ConRAG: Consensus-Driven Multi-View Retrieval for Multi-Hop Question Answering

ConRAG, a new consensus-driven multi-view retrieval framework, enhances retrieval-augmented generation (RAG) for multi-hop question answering (QA) by optimizing both query and corpus sides and integrating multi-view evidence such as relation, entity, and text signals. Experimental results demonstrate that ConRAG significantly outperforms existing methods, achieving up to a 26.9% average performance increase over standard RAG and setting a new state-of-the-art with the Gemma-4-31B model on the MuSiQue benchmark. This advancement is crucial for practitioners as it addresses the limitations of current multi-hop QA approaches, enabling more accurate and effective retrieval in complex tasks.