Apprentissage d'un ensemble pré-structuré de concepts d'un domaine : l'outil Galex

Abstract

The huge amount of electronic textual information increases exponentially just as easily as archives and working documents in academic organizations, in administration and in firms. A solution for structuring this mountain of textual database is to build a knowledge model to index this information. One way can be obtained by data extraction and classification producing conceptual indexing by knowledge acquisition. Traditionally the classification methods of Data Analysis were adapted while used for the classical table of data under an object/characteristics/value format. We present Galex (Graph Analyzer for LEXicometry) which develops structuration of knowledge by a term clustering method. This structuration synthetizes the content of information providing the mapping data to information filtering or hypertextual navigation on similar documents. Galex aims at taking into account the nature of the data to which it is applied : natural language. The complexity of natural language is well known: sense ambiguity, multiple grammatical construction of sentence, style, term creationaWe show through integration of poorly defined, though useful as concept, ontology, term and corpus, notions that clustering can be improved by adding linguistic knowledge. We base our approach on typical phenomena such as graph-statistical relations between terms, scheme relations in a context and canonical reduction of variants.