Automatic Discovery And Composition Of Information-Providing Web Services

As the wide adoption of Service-Oriented Architecture (SOA) and Web Service technology, reusing existing facilities in the service level will play a critical role during the development of applications in future. At the same time, as the number of services increases and integration of different systems prevails, it will be more and more important for the reuse procedure that the ability to discover and compose services automatically.In recent nearly ten years, the problem of automatic discovery and composition of services has been intensively studied under the name of Semantic Web Services, which leaded to a relatively complete theory and large amounts of methods. However, most of these theory and methods are targeted to all types of Web Services in the same manner. Although such treatment is general enough, it caused the system inefficient because of the complexity of the problem itself, which made the automatic discovery and composition hard to be employed in reality. In recent several years, some more focused research are carried out that gave up some certain expressive strength in return for better efficiency and scalability. This paper takes this style of research and focuses only on the information-providing type of Web Services. The purpose is to present out capability models and algorithms of matchmaking and composing more fitted to specific requirements in particular scenarios for the automatic discovery, invocation and composition. Detailedly, the main work and innovations are as follows:(1) For current usage environment of Web Services, we utilize the TF-IDF information retrieval technology in conjunction with the WordNet semantic dictionary to design a service search method capable of identifying simple semantic and not relying on centric registry:Although the automatic discovery isn't gotten, this method avoids the non-practical yet assumption of annotating services with the ontology which is taken and depended heavily on by all Semantic Web-based technologies. It improves the recall and precision of service retrieval in real environment comparing to keywords-based methods, and could be used as an auxiliary means in the manual development circle.(2)For the Semantic Web environment, we put forward a capability model which uses OWL ontology to express the precondition and effect of services. Although the model doesn't apply to the world-altering type of services, it can depict the functionality of information-providing services well, and can be supported very extensively in Semantic Web. The corresponding matchmaking criterion between two services is also presented that fits the procedure of automatic discovery and invocation. In addition, we study a specific scenario which could utilize such procedure to bring more values to users.(3) We identify the similarity between information-providing services and the data queries and accordingly study another important application of automating various tasks of such type services, i.e. to add the information-providing services into the data integration system which originated from the database field. In such scenario, the service is treated as a special database view and invoked dynamically, along with ordinary data sources, to answer the data query committed at runtime. Aimed at this scenario we put forward another capability model, Uniform Query, of information-providing services in which such type service is expressed as a query on the OWL data-type properties of some individuals which are connected by object properties, the inputs and outputs of the service are annotated as data-type properties of individuals rather than the individuals themselves. This model made the automatic service discovery and invocation more precise and easier to understand than normal service capability models, which made it more suited to the data integration scenario. Compared to SPARQL, the data query language in Semantic Web, this model keeps most frequently-used expressive properties and facilitates significantly the design of the user interface to input queries.(4) We analyze several main formal systems of dynamic knowledge, realize their accordance with respect to the operational semantic under their varied external forms, and then put forward a comprehensive capability model for information-provi-ding services based on the accordance. As to the form, this model relaxes the requirement exerted by normal models in literature that input and output parameters should have semantic types and then combines such information, if presented, into the precondition and effect; as to the semantic, this model utilizes the special features of such type services and links the effect to the precondition directly with the implication operator to form a First-Order Logic formula, which is succinct but able to erect a rigid axiomatic semantic for the model. The model obviously expresses out two features of information-providing services that they never modify the world states and inputs/outputs might not have semantic types. It is completely compatible with the two models in (2) and (3) for two different scenarios, accords with normal models in the operational semantic in the sense that information-providing services are regarded as data queries in a given world state, and keeps basic compatibleness with main Semantic Web Services standard proposals OWL-S and WSMO. Based on this model, we derived out an explicit criterion for the correctness of composition plans in the composition problem of information-providing services.(5) On the base of above comprehensive capability model, we put forward an automatic composition method for information-providing services that uses a query rewriting algorithm, MiniCon, in data integration research. Concretely we change MiniCon algorithm a little to support the subClasOf/subProperty reasoning in OWL, check the rewriting results with respect to inputs and preconditions of services according to the correctness criterion in (4), and can generate BPEL scripts capable of being deployed and invoked directly for found composite services. This method adequately utilizes the two features of information-providing type of Web Services, employs an "enumerate-combine" procedure which matches predicates first and then binds variables. Compared to normal direct search-based composition methods, it has a complemented performance characteristic and especially fits to the data integration type of scenarios in which many inputs/outputs have no semantic types. The result of performance test on a typical class of problems shows that the method has outstanding practicability.