Using combined list hierarchy and headings of HTML documents for learning domain-specific ontology

Abstract

HTML pages contain unstructured and diverse information. However, these documents lack semantics and are not machine understandable. Semantic webs aim to add formal semantics to web data, whereas ontology provides formal semantics to a domain and is thus considered a foundation of semantic webs. Domain ontologies can be constructed manually, but this process is tedious and inefficient. Thus, this study presents an ontology learning (OL) model to create domain ontologies automatically from a set of HTML pages. The key insight of this research is that it combines the list structure and headings of HTML pages to recognize the ontology vocabulary. The approach also incorporates synonym relationships with ontology and allows the semantic interpretation of ontology concepts. We implement the proposed OL approach to build sports ontology from a collection of sports domain HTML documents. The new sports ontology is tested using FaCT++ reasoner; results show no inconsistency in the ontology. Furthermore, experts evaluate the successful mapping of HTML lists and headings to the ontology vocabulary. The proposed OL approach performs effectively and achieves 92.7% and 95.4% precision values for list and heading mapping, respectively.