Genre-based Link Prediction in Bipartite Graph for Music Recommendation☆

Abstract The bipartite graph method of link prediction can apply in many fields of recommendations, with the nodes (users and items) and links (interactions between users and items). However, that links cannot represent the users’ dual preferences (like and dislike). Some researchers improved that limits by complex number representations, but still not consider the influence of users’ similarity recommendation performance. Here, we proposed an improved method to cope with this deficiency, build the relational dualities by complex number representations and computing the users’ similarity by genres weight relations. In experiments with the Xiami.com music dataset, the proposed music genre weight-based music recommendation model (MGW) performances better than the CORLP method.