Emotional Commonsense Acquisition

Sentiment mining and classification deals with binary classes, whereas Emotion mining and classification deals with multiple classes of emotions like anger, despair, happiness etc. Existing approaches to Emotion mining learn emotion classification from Text without relying on the background commonsense knowledge that humans would usually rely on. We investigate detection and classification of emotions from texts using Commonsense knowledge. To achieve this, we derive Emotional Commonsense knowledge automatically, like roller-coaster is a happy and exiting experience; or that rejection is sad. We used a very large scale n-gram dataset to construct a knowledge base which was later fine grained to a set of sentiments (happy, sad, fear, anger, surprise). We start out with a knowledge base where nouns and adjectives were classified into one of the above sentiments. We then used this knowledge on sentences to estimate the emotions it exhibits.