A Multivalued Emotion Lexicon Created and Evaluated by the Crowd

Sentiment analysis aims to uncover emotions conveyed through information. In its simplest form, it is performed on a polarity basis, where the goal is to classify information with positive or negative emotion. Recent research has explored more nuanced ways to perform emotion analysis. Unsupervised emotion analysis methods require a critical resource: a lexicon that is appropriate for the task at hand, in terms of the emotional range and diversity captured. Emotion analysis lexicons are created manually by domain experts and usually assign one single emotion to each word. We propose an automated workflow for creating and evaluating a multi-valued emotion lexicon created and evaluated through crowdsourcing. We compare the obtained lexicon with established lexicons and appoint expert English Linguists to assess crowd peer-evaluations. The proposed workflow provides a quality lexicon and can be used in a range of text property association tasks.