Reconhecimento de Sentimento em Texto Abordado Através da Computaçõa Afetiva
Publicações do PESC
This thesis proposes, implements and evaluates a new paradigm for sentiment analysis in text. Sentiment recognition is the phrase coined to reference it, because it considers that the externalization of an emotion is subject to noise, due to the author's objectives, target audience and communication venue to transmit the message. The sentiment recognition problem is solved by adapting the noisy channel proposed by Shannon for message communication. To enable this goal two new models are proposed. One is the linguistic emotional model that computes the likelihood of an observed text, given its underlying emotion (hidden). The other one computes the a priori probability of a sentiment in an annotated corpus. The resulting sentiment is the argument of the maximum function, considering three possible polarities (positive, negative and neutral), that multiplies the probabilities from both models. The results achieved demonstrate the efficiency of this new approach while enumerating points of enhancement.