摘要
针对现有差分隐私推荐算法中评分矩阵稀疏、相似性计算依赖于共同评分项且忽略负相似性的问题,提出了一种基于树相似性聚类的差分隐私推荐算法。利用决策树信息熵变化量构建用户间树相似性,采用改进的布谷鸟搜索K-means算法基于树相似性矩阵对用户进行聚类,并通过差分隐私指数机制在目标用户所在簇中选取相似邻居用户集合,根据邻居集合推荐预测分值最高的项目。实验结果在MovieLen 100K和Yahoo Music数据集上显示,该算法相比现有算法,F_(1)值分别提高了7.3%和5.4%,在保护用户隐私的前提下有效提升了推荐精确度。
To address issues such as the sparsity of rating matrices,reliance on common rating items for similarity computation,and the neglect of negative similarity in existing differential privacy recommendation algorithms,a Tree Similarity Clustering-based Differential Privacy Recommendation Algorithm is proposed.Tree similarity between users is constructed using the change of information entropy in decision trees.An improved Cuckoo Search K-means algorithm is employed to cluster users using the tree similarity matrix.A differential privacy exponential mechanism is utilized to identify a set of similar neighbor users within the target user’s cluster,and the highest-rated items are recommended based on the neighbor set.Experimental results on the MovieLens 100K and Yahoo Music datasets show that this algorithm improves the F1 score by 7.3%and 5.4%,respectively,compared to existing algorithms,significantly enhancing recommendation accuracy while maintaining differential privacy guaran-tees.
作者
尹春勇
王硕
YIN Chun-yong;WANG Shuo(School of Computer Science,School of Cyber Space Security,Nanjing University of Information Science and Technology,Nanjing 210044,China;School of Software Engineering,Nanjing University of Information Science and Technology,Nanjing 210044,China)
出处
《计算机工程与设计》
北大核心
2025年第8期2240-2247,共8页
Computer Engineering and Design
基金
国家自然科学基金面上基金项目(6177282)。
关键词
差分隐私
推荐系统
决策树
树相似性
隐私保护
聚类模型
协同过滤
differential privacy
recommender system
decision tree
tree similarity
privacy protection
clustering model
collab-orative filtering
