Music Recommendation Based on Label Correlation

The Web is becoming the largest source of digital music and users often find themselves exposed to a huge collection of items. How to effectively help users explore through massive music items creates a significant challenge that must be properly addressed in the era of E-Commerce. For this purpose, a number of music recommendation systems have been proposed and implemented, which can identify music items that are likely to be appealing to a specific user. This paper presents a hybrid music recommendation system based on the labels associated with each music album, which also explicitly takes into account the correlation among labels. Experimental results on a real-world sales dataset show that our approach can achieve a clear advantage in terms of precision and recall over traditional methods in which labels are treated as independent keywords.