Presented at: 9th Extended Semantic Web Conference (ESWC2012)
This paper explores the issue of detecting concepts for ontology learning from text. We investigate various metrics from graph theory and propose various voting schemes based on these metrics. The idea draws its root in social choice theory, and our objective is to mimic consensus in automatic learning methods and increase the confidence in concept extraction through the identification of the best performing metrics, the comparison of these metrics with standard information retrieval metrics (such as TF-IDF) and the evaluation of various voting schemes. Our results show that three graph-based metrics Degree, Reachability and HITS-hub were the most successful in identifying relevant concepts contained in two gold standard ontologies.
Keywords: concepts, machine learning, ontology, voting theory
Resource URI on the dog food server: http://data.semanticweb.org/conference/eswc/2012/paper/research/218
Explore this resource elsewhere: