基于时间加权连接的完全三部图推荐算法  

Complete Tripartite Graphs Recommendation Algorithm Based on Time-weighted Connections

作　　者：朱永华[1] 林举[1] 吴志国[1] 沈熠[1] 

机构地区：[1]上海大学计算机工程与科学学院,上海200444

出　　处：《计算机技术与发展》2015年第10期44-48,共5页Computer Technology and Development

基　　金：上海市科学技术委员会资助项目(14590500500);上海市自然科学基金(15ZR1415200)

摘　　要：基于社会化标签的个性化推荐已成为推荐领域关注的热点问题,但面临着用户信息丢失、时间效应和用户兴趣迁移等一系列挑战。文中基于用户行为数据建立用户-物品-标签完全三部图模型,并基于此提出个性化物品推荐算法。该方法首先对用户兴趣动态迁移现象进行分析,其次综合考虑用户-物品-标签三者关系,提出了完全三部图模型,接着引入时间加权连接权重来构建新的连接关系矩阵,最后在此基础上运行Mass Diffusion推荐算法,通过综合两个方向的物质扩散来获得推荐结果。实验结果表明,文中算法能够通过反映用户兴趣的动态迁移,有效地提高推荐的准确性和多样性。Personalized recommendation based on social tagging has become a key research topic in the field of recommendation. Current recommending methods, however, are facing a series of challenges, such as the loss of user information, the effect of time and user interest migration. A new personalized recommendation algorithm based on user-item-tag complete tripartite graph model derived from the user bebavior is proposed. Firstly, research on dynamic migration of user interest is carried out. Secondly, user-item-tag complete tripartite graph model is proposed with comprehensive consideration of user-item-tag relationships. Time-weighted connections is employed to construct the new connection matrix. Finally, MassDiffusion algorithm is executed to carry out personalized recommendation based on the model through combining two directions of mass diffusion. Experimental results demonstrate that the algorithm can effectively improve the accuracy and diversity of recommendation through reflecting, the dynamic migration of user interest.

关 键 词：个性化推荐 社会化标签 完全三部图 时间加权连接 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]