摘要
针对推荐系统中相似偏好用户数量较少情况下的一类新群体冷启动问题开展研究,基于多元相关分析,对传统的尺度与平移不变(Scale and Translation Invariant,STI)的协同过滤推荐方法进行改进,提出一种基于项目相关度的STI推荐方法,以应对推荐系统中的新群体冷启动问题.在此基础上,基于Movie Lens数据集对所提出的方法进行了性能分析,结果表明,所提出的方法较Pearson方法及ST1N1方法在解决新群体冷启动推荐的过程中具有更高的推荐准确率.
The cold start problem recently becomes a hot topic on Recommender systems ( or RS ) ,especially for new community cold start problem. The number of similar users for the new community cold start user is very low. It makes the traditional Collaborative Fil- tering ( or CF ) based recommendation method cannot meet the requirement of accuracy for new community recommendation. In this paper, a degree of item correlation based Scale and Translation Invariant ( or STI ) method is proposed to solve this problem. It com- bines the degree of item correlation and STI so as to predict the score of un-voted items for new community cold start user. The related experimental results show that,the method has a high recommendation accuracy.
出处
《小型微型计算机系统》
CSCD
北大核心
2015年第3期450-453,共4页
Journal of Chinese Computer Systems
基金
国家自然科学基金面上项目(71373125)资助
教育部博士点基金新教师基金项目(20113204120011)资助
教育部人文社会科学基金项目(10YJC790395)资助
江苏省高校哲学社会科学基金项目(2013SJB6300051)资助
国家级大学生实践创新训练计划项目(201310298025Z)资助
关键词
推荐系统
冷启动
新群体
项目相关度
尺度与平移不变
recommender system
cold start
new community problem
degree of item correlation
scale and translation invariant ( STI }