A Dual-Stage Framework Integrating LLM Summarization and Knowledge Graph Embeddings for Health Misinformation Detection
Zhiteng Song, G. Y. Zhang, L Zhang, Tongxuan Zhang
Lecture notes in computer science
Problems Identified (5)
Health Misinformation Governance: Widespread health misinformation on social media creates a significant challenge for public health governance.
Long-Text Information Capture: Existing deep learning methods often fail to capture critical information in lengthy texts, reducing detection performance.
Insufficient Domain Knowledge Integration: Existing methods insufficiently integrate domain-specific medical knowledge, hindering identification of misleading entities and illogical relationships.
Health Misinformation Governance: Widespread health misinformation on social media creates a significant challenge for public health governance.
Long-Text Information Capture: Existing deep learning methods often fail to capture critical information in lengthy texts, reducing detection performance.
Proposed Solutions (4)
DS-SumKG Dual-Stage Framework: The paper proposes DS-SumKG, a dual-stage framework integrating LLM summarization with knowledge graph embeddings for health misinformation detection.
LLM Summaries with Health KG Embeddings: The framework combines LLM-generated summaries with health-related knowledge graph embeddings to reduce noise and improve semantic understanding of medical entities.
DS-SumKG Dual-Stage Framework: The paper proposes DS-SumKG, a dual-stage framework integrating LLM summarization with knowledge graph embeddings for health misinformation detection.
LLM Summaries with Health KG Embeddings: The framework combines LLM-generated summaries with health-related knowledge graph embeddings to reduce noise and improve semantic understanding of medical entities.
Results (3)
High F1 Health Misinformation Detection:
Outperforms State of the Art:
High F1 Health Misinformation Detection:
Research Domain
Health misinformation detection on social media