Research And Implementation Of Music Recommendation System Based On Mobile User's Behavior Perception

The rapid development of digital media technology to make numbers of digital music resources on the network greatly improved. It makes difficult for people to find their favorite music from a huge music database. The music recommendation system can help users find their favorite music. Traditional recommendation system builds user interest model based on user's history listening record and generate a recommend list the user has not listen based on the model. This method can only reflect the long-term interests of users. It cannot aware user's interest changing in a short period. And there is a great relationship between user's interest in a short period of time and user's current behavior. Today, the client of user listen music has been turned to the mobile terminal by the PC. This provides the possibility of combining user's behavior perception with the recommendation system because mobile terminal usually equipped with sensors. By analyzing the sensor data, we can analyze the current behavior of users at any time and at any time.In this paper, the combination of mobile user behavior perception and music recommendation system is studied, and the following results are obtained:Proposed a new tagging way that using semantic analysis the playlist to auto tag user behavior tag.Proposed an algorithm of semantic similarity computation based on Word Net. The traditional semantic similarity algorithm based on Word Net is based on the organization structure of noun. In this paper, based on the organizational structure verbs, the paper calculates the similarity between two concepts based on the relative position of the two concepts in the semantic tree.A classification algorithm is proposed to realize the prediction of user's behavior. The method is based on learning the sensor data and predicting the user's behavior through the classification model. We use naive Bayes, KNN, decision tree classification algorithm of contrast experiments, experiments show that the predict user behavior by classify the sensor data is a kind of feasible method.In order to generate different recommendation results according to the same user behavior has different musical interest. In this paper, a collaborative filtering recommendation algorithm based on clustering is proposed. Firstly, the algorithm cluster the behavior of the same user generate user clustering center preference matrix. Based on this matrix, the results are recommended.At last, this paper proposes a real-time recommendation framework based on the above methods, which can be perceived in real time to change the user's interest. Based on this framework, a prototype system is implemented.