一种基于改进置信度传播的个性化推荐算法  

A Personalized Recommendation Algorithm Based on Improve Belief Propagation

作　　者：龚安[1] 孙育红 GONG An;SUN Yuhong(College of Computer & Communication Engineering, China University of Petroleum (East China), Qingdao 266580)

机构地区：[1]中国石油大学(华东)计算机与通信工程学院

出　　处：《计算机与数字工程》2019年第9期2137-2140,2195,共5页Computer & Digital Engineering

基　　金：国家油气重大专项(编号:2017ZX05013-001)资助

摘　　要：基于RWR的方法作为一种TOP-N推荐解决了传统推荐算法遇到的大多数问题,但它只考虑了均匀结点,并且在矩阵分解时需要极大的空间代价。若将置信度传播算法引入则不会存在这些问题,然而传统置信度传播算法用全局结点来计算目标结点置信度,时间复杂度较大,因此,提出将置信度传播算法改进优化后应用于个性化推荐算法中,以用户和项目为两个结点集合,通过自适应大小区域内的结点计算目标结点的置信度,根据最终结点置信度推荐相应项目给目标用户。通过与传统置信度传播算法的对比实验,得到了在最优精度下的相应参数设置,然后与基于RWR的方法和基于项目的协同过滤算法进行比较,实验结果表明,提出的算法要优于上述算法。The RWR-based method as a TOP-N recommendation solves most of the problems encountered by the traditional recommendation algorithm,but it only considers homophily of nodes and requires a great space cost in the matrix decomposition. It would not exist these problems if applied Belief Propagation algorithm to the personalized recommendation system. However,the time complexity is large when the traditional Belief Propagation algorithm calculates target node’s confidence through the global nodes. Therefore,this paper optimizes the BP algorithm and applies it to the personalized recommendation algorithm. The users and the projects are two sets of nodes. Confidence of the target node is calculated by the nodes in the adaptive size region. Then the corresponding project is recommended to the target user according to the final node confidence. In the experiment,the corresponding parameter setting under the optimal precision is obtained through the comparison with the traditional Belief Propagation algorithm,and the algorithm of this paper with the RWR-based method and the project-based collaborative filtering algorithm are compared,the results show that the proposed algorithm is superior to the above algorithm.

关 键 词：个性化推荐 置信度传播算法 结点置信度 自适应大小区域 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]