Graph Memory & Fact-Checking
Platform for Health Claims
TrumorGPT combines GraphRAG, vector similarity, and claims reasoning to extract query graphs, retrieve ground-truth medical knowledge, and deliver verifiable verdicts.
The 5 Paper Components
Designed according to IEEE TAI 2025 mathematical specifications for graph-augmented fact-checking.
1. Claim Normalization
Fuses Sentence-BERT semantic embeddings with LDA topic distributions using parameter η = 0.7 for optimal sentence centrality scoring.
2. Dual Storage Engine
Combines dense vector embeddings with explicit Knowledge Graph triples (h, r, t) stored in structured DBpedia/CDC schemas.
3. GraphRAG Traversal
Computes similarity score S(G_x, G_i) = max(Jaccard, Containment Ratio) over multi-hop relational path intersections.
4. Verification & Scoring
Leverages local Ollama (LLaMA 3.1:8B) with zero-shot triple extraction and strict relational contradiction checking.
5. Live Provenance & Citation Tracking
Delivers full factual transparency by mapping every verified claim directly back to its source reference knowledge graph (CDC, FDA, DBpedia guidelines).
REST API & Quickstart Integration
Trigger GraphRAG verification programmatically from any backend or CLI.
curl -X POST "http://127.0.0.1:8000/api/v1/fact-check" \
-H "Content-Type: application/json" \
-d '{"query": "mRNA vaccines reduce severe COVID-19 hospitalizations"}'