Situational Music Recommenda Tion In Mobile Environment

In this paper we study a new type of music recommendation, namely situational music recommendation in mobile environment. Different from traditional music recommendation systems, we will take user’s current geographical ambience into consideration in situational music recommendation. Mobile user’s location is constantly changing as well as his geographical ambience. User’s feeling and expectation of music is significantly influenced by his inner-state which in turn is greatly affected by his geographical ambience, thus we have to make different recommendations according to user’s geographical ambience.In this paper we propose a new music recommendation prototype named Pictune. Pictune utilizes location-based image retrieval and other Web services to recommend music for users virtually located at any position. We analyze the architecture and recommendation process of Pictune in depth. We tackle mainly two difficult problems in this paper, the first is arbitrary English word similarity calculation, the effectiveness and efficiency of word similarity calculation is very important to pictune, we propose three different methods to solve this problem, namely Lucene-Base method, Memory-Resident method and MultiCore-Based method; Another problem is the design and implementation of recommendation algorithm, we devise a variant generative topic model for the photo data taking photos’GPS coordinates into consideration, this model is based on the famous LDA topic model. We use Expectation Maximization algorithm for parameter estimation, and design an efficient recommendation algorithm.Extensive experiments on real data sets confirm the effectiveness and efficiency of our proposed algorithms and demonstrate the practical value of Pictune.