基于动态标签偏好信任概率矩阵分解模型的推荐算法

Recommendation Algorithm Based on Dynamic Label Preference Trust Probability Matrix Decomposition Model

为提高推荐算法性能,解决数据稀疏和冷启动因素造成的推荐精度不高的问题,提出一种改进的协同过滤推荐算法。基于三元组表示形式,利用标签集、用户集和项目资源集构建标签、用户以及项目之间的动态联系,并进行信任值评分矩阵的计算,使用信任评分矩阵融合协同推荐过程,构建概率矩阵分解模型,并基于期望最大法进行模型的求解。实验结果表明,与采用基于余弦、皮尔逊相关系数和启发式相似度模型的算法相比,该算法具有较低的绝对误差均值以及较高的覆盖率、精度与召回率。

In order to improve the collaborative performance of recommendation algorithms and solve the problem of low recommendation accuracy caused by sparse data and cold start, an improved collaborative filtering recommendation algorithm is proposed in this paper. Based on three tuple representation, it uses the tag set, the user set and the project resource set to construct the dynamic relationship among the labels, the users and the project, and it also computes the trust value score matrix. It uses the trust rating matrix fusion collaborative recommendation process to construct probability matrix decomposition model and solves the model the expectation maximization method. Experimental results show that compared with other algorithms which are based on cosine, Pearson correlation coefficient and heuristic similarity model, this algorithm has lower absolute mean error as well as higher coverage rate ,precision and recall rate.