Abstract

With the growth and complexity of heterogeneous music information network, it becomes necessary to have a deep understanding of the interaction reasons to discover how the users choose their preferences in order to build more effective recommendation systems. Moreover, even when the recommendation model embeds user-item interactions, if we analyze the user preferences, we miss references to important content-based information or actual semantics of recommended items, which support the interpretation of a recommendation process. In this thesis, we will build a hybrid recommendation model based on content-based, context-based and collaborative filtering recommendation methods to leverage heterogeneous music information graph. To achieve this goal, a Music Information Knowledge Graph (MKG) is first constructed to encode different kinds of heterogeneous information, including the interactions between users, tracks and artists. Besides, we propose a hybrid music recommendation approach for MKG using factorization machines. In the experimental part, we collect data from a wide and popular music platform called LastFM which contains a detailed profile of each user's musical taste and details of the tracks the user listens to. Moreover, we extend our dataset to be more effective by involving data about the content characteristic of the collected tracks (tracks content data) and the artists who sing these tracks (artists context data). Therefore, we get the content characteristic of the tracks from Spotify and we collect the artist details from DBpedia. Finally, all these data resources are integrated into MKG. Building Hybrid recommender based on MKG and factorization machines techniques shows that dealing with relevant content and context information can be used to increase the recommendation accuracy and to improve the awareness of the reasons behind user's preferences.

Citation

BibTeX

@misc{Azzam2021mastersthesis,
    title = {Hybrid Music Recommendation Approach for Heterogeneous Information Network using Factorization Machines},
    author = {Azzam, Majd},
    url = {https://resolver.obvsg.at/urn:nbn:at:at-ubl:1-44265},
    school = {Johannes Kepler University Linz},
    year = {2021}
}