摘要
认为建立种子集引导用户评分是解决协同过滤推荐系统新用户冷启动问题的方法之一。尝试将关联度引入种子集的构建策略,提出基于多属性综合评价的种子集策略,并利用公开数据集MovieLens设计实验,模拟推荐系统的新用户环境,对比不同种子集策略的预测准确度和成功率。实验结果表明,在更符合实际推荐系统需求的少量种子集情况下,考虑种子之间的关联性可以改善推荐效果。
This paper believes that building a seed set to guide users＇ evaluation is one of the solutions to resolve the cold start problem of new users. An integrated approach based on multiple - attribute comprehensive assessment is brought up by bringing the relevance into the construction strategy of seed set. User environment of the recommender system is simulated through the experiment designed according to public data sets, in order to compare the accuracy and success rate of forecast. The result shows that the recommender effectiveness is improved by considering the relevance of seeds, if a small amount of seeds sets which are more suitable to the real recommendation system are adopted.
出处
《图书情报工作》
CSSCI
北大核心
2013年第5期124-128,共5页
Library and Information Service
基金
中国石油大学(北京)研究生教育质量与创新工程基金项目"机构知识仓储建设及其长效服务机制研究"(项目编号:2011B008)研究成果之一
关键词
协同过滤
推荐系统
新用户冷启动
评分引导
collaborative fihering
recommender system
cold start of new users
rating eliciting
