Measuring conceptual distance using WordNet: the design of a metric for measuring semantic similarity

Abstract

This paper describes the development of a metric for measuring the semantic distance or similarity of words using the WordNet lexical database. Such a metric could be of use in development of search engines and text retrieval systems, tasks for which the richness of natural language can cause difficulty. Further, such a metric can prove invaluable to psycholinguists who wish to study lexical semantic similarity or speech errors (specifically malapropisms). The paper first explores an adjusted distance metric, a la Rada et al. 1989, and the problems such a metric presents. Additional analysis shows that adjustments can be made to such a distance metric using density calculations, both based on depth within the network and based on local density. The paper ends with a discussion about automating the task of identifying regions within the semantic space over which density calculations can be made.