An ontology-based information extractor for data-rich documents in the information technology domain

Abstract

This paper presents an information extraction method, suitable for data-rich documents, based on the knowledge represented in a domain ontology. The extractor combines a fuzzy string matcher and a word sense disambiguation (WSD) algorithm. The fuzzy string matcher finds mentions of terms combining character-level and token-level similarity measures dealing with non-standardized acronyms and inconsistent abbreviation styles. We propose a new character-level edit distance sensitive to prefixes called root distance and a token-level similarity algorithm for fuzzy acronym detection. Additionally, a WSD strategy using an ontology-based semantic relatedness measure is used to solve the inherent ambiguity of some entities. The WSD module finds a sense combination over all the document length optimizing the document semantic coherence. Our approach seems to be suitable to extract information from data-rich documents describing Orly one main object (i.e. product) by document. The results showed a precision of 78.9% with 99.5% recall using documents and an ontology related to laptop computers domain.