ChatRetriever: Adapting Large Language Models for Conversational Dense Retrieval
Source: arXiv:2404.13556 | Date: 2024-04-21 | Authors: Kelong Mao, Chenlong Deng, Haonan Chen, Fengran Mo, Zheng Liu, Tetsuya Sakai, et al.
Overview
Conversational search requires accurately interpreting user intent from complex multi-turn contexts. Traditional pipelines typically handle this by rewriting conversational queries into standalone queries before retrieval, which adds latency and can introduce errors.
ChatRetriever instead inherits the strong generalization capability of large language models to robustly represent complex conversational sessions directly for dense retrieval.
Method
The paper proposes a simple and effective dual-learning approach:
1. Contrastive learning — adapting the LLM for retrieval by training it to align conversational session representations with relevant documents. 2. Masked instruction tuning — enhancing the model's understanding of complex sessions through masked instruction tuning on high-quality conversational instruction tuning data.
Results
- Evaluated on five conversational search benchmarks.
- ChatRetriever substantially outperforms existing conversational dense retrievers.
- Achieves state-of-the-art performance on par with LLM-based rewriting approaches, without requiring a separate query rewriting stage.
- Shows superior robustness in handling diverse conversational contexts.
- ConvGQR: Generative Query Reformulation for Conversational Search
- CoSearchAgent: A Lightweight Collaborative Search Agent with Large Language Models
- Engineering Conversational Search Systems: A Review of Applications
- Agentic Conversational Search with Contextualized Reasoning
Significance
The work demonstrates that LLMs can be directly adapted to encode complex inputs such as conversational search sessions for retrieval, providing an effective and robust alternative to rewrite-then-retrieve pipelines.
Original Abstract
> Conversational search requires accurate interpretation of user intent from complex multi-turn contexts. This paper presents ChatRetriever, which inherits the strong generalization capability of large language models to robustly represent complex conversational sessions for dense retrieval. To achieve this, we propose a simple and effective dual-learning approach that adapts LLM for retrieval via contrastive learning while enhancing the complex session understanding through masked instruction tuning on high-quality conversational instruction tuning data. Extensive experiments on five conversational search benchmarks demonstrate that ChatRetriever substantially outperforms existing conversational dense retrievers, achieving state-of-the-art performance on par with LLM-based rewriting approaches. Furthermore, ChatRetriever exhibits superior robustness in handling diverse conversational contexts. Our work highlights the potential of adapting LLMs for retrieval with complex inputs like conversational search sessions and proposes an effective approach to advance this research direction.