摘要
如何在海量的信息中找到用户感兴趣的个性化信息是推荐系统要解决的核心问题。基于协同过滤的思想,从传统的矩阵分解到深度学习,推荐系统走过了漫长的发展历程。现有的方法不能显式地从数据中建模协同过滤信号。图卷积神经网络可解决这个问题,并且得到高质量的嵌入表示。传统的图卷积方法没有考虑不同项目对用户的差异性影响,通过引入注意力机制捕获具有差异性的用户与项目特征信息,可以进一步提高嵌入质量。实验结果显示,该方法具有更好的推荐效果。
Finding the information that users are interested in the massive amount of information is the core problem to be solved by the recommendation system.Based on the idea of collaborative filtering,from traditional matrix decomposition to deep learning,recommendation systems have gone through a long development process,The existing methods cannot explicitly model collaborative filtering signals from data.Graph convolutional neural networks can be used to solve this problem well and get high-quality embedding representations.The traditional graph convolution method does not consider the differential impact of different items on users.Therefore,attention mechanism is introduced to capture different users and item characteristic information,which further improves the embedding quality.The experimental results show that this method has better recommondation effects.
作者
黄俊辉
杨艳
HUANG Jun-Hui;YANG Yan(Heilongjiang University School of Computer Science and Technology,Harbin 150080,china;Heilongjiang University Key Laboratory of Database and Parallel computing of Heilongjiang Province,Harbin 150080,china)
出处
《黑龙江大学工程学报》
2022年第1期60-67,共8页
Journal of Engineering of Heilongjiang University
基金
黑龙江省自然科学基金联合引导项目(LH2020F043)。
关键词
图卷积神经网络
深度学习
注意力机制
推荐系统
Graph convolutional neural network
deep learning
attention mechanism
recommendation system
