Social Tag-based Mobile Music Retrieval

With the prevalence of mobile phones, mobile services in China gradually have promising market prospect. With the continuous development of Mobile technology, as well as the changes in people's minds, mobile applications are no longer just a fashion. But relative to Internet applications, applications on mobile devices face difficulties to quickly and accurately find the information they need in the face of a great deal of information. Mobile users propose higher demands on the retrieval accuracy rate and the needs of personalization. Therefore, mobile retrieval has become a hotspot research field with wide market prospect. In people's leisure activities, music, as an important leisure activity, is indispensable in mobile applications. As a result, how to provide users with accurate and personalized music retrieval service with the hardware-constrained conditions on mobile phones and other mobile devices is the most important research field in mobile music retrieval.This paper first introduces the practical significance of mobile music retrieval. And the state-of-the-art and implementations of mobile retrieval are presented. In light of social network analysis technique currently used in mobile retrieval, folksonomy and social tags are given a detailed introduction. The primary methods of music retrieval are also analyzed.In addition, the concept of music genome is introduced into this paper, based on which the characteristics of music are analyzed. Users' preference on features of music is then obtained during retrieval using the statistical analysis of social tags. A weighted undirected graph is built based on the relationship of social tags between music. Random walk and the PageRank algorithm are utilized to calculate the musical heat based on social tags.In this paper, a retrieval model based on social tags is proposed, and through the analysis of user behavior, a function called label attenuation which can reflect the importance of secondary labels entered by users is introduced. The music retrieval model based on simply matching approach is improved by combining with characteristics of the user's preferences on music and musical heat. On this basis, a social tags-oriented mobile music retrieval system is designed. Involved in the use of artificial methods for evaluation, the paper carries out comparative experiments on the system to evaluate performance. Experimental results show that the system can achieve a higher accuracy and meet the user's personalized requirements with the hardware-constrained conditions on mobile phones and other mobile devices.