中文摘要：

随着协同标注功能的普及,用户可以通过对自己感兴趣的音乐进行标注从而实现个性化的分类管理,因此音乐共享系统中的社会化标签已成为互联网的重要资源。考虑到社会化标签的特性及其对音乐信息检索系统的影响并综合考虑了用户的检索行为、歌词和音乐标签,利用MT-LDA方法对标签进行聚类以获取主题类别,从而进行分析得出检索主题,提高音乐信息检索系统的效率和性能。实验结果表明：在没有属性数据信息的检索情况下,基于标签主题的MT-LDA检索模型相比于基于标签关键词检索模型,尤其是在音乐标签稀疏和非正规的情况下,在一定程度上更能够提高音乐信息检索性能。

英文摘要：

Music sharing systems with collaboratively tagging function have been important parts on the Internet. They make the system users annotate and categorize their own interests and thoughts about the resources possible. Considering the characteristics of social tagging and its influence on Music Information Retrieval（ MIR） system,MT-LDA method by jointly considering lyrics,tags and searching behavior can be used to analyze collaboratively generated tags and the topic category of tags to catch the retrieval topic,so as to improve the efficiency and performance of MIR system. The experiment shows that MT-LDA retrieval model based on tags topic performs better than keywords retrieval model based on tags into improving the MIR system performance especially tags for tracks are extremely sparse and informal,when retrieval information have no attribute data.