基于近邻协同过滤算法的相似度计算方法研究

Research on Similarity Measure in Neighborhood Based Collaborative Filtering

基于近邻的协同过滤算法是目前最为成功的个性化推荐算法之一,相似度计算方法是其核心内容.然而,该算法通常存在冷启动,数据稀疏性以及在相似度计算的时候只能使用共同评分之类的问题.为了有效改善上述算法推荐质量不高的问题,本文介绍了巴氏系数(Bhattacharyya Coefficient),结合修正余弦相似度、杰卡德相似度和皮尔森相关系数的优点,提出一种考虑了项目全局评分信息和局部评分信息的多重相似度融合的协同过滤算法(CFSI).在数据集MovieLens上的实验表明,CFSI在一定程度上有效缓解了数据稀疏性和冷启动问题,并且提高了推荐质量.

Similarity measure is the core for neighborhood based Collaborative Filtering (CF) which is one of the most successful approaches for personalized recommendation algorithm.In order to improve the problems of neighborhood based CF in terms of user cold-start, Data sparsity and similarity calculation method which relies only on co-rated ratings, by utilizing Bhattacharyya coefficient and combining the advantages of adjusted-cosine similarity, Jaccard similarity and Pearson correlation, we proposed an collaborative filtering algorithm of similarity integration(CFSI) considering both the global ratings and the local ratings in this paper.We experimented with datasets in MovieLens.Experimental results showed that the new approach has improved user cold-start and data sparsity problem and the accuracy of recommendation.