Conversational Query Reformulation with the Guidance of Retrieved Documents

Given a multi-turn conversational context and a raw user query, the goal of Conversational Query Reformulation (CQR) is to transform the query into a de-contextualized form that maximizes retrieval effectiveness for a downstream passage retriever. Conversational search seeks to retrieve relevant passages for the given questions in a conversational question answering system. Conversational Query Reformulation (CQR) improves conversational search by refining the original queries into de-contextualized forms to address issues such as omissions and coreferences. Previous CQR methods focus on imitating human-written queries, which may not always yield meaningful search results for the retriever. In this paper, we introduce GuideCQR, a framework that refines queries for CQR by leveraging key information from the initially retrieved documents. Specifically, GuideCQR extracts keywords and generates expected answers from the retrieved documents, then unifies them with the queries after filtering to add useful information that enhances the search process. Experimental results demonstrate that our proposed method achieves state-of-the-art performance across multiple datasets, outperforming previous CQR methods. Specifically, GuideCQR achieves MRR gains of 5.4% over LLM4CS on CAsT-19 and NDCG@3 gains of 29.2% on QReCC, and consistently improves retrieval across CAsT-19, CAsT-20, and QReCC benchmarks, demonstrating strong adaptability to various query types including human-rewritten queries. Additionally, we show that GuideCQR can get additional performance gains in conversational search using various types of queries, even for queries written by humans.