Research And Implementation Of Hybrid Music Recommendation Algorithm Based On Time Features

With the rapid development of mobile Internet and multimedia technology,all kinds of information in various fields is explosive growth,the amount of network load information is becoming more and more huge,and the problem of information overload is becoming increasingly serious.Recommender system can solve this problem effectively.It can mine users’ potential preferences and provide personalized services for users by analyzing users’ historical behavior data.According to the service scenario of music recommendation platform,based on the analysis of the basic elements of music recommendation and the existing algorithm research,this thesis optimizes the traditional recommendation algorithm,mainly including the following aspects:(1)In the research of solving the problem of music cold start,a method of music type prediction based on CGABC-SVM is proposed.Four different audio features are extracted from the original audio signal to form a multi feature fusion matrix.According to these features,support vector machine(SVM)is used to classify music automatically.In order to solve the problem that it is difficult to select the parameters of SVM,the cross global artificial bee colony algorithm(CGABC)is used to optimize the parameters,and a music classification model is constructed to achieve the purpose of music type prediction and reduce a lot of manual annotation work.(2)This thesis improves the traditional collaborative filtering algorithm.In the construction of user-music score matrix,the interest calculation rules and score standardization function are set to eliminate the "Matthew effect" caused by too large or too small score value,so that the user’s interest score of music is more real.In order to solve the problem that the user’s personal interest will change with time,this thesis introduces a nonlinear time decay function based on the previous research,optimizes the user-music score matrix based on the time characteristics,and reflects the change of user’s interest.The penalty factor is introduced to improve the user similarity calculation formula,in order to reduce the impact of popular items on the calculation results.(3)This paper adopts hybrid recommendation strategy.The constructed music classification model is used to predict music types and automatically generate tags.User preferences are obtained through the user music score matrix information optimized by the interest model and time decay function,and then recommendation results are generated according to the user’s preference for music tags,and recommendation results based on music tags are established.The improved collaborative filtering algorithm and the music tag based recommendation algorithm are mixed,and the advantages of the two algorithms are combined to avoid the problems of a single algorithm and improve the recommendation accuracy.(4)The hybrid recommendation strategy proposed in this thesis is applied to the actual music recommendation scene.This thesis analyzes the core requirements of the music service platform,designs the architecture,function and database of the system,determines the relevant functional details and technologies,and completes the design and implementation of the music recommendation system.