Supervised learning of term similarities

Abstract

In this paper we present a method for the automatic discovery and tuning of term similarities. The method is based on the automatic extraction of significant patterns in which terms tend to appear. Beside that, we use lexical and functional similarities between terms to define a hybrid similarity measure as a linear combination of the three similarities. We then present a genetic algorithm approach to supervised learning of parameters that are used in this linear combination. We used a domain specific ontology to evaluate the generated similarity measures and set the direction of their convergence. The approach has been tested and evaluated in the domain of molecular biology.