Investigating a Method for Automatic Construction and Population of Ontologies for Services: Performances and Limitations

摘要

Ontological engineering is a complex process, involving multidisciplinary skills. The Semantic Web, and more specifically Semantic Web Services spreading suffer from the difficulty of producing an ontology sufficiently detailed to be able to correctly describe the data flows exchanged between services. These data are often described using sector-specific vocabulary. Linking these descriptions to external knowledge sources capable of unifying them is often a complex process, requiring adequate sources to be found and properly used. In this paper, we investigate a method combining existing string distance measurement, NLP-analysis and clustering algorithms for automatic construction and population of an ontology. This method takes services capacities descriptions as only input, without external sources of knowledge. It is tested on a set of more than 10,000 services for 106,000 different measures to classify in an ontology, performances and limitations are exposed.