面向音乐推荐的全变差图非负矩阵分解方法

Facing music recommended total variation non-negative matrix decomposition method

目前的音乐推荐系统以考虑歌曲特征和情景上下文因素为主来进行推荐,但选取特征的干扰因素较多,使得噪声干扰较大。为此,提出一种面向音乐推荐的全变差非负矩阵分解方法,通过综合考虑众多因素的影响并借助全变差减少噪声误差。该方法以优化损失函数为目标,在达到全局最优的同时,提高预测的准确度。通过在真实数据集的实验表明,预测的准确性上有显著提高,尤其对于模糊类型的歌曲也能有较好的推荐效果,更好地满足了移动音乐服务的个性化需求。

The current music recommendation system is based on the characteristics of the song and the contextual factors.However,there are many interference factors in the feature selection,which makes the noise jamming.To solve this problem,this paper proposed a method of total variation non-negative matrix factorization for music recommendation,by taking into account the influence of many factors and by means of total variation to reduce noise error.In order to improve the accuracy of prediction,the objective of the method was to optimize the loss function.The real data set experiments show that significantly improve the prediction accuracy,especially for fuzzy types of songs can also have a better recommendation effect,better meet the personalized needs of mobile music service.