Approximate Matching of Service Description Using Ontologies and Rules

Web services, which are heterogeneous application accessible over the Internet, may be integrated into more sophisticated compound services. Features of a service such as its functionalities can be characterized and recorded in its service description as a basis for determining if the service satisfies some particular needs. Expressed with a suitable ontology language, service descriptions become machine-interpretable, making it possible to discover and compose services automatically. We are then faced with a situation when no perfect services can be found because the requirement is too strict. Fortunately, perfect matches may not be necessary in most cases, and the requester may be satisfied with “good enough” services. In this thesis, we propose an approximate matching scheme, which returns reasonable substitute services when no exact matches can be found. In our scheme, service descriptions and requirements are both expressions coded with concepts predefined in some ontology language, and service matching is reduced to subsumption checking. To find approximate matches, a requirement is loosened by replacing primitive concepts in the original expression with substitute concepts. Substitute concepts are selected against the similarity values derived from how many features they have in common with the original ones. Substitute concepts are also found by relation compositions which can be revealed with domain-specific inference supported by rule engines. We apply the approximate scheme in a trip planning system where service descriptions are coded with OWL-DL and subsumption checking is done by Racer. OWL-DL, which implements concrete domains with datatypes, disallows user defined datatypes and therefore concepts cannot be constrained with upper/lower bounds of concrete domains. To represent intervals in a service description, we propose two approaches to model quantitative relations where upper/lower bounds are defined with concept subsumptions and object properties, and inference problems in concrete domains such as linear inequality is translated into subsumption checking. With our approximate matching scheme, the trip planning system is able to find exactly matched or approximately matched services, and chooses among them against the ranks automatically produced by our ranking mechanism to compose his own trip.