation, Fake news, Neural networks, Natural
language understanding, Text generation.”)

[4]. UAgent Development Team, “UAgent Framework: Building Autonomous Intelligent

Agents,” Signals Dev, 2024.
[Link]/signals-dev/UAgent (accessed July 8, 2025)
(Keywords: “UAgent, Intelligent agents, Modular backend, REST agents, Agent
architecture, Python, AI frameworks.”)

[5]. OpenAI Communications Team, “GPT-4 API Documentation,” OpenAI Developer

Docs, 2024.
[Link]/docs (accessed July 8, 2025)
(Keywords: “GPT-4, Prompt engineering, Natural language processing, Reasoning,
Summarization, Language models.”)

[6]. ReactJS Core Team, “React: A JavaScript Library for Building User Interfaces,” Meta
Platforms Inc., 2024.
[Link] (accessed July 8, 2025)
(Keywords: “React JS, Component-based UI, JavaScript, Frontend development, Web
apps.”)

[7]. Tavily Technologies Documentation Team, “Tavily API for Web Search,” Tavily AI
Docs, 2024.
[Link] (accessed July 8, 2025)
(Keywords: “Tavily, Real-time web search, API integration, AI data retrieval, Search
indexing.”)
[8]. Y. Liu, M. Ott, N. Goyal, J. Du, M. Joshi, D. Chen, and V. Stoyanov, “RoBERTa: A
Robustly Optimized BERT Pretraining Approach,” Facebook AI Research (FAIR), 2019.
[Link]/abs/1907.11692
(Keywords: “RoBERTa, NLP, Pretraining, Transformers, Text classification.”)

[9]. Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova, “BERT: Pre-
training of Deep Bidirectional Transformers for Language Understanding,” Google AI
Language, 2019.
[Link]/abs/1810.04805
(Keywords: “BERT, Transformers, Text understanding, NLP, Transfer learning.”)

[10]. Hugging Face Team, “Transformers Library for NLP,” Hugging Face
Documentation, 2024.
[Link]/docs/transformers (accessed July 8, 2025)
(Keywords: “Transformers, BERT, RoBERTa, T5, NLP models, HuggingFace.”)

[11]. Railway Engineering Team, “Railway: Infrastructure for Developers,” Railway App
Docs, 2024.
[Link] (accessed July 8, 2025)
(Keywords: “Railway, Backend hosting, Environment deployment, Agent execution.”)

[12]. AgentVerse Development Team, “Deploying Agents using AgentVerse,”

[Link] Documentation, 2024.
[Link] (accessed July 8, 2025)
(Keywords: “AgentVerse, Agent hosting, Agent deployment, UAgent cloud integration.”)

[13]. Poetry Development Team, “Dependency Management with Poetry,” python-

[Link] Documentation, 2024.
[Link] (accessed July 8, 2025)
(Keywords: “Poetry, Virtual environments, Python dependencies, Package management.”)

[14]. Axios Contributors, “Axios: Promise-based HTTP Client for the Browser,” Axios
Documentation, 2024.
[Link] (accessed July 8, 2025)
(Keywords: “Axios, HTTP client, REST communication, API requests, JavaScript.”)

[15]. Git Special Interest Group, “Version Control with Git,” [Link] and GitHub
guides, 2024.
[Link]/[Link] (accessed July 8, 2025)
(Keywords: “Git, GitHub, Version control, Code collaboration, Repositories.”)

[16]. Microsoft Corporation, “Visual Studio Code Editor,” [Link]

Documentation, 2024.
[Link] (accessed July 8, 2025)
(Keywords: “VS Code, IDE, JavaScript editor, Python development, Extensions.”)

[17]. Natural Language Toolkit Development Team, “NLTK Python Library,” [Link]
Documentation, 2024.
[Link] (accessed July 8, 2025)
(Keywords: “Tokenization, Lemmatization, Stopword removal, Text preprocessing, NLP.”)