面向稀疏数据的协同过滤算法相似度  

Similarity of collaborative filtering algorithm in sparse data

作　　者：赵文涛[1] 冯婷婷 ZHAO Wen-tao;FENG Ting-ting(School of Software,Henan Polytechnic University,Jiaozuo 454003,China;School of Computer Science and Technology,Henan Polytechnic University,Jiaozuo 454003,China)

机构地区：[1]河南理工大学软件学院,河南焦作454003 [2]河南理工大学计算机科学与技术学院,河南焦作454003

出　　处：《计算机工程与设计》2024年第6期1728-1734,共7页Computer Engineering and Design

基　　金：国家自然科学基金项目(61503124);河南省科技厅科技攻关基金项目(182102310935)。

摘　　要：针对数据稀疏加剧导致传统相似度模型的推荐准确性低的问题,提出一种混合的协同过滤相似度模型。引入Jensen-Shannon(JS)散度作为基函数,利用全局评级概率分布衡量用户间评级偏好相似度。定义融合评级值的结构型相似度作为权重因子,针对用户的共同评级项目设计差异化的相似度计算方式,提高相似用户的区分度,得到基于相对区间跨度的相似度。在不同稀疏度数据集上与7种具有代表性的相似度方法进行对比实验,其结果表明了所提方法在预测和推荐准确性指标上均有良好性能。Aiming at the problem of the low recommendation accuracy of traditional similarity models due to the increasing data sparsity,a hybrid similarity model was proposed.The Jensen-Shannon(JS)divergence was introduced as the basis function,and the global rating probability distribution was used to measure the similarity of rating preferences among users.The structural similarity containing rating values was defined as a weighting factor.The differentiation similarity calculations were designed for the co-rated items to improve the differentiation degree of similar users,the similarity based on the relative interval span was obtained.Experiments were conducted with seven representative similarity methods on different sparsity datasets.Compared with seven representative similarity methods on different sparsity data sets,the results verify that the proposed method has good performance in both prediction and recommendation accuracy indicators.

关 键 词：稀疏数据 协同过滤 相似度 散度 用户评级偏好 全局结构 相对区间跨度 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]