Music Recommendation System Based On Audio Features And Social Tags

Digital music service has become one of common applications in the Internet. However, enormous amount of Internet music makes it quite hard for users to find out interesting music while a lot of music in the long-tailed region has seldom been visited, which implies that inefficient utilization of music information. Music recommendation helps to filter music information and to push music to probably interested users, which has been seen as an effective solution to the problem of efficient utilization of enormous music information.This thesis focuses on the music recommendation technology which adopts hybrid recommendation algorithm, applying combination of contextual and content-based features of each of tracks, these features include both audio descriptors and social tags. Firstly, this thesis gathers mass of audio features and social tags through data mining and digital signal processing, after preprocessing of musical features, recommender system calculates the similarity of music by means of dimensionality reduction, and uses visualization technologies to map the similarity into a two-dimensional musical space.This thesis gained a large number of user tags from commercial websites, analyzed eighty audio features of tracks, conducted a database of musical features, implemented hybrid music recommendation algorithm and measured the performance of algorithm. The experimental results showed that social tags are one of effective music resource descriptors, and audio features are useful complement for social tags to increase the accuracy and comprehensiveness of resources. The visualization effects afforded friendly and flexible interactive interface, giving the users very good visual experience.The results of this thesis can be used in Internet music recommendation system and provide effective technical support for efficient utilization and access of enormous music resources.