NeuroFact: Hybrid NLP and Knowledge Graph Model for Real-Time Fact Verification
Authors-Saranya, Dr. Swarna Surekha
Keyword-Fact Verification, Knowledge Graphs, Graph Neural Networks, Hybrid AI Systems, Misinfor-mation Detection, Real-Time Verification.
With the growing amount of misinformation on digital channels, there is a need for automated verification methods that can provide high accuracy while remaining explainable and computa-tionally efficient. In this work, we propose NeuroFact, a novel framework that combines NLP approaches with KG reasoning to perform real-time fact verification. Our proposed model con-sists of three interconnected components: (1) a neural engine for decomposing natural language claims into atomic subclaims, (2) a dual-path retriever that combines dense passage retrieval and knowledge graph-based search over DBpedia, and (3) a fact verification model using graph neural networks that captures multi-hop relations. Our experiments show that our method attains a 0.93 F1 score on FEVER dataset's Supported/Refuted classification task without any further fine-tuning. Our comparison with state-of-the-art models shows that NeuroFact provides a 7.1% increase over baselines' LLMs, with 78% evidence support and real-time inference less than 500 ms.
Doi-[https:/doi.org/10.5281/zenodo.20813118]