Effects of Feature-based Explanation and Its Output Modality on User Satisfaction with Service Recommender Systems

Recent advances in natural language based virtual assistants have attracted more researches on application of recommender systems (RS) into the service product domain (e.g., looking for a restaurant or a hotel), given that RS can assist users in more effectively obtaining information. However, though there is emerging study on how the presentation of recommendation (vocal vs. visual) would affect user experiences with RS, little attention has been paid to how the output modality of its explanation (i.e., explaining why a particular item is recommended) interacts with the explanation content to influence user satisfaction. In this work, we particularly consider feature-based explanation, a popular type of explanation that aims to reveal how relevant a recommendation is to the user in terms of its features (e.g., a restaurant's food quality, service, distance, or price), for which we have concretely examined three content design factors as summarized from the literature survey: feature type, contextual relevance, and number of features. Results of our user studies show that, for explanation presented in different modalities (text and voice), the effects of those design factors on user satisfaction with RS are different. Specifically, for text explanations, the number of features and contextual relevance influenced users' satisfaction with the recommender system, but the feature type did not; while for voice explanations, we found no factors influenced user satisfaction. We finally discuss the practical implications of those findings and possible directions for future research.