The Key Technoloy Research On Trustworthy Web Services Discovery Based On Machine Learning Methods

With the rapid development of Internet and clound computing, more and more researchers focus on web service technology. Among the research topics of Web service, the trustworthy web service discovery has become an increasingly important and challenging subject. Due to the distributed, open, uncertain and uncontrollable properties of the network environment where web service deploys, how to rapidly provide accurate, secure and effective service for users becomes especially important. Thus, the research on trustworthy web service discovery is a prospective, basic and strategic task, and plays a significant role in the software development.This dissertation focus on the trustworthy WSDL(Web Service Description Language)-based web service discovery, and study the several key issuses such as service representation, service clustering, service matching and service Qo S prediction by introducing the popular kernel-based learning, metric learning and sparse learning theories in the machine learning field. The main research work and contributions of this dissertation are as follows:(1) A lightweight semantic web service discovery model based on WordNet and concept semantic dimension reduction is proposed. Firstly, by introducing the WordNet and Latent Semantic Index(LSI) into the lexical vectors in Vector Space Model, this model obtains the low-dimensional compact semantic vectors which well describe web services’ true semantic characterization; Secondly, this model employs a flexible kernel-based similarity matching mechanism to better evaluate the similarity between the real-world web services, which overcome the traditional matching mechanism’s limitation of ignoring the statiscal properties of the samples distribution.(2) A Kernel Batch SOM neural network(KBSOM) algorithm is presented, and then based the KBSOM neural network, a clustering facilitated web services discovery model is constructed. This model employs the KBSOM neural network to automatically cluster those web services with the same or similar function into one class, and then extract the class ID service to facilitate web services registry and discovery. In this way, the model minimizes the services discovery duration, and ultimately alleviates the efficiency of web service discovery.(3) A supervised feature extraction and adaptive similarity evaluation mechanism for web service is propsed. Firstly, by introducing the inverse category frequency factor into the traditional unsupervised TF-IDF term weighting mechanism, we present a supervised term weighting method integrating with the prior class information; Secondly, by employing the large-margin metric learning theory, we propose a web service similarity learning method based on the prior class information to well assess the similarity of the real-world web services. Both the two proposed methods can efficiently distinguish the relevant and independent features, and reasonably reveal the intrinsic data distribution. Finally, the constructed web service discovery model based on the supervised service representation and adaptive similarity evaluation mechanism verifies the validity of these two methods.(4) A trustworthy web service Qo S prediction mechanism based on matrix completion with structural noise is proposed. Firstly, by introducing the L2,1-norm regularization constraint, we formulate the problem of web service Qo S prediction as the problem of matrix completion with structural noise. Secondly, to efficiently solve the problem of matrix completion with structural noise, this dissertation proposes a robust OSMCSN algorithm(Operator Splitting based Matrix Completion with Structural Noise). The proposed algorithm not only can exactly detect the positions where the data is contaminated, but also can effectively predict the missing Qo S properties. Finally, experimental results performed on a real public dataset demonstrate the feasibiligy of our proposed QoS prediction mechanism.(5) An asymptotically stable multi-valued many-to-many Gaussian associative memory model(M3GAM) is proposed, and the M3GAM’s asymptotical stability is proved theoretically in both synchronous and asynchronous update modes. In addition, the M3GAM’s storage capacity and error-correcion capability are also analyzed and investigated. Finally, the experimental results about association based image retrieval verify the M3GAM’s robust performance. In the future, we will focus on employing the proposed M3 GAM to facilitate the intelligent management of the concept semantic network for Web service.