Crowdsourcing a More Realistic Emotional Lexicon Process

Abstract

Even though sentiment analysis and emotion analysis are relatively new subfields of natural language processing, they are increasingly being used by businesses and academics. Their primary objective is to automatically capture the emotions and opinions of people as they wrote their text. Whether the text is from social media, literature, or any other form of text, it is a difficult problem. We present a novel approach by introducing the idea of capturing a range of universal primary emotions (i.e., sadness, happiness, anger, disgust, surprise, and fear). By crowdsourcing the emotion lexicon, it creates a diverse lexicon of emotions. This type of lexicon shows how many people within the same culture and language can have both convergent emotions where everyone universally every agrees that a word has a single emotion behind it (e.g., yummy = happy) and divergent emotions where there is a range of feelings for the same word (e.g., hated = angry, disgusted, and sad). Finally, we empirically prove that emotions are not binary as is often used by modern approaches to this problem (e.g., not sad ≠ happy), showing that people’s emotions are complex and nuanced.