Enhancement of a subjective word lexicon by semantic tests

Recent considerations in opinion mining, oriented by real applicative tasks, require creating lexical and semantic resources quantitatively and qualitatively rich. In this context, we present a method to automatically enhance a lexicon of subjective terms. The method relies on the indexing of Web documents by a search engine and large number of requests automatically sent. The construction of these requests, linguistically motivated, can infer the value of semantics and axiological aspect of a large number of adjectives, nouns and noun phrases, verbs and verbal phrases of French. We then evaluate the lexicon enhancement by testing the detection of local evaluation in a corpus of 5 000 blogs notes and comments. MOTS-CLÉS : fouille d’opinions, subjectivité, ressource lexico-sémantique, apprentissage.