摘要
为了解决传统协同过滤算法的准确度因评分缺失产生的剧烈变化以及冷启动问题,本文提出了一种新的降噪自编码器推荐算法。该方法将注意力机制与辅助信息共同融入降噪自编码器中对评分与交互数据进行处理。首先针对用户交互项目动态分配注意力以学习用户偏好,然后再次通过注意力机制学习隐藏层向量、用户偏好、辅助信息的权重以获得完整评分矩阵。在公开数据集上对该算法进行实验仿真,观察算法性能。结果表明,该算法有效利用了辅助信息且准确度有明显提高。
In order to solve the problem that the changes in accuracy of the traditional collaborative filtering algorithm due to the lack of ratings and cold start,a new denoising autoencoder algorithm is proposed for recommendation.In this algorithm,attention mechanism and auxiliary information are integrated into the denoising autoencoder model to process the ratings and interactive data.Attention is dynamically allocated to user interaction items to learn user preferences,and then the weights of hidden layer vector,user preferences,and auxiliary information are learned again through the attention mechanism to obtain the complete ratings matrix.The algorithm is simulated on the open data set to observe its performance.The results show that the algorithm uses the auxiliary information effectively and the accuracy is improved obviously.
作者
王倩雯
张延华
付琼霄
李萌
李庆
Wang Qianwen;Zhang Yanhua;Fu Qiongxiao;Li Meng;Li Qing(Faculty of Information Technology, Beijing University of Technology, Beijing 100124)
出处
《高技术通讯》
EI
CAS
北大核心
2020年第12期1234-1242,共9页
Chinese High Technology Letters
基金
国家自然科学基金(61571021,61901011)
中国博士后科学基金面上项目(2018M640032)
北京市博士后科研活动经费(ZZ2019-73)
朝阳区博士后科研活动经费(2019ZZ-4)
北京工业大学基础研究基金资助项目。
关键词
降噪自编码器
注意力机制
推荐算法
辅助信息
深度学习
denoising autoencoder
attention mechanism
recommendation algorithm
auxiliary information
deep learning
