基于社会网络信息流模型的协同过滤算法

Social network information flow model based collaborative filtering algorithm

为提高个性化推荐技术的准确率,首先在多维半马氏过程的状态空间中定义'空状态',得到扩展多维半马氏过程,将其与社会网络分析理论结合,得到社会网络信息流模型,该模型描述了社会网络成员间的信息流动过程。然后基于社会网络信息流模型,提出协同过滤算法SMRR(Semi-Markov and reward renewal)。实验表明,由于综合考虑用户自身偏好和社会网络中其他成员的影响,SMRR的预测准确率明显高于原有算法。

In order to improve the accuracy of personalized recommendation, ＂idle state＂ was firstly introduced into multi-dimension Semi-Markov process, therefore extended multi-dimension Semi- Markov process was obtained. Then by combing the extended multi-dimension Semi-Markov process with social network analysis theory, social network information flow model was established. This model describes the information flow process among social actors. Based on the social network information flow model, a collaborative filtering algorithm, Semi-Markov and Reward Renewal （SMRR）, was proposed. Experiments demonstrate that, regarding accuracy, SMRR performs better than existed algorithms. Considering personalized behavior patterns of actors and the influence of interacting social actors together can improve the accuracy of personalized recommendation.