Abstract
Conversational search seeks to retrieve relevant passages for the givenquestions in conversational question answering. Conversational QueryReformulation (CQR) improves conversational search by refining the originalqueries into de-contextualized forms to resolve the issues in the originalqueries, such as omissions and coreferences. Previous CQR methods focus onimitating human written queries which may not always yield meaningful searchresults for the retriever. In this paper, we introduce GuideCQR, a frameworkthat refines queries for CQR by leveraging key infFormation from the initiallyretrieved documents. Specifically, GuideCQR extracts keywords and generatesexpected answers from the retrieved documents, then unifies them with thequeries after filtering to add useful information that enhances the searchprocess. Experimental results demonstrate that our proposed method achievesstate-of-the-art performance across multiple datasets, outperforming previousCQR methods. Additionally, we show that GuideCQR can get additional performancegains in conversational search using various types of queries, even for querieswritten by humans.