Research On Key Technologies For Web Service Selection Based On Multi-dimensional Information Mining

With the development of Web service technology, many enterprises encapsulate their internal application systems as Web services, and publish them onto the network. According to their own business needs, users can invoke proper Web services directly, or composite them as a new application system. However, as the increasing number of similar function Web services in practical applications, combined with various complicated factors, such as distribution characteristics (including distribution characteristics of published information and location), coarse grain of composite services, users’preferences and so on, these can seriously affected the quality of Web service selection. Therefore, taking exploit of the existing multi-dimensional information on network, we perform the research on key technologies of efficient and effective Web services selection. The main achievements are as follows:First, improve the efficiency of candidate services search. In this paper, we propose a semi-supervised learning method for Web service functional category mining (SLM-WFCM for short), based on Web service description files (WSDL) and a small amount of classified information. We employ the elements related with service function in WSDL, including portType, operation, message and type as the processing object, and utilize the reference relationship between these elements to calculate the similarity between Web services and their operations. Meanwhile, combined with a small amount of prior knowledge, we propose a semi-supervised co-cluster algorithm to mine the functional category of Web services. Experimental results demonstrate that SLM-WFCM can significantly improve the average accuracy of Web service functional category mining with a small amount of monitoring information, and it can effectively support user to search the candidate services set.Second, predict the missing QoS. In this paper, we propose a QoS prediction method based on the multi-dimensional feature mining (QPM-MFE for short), by taking exploiting of multi-dimensional historical QoS information. QPM-MFE first tackles the multi-dimensional historical QoS information with Gaussian normalization, and extracts the features of Web services from these normalized QoS information with non-negative matrix factorization. Then, it learns the mapping relationship between the extracted features and historical QoS information of the target user with multi-output support vector machine based on differential evolution. At last, it predicts the missing QoS information for target user. Comprehensive empirical studies demonstrate that QPM-MFE can not only predict the multi-dimensional QoS information simultaneously, but also improve the accuracy of prediction.Third, improve QoS of selected composition on condition of real-time performance guarantee. In this paper, we propose a multi-constraint service selection method based on local approximate filter (MSSM-LAF for short), by taking exploiting of the potential relationship among candidate services. We first filter out part of the unsatisfied constrain services with local approximate filter, which can not only reduce the search space, but also estimate the local fitness of the rest candidate services. Then we design a directed particle swarm algorithm for searching. In order to improve the search ability of our algorithm, we redefine the particles update operator, and design a dynamic parameter adjustment strategy, a fitness function, and local fit first mutation mechanism. Finally, taking local fitness of candidate services as boot information, we can search for the optimal combination with the proposed particle swarm algorithm. Experimental results demonstrate the effectiveness of MSSM-LAF.