摘要
为提高基于项目的协同过滤推荐SlopeOne算法的预测结果,在算法的项目相异性计算和评分预测过程中引入个性化情境因素.首先对基于项目的协同推荐方法进行综述,然后针对不同情境下的评分记录进行项目间相异性计算,根据此结果计算检验集中的项目在不同情境下的预测评分,并以预测结果为依据为每个用户得到个性化情境,进而为用户对新的资源项目进行评分预测.最后在标准的Movielens数据集上进行实验,其中,U2的训练集和测试集用来训练个性化情境,其他数据集用来检验算法的预测结果.通过对改进的推荐算法与经典的基于项目的协同过滤算法SlopeOne进行比较,实验数据表明改进后算法的推荐结果有较大提高.
In order to improve the result of item-based collaborative filtering(CF) recommendation approach,this paper incorporates personalized context into the computation of item differences and rating prediction.First,the prior research and the problem of item-based CF approaches are reviewed.Then,item differences according to the ratings in different contexts are calculated.Based on the item differences,ratings are extrapolated for the items in examining dataset.The personalized context for every user is further identified and new items in training dataset are predicted according to the result of the prediction.Finally,an experiment is given to evaluate the proposed approach and it is compared with a typical item-based SlopeOne CF using Movielens dataset.In which,the training and test datasets of U2 are used to obtain personalized context;the other datasets are applied to check the final prediction results.The experimental results show that the proposed approach provides better quality than SlopeOne.
出处
《东南大学学报（自然科学版）》
EI
CAS
CSCD
北大核心
2009年第S1期27-31,共5页
Journal of Southeast University：Natural Science Edition
基金
国家社科基金重大资助项目(ACA07004-08)
中国博士后科学基金资助项目(20080440699)
重庆市自然科学基金资助项目(2008BB2183)
重庆市教育委员会科学技术研究资助项目(KJ071601)
重庆市教育科学"十一五"规划资助项目(2008-ZJ-064)
关键词
协同过滤
推荐算法
项目相异性
情境
个性化
collaborative filtering
recommendation algorithm
item difference
context
personalization
