Learning Affect with Distributional Semantic Models

Abstract

The affective content of a text depends on the valence and emotion values of its words. At the same time a word distributional properties deeply influence its affective content. For instance a word may become negatively loaded because it tends to co-occur with other negative expressions. Lexical affective values are used as features in sentiment analysis systems and are typically estimated with hand-made resources (e.g. WordNet Affect), which have a limited coverage. In this paper we show how distributional semantic models can effectively be used to bootstrap emotive embeddings for Italian words and then compute affective scores with respect to eight basic emotions. We also show how these emotive scores can be used to learn the positive vs. negative valence of words and model behavioral data.