Propagation de polarités dans des familles de mots: impact de la morphologie dans la construction d’un lexique pour l’analyse de sentiments
Nuria Gala Pavia, Caroline Brun
Lexical resources are essential for many natural language applications (for example, opinion mining from corpora). However, building them entails different problems (cost, coverage, etc.). In this paper, we wanted to verify whether morphological information about derivation could be used to automatically annotate semantic information. Starting from a resource that groups words into morphological families in French, we have built a lexicon with polarities for 4 065 words from an initial seed set of manual annotated adjectives. The results obtained show that spreading polarities is accurate for 78.89% of the families with a unique adjective. The lexicon obtained also improves the results of the opinion mining system on different corpora.
TALN Conference, Grenoble, France, 4-8 June 2012.