一种局部和全局用户影响力相结合的社交推荐算法

Local and global user influence combined social recommendation algorithms

传统的协同过滤推荐系统认为用户之间的行为相互独立,忽视了用户之间的影响关系.而用户的历史行为数据不同、社交网络关系不同,其相互之间存在的影响力不同.为了分析用户的社交影响力在推荐中所起到的作用,通过历史行为数据和社交网络结构分析用户的局部影响力和全局影响力,分别提出基于局部影响力和基于全局影响力的模型以及两种影响力综合的模型.通过在真实的数据集上的实验表明,与以往方法相比,本文提出的基于影响力的三种模型在推荐精度上有一定提升,且在稀疏的数据集上基于全局影响力的模型和综合模型提升效果比更明显.

With the rapid development of e-commerce,it becomes more and more difficult for users to find their favorite products and recommender systems play an important role in solving that problem.Traditional collaborative filtering recommender systems did not consider the relationship among different users.However,different users have different historical behaviors and social relations which lead to various influences.Due to the fact that the influence from users with similar interest are more powerful,the user similarity can be used to describe local influences on users.And many users like following in the footsteps of their leaders or friends,so the importance of users in the social relation can be used to describe global influence in the network.In order to analyze the users social influence on recommendation,this paper proposes three kinds of recommendation models considering users historical behaviors and social relations.The three kinds of models are respectively local-based influence model,global-based influence model and combining model.Experimental results on real datasets show that these new algorithms are superior to the com-parison algorithms on recommendation precision.Moreover,the global-based influence model and the combining model have obtained greater improvement on sparse datasets.