基于概率图模型的音乐推荐方法

Music recommendation method based on the probabilistic graph mode

提出了一种基于概率图模型的音乐推荐方法。该方法利用歌曲相似度确定歌曲间的关系,从而建立歌曲网络,在此基础上利用主题模型得到的歌曲主题概率分布,建立包含局部属性(主题概率分布)和全局结构(歌曲网络)的概率图模型,转化为因子图后,再利用推理算法对概率图模型进行计算,最终获得歌曲在不同主题下的推荐列表。实验表明,本方法能够获得较好的推荐效果。

This paper proposes a music recommendation method based on probabilistic graphical model.This method utilizes similarity between songs to define the relationship between songs,thereby establishing song network,and utilizes the topic distributions obtained by topic modeling,therefore establishing a probabilistic graphic model utilizing both the local attributes（topic distribution）and the global structure（network information）,and then transform the probabilistic model into the factor graph model,the approximate inference algorithm is then be used to infer the result.In the end,the model gives a different list of recommendations under different topic.Experimental results show that this method performs better on topical recommendation and is a beneficial exploration on recommender area.