面向推荐多样性改进的概率传播模型

针对基于二部图的概率传播( ProbS)模型以优化推荐列表的精确度为目标，而忽略了推荐多样性的问题，提出了改进的概率传播( iProbS)模型。 iProbS将项目得分预测过程分解为资源的3步传播过程，每步传播包含传播概率和传播损耗。设计传播概率时，考虑的因素是用户评分；设计传播损耗时，则分别考虑了项目的度、用户熵和邻居项目。通过在2个常用数据集MovieLens和Netflix上的大量不同实验，证明了iProbS算法在推荐准确率、推荐整体多样性、推荐个体多样性以及销售平衡4个方面均比ProbS模型性能更好。最后按不同的推荐步骤分析了iProbS算法的计算复杂度。

Probabilistic Spreading Models for Improving Recommendation Diversity

Bipartite-graph based probabilistic spreading ( ProbS) algorithms often focus on optimizing the accuracy of recommendation lists while ignoring diversity, another key property to evaluate the quality of recommendation results. In order to deal with this problem, an improved probabilistic spreading ( iProbS) algorithm is proposed in the present paper. The iProbS algorithm divides the recommendation process into three steps of resource spreading, and each resource spreading step constrained by spreading probability and spreading cost simultaneously. Users ’ scores rating on items are applied to compute spreading probability, at the same time, the degree of items, entropy of users, and the neighbors of items are considered for computing spreading costs. Extensive experiments on two widely used data sets ( from MovieLens and Netflix) show that iProbS can effectively improve recommending accuracy, aggregate di-versity, individual diversity, and sales balance of recommendation lists. Finally, computational complexi-ties of iProbS are studied from its different computing steps.