Title: "Recommender Systems for Music Retrieval Tasks"
Abstract:
Music is ubiquitous in today’s world and with the rise of streaming platforms, listeners now have access to more music than ever before. Music recommender systems aim to help users discover and retrieve music they like and enjoy. This requires incorporating information about users and their preferences into retrieval and recommendation algorithms.
Most importantly, such user-centric retrieval approaches need to capture aspects that influence the user’s perception and preference for music. Aspects that influence human perception of music include music content (e.g. a song's tempo), music context (e.g., lyrics), user properties (e.g., a user's general music preferences), and user context (e.g., the current activity, or emotional state). These aspects are typically captured by a user model. To compute items that best meet the user’s needs and preferences, these user models are compared with item models, which capture the characteristics of individual items.
Current user models for personalized retrieval tasks are typically modeled rather simplistic, mostly focusing on single aspects of the user. Comprehensive user models are rare in music information retrieval. Consequently, there is not only a lack of comprehensive user models, but also a lack of retrieval and recommendation approaches that allow integrating and combining multi-faceted user and item models.
This habilitation thesis contributes to the field of (music) recommender systems in the following aspects: (1) We present novel rich user and item models to capture the characteristics of users, their context, and musical items. (2) We jointly leverage these user and item models in newly designed context-aware recommender systems. (3) We investigate biases in state-of-the-art recommender systems based on the proposed user and item models. (4) We propose an evaluation framework to conceptualize the recommendation evaluation space, enabling a comprehensive assessment of the factors influencing recommendation performance.