A Hybird Graph-based Algorithm Rresearch On The Music Recommendation

At present, the endless information has penetrated into all walks of life and everyone's daily activities. Various music websites and applications sprung into being and vastly facilitate people's music needs.However, each of the major music websites contains millions of songs, whose information is a great overload beyond an individual can hold. In order to find songs appealing to everyone's taste, a lot of personalized music recommendation systems emerged and has been widely used in the modern society.Content-based algorithm and Collaborative Filter-based algorithm are two popular music algorithms. A mount of research has been put on these two algorithms and get high accuracy. However, with the in-depth study of recommendation system, more and more research found that Accuracy is not enough to measure users' satisfaction. They found the fact that recommendations can be accurate in terms of relevancy but also probably being familiar to the users. Thus, Novelty began to get high attention. Particularly, the Fifth ACM Conference on Recommendation Systems organized a special seminar to discuss it.Therefore, this paper proposed a graph-based novelty framework of music recommendation, which applied users' rating records to create a preference-directed graph and a positively-linked undirected graph. Then these two graphs were combined together to create a hybrid graph. Finally,with using the concept of entropy, the proposed system and a traditional collaborative filtering system were tested on Douban data. And the results showed that the proposed system indeed gained a good recommendations that were both relevant and novel.