兼顾用户话语权的改进加权Slope-One协同过滤推荐  

Improved Weighted Slope-One Collaborative Filtering Recommendation Algorithm Considering User Discourse Power

作　　者：陈梅梅[1] 董晨光 王淇 戴伟辉[2] CHEN Mei-mei;DONG Chen-guang;WANG Qi;DAI Wei-hui(Glorious Sun School of Business Management,Donghua University,Shanghai 200051,China;School of Managenent,Fudan University,Shanghai 200433,China)

机构地区：[1]东华大学旭日工商管理学院,上海200051 [2]复旦大学管理学院,上海200433

出　　处：《小型微型计算机系统》2022年第9期1814-1819,共6页Journal of Chinese Computer Systems

基　　金：国家社会科学基金项目(20BGL284)资助;教育部哲学社会科学研究重大课题攻关项目(19JZD010)资助.

摘　　要：针对协同过滤算法中数据稀疏性导致的推荐结果精确度不高的问题,本文提出一种改进的加权Slope-One算法填充评分矩阵.首先,利用用户的评论次数信息区分用户活跃度,然后,在加权Slope-one算法考虑不同项目之间评分用户数量差异影响的基础上,进一步考虑不同活跃度的用户话语权差异对评分预测的影响,提出了兼顾用户话语权的加权Slope-One算法,最后,基于Movie-Lens和Amazon-Clothes两个不同商品品类的数据集,对4种协同过滤算法进行了不同填充比例和不同最优近邻数情况下的仿真实验.仿真对比发现:在仿真实验确定的最优矩阵填充比例和最优近邻数的情况下,相比加权Slope-One协同过滤、原始协同过滤、基于奇异值分解的协同过滤等推荐算法,引入本文所提出的改进加权Slope-One的协同过滤推荐算法,在数据稀疏度不同的两个数据集上的MAE值都更低,说明本文算法能够有效降低数据稀疏性并达到了提高推荐精确度的目的.In view of the inaccuracy of recommendation results caused by data sparsity in collaborative filtering algorithm,the matrix filling method still needs to be optimized.The user activity is distinguished by the frequency of user′s comments firstly.Based on weighted Slope-One algorithm which considers the influence of the number of scoring users between different items,further considering that users with different activity have different discourse power in rating prediction,an improved weighted slope one algorithm based on user discourse power is proposed in this paper.The simulation experiments based on two datasets with different data sparsity,Movie-Lens and Amazon-Clothes,show that:compared with algorithms of weighted Slope-One,original collaborative filtering and SVD collaborative filtering,the proposed algorithm based on the optimal matrix filling ratio and optimal neighbor number determined by simulation experimen has the lowest MAE values under the two datasets.The proposed algorithm improves the accuracy of recommendation by effectively overcoming the influence of data sparsity.

关 键 词：评分矩阵 数据稀疏性 Slope-One算法 协同过滤 用户话语权 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]