Automatic Expansion of a Social Judgment Lexicon for Sentiment Analysis

Reviewed by Francisco CoutoWe present a new method for automatically enlarging a sentiment lexicon for mining social judgments from text, i.e., extracting opinions about human subjects. We use a two-step approach: first, we find which adjectives can be used as human modifiers, and then we assign their polarity attribute. To identify the human modifiers, we developed a set of hand-crafted lexico-syntactic rules representing elementary copular and adnominal constructions where such predicates can be found. These are applied to a large n-gram collection, gathering evidence about human/not-human adjective behavior. The adjective and rule frequencies are then used as input features to a statistical classifier. In the second and final stage, we assign polarities to human adjectives, by exploring the graph of synonyms. We calculate the shortest distances of each adjective with unknown polarity to the human adjectives of each sentiment class with prior polarity in the initial lexicon. The computed distances for each adjective are then used as the input features of a statistical polarity classifier. The application of the proposed method to a collection of manually annotated adjectives showed its effectiveness for developing a fine-grained sentiment lexicon for social judgment, which can be seen as a human domain-specific alternative to general-purpose lexicons. We focus on the polarity classification of Portuguese human adjective predicates, but the studied method could be applied to other languages and semantic predicates. The lexicon of human adjectives with polarities produced with this method is now available as an open resource