A new music recommendation system has been developed by information scientists in Korea. They report details in the latest issue of the International Journal of Intelligent Information and Database Systems.
The addition of emotional tags, evocative keywords, to the music files you listen to could improve music recommendations in a way that is not possible with the standard recommendation approaches used by well-known social music sites where number of plays and “likes” are the only factors taken into account. The approach gets around the cold-start problem for new artists and new users alike allowing music that has not had a chance to become popular to be tagged and if that if lots of users tag it positively it will become more highly recommended.
Hyon Hee Kim, Donggeon Kim and Jinnam Jo of the Department of Statistics and Information Science, at Dongduk Women’s University, in Seoul, explain that a combination of listening habits and meaningful, semantic, tagging of music files, with terms such as melancholy, tragic, joyful, awesome, unexpected, boring, annoying and many others, has allowed them to develop a unified music recommendation system. The team has tested their approach on 1,000 users, 12,600 tags added to 18,700 music items that they listened to and randomly collected from the well-known online music service last.fm. They report that their approach performs better than the conventional recommendation systems based on simple frequency of play metrics or on a simple positive-negative grading system.
The team’s approach classifies tags as being organizational, genre classifying and emotional and then breaks down the emotional into positive and negative as well as adding a greater statistical weight to those tags, such as perfect or boring, than to the ones that simply define a track as, for instance, progressive rock or jazz.
“Our proposed approach gives a good solution to the conventional cold start problem,” the researchers explain, “Collecting users’ listening habits takes time for users to listen to the music items for a long duration.” The use of emotional tags makes it much easier for users to select out music tracks based on mood or other evocative factors and avoid those that are not likely to please them, in this way not only is the new system acting as a kind of “word-of-mouth” recommendation but it also overcomes the data sparseness problem facing new users and new musicians.
“A unified music recommender system using listening habits and semantics of tags” in Int. J. Intelligent Information and Database Systems, 2014, 8, 14-30