融合用户相似度与项目相似度的加权Slope One算法

Integrating User Similarity and Item Similarity into Weighted Slope One Algorithm

个性化推荐技术为人们处理信息过载问题提供了一种有效的解决方式.协同过滤是推荐技术常用的算法之一,本文研究的Slope One算法就是一种基于项目的协同过滤推荐算法,但是,它并未考虑到用户相似度及项目相似度的问题.因此,本文提出5种新的融合用户相似度与项目相似度的加权Slope One算法,即分别使用信任因子和Jaccard方法找出具有影响力的用户,使用Pearson方法找出当前项目的相似项目.最后,在Epinions和Movielens数据集上的对比实验结果表明,融合Jaccard和Pearson的混合算法在数据集稀疏以及邻居数目较少的情况下,仍能获得较高的推荐准确度.

The personalized recommendation technology provide people an effective method to solve the problem of information overload. Collaborative filtering is one of the key algorithm of recommendation technology, In this paper, the Slope One algorithm is a kind of item-based collaborative filtering recommendation algorithm, However, it doesn＇t take user similarity and item similarity into consideration when it works. Therefore, this article puts forward five new method to integrate user similarity and item similarity into weighted Slope One algorithm. Using trust and Jaccard to find the influential users, Pearson to find the similar items of current item, respectively. Experiments on the well-known datasets Epinions and Movielens show that the algorithm weighted by Jaccard and Pearson in the case of sparse datasets and less neighbor achieves great improvement of prediction accuracy.