Research On Intelligent Recommendation Method Based On Music Feature

The current multimedia resources represented by music have growing rapidly on the Internet.It becomes a problem for users to get interested musics effectively,while the current music personalized recommendation system recommends musics based on varieties of different ways and the results is also so-so, it has great significance for the study of music recommendations. From the view of music,to develop an intelligent recommended way by combining feature extraction and research of current popularity of recommended methods.Users can obtain the actual operating results dynamically which increases the interaction, participation and improves the quality and experience.In this paper,as a starting point of music acoustic base,to analyze three main features:loudness,pitch and timbre,using melody contour features to characterize the music features.We regard MIDI musics as an object of stydy,analyze head tracks,block tracks and MIDI message as start,extract five feature amonuts combining with the name of music,artist,year,style and other labels to participate feature similarity calculation. Then,study the current mainstream method and analyze the advantages and disadvantages of the recommended policy,select collaborative filtering method as a music recommendation engine. Based on the analysis of user-based,content-based,model-based collaborative filtering,each of them has its own limitation,which presented three recommended ways to integrate and make a joint recommendation methods.Finally,combine the characteristic data,recommendation engine with recommended methos to provided users with results in demand.This paper is divided into two parts:research and MIDI music recommendation feature extraction methods. Based on the study of music features and recommendation system, select 400 MIDI music as object of study, extracted melody features into five dimensions,as a data source to a joint approach to recommend songs. At the same time,users’operation such as listen, download, search, evaluation, will dynamically adjust the user preference of the song, and thus adapt to the user’s tastes.