Modeling User Music Preference Through Usage Scoring and User Listening Behavior for Generating Preferred Playlists

Abstract

Recommending the most appropriate music is one of the most studied fields in the context of Recommendation systems with the growing number of content available to users and consumers alike. As it is an important aspect in the use of multi-media systems and the music industry, it is important to note that the typical approach is through collaborative-filtering.

In this paper, the study considered a more personalized view and examined to which degree a user’s music preference can be modeled using information gathered from the user with respect to their listening behavior and music selected. The study proposes an approach to modeling a user’s music preference using a series of usage scores obtained from a user’s listening behavior and to generate a playlist derived from the obtained model.

Using a novel data set, the proposed approach resulted to an average True-Positive rating of 54.43% in predicting music files that the user will select for the month given the previous month’s data and an overall performance of 82.53% in producing entries to a preferred playlist, showing the possibility of more refinements and further study.