Wikipedia HTML']HTML Structure Analysis for Ontology Construction

Previously, the main problem of information extraction was to gather enough data. Today, the challenge is not to collect data but to interpret and represent them in order to deduce information. Ontologies are considered suitable solutions for organizing information. The classic methods for ontology construction from textual documents rely on natural language analysis and are generally based on statistical or linguistic approaches. However, these approaches do not consider the document structure which provides additional knowledge. In fact, the structural organization of documents also conveys meaning. In this context, new approaches focus on document structure analysis to extract knowledge. This paper describes a methodology for ontology construction from web data and especially from Wikipedia articles. It focuses mainly on document structure in order to extract the main concepts and their relations. The proposed methods extract not only taxonomic and non-taxonomic relations but also give the labels describing non-taxonomic relations. The extraction of non-taxonomic relations is established by analyzing the titles hierarchy in each document. A pattern matching is also applied in order to extract known semantic relations. We propose also to apply a refinement to the extracted relations in order to keep only those that are relevant. The refinement process is performed by applying the transitive property, checking the nature of the relations and analyzing taxonomic relations having inverted arguments. Experiments have been performed on French Wikipedia articles related to the medical field. Ontology evaluation is performed by comparing it to gold standards.