ConvEx-DS: A dataset for conversational explanations in recommender systems

Hernandez-Bocanegra, D. C., & Ziegler, J. (to appear). In IntRS’21: Joint Workshop on Interfaces and Human Decision Making for Recommender Systems.

Abstract

Conversational explanations are a novel and promising means to support users’ understanding of the items proposed by a recommender system (RS). Providing details about items and the reasons why they are recommended in a conversational, language-based style allows users to question recommendations in a flexible, user-controlled manner, which may increase the perceived transparency of the system. However, little is known about the impact and implications of providing such explanations, using for example a conversational agent (CA). In particular, there is a lack of datasets that facilitate the implementation of dialog systems with explanatory purposes in RS. In this paper we validate the suitability of an intent model for explanations in the domain of hotels, collecting and annotating 1806 questions asked by study participants, and addressing the perceived helpfulness of the responses generated by an explainable RS using such intent model. Thus, we release an English dataset (ConvEx-DS), containing intent annotations of users’ questions, which can be used to train intent classifiers, and to implement a dialog system with explanatory purpose in the domain of hotels.

Resources

Related publications

Argumentative explanations for recommendations - Effect of display style and profile transparency

More »

Focus area