Explainability in music recommender systems

Abstract

The most common way to listen to recorded music nowadays is via streaming platforms, which provide access to tens of millions of tracks. To assist users in effectively browsing these large catalogs, the integration of music recommender systems (MRSs) has become essential. Current real-world MRSs are often quite complex and optimized for recommendation accuracy. They combine several building blocks based on collaborative filtering and content-based recommendation. This complexity can hinder the ability to explain recommendations to end users, which is particularly important for recommendations perceived as unexpected or inappropriate. While pure recommendation performance often correlates with user satisfaction, explainability has a positive impact on other factors such as trust and forgiveness, which are ultimately essential to maintain user loyalty. In this article, we discuss how explainability can be addressed in the context of MRSs. We provide perspectives on how explainability could improve music recommendation algorithms and enhance user experience. First, we review common dimensions and goals of recommenders explainability and in general of eXplainable Artificial Intelligence (XAI), and elaborate on the extent to which these apply-or need to be adapted-to the specific characteristics of music consumption and recommendation. Then, we show how explainability components can be integrated within a MRS and in what form explanations can be provided. Since the evaluation of explanation quality is decoupled from pure accuracy-based evaluation criteria, we also discuss requirements and strategies for evaluating explanations of music recommendations. Finally, we describe the current challenges for introducing explainability within a large-scale industrial MRS and provide research perspectives.Darius Afchar and Alessandro B. Melchiorre contributed equally to this study.

Transparency in Music-Generative AI: A Systematic Literature Review

“…However, explanations may have multiple content and representations. This is why many of the identified publications in this review mention the relevance of focusing the explanation to the specific end-user 16 (Abdollahi and Nasraoui, 2018;Arrieta et al, 2020;Afchar et al, 2022;Adadi and Berrada, 2018). It is essential to centre the explanations towards the corresponding audience profile and state why the explanation is requested (Arrieta et al, 2020).…”

“…For example, in music recommender systems, to understand why a song was suggested in a playlist, listeners and developers require different information. At the same time, taking into account the aimed audience is important to avoid overwhelming, as providing too much information would contribute to cognitive saturation (Afchar et al, 2022). That is, it is better to provide accurate and relevant explanations, rather than all the available information.…”

“…In addition, most of the identified publications point out the need to evaluate the performance of the explanations. The lack of formality and clear definition on an explanation task, does not allow comparing performance between strategies nor guaranteeing that the explanations are understandable by the set audience (Clinciu et al, 2021;Afchar et al, 2022). Towards solving this issue, Clinciu et al (2021) proposes an evaluation strategy to analyse how effective explanations are, taking into account the user's understanding.…”

Recommender systems: Trends and frontiers

“…Two other papers, those by Afchar et al. (2022) and Massimo and Ricci (2022) focus on specific application domains, music and tourism, and thus aim at improving our understanding of particular problems in these areas, such as, the impact of item popularity in the recommender systems and in the users' evaluation of recommendations. Afchar et al.…”

“…Afchar et al. (2022) specifically address the problem of explainability of music recommendations and discuss questions of how to integrate such explanations within a large‐scale industrial music streaming platform. Massimo and Ricci (2022), on the other hand, investigate the problem of recommending the next point‐of‐interest (POI) to tourists.…”