CHIQ: Contextual History Enhancement for Improving Query Rewriting in Conversational Search

Overview

• Paper: CHIQ: Contextual History Enhancement for Improving Query Rewriting in Conversational Search
• Authors: Fengran Mo, Abbas Ghaddar, Kelong Mao, Mehdi Rezagholizadeh, Boxing Chen, Qun Liu, et al. (7 authors in total)
• Published: 2024-06-07
• arXiv: https://arxiv.org/abs/2406.05013
• Code: https://github.com/fengranMark/CHIQ (to be publicly released upon acceptance)
• Category: Multi-Turn / Conversational Search

Abstract (original)

> In this paper, we study how open-source large language models (LLMs) can be effectively deployed for improving query rewriting in conversational search, especially for ambiguous queries. We introduce CHIQ, a two-step method that leverages the capabilities of LLMs to resolve ambiguities in the conversation history before query rewriting. This approach contrasts with prior studies that predominantly use closed-source LLMs to directly generate search queries from conversation history. We demonstrate on five well-established benchmarks that CHIQ leads to state-of-the-art results across most settings, showing highly competitive performances with systems leveraging closed-source LLMs. Our study provides a first step towards leveraging open-source LLMs in conversational search, as a competitive alternative to the prevailing reliance on commercial LLMs. Data, models, and source code will be publicly available upon acceptance at https://github.com/fengranMark/CHIQ.

Key Points

• Problem: In conversational search, user queries are often ambiguous because they depend on conversation history. Prior approaches mostly use closed-source (commercial) LLMs to directly generate rewritten search queries from the raw history.
• Approach: CHIQ is a two-step pipeline that first uses an LLM to resolve ambiguities in the conversation history, then performs query rewriting with the disambiguated history. This explicitly separates ambiguity resolution from query generation.
• Open-source focus: The method demonstrates that open-source LLMs can be effectively deployed for this task, reducing reliance on commercial APIs.
• Results: Across five well-established conversational search benchmarks, CHIQ achieves state-of-the-art results in most settings and is highly competitive with systems that use closed-source LLMs.

Method Outline

The paper's technical route, as described in its abstract, follows a two-step design:

1. Contextual history enhancement: An LLM analyzes the conversation history to resolve coreference and ambiguity, producing a clearer, self-contained context. 2. Query rewriting: The enhanced history is used to generate a standalone search query suitable for retrieval.

Related Work Cross-References

• A Survey on Multi-Turn Interaction Capabilities of Large Language Models (arXiv:2501.09959)
• Evaluating LLM-based Agents for Multi-Turn Conversations: A Survey (arXiv:2503.22458)
• Aligning Query Representation with Rewritten Query and Relevance Judgments (ACM DOI: 10.1145/3627673.3679534)
• An Empirical Analysis on Multi-Turn Conversational Recommender Systems (ACM DOI: 10.1145/3626772.3657893)
• Beyond Whole Dialogue Modeling: Contextual Disentanglement for Conversational Search (arXiv:2504.17427)
• Few-Shot Conversational Dense Retrieval, SIGIR 2021 (ACM DOI: 10.1145/3404835.3462856)

Takeaways for Practitioners

1. Researchers: Reproduce the benchmark comparisons; note whether statistical significance and compute costs are reported. 2. Engineers: The ambiguity-resolution step is a pluggable module that can be added to existing conversational retrieval stacks without switching to closed-source LLMs. 3. Product teams: Open-source LLM-based query rewriting can reduce API costs while retaining competitive retrieval quality across multiple turns of dialogue.

> Note: Specific quantitative results should be verified against the original PDF before citation; this post is based on the abstract and public metadata.