[1]黄巧文,周宽久,费 铮,等.基于模糊聚类的多视图协同过滤推荐算法[J].计算机技术与发展,2023,33(08):14-22.[doi:10. 3969 / j. issn. 1673-629X. 2023. 08. 003]
　HUANG Qiao-wen,ZHOU Kuan-jiu,FEI Zheng,et al.Fuzzy Clustering Based Multi-view Collaborative Filtering Recommendation Algorithm[J].,2023,33(08):14-22.[doi:10. 3969 / j. issn. 1673-629X. 2023. 08. 003]

点击复制

基于模糊聚类的多视图协同过滤推荐算法()

分享到：

《计算机技术与发展》[ISSN:1006-6977/CN:61-1281/TN]

卷:

33

期数:

2023年08期

页码:

14-22

栏目:

大数据与云计算

出版日期:

2023-08-10

文章信息/Info

Title:

Fuzzy Clustering Based Multi-view Collaborative Filtering Recommendation Algorithm

文章编号:

1673-629X(2023)08-0014-09

作者:

黄巧文¹ ; 周宽久² ; 费 铮¹ ; 崔云鹏¹

1. 北京商业机械研究所 信息化研究部,北京 100070;

2. 大连理工大学 软件学院,辽宁 大连 116081

Author(s):

HUANG Qiao-wen1 ; ZHOU Kuan-jiu2 ; FEI Zheng1 ; CUI Yun-peng1

1. Department of Information Technology,Beijing Commercial Machinery Research Institute,Beijing 100070,China;

2. School of Software,Dalian University of Technology,Dalian 116081,China

关键词:

协同过滤; 多视图聚类; 最大熵方法; 推荐算法; 模糊聚类

Keywords:

collaborative filtering; multi-view clustering; maximum entropy method; recommendation algorithm; fuzzy clustering

分类号:

TP181

DOI:

10. 3969 / j. issn. 1673-629X. 2023. 08. 003

摘要:

传统的协同过滤推荐算法在面对多源异构数据时推荐效果差、执行效率低,且难以挖掘用户的关键行为。 针对这种情况,提出基于模糊聚类的多视图协同过滤

推荐算法。 该算法旨在收集用户的多种行为偏好来构建多视图数据,并通过对每种行为进行加权来挖掘多种行为之间的关联信息,提取关键行为信息,进而提升

推荐效果。 为提高推荐结果的精确度,利用多个视图的行为权重和偏好值,提出项目的加权相似性度量方法,同时引入项目同现矩阵以进一步提高相似性度量的

准确性。 为优化相似项目的搜索空间,结合多视图聚类的思想,在传统模糊聚类方法的基础上引入质心约束和最大熵理论,提出一种基于质心约束的多视图熵加

权模糊聚类算法。 此外,为提高算法对非线性数据的处理能力,引入核映射技术将线性不可分的低维特征映射到高维核空间使其变得线性可分,从而提出一种基

于质心约束的多视图加权核模糊聚类算法。 在与较先进的基于聚类的协同过滤推荐算法的比较实验中,所提算法的平均绝对误差提升了 1. 95 百分点,召回命中率提升了 1. 54 百分点。 实验结果表明,所提算法有效地提升了推荐结果的命中率和准确性。

Abstract:

The traditional collaborative filtering recommendation algorithm has poor recommendation effect and low execution efficiencyin the face of multi-

source heterogeneous data, and it is difficult to mine essential information about user behavior. To address thisproblem,a multi-view collaborative?

filtering recommendation algorithm based on fuzzy clustering is proposed. The algorithm aims tocollect various behavioral preferences of users to?

construct multi- view data and mine the correlation information between variousbehaviors by weighting each behavior, thereby improving the recommendation effect. In this algorithm, a weighted similaritymeasurement method for items is proposed by referring to the multi-view behavior?

weights and item preference values,in which a co-occurrence matrix is introduced to further improve the accuracy of similarity measurement. To?

optimize the search space of similar items,inspired by multi-view clustering,a centroid constraint-based multi-view entropy weighted fuzzy clustering algorithm is proposed by introducing centroid constraint and maximum entropy theory. Besides,for improving the algorithm’s ability to process nonlinear data,a kernelmapping technique is introduced to map the linearly inseparable low-dimensional features to the high-dimensional kernel space to makethem linearly separable. In this way, a centroid constraint - based multi - view weighted kernel fuzzy clustering algorithm is furtherpresented. In the comparison experiment with the state-of-the-art clustering-based collaborative filtering recommendation algorithms,the mean absolute error of the proposed algorithm?

is increased by 1. 95 percentage points,and the recall hit rate is increased by 1. 54 percentage points. The experimental results reveal that the proposed algorithm effectively improves the hit rate and accuracy of the recommendation results.

常用功能