融合信任用户间接影响的个性化推荐算法

A personalized recommendation algorithm incorporating trusted users' indirect influence

为了解决推荐系统中固有的数据稀疏性和冷启动问题,通常会采用一些额外的与用户或是项目有关的信息。提出了一种新颖的基于矩阵因子分解的推荐算法,其结合了其他用户对于活动用户未来评分的间接影响作用,并进一步将社交网络中的信任关系融入到算法中。同时,为了避免学习参数时过度拟合,引入了一种加权的正规化因子。最后针对一般情况和冷启动情况,分别在Epinions数据集和Ciao数据集上进行了实验。实验结果表明,相比于其它相关算法,本算法在推荐准确性上有了很大的提高,同时能更好地解决相关问题。

To address the inherent data sparsity and cold-start problem in the recommender system, additional sources of information about users or items are usually adopted. We propose a novel matrix factorization based recommendation algorithm which integrates the indirect influence of other users on active user＇s future ratings. Furthermore, we incorporate the trust relation in social network into our algorithm. In addition, we introduce a weighted regularization factor to avoid over-fitting when learning parameters. We conduct experiments on Epinions dataset and Ciao dataset in normal cases and cold-start cases. Experimental results demonstrate that the recommendation accuracy of our algorithm is greatly improved in comparison with other counterparts and it can better handle the problems concerned.