Dense Passage Retrieval in Conversational Search

TL;DR

This work investigates applying dense passage retrieval (DPR) to conversational search using the CAsT benchmark. It develops an end-to-end GPT2QR+DPR system and analyzes multiple query reformulation strategies to enhance retrieval with dual-encoder dense representations, requiring offline passage embeddings and runtime query embeddings. Across MSMARCO and CAsT data, dense retrieval demonstrates competitive performance with BM25 and gains further from effective query reformulation, even under limited fine-tuning. The study highlights the practicality and potential of neural-based retrieval for conversational information seeking and outlines directions for improving hard-negative training and reranking with additional models. Overall, the results indicate that dense retrieval can significantly improve retrieval accuracy in conversational search and offers a scalable, context-aware alternative to traditional term-based methods.

Abstract

Information retrieval systems have traditionally relied on exact term match methods such as BM25 for first-stage retrieval. However, recent advancements in neural network-based techniques have introduced a new method called dense retrieval. This approach uses a dual-encoder to create contextual embeddings that can be indexed and clustered efficiently at run-time, resulting in improved retrieval performance in Open-domain Question Answering systems. In this paper, we apply the dense retrieval technique to conversational search by conducting experiments on the CAsT benchmark dataset. We also propose an end-to-end conversational search system called GPT2QR+DPR, which incorporates various query reformulation strategies to improve retrieval accuracy. Our findings indicate that dense retrieval outperforms BM25 even without extensive fine-tuning. Our work contributes to the growing body of research on neural-based retrieval methods in conversational search, and highlights the potential of dense retrieval in improving retrieval accuracy in conversational search systems.

Figures (1)

• Figure 1: Complete model architecture of GPT2QR+DPR