Concept-enhanced heterogeneous graph network for fact verification.

Area of Science:

• Natural Language Processing
• Artificial Intelligence
• Information Retrieval

Background:

• Fact verification is a complex NLP task requiring evidence retrieval from reliable sources.
• Existing methods often overlook multi-granularity information and lack inherent concept understanding.
• Challenges include aggregating scattered textual clues and leveraging entity concepts for accuracy.

Purpose of the Study:

• To propose a novel Concept-Enhanced Heterogeneous Graph Network (Concept-HGN) for improved fact verification.
• To address limitations in current fact verification models concerning multi-granularity and concept information.
• To enhance the accuracy and robustness of automated fact-checking systems.

Main Methods:

• Constructing a heterogeneous graph to aggregate information from multiple evidence sentences.
• Implementing hierarchical node granularity within the graph for effective clue aggregation.
• Leveraging intrinsic entity concepts from YAGO to guide the fact verification process.

Main Results:

• Concept-HGN achieved 80.26% (LA) and 77.68% (FS) on the FEVER dataset.
• On the UKP Snopes dataset, accuracy reached 65.7% and macro F1 reached 61.9%.
• The proposed model demonstrated superior performance compared to baseline models.

Conclusions:

• Concept-HGN effectively integrates multi-granularity and concept information for fact verification.
• The model achieves state-of-the-art performance, outperforming existing methods.
• This approach offers a promising direction for advancing automated fact-checking technologies.