综合项目权值分配与时间相关的协同过滤模型

A collaborative filtering recommendation algorithm combining items weight allocation and time dependence

根据长尾理论,被反馈次数少的项目所包含的反馈信息并不少于被反馈次数较高的,传统的协同过滤算法中缺乏考虑冷门项目在最终的项目推荐过程中的影响力,对此,提出了一种改进的协同过滤推荐模型。通过对冷门项目的分析筛选,在用户相似性计算时提高冷门项目所占的比重,以体现用户的个性和兴趣。此外,考虑到时间效应的影响,在兴趣预测过程中引入时间因子。实验结果表明,提出的算法能提高寻找最近邻居的准确性,从而改善协同过滤的推荐质量。

According to the long tail theory,the items with fewer feedbacks do not necessarily contain less information than those with more feedbacks.In the traditional collaborative filtering algorithms,the influences from unpopular items are usually ignored in the process of the eventual recommendation.To address this problem,an improved collaborative filtering recommendation model is proposed.By evaluating the unpopular items analytically,the weight of these items should be improved in calculating users＇ similarities,so as to reflect users＇ personalities and interests.Moreover,taking into account the impact of the time dependence,the time factor is introduced during the prediction of interests.Experimental resuits show that the algorithm can raise the accuracy of searching the nearest neighbors,and improve the recommendation quality of the collaborative filtering.