Theory-driven Measurement of Emotion (Expressions) in Social Media Text

Abstract

Sentiment analysis, or the measurement of opinions, attitudes and emotions in text, has exploded in the past 10 years. The number of academic articles published that mention “sentiment analysis” grew from 78 in 2008 to 1,876 in 2018 (Web of Science). This growth brought notable advances in sentiment classification whether it be measuring emotional valence (i.e., negative vs. positive sentiment) or the measurement of discrete, specific emotions (e.g., “anger”). In particular, there has been recent interest in measuring emotion expressions in social media text, as there are now over 3 billion users expressing emotions online on a regular basis. Although there is now a wealth of emotion expression data that exists on social media, only recently have psychological scientists begun to make use of it for the purposes of observational research and psychological hypothesis testing. In this chapter, we aim to provide psychologists with a practical guide for making use of advances in sentiment analysis for the purpose of measuring specific emotions in social media text, and ultimately testing hypotheses. We begin with an examination of the type of emotion-related information researchers can expect to measure in social media text. Next, we present a practical guide for the theory-driven application of supervised machine learning to measure specific emotion expressions on social media using the measurement of moral outrage expression as a case study. We end with considerations for researchers who wish to apply sentiment analysis of social media text for the purposes of psychological hypothesis testing.