基于深度学习的网络资源优先协同过滤推荐

Web Resource Priority Collaborative Filtering Recommendation Based on Deep Learning

为帮助用户快速、准确地获取所需的网络资源为目的,提出基于深度学习的网络资源优先协同过滤推荐方法。首先分析推荐过程的组成架构,将其划分为信息处理、用户建模、推荐算法等多个功能模块。然后通过共现关系分别描述网络资源与用户之间的关联性,从而建立资源-用户特征矢量模型,获取表示全面特征的目标函数。将能够反映丰富物理量的张量引入到神经网络中,合并一阶张量与二阶张量,得出神经网络的输出信号,再采用反向传播算法对神经网络做深度学习,获得输出层、隐含层与输入层误差。计算整体损失函数的偏导数,直到损失函数收敛,结束学习过程,从而生成优先协同过滤推荐结果。仿真结果证明,上述方法可以更有效的获取资源与用户特征,可为用户推荐合适的网络资源。

A network resource priority collaborative filtering recommendation method based on deep learning is studied in order to help users obtain the required network resources quickly and accurately. First of all, the constituent architecture of the recommendation process was systematically investigated. The architecture consists of information processing, user modeling, recommendation algorithm and other modules. Secondly, based on the co-occurrence relationship, the correlation between network resources and users was described, respectively, thus founding the resource user feature vector model for obtaining the objective function representing the comprehensive features. Then, the tensor reflecting rich physical quantities was introduced into the neural network. The first-order tensor and the second-order tensor were combined to obtain the output signal of the neural network. The backpropagation algorithm was introduced to do deep learning for the neural network in order to obtain the errors of the output layer, hidden layer and input layer. The partial derivative of the overall loss function was calculated circularly until the loss function converges, which meant the termination of the learning process, generating the preferred collaborative filtering recommendation results. The simulation results show that this method can effectively obtain resources and user characteristics, and can recommend appropriate network resources for users.