基于双注意力深度学习的在线资源推荐  

Online resource recommendation based on double attention deep learning

作　　者：李会芬[1,2] 焦小刚 黄丽霞 Li Huifen;Jiao Xiaogang;Huang Lixia(School of Information Technology,Ningxia Industrial and Commercial College,Yinchuan 750021,China;School of Software,Tongji University,Shanghai,200092,China;School of Information Engineering,Ningxia University,Yinchuan 750021,China;Business School,Beifang University for Nationalities,Yinchuan 750021,China)

机构地区：[1]宁夏工商职业技术学院信息技术学院,宁夏银川750021 [2]同济大学软件学院,上海200092 [3]宁夏大学信息工程学院,宁夏银川750021 [4]北方民族大学商学院,宁夏银川750021

出　　处：《南京理工大学学报》2023年第2期221-227,共7页Journal of Nanjing University of Science and Technology

基　　金：宁夏教育厅项目(宁职教成办[2018]102号)。

摘　　要：为了提高在线资源推荐的性能,采用深度学习卷积神经网络(Convolutional neural network,CNN)进行资源推荐,同时对资源-用户特征进行双注意力机制特征提取,以进一步提高推荐精准度。对资源-用户特征进行编码并初始化,分别进行通道注意力机制运算和空间注意力机制运算。将两个注意力机制的运算结果加权求和得到新的用户-资源特征。建立基于CNN的在线资源推荐模型,并以资源和用户的最小特征差作为损失函数进行迭代优化,从而求解出CNN网络参数。通过双注意力机制的用户-资源特征输入到CNN模型,并执行训练以获得符合用户需求的推荐结果。试验结果表明,通过合理设置双注意力机制通道数及卷积核尺寸等参数,可以有效提高双注意力CNN的推荐性能。与常用资源推荐算法相比,所提算法在推荐准确度及稳定性方面均具有一定的提升。To improve the performance of online resource recommendation,a deep learning convolutional neural network(CNN)is used for resource recommendation,while dual-attention mechanism feature extraction is performed on resource-user features to further improve recommendation accuracy.First,the resource-user features are encoded and initialized,and the channel attention mechanism operation and the spatial attention mechanism operation are performed respectively.The results of the two attention mechanisms are weighted and summed to obtain the new user-resource features.Then,a CNN-based online resource recommendation model is built and the minimum feature difference between resource and user is used as the loss function for iterative optimisation to solve the CNN network parameters.Finally,the user-resource features by the dual-attention mechanism are fed into the CNN model and training is performed to obtain recommendation results that meet the user’s needs.The experimental results show that the recommendation performance of the dual-attention CNN can be effectively improved by reasonably setting the parameters such as the number of channels of the dual-attention mechanism and the size of the convolutional kernel.Compared with the commonly used resource recommendation algorithms,the proposed algorithm has a certain improvement in recommendation accuracy and stability.

关 键 词：资源推荐 卷积神经网络 双注意力 通道注意力 空间注意力 

分 类 号：TP391.3[自动化与计算机技术—计算机应用技术]