Tag-based Artist Similarity and Genre Classification

Social tags are becoming more and more popular in Web2.0 recently. Tags defined by users are of high-level semantic for music. In this paper, we present a similarity calculation and genre classification measure for music artists with the use-defined tags from Last.fm. Similarities between artists are calculated based on tag co-occurrence. The k-nearest neighbor algorithm (k-NN) has been used to classify the music genre. Experiments show that tags are effective to characterize similarities between artists and the proposed approach outperforms the previous web-based approaches in artist genre classification with the highest average accuracy of 95%, compared with 89.5% of Schedl et al. and 81.2% of Knees et al.