Abstract: In response to the problem that sources of index weight assignment in traditional quality evaluation methods are not comprehensive, the service description ontology is divided into two levels（the shared ontology, the exclusive ontology） and QoS ontology with the multilayer structure “abstract-application-metric” is constructed for the object description & data acquisition of QoS metrics. The double-direction metric model DM-QSM based on depth belief network with regression model is established, in which the service description information and similar service history data are used as training sample data set to train DM-QSM forward by combining with user feedback on the DM-QSM reverse tuning, to achieve the adaptive adjustment of indexes weights with user preference degree. Finally, by using the programmable modeling environment NetLogo as the experimental platform, the public service data set QWS as the training sample set and the e-commerce application service as the test sample set, the feasibility and effectiveness of DM-QSM are verified.

Key words: multilayer ontology, Quality of Service（QoS） metrics, depth belief network

摘要： 针对传统质量评估模式中指标权重赋值依据单一的问题，首先将服务描述本体分为共享本体和专属本体两个抽象层次，构建具有“抽象-应用-度量”多层结构的QoS本体，用于QoS度量的对象描述和数据采集；然后建立基于深度信任网络和回归模型的双向度量模型DM-QSM，将服务描述信息和类似服务历史数据作为训练样本数据集对DM-QSM进行正向训练，再结合用户反馈对DM-QSM进行逆向调优，以实现QoS度量指标权重及其偏好度的自适应调节。最后选用可编程建模环境NetLogo为实验平台、公共服务数据集QWS为训练样本集、电子商务应用服务为测试样本集，验证了DM-QSM的可行性和有效性。

关键词: 多层本体, 服务质量（QoS）度量, 深度信任网络