Music Recommendation System

Music distribution over the Internet has increased rapidly, greatly improving access to music. However, it remains difficult to find one’s favorite music in such a large market. To address this problem, various music recommendation systems have been proposed. Previous music recommendation systems are of two types. One method involves recommending the favorite music of other users who have similar preferences; another involves tagging music, recommending material with the same tags as one’s favorites. Tagging is done by experts or by machines. But music preferences are complex. You may not necessarily find music that is similar to your favorites, and the classification by expert or machine may differ from your subjective view.


Comventional Music Recommendation System

To address this problem, we have proposed a music recommendation system based on personal subjectivity, developed using the Kansei retrieval system.

Outline of the music recommendation system

This system depends on imitating the user’s Kansei, using Kansei agents to achieve this.


Proposed System

The Kansei agents represent the user in the system and can output evaluations of music. If the Kansei agents can output the same evaluation as the user for all the music, they can retrieve music from the database on the user’s behalf, returning recommendations based on the user’s Kansei. To do this, the Kansei agents must learn the user’s Kansei.

How is this done? The agents learn the user’s Kansei based on the error between user and agent evaluations. By repeating this process, the Kansei agents learn to imitate the user’s Kansei, so implementing a music recommendation system based on personal subjectivity.

The effectiveness of this system has been verified by simulation. However, the system has a technical problem in that it requires a large number of user evaluations in order to function correctly.

References

M.Inoue, H.Takenouchi, M.Tokuamaru, ” Music Recommendation System Improvement Using Distributed Genetic Algorithm”, 2016 Joint 8th International Conference on Soft Computing and Intelligent Systems and 2016 17th International Symposium on Advanced Intelligent Systems (SCIS&ISIS 2016), pp.627-630, 2016-08 (Sapporo, Hokkaido, Japan).

M.Inoue, H.Takenouchi, M.Tokuamaru, ” Music Recommendation System Using Kansei Agent and Music Fluctuation Properties”, 16th International Symposium on Advanced Intelligent Systems (ISIS2015), F1d-3, pp.760-768, 2015-11 (Mokpo, Korea).