Recherche improbable d'une homogène diversité: Le débat sur l'identité nationale

Abstract

In this paper, we compare the effects of two methods of morphological correction of corpus coming from the web on ALCESTE analysis made with the IRAMUTEQ software. From the 18 240 contributions to the debate on national identity, we compare the initial corpus with a manually corrected one and with a semi-automatic correction method based on a particular used of the Hunspell corrector. The three corpora (initial, automatic and manual) are used in two different hierarchical clustering: one that retain the 1 500 most frequent words and one that retain the 3 000 most frequent words. The comparison of results obtained on each corpus shows that the automatic correction that we proposed allow to come significantly closer to a manual one.