Intent-aware Music Recommender Systems

The music that we prefer, such as a particular artist or genre, depends on our intended purpose for listening. For example, we may choose to listen to calming music after a long day at work but opt for upbeat music when preparing for a party. While music psychologists have identified several general intents for music listening, such as self-expression, mood regulation, and social connectedness, these intents are more nuanced. Unfortunately, current music recommender systems do not take user intents into account when making recommendations. Additionally, since music is a multifaceted and multimodal cultural commodity, the way a music item is represented (e.g., music video clip, digital score sheet, or audio file) limits its possible uses and the purpose it may serve for the listener. For instance, a music video clip may not be suitable for practicing piano, while a digital score sheet might be. Against this background, the project aims to research (1) novel algorithms that recognize a listener`s intent when accessing music-related content and (2) create a new generation of music recommender systems that recognize and adapt to user intents. Specifically, we will first identify and investigate the intents that drive users when accessing multimedia collections of music-related content. Then, we will create user preference models that integrate various content descriptors, including audio, textual, and visual characteristics of music items. Third, we will study which multimedia content descriptors are best suited to predict listening intents. In addition to analyzing content descriptors, we will explore the potential benefits of incorporating user and contextual aspects, such as demographics, personality traits, mood, activity, and social context, into the prediction model to improve its performance. Ultimately, we will integrate the user preference profiles and predicted intents to devise a new music recommendation algorithm that reacts dynamically to changing intents and explains its recommendations to the listener. The algorithm will be implemented in a prototype intent-ware music recommendation system.