Interactive Music Recommendation: Context, Content and Collaborative Filtering
COM2 Level 4
Executive Classroom, COM2-04-02
Abstract
Music recommendation systems predict users' preferred songs and thus greatly ease the process of music selection and also boost the revenue of online music merchants. However, results produced by existing music recommenders are still not satisfactory because of their ignorance of relevant information or the drawbacks of their underlying modeling techniques.
To better satisfy users' music needs, this thesis strives to improve recommendation performance from three aspects. First, we developed the first context-aware music recommendation system that recommends songs to match the target user's daily activities including sleeping, running, studying, working, walking and shopping. Second, we developed a model that simultaneously learns features from audio content and makes personalized recommendations based on deep belief network. The features are then incorporated into collaborative filtering to form an effective hybrid recommendation method. Third, we present a new approach to music recommendation by formulating the exploration-exploitation trade-off in music recommendation as an interactive reinforcement learning task. Extensive evaluations are conducted to demonstrate the effectiveness of the developed methods.