Session-based Track Embedding for Context Aware Recommendation

Nowadays, music streaming platforms are a popular way to consume music. Recommender systems are an important feature for these platforms since they are used to attract users and keep them using the service by providing users with recommendations for tracks they might enjoy. Hence, the performance of such a recommender system for these platforms is substantial. Due to their popularity, such platforms provide a large source of information about the listening behavior of their users which can be used for their recommender systems. The goal of this thesis is to develop a context-aware track recommender system that exploits the playing history of a user as context. To achieve this, a dataset containing the listening histories of users is split into single listening sessions of each user. The historical playing sequences of the sessions are used to learn the embeddings of songs with the help of neural networks. The embedding is used for context-aware recommending tracks to users. The evaluation shows, that our proposed track recommender system outperforms a collaborative ltering algorithm which is utilized as baseline.