基于概率主题模型的物联网服务发现

发布时间：2018-03-28 04:30

本文选题：物联网　切入点：服务发现　出处：《软件学报》2014年08期

【摘要】：针对物联网中服务数量的大规模性、服务描述的异构性以及设备服务的资源高度受限性和移动性等特点,提出了一种基于概率主题模型的物联网服务发现方法.该方法的主要特点是:1)利用英文Wikipedia构建高质量的主题模型,并对类似短文本的服务文本描述进行语义扩充,使主题模型能够更有效地估计服务文本描述的隐含主题;2)提出利用非参数主题模型学习服务文本的隐含主题,降低模型训练时间;3)利用服务隐含主题对服务进行自动分类和文本相似度计算,快速减少服务匹配数量,加速服务文本相似度计算;4)提出能够同时支持WSDL-based和RESTful两种物联网服务的signature匹配算法.实验结果表明:与现有的物联网服务发现方法相比,该方法的准确率(precision)和归一化折损累积增益(NDCG)都有较大幅度的提高.
[Abstract]:Aiming at the large scale of service quantity in the Internet of things, the heterogeneity of service description, the resource limitation and mobility of device service, etc. In this paper, a probabilistic topic model based service discovery method for Internet of things is proposed. The main feature of this method is: (1) using English Wikipedia to construct a high quality topic model, and extending the semantic representation of service text similar to short text. So that the topic model can more effectively estimate the implied topic described by the service text. (2) A non-parametric topic model is proposed to learn the implied topic of the service text. To reduce the training time of the model, the service can be classified automatically and the text similarity can be calculated by using the hidden topic of the service, and the number of service matching can be reduced quickly. (4) A signature matching algorithm, which can support both WSDL-based and RESTful, is proposed. The experimental results show that the proposed algorithm is more efficient than the existing methods. Both the accuracy of this method and the normalized loss cumulative gain (NDCG) are greatly improved.
【作者单位】：中国科学院数学与系统科学研究院;中国科学院大学;高可信软件技术教育部重点实验室(北京大学);
【基金】：国家自然科学基金(61232015,91318301)
【分类号】：TP391.44;TN929.5

【相似文献】