面向推荐系统的音乐特征抽取

音乐推荐系统是指根据用户的历史浏览数据，从候选库中推荐给用户可能喜欢的音乐的一种新型网络服务。该系统的关键在于需要对整个数据库按照音乐风格进行分类，基于此提出一种新的音乐特征处理方法来完成音乐库分类，以有效实现音乐推荐。该方法首先为候选音乐库构建常规的音乐特征数据集，然后基于分形理论对数据集进行属性约简，获取每一首音乐的推荐特征向量，并且依据特征向量的特点，定义了一种新的距离度量方法。在包含六种风格的音乐数据库的实验中，仿真结果证明了提出的音乐推荐特征和距离度量的有效性，与现有的基于内容的音乐检索研究相比，音乐推荐特征的使用极大地降低了对数据库存储量的需求，对音乐推荐系统的网络开发具有很好的应用价值。

Music feature extraction method for recommendation system

Music recommendation system is to provide needful music for users from the candidate music library based on analyzing the user’s browsing history data.The classification of whole database according to music style is the key point of this system.Then this paper proposes a novel music feature for achieving music classification.Firstly the regular data sets of music feature are constructed.Then the attribute of data sets is reduced based on fractal dimension theory and the recommendation features are obtained for each music.Also a new distance measure is given for the similarity of this recommendation feature.During the experiment based on database including six styles of music,simulation results demonstrate the effectiveness of the proposed music recommendation feature and the new distance.Compared with the study of current content-based music retrieval,music recommendation feature can greatly reduce the requirement of database memory and provide a promising application for the network exploitation of music recommendation system.