摘要
【目的】改进基于用户的协同过滤算法以缓解因数据稀疏、用户共同评分稀少所导致的问题,进而提高评分预测的精度。【方法】提出结合用户打分时间发现具有相似打分行为的用户,并将用户评分方差相似性融入到相似度的计算中,使得目标用户在最近邻的选取上更加合理。【结果】实验结果表明,相较基于用户的协同过滤算法,新算法的平均绝对误差降低约2%,在一定程度上改善了推荐系统的推荐效果。【局限】该算法仅在MovieLens数据集上进行了实验测试,还需要在其他数据集上进行检验。【结论】本文算法能够有效地提高推荐精度,具有一定的可行性和现实意义。
[Objective] This paper tries to solve the problems facing traditional collaborative filtering algorithm due to sparse data and few users＇ common scores, and then improve the accuracy of the score prediction systems. [Methods] First, we identified users with similar scoring behaviors based on their scoring time. Second, we integrated the similarity of user score variance to the calculation of similarity. [Results] The new algorithm, which reduced the MAE by 2% compared to the traditional algorithm, improved the performance of recommendation system. [Limitations] The proposed algorithm was only examined with the MovieLens dataset, which needed to be expanded to other datasets. [Conclusions] The proposed algorithm can improve the effectiveness of recommendation systems.
作者
李道国
李连杰
申恩平
Li Daoguo Li Lianjie Shen Enping(School of Information Engineering, Hangzhou Dianzi University, Hangzhou 310018, China School of Management, Hangzhou Dianzi University, Hangzhou 310018, China)
出处
《现代图书情报技术》
CSSCI
2016年第9期65-69,共5页
New Technology of Library and Information Service
基金
浙江省自然基金项目"技术知识特性
整合
知识能量与组织学习对企业间合作创新能力关联性研究"(项目编号:LY12G01002)的研究成果之一
关键词
协同过滤
数据稀疏
相似评分用户评分方差相似性
最近邻
Collaborative filtering
Data sparsity
Similarity score
User rating variance similarity
Nearest neighbor
