Researchers have introduced CoG (Cognition on Graph), a novel framework designed to enhance how large language models (LLMs) navigate and utilize extensive knowledge graphs and text corpora for complex reasoning tasks. Unlike existing methods that rely on reactive exploration, CoG employs a proactive, cognitive-inspired cycle of planning, exploring, and reflecting to adaptively search for information. This approach fosters a deep, bidirectional synergy between structured graph data and unstructured text, enabling entities identified in text to dynamically guide the graph exploration process and bridge knowledge gaps. Experiments on multiple question-answering benchmarks show that CoG significantly surpasses current state-of-the-art methods in both accuracy and exploration efficiency. AI
IMPACT This framework could significantly improve LLM performance on knowledge-intensive tasks by enabling more efficient and adaptive navigation of complex information landscapes.
RANK_REASON The item describes a new research paper detailing a novel framework for LLMs. [lever_c_demoted from research: ic=1 ai=1.0]
- arXiv
- Cognition on Graph
- Hugging Face
- knowledge graph
- large-language models
- QA
- retrieval-augmented generation
- text corpora
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
