Abstract in English:

Providing explanations based on user reviews in recommender systems (RS) can increase users' perception of system transparency. While static explanations are dominant, interactive explanatory approaches have emerged in explainable artificial intelligence (XAI), so that users are more likely to examine system decisions and get more arguments supporting system assertions. However, little attention has been paid to conversational approaches for explanations targeting end users. In this paper we explore how to design a conversational interface to provide explanations in a review-based RS, and present the results of a Wizard of Oz (WoOz) study that provided insights into the type of questions users might ask in such a context, as well as their perception of a system simulating such a dialog. Consequently, we propose a dialog management policy and user intents for explainable review-based RS, taking as an example the hotels domain.