摘要
针对传统基于单分类的推荐算法容易陷入"单指标最优"的困境和推荐精度低的问题,提出一种融合K-最近邻(KNN)和Gradient Boosting(GBDT)的协同过滤推荐算法。该算法利用K-最近邻法过滤出目标用户的多组候选最近邻居集,并综合集成学习的优点,采用多分类器对多组推荐结果进行集成。在相似度计算公式中引入了若只有单个用户评价的物品权重,以此获得更多目标用户的潜在信息。实验结果表明,该算法有效缓解了目标用户与候选最近邻居集之间的数据集稀疏性,提升了推荐精度。
In order to solve the problem of easy to fall into the trouble of"single indicator optimal"and low recommendation accuracy for tradition collaborative filtering recommendation algorithm based on single-class classification,hybrid collaborative filtering recommendation algorithm based on KNN-GBDT is proposed.The algorithm used K-nearest neighbor method to filter out multiple candidate nearest neighbors and comprehend advantages of ensemble learning.It uses multiple classifier to ensemble multiple recommendation results.In order to get more potential information about the target user,consider items weight with only single user rating into the similarity calculation formula.Experimental results show that the algorithm can reduce the sparse of data set which is formed by the target user and the candidate nearest,and improves the recommendation accuracy.
作者
王永贵
李倩玉
WANG Yonggui;LI Qianyu(College of Software,Liaoning Technical University,Huludao,Liaoning 125105,China)
出处
《计算机工程与应用》
CSCD
北大核心
2021年第9期103-108,共6页
Computer Engineering and Applications
基金
国家自然科学基金面上项目(1772249)。
关键词
协同过滤
推荐算法
多分类器
相似度
collaborative filtering
recommendation algorithm
multiple classifier
similarity