Tag
The paper finds that multi-hop RAG methods amplify ASR corruption in queries, reducing robustness in speech-based retrieval systems compared to naive methods.
Introduces CMT-RAG, a complementary memory framework for multi-turn multi-hop conversational RAG that aligns conversational memory with retrieval using sub-question-level reasoning traces. Also presents MuMu-QA, a benchmark with cross-turn sub-question dependencies.
DynaKRAG introduces a unified framework for multi-hop retrieval-augmented generation that learns a state-conditioned policy to select evidence operations, outperforming baselines on HotpotQA, 2Wiki, and MuSiQue.
Presents a training-free method for multi-hop retrieval-augmented generation that avoids costly graph rebuilds when underlying data changes, tackling the staleness issue in dynamic environments.
MOTHRAG is a multi-hop RAG system that matches the performance of top GPU-dependent systems (HippoRAG 2, CoRAG, NeocorRAG) using only commodity API calls, with no GPU, no fine-tuning, and deployment via pip install plus API keys.