Abstract: A method is provided for data processing with a view to determining the main attributes of a product defined by a description including a plurality of words in which, for each word, one determines whether it belongs to one or more predetermined sets or glossaries. Then, for each word that belongs to one or more glossaries: one assigns to the word a plurality of codings produced depending on the predetermined glossary to which the word belongs and on the other words in the description and/or the glossaries to which the other words belong; and one uses a dual-class categoriser to analyse all these codings produced for the word in order to determine whether the word is a main attribute of the product.