Expanding the Range of Automatic Emotion Detection in Microblogging Text

Abstract

Detecting emotions on microblogging sites such as Twitter is a subject of interest among researchers in behavioral studies investigating how people react to different events, topics, etc., as well as among users hoping to forge stronger and more meaningful connections with their audience through social media. However, existing automatic emotion detectors are limited to recognize only the basic emotions. I argue that the range of emotions that can be detected in microblogging text is richer than the basic emotions, and restricting automatic emotion detectors to identify only a small set of emotions limits their practicality in real world applications. Many complex emotions are ignored by current automatic emotion detectors because they are not programmed to seek out these “undefined” emotions. The first part of my investigation focuses on discovering the range of emotions people express on Twitter using manual content analysis, and the emotional cues associated with each emotion. I will then use the gold standard data developed from the first part of my investigation to inform the features to be extracted from text for machine learning, and identify the emotions that machine learning models are able to reliably detect from the range of emotions which humans can reliably detect in microblogging text.