融合社交行为和标签行为的推荐算法研究

针对传统推荐算法忽略用户社交影响、研究角度不全面和缺乏物理解释等问题,提出一个融合社交行为和标签行为的推荐算法。首先用引力模型计算社交网络中用户节点之间的吸引力来度量用户社交行为的相似性;其次通过标签信息构建用户喜好物体模型,并使用引力公式计算喜好物体之间的引力来度量标签行为的相似性。最后,引入变量融合两方面信息,获取近邻用户,产生推荐。采用Last.fm数据集进行实验研究,结果说明推荐算法的准确率和召回率更高。

A Study of Recommended Algorithms Integrating Social Behavior and Labeling Behavior

In view of the traditional recommendation algorithm ignoring the impact of social behavior of users, the incomprehensive research perspective and lack of physical explanation, a recommendation algorithm was proposed that integrated social behavior and tagging behavior of users. Firstly, the attractiveness between user nodes in social network was calculated by gravity model to measure the similarity of users" social behavior. Secondly, the user"s favorite object model was constructed by label information, the gravitation formula was also used to calculate the gravitation between favorite objects to measure the similarity of tagging behavior. Finally, the paper introduced the variables to weigh the proportion of two similar values, and then got the set of neighbors and generated recommendations. Experimental results using Last.fm dataset showed that the proposed algorithm had higher precision and recall.