摘要
为了探索深度神经网络在推荐模型中的应用前景,文中构建了一个集成了深度神经网络和矩阵分解的推荐模型,通过与用户的反馈配合使用实现协作过滤和评分预测。用户评分被视为对商业项目的用户明确反馈,然后将其投影到用户和项目的共享空间中,将表示每个用户和每个项目的潜在特征向量作为嵌入形式传递到矩阵分解和多层感知器模型中,最后将这两个结果嵌入进行组合,并传递到另一个神经体系结构中依次为每个用户的商品项目进行评分预测。实验表明,相对于一般的协同过滤模型,文中提出方法的MAE和RMSE分别明显下降,具有良好的预测性能。
In order to explore the application prospects of deep neural networks in recommendation models,a recommendation model is built by integrated deep neural networks and matrix factorization.And collaborative feedback and user feedback is used to achieve collaborative filtering and scoring prediction.User ratings are considered as explicit feedback to users of commercial projects,then projected into the shared space of users and projects.Finally,the potential feature vectors representing each user and each project are passed to the matrix decomposition and multi-layer as an embedded form In the perceptron model,these two results are finally embedded and combined,and passed to another neural architecture to make score predictions for each user’s product items in turn.Experiments show that compared with the general collaborative filtering model,the MAE and RMSE of the proposed method are significantly reduced,and have good prediction performance.
作者
王莉
徐亮
WANG Li;XU Liang(Jiangsu United Vocational and Technical College,Xuzhou Medical Branch,Xuzhou 221116,Jiangsu Province,China)
出处
《信息技术》
2020年第9期66-69,共4页
Information Technology
基金
江苏联合职业技术学院科研项目(B/2018/07/037)。
关键词
神经网络
深度学习
协同过滤
矩阵分解
neural network
deep learning
collaborative filtering
matrix decomposition
