摘要
协同过滤推荐算法中,相似性计算对推荐质量起着至关重要的作用.针对传统算法相似性度量方法的不足,提出一种融合用户兴趣相似性和评分相似性的协同过滤推荐算法.算法将用户的评分项目信息映射为兴趣向量,计算用户的兴趣相似性,并使用用户兴趣相似性和评分相似性进行两次融合,从而对传统相似性度量仅仅依靠用户评分进行相似性计算引起的误差进行修正.在计算相似性过程中,通过引入专家信任度的概念对用户未评分项目进行评分预测填充,从而降低由于数据稀疏性引起的评分相似性计算误差.实验结果表明,该方法在推荐覆盖率和准确性上相对传统算法有所提升.
In collaborative filtering recommendation algorithm, the similarity calculation plays a vital role in the recommendation quali- ty. Aiming at the shortcomings of the traditional method of similarity measurement algorithm, this paper proposes a collaborative filte- ring recommendation algorithm based on the interest similarity of users and the similarity of scores. The algorithm maps the user's score project information to the interest vector, calculates the user's interest similarity, and uses the user's interest similarity and the score similarity for two times fusion, thus, fix the error caused by the similarity calculation of the traditional similarity measure only depends on the user's score. In the process of computing the similarity,predicts and fills of the item did not score by introducing the concept of trust expert, thereby reducing the error caused by the similarity of the data sparsity. Experimental results show that the method recommended in the coverage and accuracy has improved compared to the traditional method.
出处
《小型微型计算机系统》
CSCD
北大核心
2017年第2期357-362,共6页
Journal of Chinese Computer Systems
基金
国家自然科学基金项目(61300216)资助
河南省科技攻关项目(132102210123)资助
河南省基础与前沿技术研究项目(132300410332)资助
河南省教育厅科技攻关计划项目(13A520321)资助
关键词
协同过滤
相似性
兴趣相似性
专家信任度
数据稀疏
collaborative filtering
similarity
interest similarity
expert trust
data sparsity