Evaluating Some Heuristics to Find Hyponyms Between Ontologies

摘要

The discovery of hyponymy relationships between elements belonging to two different ontologies is a common task when integrating semantic information from different sources. Most of the existing work combines several strategies or criteria to take a global decision (if there is an hyponymy relationship or not) for each pair of entities. However, such heuristics are typically not separately evaluated. In this paper, we evaluate two techniques used in the discovery of hyponymy relationships based on shared properties and on similar entity names. We also use translation dictionaries to deal with cross-lingual ontology pairs and lexical resources to increase the number of names of an entity. Our experiments make it possible to identify some limitations of ontology sets commonly used as benchmarks and to argue that more complex lexical similarity measures are needed.