Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

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

Fengran Mo, Abbas Ghaddar, Kelong Mao, Mehdi Rezagholizadeh, Boxing Chen, Qun Liu, Jian-Yun Nie

TL;DR

CHIQ, a two-step method that leverages the capabilities of LLMs to resolve ambiguities in the conversation history before query rewriting, is introduced, a two-step method that leverages the capabilities of LLMs to resolve ambiguities in the conversation history before query rewriting.

Abstract

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.

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

TL;DR

CHIQ, a two-step method that leverages the capabilities of LLMs to resolve ambiguities in the conversation history before query rewriting, is introduced, a two-step method that leverages the capabilities of LLMs to resolve ambiguities in the conversation history before query rewriting.

Abstract

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.
Paper Structure (38 sections, 3 equations, 1 figure, 10 tables)

This paper contains 38 sections, 3 equations, 1 figure, 10 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Methodology
  4. Task Formulation
  5. History Enhancement
  6. Question Disambiguation
  7. Response Expansion
  8. Pseudo Response
  9. Topic Switch
  10. History Summary
  11. Ad-hoc Query Rewriting
  12. Search-Oriented Fine-tuning
  13. Experimental Setup
  14. Datasets and Evaluation Metrics
  15. Baselines
  16. ...and 23 more sections

Figures (1)

  • Figure 1: An illustrative example of a conversational history (left box) and the gold positive passage relevant to the last user turn. The enhanced history obtained using our method described in § \ref{['sec:History Enhancement']} is in the middle box. The right box shows the three search queries generated by LLaMA-2-7B conditioned on the original history, and our CHIQ-FT and CHIQ-AD methods described in § \ref{['sec:QR Finetuning']} and § \ref{['sec:Ad-hoc QR']}, respectively. Underlined terms in the gold passages are those that appear in the query generated by our approaches, which is conditioned on the enhanced history and did not appear in the query generated by the method that uses the original history.