Mobile Context-aware Music Recommendation System

Mobile context-aware music recommendation system makes contribution to research and application area to a certain extent, however, research on mobile context-aware music recommendation is very rare. There is also no such existing APP in domestic application market. This paper proposed and implemented an improved mobile context-aware music recommendation system, which established the relationship between music and context from existing playlist data to avoid high cost and inaccuracy brought by data annotation manually. It certainly can be said that this paper provided theoretical research and practical application value.As known, there was one typical related work using context information recommended music for daily activities like running, working. It researched the relationship between sensor, context and music to recommend. The proposed system can effectively use the mobile context information, avoiding the data sparsity and cold start problem, and it also integrated incremental learning. However, it need experts to tag each music manually in advance which brought high cost and inaccurate description for the relationship between music and context.The system proposed in this paper based on the advantages and make up for shortcomings. It crawled playlists from the Internet to establish the relationship between music and context, which avoid the high cost problem by annotation manually, and also descript the relationship between music and context more accurate. Besides, it redesigned the context setting, which improve the distinction between the context states and makes it more in line with the Chinese user’s habits. In algorithm design, this paper used Naive Bayesian algorithm to build the sensor-context model, crawled the playlist and calculated probability statistics to build the music-context model, and then used Bayesian framework to combine the two model to obtain user’s preference probability for each music in the user’s current context. This paper also integrated incremental learning to use user’s feedback to improve performance.This paper implemented a complete system, including the server side part and a mobile phone APP. 52482 samples of sensor data,40871 music information, and 1380 playlists are collected and used to build the music recommendation model. The server part provides RESTful Web API based on Jersey, MySQL, Hibernate. The client is for the Android platform, using three party library assistant like Ormlite, Volley.The questionnaire survey results showed that the mobile context-aware music recommendation system has good research and application values. Performance analysis results showed that compared with the existed algorithm, the improved algorithm proposed by this paper improved the recommendation performance and the training speed.