Keyword Fusion Optimization: A Simple and Efficient Query Rewriting Framework

User queries in conversational systems often rely on historical conversational context for interpretation. Conversational query rewriting transform user queries into independent query statements, removing contextual dependencies to clarify the user's true intent. However, existing approaches face challenges such as high model fine-tuning costs, difficulty adapting to various scenarios, and susceptibility to information forgetting in long or complex conversations. To address these issues, we present a simple yet effective conversational search framework, KwRewriter, which utilizes keyword-assisted query rewriting to emphasize the core elements of user intent and mitigate information forgetting. KwRewriter comprises two components: a query rewriting module that combines keywords with query rewriting, and a passage keyword extractor that incorporates passage keyword information. Experimental results on two large question-answering datasets, QReCC and TopioCQA, demonstrate that KwRewriter significantly improves performance with minimal additional cost.