Research On Domain Topic Oriented On-demand Services Discovery Approach

With the rapid development of SOA and SaaS technology, service-oriented software development is gradually becoming the mainstream of software development on the Internet. Under this trend, a growing number of Web services are available on the network. Meanwhile, these service resources have the obvious heterogeneous characteristics due to the following two reasons. One is diverse protocols (such as SOAP, REST) followed by services, the other is various service description languages, such as WSDL, OWL-S, natural language text, and so on. The sharp increase of service scale and the heterogeneity of services increase the difficulty in on-demand services discovery for mass users, and also bring a great challenge for software developers to effectively discover and reuse services. Therefore, faced with the abundant and heterogeneous service resources, how to conduct on-demand services discovery becomes a key issue in service-oriented software engineering.However, with the diversification of service description, service type is also gradually diversified. While existing services discovery approaches are short of effective organization of candidate services, the adopted matching strategy are only suitable for one type of service query, and put less emphasis on RESTful services. In addition, the increase of network users leads to users' requirements showing a trend of personalization and diversity, which further complicates on-demand services discovery. To address the problems mentioned above, the key problem that the thesis explores is:faced with widely distributed heterogeneous service resources on the Internet and personalized requirements of users, how to organize services to effectively support on-demand services discovery. Considering this key problem, the main contributions of the thesis are summarized as follows.(1) An iterative services classification approach based on support vector machine is proposed, to effectively organize services from the perspective of domain that each service belongs to. Based on vector space model, different types of service description documents are transformed to the corresponding vectors, and then support vector machine is used to conduct iterative services classification. The proposed approach can support domain oriented services classification for different types of services, and lay foundation for the subsequent services clustering in a specific domain.(2) A services clustering approach based on probability is proposed, to organize services in a specific domain from the perspective of topic that each service belongs to. On the basis of domain oriented services classification, LDA topic model is used to model different types of service description documents, then topic oriented services clustering is conducted by probability. Moreover, according to the relevance between each service and its belonging topic cluster, it ranks services from each topic cluster in a specific domain. This proposed approach can effectively cluster different types of services within a specific domain, which is helpful for services organization from different levels and different perspectives, thus contributing to on-demand services discovery. In addition, it can also provide support for services composition and services recommendation.(3) An on-demand services discovery approach based on "domain-topic cluster-service" matching is proposed. On the basis of domain oriented services classification and topic oriented services clustering, service matching is conducted from different perspectives and different levels, to gradually locate services relevant to user's requirements. More specifically, according to functional similarity computation, user's requirements are located from a specific domain to a specific topic cluster in the located domain. Then, according to functional and nonfunctional similarity computation, as well as the corresponding service filtering strategy, to discover user desired services from the located topic cluster. The proposed approach gradually narrows the size of candidate services by the combination of multi-strategy, which is beneficial for users to rapidly discover the services that satisfy their personalized requirements.(4) The application in cloud services supermarket demonstrates the practical value of on-demand services discovery approach based on services classification and services clustering; furthermore, two potential applications, namely, construction of domain ontology and service tag recommendation, show that the proposed on-demand services discovery approach has good application prospects.In summary, on the basis of domain oriented services classification and topic oriented services clustering, this thesis proposed an on-demand services discovery approach based on "domain-topic cluster-service" matching, to gradually locate user desired services by multi-strategy, so as to effectively support on demand services discovery. Meanwhile, such services organization contributes to the management of service resources from different perspective, different level and different granularity, and provides support for services composition and services recommendation.