Reproduce “CMT-RAG: Complementary Memory Traces for Multi-turn Multi-hop RAG” 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
Lang Zhou, Yingjian Chen, Shuxuan Li, Kun-Yu Lin, Zhilin Zhao
CMT-RAG represents conversational retrieval steps as reusable traces containing sub-questions, dependencies, keywords, answers, and supporting paragraphs. Its recurrent state and trace graph resolve follow-ups and combine earlier evidence with newly retrieved passages. On MuMu-QA’s long-dialogue development split, it reports 41.73 Exact Match and 55.63 token-level F1, with 13.99 mean unique context paragraphs; gold-evidence recall remains below two baselines.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.