ICR: Iterative Clarification and Rewriting for Conversational Search

2025-09-05

AI Summary

In conversational query rewriting, multi-turn interactions often contain multiple ambiguous expressions, making it difficult for end-to-end models to simultaneously identify and rewrite all ambiguous terms—thereby limiting retrieval performance. To address this, we propose the Iterative Clarification and Rewriting (ICR) framework, which decomposes the complex rewriting task into two alternating stages: clarification question generation and query rewriting, both implemented via prompt-driven generative models to enable dynamic feedback optimization. Our key contribution is a controllable interaction mechanism that alleviates modeling pressure induced by compound ambiguity. Evaluated on two benchmark datasets—QReCC and CANARD—ICR achieves state-of-the-art retrieval performance, substantially outperforming existing methods. Ablation studies confirm that the iterative process delivers consistent, incremental gains in effectiveness.

Technology Category

• Data Mining & Knowledge Management: Conversational Systems for Recommendation & Retrieval
• Natural Language Processing: Question Answering
• Search and Optimization: Learning to Search

Application Category

• Search and Retrieval-Augmented AI: Assisted, interactive, and conversational search
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Most previous work on Conversational Query Rewriting employs an end-to-end rewriting paradigm. However, this approach is hindered by the issue of multiple fuzzy expressions within the query, which complicates the simultaneous identification and rewriting of multiple positions. To address this issue, we propose a novel framework ICR (Iterative Clarification and Rewriting), an iterative rewriting scheme that pivots on clarification questions. Within this framework, the model alternates between generating clarification questions and rewritten queries. The experimental results show that our ICR can continuously improve retrieval performance in the clarification-rewriting iterative process, thereby achieving state-of-the-art performance on two popular datasets.

Problem

• Addresses multiple fuzzy expressions in conversational queries
• Improves retrieval through iterative clarification and rewriting
• Overcomes limitations of end-to-end query rewriting paradigms

Innovation

• Iterative clarification and rewriting framework
• Alternates question generation and query rewriting
• Improves retrieval through iterative clarification process