Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Neural Approaches to Conversational Information Retrieval

Jianfeng Gao, Chenyan Xiong, Paul Bennett, Nick Craswell

TL;DR

This work surveys neural approaches to conversational information retrieval (CIR), framing CIR as a task-oriented dialogue problem that combines contextual understanding, multi-modal retrieval, and natural language generation. It details architectures, benchmarks, and learning paradigms, including dense and sparse retrieval, QMDS, CMC, and C-KBQA, along with evaluation strategies and proactive conversational techniques. The contributions include a comprehensive taxonomy of CIR components, neural models for end-to-end CIR pipelines, and case studies of commercial systems, highlighting how modern PLMs and neural readers enable human-centric, interactive information access. The findings illustrate the centrality of contextual understanding, task-oriented dialog management, and grounded generation in achieving effective, scalable CIR across open-domain and knowledge-base settings, with strong emphasis on evaluation, safety, and continual learning.

Abstract

A conversational information retrieval (CIR) system is an information retrieval (IR) system with a conversational interface which allows users to interact with the system to seek information via multi-turn conversations of natural language, in spoken or written form. Recent progress in deep learning has brought tremendous improvements in natural language processing (NLP) and conversational AI, leading to a plethora of commercial conversational services that allow naturally spoken and typed interaction, increasing the need for more human-centric interactions in IR. As a result, we have witnessed a resurgent interest in developing modern CIR systems in both research communities and industry. This book surveys recent advances in CIR, focusing on neural approaches that have been developed in the last few years. This book is based on the authors' tutorial at SIGIR'2020 (Gao et al., 2020b), with IR and NLP communities as the primary target audience. However, audiences with other background, such as machine learning and human-computer interaction, will also find it an accessible introduction to CIR. We hope that this book will prove a valuable resource for students, researchers, and software developers. This manuscript is a working draft. Comments are welcome.

Neural Approaches to Conversational Information Retrieval

TL;DR

This work surveys neural approaches to conversational information retrieval (CIR), framing CIR as a task-oriented dialogue problem that combines contextual understanding, multi-modal retrieval, and natural language generation. It details architectures, benchmarks, and learning paradigms, including dense and sparse retrieval, QMDS, CMC, and C-KBQA, along with evaluation strategies and proactive conversational techniques. The contributions include a comprehensive taxonomy of CIR components, neural models for end-to-end CIR pipelines, and case studies of commercial systems, highlighting how modern PLMs and neural readers enable human-centric, interactive information access. The findings illustrate the centrality of contextual understanding, task-oriented dialog management, and grounded generation in achieving effective, scalable CIR across open-domain and knowledge-base settings, with strong emphasis on evaluation, safety, and continual learning.

Abstract

A conversational information retrieval (CIR) system is an information retrieval (IR) system with a conversational interface which allows users to interact with the system to seek information via multi-turn conversations of natural language, in spoken or written form. Recent progress in deep learning has brought tremendous improvements in natural language processing (NLP) and conversational AI, leading to a plethora of commercial conversational services that allow naturally spoken and typed interaction, increasing the need for more human-centric interactions in IR. As a result, we have witnessed a resurgent interest in developing modern CIR systems in both research communities and industry. This book surveys recent advances in CIR, focusing on neural approaches that have been developed in the last few years. This book is based on the authors' tutorial at SIGIR'2020 (Gao et al., 2020b), with IR and NLP communities as the primary target audience. However, audiences with other background, such as machine learning and human-computer interaction, will also find it an accessible introduction to CIR. We hope that this book will prove a valuable resource for students, researchers, and software developers. This manuscript is a working draft. Comments are welcome.
Paper Structure (202 sections, 83 equations, 78 figures, 12 tables, 1 algorithm)

This paper contains 202 sections, 83 equations, 78 figures, 12 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Book Organization
  3. How People Search
  4. Information Seeking Tasks
  5. Information Seeking Models
  6. CIR as Task-Oriented Dialog
  7. CIR System Architecture
  8. An Example: Macaw
  9. Remarks on Building an Intelligent CIR System
  10. Early Works on CIR
  11. System-Oriented and User-Oriented IR Research
  12. System Architecture
  13. Search Result Presentation
  14. Relevance Feedback Interactions
  15. Exploratory Search
  16. ...and 187 more sections

Figures (78)

  • Figure 1: A modular architecture for multi-turn task-oriented dialog systems. It consists of the following modules: NLU (natural language understanding), DM (dialog manager), and NLG (natural language generation). DM contains two sub-modules, DST (dialog state tracker) and POL (dialog policy). The dialog system, indicated by the dashed rectangle, has access to an external database or Web collection. Adapted from gao2019neural.
  • Figure 2: An example of a task-oriented dialog. (Top) A user goal and a task description. (Bottom) Multiple turns of user-system utterances, and the dialog belief states and database states at Turns 2 and 8. Adapted from gao2020robust.
  • Figure 3: An example of Bing Web search interface. The system makes search more effective by auto-completing an input query (Top-Left), organizing search results in the SERP (Right), and suggesting related queries that people also ask (Bottom-Left).
  • Figure 4: A reference architecture of CIR systems.
  • Figure 5: An example conversational search session, where the contextual query understanding modules rewrite user queries into context-independent queries as indicated by the arrows. Adapted from zhou2020design.
  • ...and 73 more figures