Research Of Context-aware Web Service Discovery Approaches Fused With Wikipedia Knowledge

The rapid growing of internet has provided massive available information and software resources for software development which has dramatically changed the pattern of software development technology. The development pattern of service composition and service reuse which can deal with the changing business requirements has gradually become the new direction of internet software development technology. Service discovery, which is an essential prerequisite for the realization of service composition and reuse and highly affects whether the true "plug and play" of service can be achieved, has a far reaching influence on the efficiency of internet software development and the quality of the internet software. With the rapid growth of the resources of internet Web Service, how to help users to discover personalized service resources quickly and effectively has become a key issue which urgently awaits to be solved. To satisfy the personalized, diversified and changing dynamicly user requirements of Web service discovery, a context-aware and muti-strategy fusion service discovery framework is proposed first. On this basis, with the intention of satisfying personalized user requirements, and aiming at the problems in current model of centralized Web service discovery, personalized Web service discovery and recommendation under different context is realized by organically integrating the mutiple strategies of Web service discovery and recommendation. The results of relevant experiments verified the feasibility and efficiency of the proposed methods. The research contents and approaches in this thesis are listed as follows:(1 Establishing a context-aware and muti-strategy fusion framework for personalized Web service discovery.Service clustering can boost the ability of service discovery, service recommendation can satisfy the personalized user requirements of service discovery, and contextual knowledge can provide more precise illustration of the user requirements. Fusing above strategies can give full play to their respective advantages and make reparation for each other. As a result, a muti-strategy fusion framework of personalized service discovery is proposed firstly which enhances the ability of service organization and classification and the effects of discovery by service clustering, predicts the preferences of users to provide personalized discovery results by technology of service recommendation, and employs contextual knowledge in the process of service discovery for further depicting users to promote the effectiveness of servcie discovery. This framework, which is user centered, can improve the users’ experience of personalized web service discovery by integrating mutiple strategies.(2) Establishing a service clustering model based on Transfer LearningDue to the semantic sparsity of current service description files, the traditional model, such as bag of words, has many limitations under the context of semantic sparsity. Therefore, a new service clustering approach enriching the semantic of the service description based on external knowledge Wikipedia is proposed to improve the clustering result. This approach which utilizes Transfer Learning to fuse Wiki knowledge, tag information and service descriptions constructs the hidden topic representations of the services in the latent space which will be applied for clustering. This approach, which can integrate the Wiki knowledge and tag information in source domain, map the knowledge in both source domain and target domain into a same feature space to facilitate the task of clustering, not only enchrich the semantic of the services but also achieve the clustering organization of service and finally establish the foundation of improving the efficiency of service discovery.(3)Proposing a time aware service recommendation approach based on implicit feedbacksComparing with that searching technology of service can only provide the same searching results for all users based on a same searching request, the recommendation technology of service often provides more personalized results for users. The methods of service recommendation based on explicit feedbacks are not widely applied in the field of service computing and one of the main reasons is that the explicit feedbacks between users and services are difficult to collect. As a result, focusing on implicit feedbacks we propose to construct a pseudo rating generating mechanism which is applied to mapping the implicit preferences of users to explicit ratings. Then the contextual information of time in the implicit feedbacks is detailed extracted and analysed to track and discover the characteristic of the implicit preferences of users evolving over the time. Finally an approach of time aware service recommendation based on probabilistic matrix factorization using implicit feedbacks is constructed for users which implicit feedbacks are used and contextual information are integrated. This approach can provide time aware and personalized service recommendation for users.(3)Proposing an online service recommendation approach based on implicit feedbacksAs then traditional recommendation systems, the recommendation system of Web service has cold start problem too in which the user cold start problem is one of the important issues. Leveraging more contextual information can help depict the profiles of users so as to provide much more user information and enhance the ability of the recommendation system. The PMF model based on online mode which can capure the dynamic changing of user preferences is utilized to provide online and personalized service recommendations. We further analyse the contextual information in the service recommendation system and construct a service recommendation system using "most popular" strategy which integrates contextual information of time and location to provide most popular service recommendations for the cold start users. After interaction between user and the recommendation system, the new implicit feedbacks generated in the interaction are streamly processed by using the online service recommendation method. Finally a context aware online service recommendation system is proposed.This thesis focuses on the key issue that facing with the diversified service resources in the internet and pluralistic user requirements, how to use contextual information to fine depict the characteristics of service and user requirements to conduct effective personalized service discovery. By the composition of mutiple strategies and using many kinds of external knowledge, a user centered approach is proposed to take various forms to provide services users need and can support the personalized service discovery effectively. Finally, based on relevant research methods, experiments are designed and conducted to verify the efficiency of the approaches. The research content 1 which is the main research framework of this thesis is the entrance of the follow-up studies. The research contents 2,3,4 which group together to content 1 provide different solutions of personalized service discovery from different aspects and using respective methods.