基于切比雪夫一阶截断式的谱卷积协同过滤  

Collaborative filtering of spectral convolution based on Chebyshev first-order truncation

作　　者：王嘉豪 梅红岩[1] 刘鑫 李晓会[1] WANG Jia-hao;MEI Hong-yan;LIU Xin;LI Xiao-hui(School of Electronic and Information Engineering,Liaoning University of Technology,Jinzhou 121001,China)

机构地区：[1]辽宁工业大学电子与信息工程学院,辽宁锦州121001

出　　处：《计算机工程与设计》2022年第12期3406-3413,共8页Computer Engineering and Design

基　　金：国家自然科学基金青年基金项目(61802161);辽宁省自然科学基金项目(20180550886)。

摘　　要：为解决个性化推荐协同过滤中存在用户关联项目过少,而引起的用户冷启动问题,提出一种基于切比雪夫优化的谱卷积协同过滤推荐算法。将用户-项目二部图转换到谱域,通过切比雪夫一阶截断式建立深度前馈神经网络,优化卷积过程,省略拉普拉斯矩阵复杂的特征分解,缩短模型训练时间,在谱域中快速发现用户与相关项目之间的隐性关联信息。经过实验验证,该方法对提升推荐结果的准确性有着较为明显的帮助,更为有效挖掘用户与项目间关联信息。To solve the problem of cold start of users caused by few user-associated items in personalized recommendation collaborative filtering,a recommendation algorithm based on Chebyshev optimization of spectral convolution collaborative filtering was proposed.The user-item bipartite graph was transferred into a spectral domain.Through Chebyshev first-order truncation,deep feedforward neural network was built,the convolution process was effectively optimized,complex eigenvalues of Laplacian model were omitted,the training time was shortened and the deep connection between the user and the project was found effectively in the domain.Experimental results show that the proposed method can improve the accuracy of the recommendation results and mine the associated information between users and projects more effectively.

关 键 词：推荐系统 图神经网络 谱域图卷积 前馈神经网络 反向传播 切比雪夫 协同过滤 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]