ConvGQR: Generative Query Reformulation for Conversational Search
- Paper: https://arxiv.org/abs/2305.15645
- Authors / Affiliations: Fengran Mo, Kelong Mao, Yutao Zhu, Yihong Wu, Kaiyu Huang, Jian-Yun Nie
- Published: 2023-05-25
- Topics: Conversational Search, Multi-Turn Retrieval, Query Understanding
- Manual query rewrites are imperfect supervision targets; combining query rewriting with answer generation yields stronger reformulations.
- Coupling reformulation quality with retrieval objectives (knowledge infusion) improves end-to-end search performance.
- The framework avoids costly re-training of the underlying query encoder/retriever.
- A Survey of Conversational Search (Sep 2025, ACM)
- Engineering Conversational Search Systems: A Review of Applications (arXiv:2407.00997)
- ChatRetriever: Adapting Large Language Models for Generalized and Robust Retrieval (arXiv:2404.13556)
- CoSearchAgent: A Lightweight Collaborative Search Agent with Large Language Models (arXiv:2402.06360)
Background and Motivation
In conversational search, the user's real search intent for the current turn depends on the previous conversation history. Determining a good search query from the whole conversation context is challenging. To avoid the expensive re-training of the query encoder, most existing methods learn a rewriting model that de-contextualizes the current query by mimicking manual query rewriting. However, manually rewritten queries are not always the best search queries — training a rewriting model on them limits the model's ability to produce good queries. Another useful hint is the potential answer to the question.
Method
ConvGQR reformulates conversational queries based on generative pre-trained language models (PLMs) with two components:
1. Query rewriting module: generates a de-contextualized standalone query from the conversation history. 2. Answer generation module: generates a potential answer to the current question.
By combining the rewritten query with the generated potential answer, ConvGQR produces better search queries than rewriting alone. In addition, a knowledge infusion mechanism relates query reformulation to retrieval performance, jointly optimizing both query reformulation and retrieval.
Evaluation
Extensive experiments on four conversational search datasets demonstrate the effectiveness of ConvGQR. Precise metric values should be verified against the original paper's tables.
Key Takeaways
Original Abstract
> In conversational search, the user's real search intent for the current turn is dependent on the previous conversation history. It is challenging to determine a good search query from the whole conversation context. To avoid the expensive re-training of the query encoder, most existing methods try to learn a rewriting model to de-contextualize the current query by mimicking the manual query rewriting. However, manually rewritten queries are not always the best search queries. Training a rewriting model on them would limit the model's ability to produce good search queries. Another useful hint is the potential answer to the question. In this paper, we propose ConvGQR, a new framework to reformulate conversational queries based on generative pre-trained language models (PLMs), one for query rewriting and another for generating potential answers. By combining both, ConvGQR can produce better search queries. In addition, to relate query reformulation to retrieval performance, we propose a knowledge infusion mechanism to optimize both query reformulation and retrieval. Extensive experiments on four conversational search datasets demonstrate the effectiveness of ConvGQR.