BICT Honors Degree 2025
A Hybrid RAG-Based Framework for Real-Time Verification of Health Claims on Social Media Using Knowledge Graphs and Vector Embeddings
Using Knowledge Graphs and Vector Embeddings to improve the accuracy, scalability, and transparency of fact verification on social media.
Aims and Objectives
Our main goal is to design and evaluate a hybrid RAG framework that improves accuracy, scalability, and transparency in fact-checking.
Our Methodology
Our hybrid RAG pipeline uses a multi-stage process to verify social media posts with high accuracy and transparency.
Intake
A post (text + optional links) is submitted and routed via WebSockets for processing.
Embedding & Categorization
The post is converted into a semantic embedding and classified into a domain (e.g., politics, health).
Retrieval & Ranking
A dual retrieval process from a Vector DB and Knowledge Graph, followed by a hybrid re-ranking of results.
Corrective Retrieval
An agentic loop creates focused searches on external sources for failed facets and re-ranks evidence.
Evidence Validation
User-submitted sources are validated for trust and entailment. An admin panel governs knowledge base growth.
RAG Generation
A RAG prompt is constructed with the post and evidence. The LLM outputs a truthfulness estimate and explanation.
Delivery & Feedback
Results are delivered back to the platform, and user feedback is collected for continuous improvement.
Meet Our Research Team
The driving force behind this research is a team of passionate and dedicated students and their supervisor.

Focus Area: Software Technology. University of Sri Jayewardenepura.
ICT/21/809

Focus Area: Software Technology. University of Sri Jayewardenepura.
ICT/21/813

Focus Area: Software Technology. University of Sri Jayewardenepura.
ICT/21/819

Lecturer, Faculty of Technology, University of Sri Jayewardenepura.
System Architecture
An overview of the system architecture for our social media oriented agentic RAG verification system.

