基于位置与信誉感知深度神经网络的云服务质量预测

Cloud Service Quality Prediction Based on Location and Reputation Aware Depth Neural Network

随着众多具有相似功能的云服务的出现,为用户选择最适合的服务的难度也越来越大。协同过滤是处理这一问题的常用方法,但是它面临着数据稀疏性和可信度这两个普遍的问题,这大大降低了其预测准确性。为了正确和系统地解决这个问题,提出了一种基于位置与信誉感知深度神经网络的云服务质量预测方法。首先,结合地理位置信息采用K-means聚类过滤不可信用户;其次,在可信用户和服务之间构建一个深度神经网络模型,采用信誉感知网络嵌入来学习用户的潜在表示,并通过全连接层输出用户特征向量和服务特征向量;最后,通过矩阵点乘求和对两者进行融合,输出预测值。在公共数据集WSDream上进行实验的结果表明,与给出的其他方法相比,所提方法在预测准确率上有明显提高。

With the emergence of many cloud services with similar functions,it is becoming increasingly difficult to select the most suitable service for users.Collaborative filtering is a common method to deal with this problem.However,it faces two common problems of data sparsity and credibility,which greatly reduce its prediction accuracy.In order to solve this problem correctly and systematically,a cloud service quality prediction method based on location and reputation aware depth neural network is proposed.First,the untrusted users are filtered by K-means clustering based on geographic location information.Then,a deep neural network model is constructed between trusted users and services,employing reputation-aware network embeddings to learn potential representations of users and outputting user feature vectors and service feature vectors through a fully connected layer.Finally,the two are fused by matrix dot product summation and the predicted values are output.The experimental results on the public dataset WSDream indicate that the proposed method has a significant improvement in prediction accuracy compared to the other methods given.