摘要： 现有的协同过滤算法在计算用户间或项目间相似度时,由于数据集稀疏导致相似度差值过小,难以找出真正的相似用户与相似项目。为此,提出一种融合影响因子的加权协同过滤算法。利用基于用户间与项目间共同评分用户个数的影响因子,修正用户及项目相似度,并分别定义基于项目与用户的预测评分计算算法,将项目间与用户间的共同评分项作为加权系数,得出最终的预测评分算法,根据最终预测评分,采用TopN算法进行推荐。在真实数据集上的实验结果表明,该算法在不同邻居数上的平均绝对误差小于0.78,明显提高了推荐质量。

关键词: 影响因子, 协同过滤, 相似度, 加权算法, 预测评分, 推荐系统

Abstract: The traditional collaborative filtering algorithms may lead to very small difference between each similarities when evaluating the similarity of users or items because of the datasets’ sparsity,which are one major cause of the problem that cannot find the real similar users and items.To solve this problem,this paper modifies the computation of similarity by introducing impact factor which is based on the amount of common ratings users and items.In addition,it presents user-based and item-based prediction computation algorithms for this new calculation method,and takes the common evaluations between users and items as the weighted factor.A weighted prediction computation algorithm based on the common evaluations is employed to compute the final evaluations which are used to get the TopN recommendation.Experimental results by using a real dataset show that the proposed algorithm can achieve lower mean absolute error than 0.78 at different amount of neighbors and better quality of recommendation.

Key words: impact factor, collaborative filtering, similarity, weighted algorithm, prediction rating, recommender system

中图分类号: 

• TP311