Semantic Categorization of Web Services Based on Feature Space Transformation

Abstract

Automatic semantic web service annotation mechanisms are required for enabling more efficient and accurate search and discovery of services on the web. In this context new mechanisms are necessary for improving the overall accuracy of the semantic characterization process, preserving the overall performance at an acceptable level. Existing semantic categorization mechanisms take into account all tokens that are included in web service description documents thus resulting in poor performing categorization tasks. This paper demonstrates how a significant improvement in performance can be achieved by applying the Bayes' theorem for transforming the feature space in order to decrease its dimension, without sacrificing prediction accuracy. Experimental evaluation of our approach with respect to other feature selection techniques shows that the former achieves better prediction accuracy, as well as performance when the categorization of a web service in an application domain is concerned.