Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Explainability for Transparent Conversational Information-Seeking

Weronika Łajewska, Damiano Spina, Johanne Trippas, Krisztian Balog

TL;DR

This paper addresses the challenge of transparency in conversational information-seeking (CIS) by proposing explanations about the information source, system confidence, and potential limitations accompanying three-sentence responses generated via retrieval-augmented generation. It conducts a crowdsourced user study across ten experimental conditions to assess how explanation quality and presentation mode affect perceived usefulness and explanation ratings, including both accurate and noisy variants. Key findings show that noisy explanations substantially reduce perceived usefulness, while high-quality explanations can improve perceived source and confidence judgments, though accuracy-related judgments remain relatively unaffected by explanations. The work contributes the first CIS-focused user study on explanations, a manually curated dataset with controlled noise, and generalizable insights about the trade-offs between explanation effort and user gain, with implications for designing trustworthy CIS experiences and informing future personalization of transparency features.

Abstract

The increasing reliance on digital information necessitates advancements in conversational search systems, particularly in terms of information transparency. While prior research in conversational information-seeking has concentrated on improving retrieval techniques, the challenge remains in generating responses useful from a user perspective. This study explores different methods of explaining the responses, hypothesizing that transparency about the source of the information, system confidence, and limitations can enhance users' ability to objectively assess the response. By exploring transparency across explanation type, quality, and presentation mode, this research aims to bridge the gap between system-generated responses and responses verifiable by the user. We design a user study to answer questions concerning the impact of (1) the quality of explanations enhancing the response on its usefulness and (2) ways of presenting explanations to users. The analysis of the collected data reveals lower user ratings for noisy explanations, although these scores seem insensitive to the quality of the response. Inconclusive results on the explanations presentation format suggest that it may not be a critical factor in this setting.

Explainability for Transparent Conversational Information-Seeking

TL;DR

This paper addresses the challenge of transparency in conversational information-seeking (CIS) by proposing explanations about the information source, system confidence, and potential limitations accompanying three-sentence responses generated via retrieval-augmented generation. It conducts a crowdsourced user study across ten experimental conditions to assess how explanation quality and presentation mode affect perceived usefulness and explanation ratings, including both accurate and noisy variants. Key findings show that noisy explanations substantially reduce perceived usefulness, while high-quality explanations can improve perceived source and confidence judgments, though accuracy-related judgments remain relatively unaffected by explanations. The work contributes the first CIS-focused user study on explanations, a manually curated dataset with controlled noise, and generalizable insights about the trade-offs between explanation effort and user gain, with implications for designing trustworthy CIS experiences and informing future personalization of transparency features.

Abstract

The increasing reliance on digital information necessitates advancements in conversational search systems, particularly in terms of information transparency. While prior research in conversational information-seeking has concentrated on improving retrieval techniques, the challenge remains in generating responses useful from a user perspective. This study explores different methods of explaining the responses, hypothesizing that transparency about the source of the information, system confidence, and limitations can enhance users' ability to objectively assess the response. By exploring transparency across explanation type, quality, and presentation mode, this research aims to bridge the gap between system-generated responses and responses verifiable by the user. We design a user study to answer questions concerning the impact of (1) the quality of explanations enhancing the response on its usefulness and (2) ways of presenting explanations to users. The analysis of the collected data reveals lower user ratings for noisy explanations, although these scores seem insensitive to the quality of the response. Inconclusive results on the explanations presentation format suggest that it may not be a critical factor in this setting.
Paper Structure (16 sections, 4 figures, 5 tables)

This paper contains 16 sections, 4 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Methodology
  4. Experimental Design
  5. Crowdsourcing Task Design
  6. User Study Execution
  7. Data
  8. Pilot Study
  9. Main Study
  10. Results
  11. User's Perception of Response and Explanations
  12. Effect of the Explanation Quality (RQ1)
  13. Effect of the Presentation Mode (RQ2)
  14. Qualitative Analysis
  15. Discussion
  16. ...and 1 more sections

Figures (4)

  • Figure 1: Information-seeking dialogue with a CIS system with explanations (sources, confidence, and limitations).
  • Figure 2: Examples of responses with explanations for the query: What was the US reaction to the Black Lives Matter movement?
  • Figure 3: High-level design of the user study.
  • Figure 4: Mean scores for response usefulness and explanation ratings for different quality of the explanations (top) and presentation mode (bottom). All differences between the ratings within a given plot are statistically significant.