Complement Recognition-Based Formal Concept Analysis for Automatic Extraction of Interpretable Concept Taxonomies from Text

Abstract

The increasing scale and pace of the production of digital documents have generated a need for automatic tools to analyze documents and extract underlying concepts and knowledge in order to help humans manage information overload. Specifically, since most information comes in the form of text, natural language processing tools are needed that are able to analyze the sentences and transform them into an internal representation that can be handled by computers to perform inferences and reasoning. In turn, these tools often work based on linguistic resources for the various levels of analysis (morphological, lexical, syntactic and semantic). The resources are language (and sometimes even domain) specific and typically must be manually produced by human experts, increasing their cost and limiting their availability. Especially relevant are concept taxonomies, which allow us to properly interpret the textual content of documents. This paper presents an intelligent module to extract relevant domain knowledge from free text by means of Concept Hierarchy Extraction techniques. In particular, the underlying model is provided using Formal Concept Analysis, while a crucial role is played by an expert system for language analysis that can recognize different types of indirect objects (a component very rich in information) in English.