融合隐含信任度和项目关联度的矩阵分解推荐算法

随着社交网络的发展，融合社交信息的推荐系统在一定程度上解决了协同过滤推荐系统的冷启动和数据稀疏等问题，但是在信任数据稀疏情况下，仍会造成推荐精度降低等问题。为此，提出了一种融合隐含信任度和项目关联度的矩阵分解推荐算法。首先，利用矩阵分解模型将信任数据进行分解，得到用户的潜在被信任矩阵，在此基础上引入用户的影响力，从而提出了基于隐含信任度的推荐模型；然后，为了更好的利用项目间的关联信息，反映项目间的有向性，提出了基于项目关联度的推荐模型；最后，综合两种推荐模型并构建了一种推荐算法TCRMF。实验结果表明，所提算法在评分数据和信任数据稀疏的情况下仍然可以有效地提高推荐算法的精度，具有良好的应用前景。

Matrix factorization recommendation algorithm combing implicit trust and item correlation

With the development of social network, recommendation system fusing social information sovles the problems of cold starting and sparse rating datas of collaborative filtering recommendation system to some extend. Therefore, it still causes problems of making recommendation accuracy decline in the case of the sparse trust datas. Thus, this paper proposed matrix factorization recommendation algorithm combing implicit trust and item correlation. Firstly, it decomposed the trust datas by the matrix factorization model, and obtained the implicit trusted matrix of users. It introduced the influence of users on this basis, and proposed the recommendation model based on implicit trust. Secondly, it proposed the recommendation model based on item correlation in order to make better use of correlation information between items and reflect on orientation between items. Finally, it combined the two kinds of recommendation models, and proposed the TCRMF algorithm. The experimental results show that the proposed algorithm still improves the accuracy of recommendation algorithm effectively in the conditions that score and trust datas are sparse, and has a good application prospect.