To explain or not to explain

Abstract

Recommender systems have been increasingly used in online services that we consume daily, such as Facebook, Netflix, YouTube, and Spotify. However, these systems are often presented to users as a "black box", i.e. the rational for providing individual recommendations remains unexplained to users. In recent years, various attempts have been made to address this black box issue by providing textual explanations or interactive visualisations that enable users to explore the provenance of recommendations, and benefits in terms of precision and user satisfaction, among others, have been demonstrated. Previous research had also indicated that personal characteristics such as domain knowledge, trust propensity and persistence may also play an important role on such perceived benefits. Yet, to date, little is known about the effects of personal characteristics when explaining recommendations. To address this gap, we developed a music recommender system with explanations and conducted an online study using a within-subject design. We captured various personal characteristics of participants and administered both qualitative and quantitative evaluation methods. Results indicated that personal characteristics have some significant influence on the interaction and perception of recommender systems and that this influence changes by adding explanations. Especially people with a low need for cognition benefited from explained recommendations. For people with a high need for cognition, we observed that explanations could lower their confidence. Based on these results, we present some first design implications for explaining recommendations.

Interactive Music Genre Exploration with Visualization and Mood Control

“…Visualization can serve as a way to explain recommendations [22], while user control, as a part of interaction [10], has also shown to benefit many aspects of recommender systems, such as improving the perceived accuracy and satisfaction of the recommendations [10] as well as helping users better understand the recommendation process [2]. MusicCube [32] allows users to select preferred tunes from 2D scatter plots in an interactive way.…”

“…The authors found that explaining a hybrid recommendation process through a visual user interface can improve user satisfaction. To investigate the effect of personal characteristics on different visual explanation techniques, Millecamp et al [22,23] designed an interactive visual explanation interface which enables users to adjust their preferences for each musical attribute with a slider. Users could also justify the recommendations by comparing the recommended tracks with their personal profile in the attribute space with bar charts and scatter plots.…”

“…3.2 for details of the two visualization approaches), to investigate which visualization can better support users to explore new music genres. The bar chart visualization (adapted from the previous work [22]) is selected as the baseline because of its effectiveness for explanation, which allows users to compare the recommended tracks with their average preferences directly. Due to its straightforward representation, users might find it easy to grasp information from the bar chart.…”

“…Earlier work has suggested that visualizations with user control can serve a way to justify the recommendations [2,22] as well as support users to explore the information space [17,18]. Through providing justification, transparency, and certain level of user control, interactive visualizations have been shown to be useful to help users understand the generated recommendations and the recommendation process better, and further increase their satisfaction [7].…”

“…Based on the existing music genre exploration system [19], we investigate how different types of visualizations as well as mood control would improve users' understanding of the recommendations from an unfamiliar new music genre, and support them to explore more. Specifically, we look into two visual explanation ways: a bar chart adapted from previous work [22], and a contour plot visualization which not only shows how the recommendations are related to the user's preferences but also to the new music genre. Additionally, we explore the benefits of mood control on two musical features related to mood (energy and valence).…”