Dynamic Multi-Hop Graph Traversal for Deep Reasoning
Replace static one-hop graph retrieval with a dynamic beam search traversal agent that explores multi-hop paths based on semantic relevance to missing information.
Source paper
Zirui Guo, Lianghao Xia, Yanhua Yu, Tu Ao, Chao Huang
Generated by AI. The paper’s authors did not write these.
Replace static one-hop graph retrieval with a dynamic beam search traversal agent that explores multi-hop paths based on semantic relevance to missing information.
Distill graph-extraction capabilities from large models into a specialized small language model to enable local, low-cost, and low-latency document indexing.
Incorporate temporal timestamps and conflict detection into graph updates to prioritize newer information and deprecate contradicted relations during adaptation.
The paper introduces LightRAG, a Retrieval-Augmented Generation system designed to overcome the limitations of flat vector retrieval and the computational costs of existing graph-based methods like GraphRAG. LightRAG employs a dual-level retrieval paradigm that combines graph structures with vector representations to capture both specific entity details (low-level) and broader conceptual themes (high-level), while supporting efficient incremental updates without rebuilding the entire index.
LightRAG utilizes a Large Language Model (LLM) to parse raw text into a Knowledge Graph containing entities and relationships, generating key-value pairs for profiling. For retrieval, it extracts low-level and high-level keywords from a user query, performs vector matching against the graph's entities and relationships, and retrieves connected subgraphs (including one-hop neighbors) to formulate a comprehensive context for answer generation.