Reproduce “CID-GraphRAG: Enhancing Multi-Turn Dialogue Systems through Dual-Pathway Retrieval of Conversation Flow and Context Semantics” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Ziqi Zhu, Tao Hu, Honglong Zhang, Dan Yang, Hangeng Chen, Mengran Zhang, Xilun Chen

CID-GraphRAG retrieves customer-service examples using both conversational intent transitions and semantic similarity, then supplies them to a language model. The authors test it on held-out vehicle-sticker conversations, where an LLM judge recorded 60 wins for responses, compared with 38 for Conversation RAG; five human experts rated it highest on average. These results do not establish transfer beyond this dataset.

Key insights

  • Intent information, even with a small weighting (10%), provides crucial structural guidance, effectively filtering out semantically similar but contextually inappropriate examples and reducing semantic drift.
  • The synergistic integration of intent transition structures and semantic retrieval outperforms individual purely semantic or purely intent-based approaches in multi-turn dialogues.
  • Human evaluators exhibited a stronger preference for intent-guided responses in aspects like style consistency and professionalism compared to LLM-as-a-Judge evaluations.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.