Discovering Semantic Sibling Associations from Web Documents with XTREEM-SP

The semi-automatic extraction of semantics for ontology enhancement or semantic-based information retrieval encompasses several open challenges. There are many findings on the identification of vertical relations among concepts, but much less on indirect, horizontal relations among concepts that share a common, a priori unknown parent, such as Co-Hyponyms and Co-Meronyms. We propose the method XTREEM-SP (Xhtml TREE Mining for Sibling Pairs) for the discovery of such binary "sibling"-relations between concepts of a given vocabulary. While conventional methods process an appropriately prepared corpus, XTREEM-SP operates upon an arbitrarily heterogeneous Web Document Collection on a given topic and returns sibling relations between concepts associated to it. XTREEM-SP is independent of domain and language and does not rely on linguistic preprocessing nor on background knowledge beyond the ontology it is asked to enhance. We present our evaluation results with two gold standard ontologies and show that XTREEM-SP performs well, while being computationally inexpensive.