Inferring Data Transformation Rules to Integrate Semantic Web Services

Abstract

OWL-S allows selecting, composing and invoking Web Services at different levels of abstraction: selection uses high level abstract descriptions, invocation uses low level grounding ones, while composition needs to consider both high and low level descriptions. In our setting, two Web Services are to be composed so that output from the upstream one is used to create input for the downstream one. These Web Services may have different data models but are related to each other through high and low level descriptions. Correspondences must be found between components of the upstream data type and the downstream ones. Low level data transformation functions may be required (e.g. unit conversions, data type conversions). The components may be arranged in different XML tree structures. Thus, multiple data transformations are necessary: reshaping the message tree, matching leaves by corresponding types, translating through ontologies, and calling conversion functions. Our prototype compiles these transformations into a set of data transformation rules, using our tableau-based $\cal ALC$ Description Logic reasoner to reason over the given OWL-S and WSDL descriptions, as well as the related ontologies. A resolution-based inference mechanism for running these rules is embedded in an inference queue that conducts data from the upstream to the downstream service, running the rules to perform the data transformation in the process.