Predicting Emotional Word Ratings using Distributional Representations and Signed Clustering

摘要: Inferring the emotional content of words is important for text-based sentiment analysis, dialogue systems and psycholinguistics, but word ratings are expensive to collect at scale across languages or domains. We develop a method that automatically extends word-level unrated using signed clustering vector space representations along with affect ratings. use our determine word’s valence arousal, which its position on circumplex model affect, most popular dimensional emotion. Our achieves superior out-of-sample rating prediction both affective dimensions three different when compared state-of-the-art similarity based methods. can assist building new improve downstream tasks such as analysis emotion detection.