Research On Personalized Music Recommendation Algorithm Based On Context Awareness

The rise of Web2.0 and the arrival of the era of big data, has become the trend of today's information era, which has made the traditional Internet model gone over. At the same time, the magnanimity data information penetrated into every corner of the Internet, the problem of information overload is becoming more and more serious. Personalized music recommendation technology, as an effective application to solve the problem of digital music information overload, has become a hot research topic in the field of music services.Traditional music recommendation algorithm mostly concerned about the "user-project" of the two-dimensional evaluation of the relationship, but ignores the context information when user listen to the music, such as time, location, weather and so on. Consider user's interests most changed by the changing of context information, this paper puts forward some improved algorithm based on context awareness. The main research work of this paper is as follows:(1) According to the data sparsity problem in the field of music recommendation, an implicit score conversion method is proposed. The implicit score conversion is based on the number of cycles in user's listen to the same music, and the number of the cycle is mapped to the score value of the music.(2) In view of the shortcomings of the traditional recommendation method in dealing with the relationship between users, based on the time series in user's listening records, a time-user-relationship model is proposed. This model based on the time sequence of user's consumption music, the user relationship is divided into influencer and the affected person,so we can get a clear directed relationship between users.(3) As the traditional recommendation method cannot meet the demand of context-awareness,this paper proposed a kind of context-reenacted model. The context-reenacted model makes use of the temporal relationship between each users, combined with the contextinformation when user listening to the music, and recommend some musics corresponds to the similar user's behavior which happend in the same context.(4) As the context-reenacted model use pre filtering technology to deal with the native data directly, which is not the real integration of model and context, this paper proposed a context-aware model based on probability matrix decomposition. In order to deal with multi-dimensional "user- context- item" model properly. This paper map "user-context-item" into "user-item" model's "item", and puts forward a context aware recommendation method based on probabilistic matrix decomposition.(5) In order to optimize the relationship between users in probabilistic matrix decomposition,combined time-user-relationship with probability matrix decomposition, and add the the context infomation when user listening to music, put forward a context-aware recommendation model based on sequential probability matrix decomposition.(6) Through experiments, it is proved that the improved method is effective. Considering the introduction of context information, the data sparsity problem is aggravated, the training set is divided into different sparse intervals, so the difference of each algorithm is evaluated comprehensively.