Word-embedding data as an alternative to questionnaires for measuring the affective meaning of concepts

• In Event: 4467 - Section on Social Psychology Refereed Roundtables (Cosponsored with Section on Sociology of Emotions)
  In Refereed Roundtable 100min: Table 3: Affective Meaning and Culture

Abstract

Affect control theory has conventionally obtained measures of the evaluation, potency, and activity of concepts using questionnaires. The vector representation of words has been a flourishing pursuit of recent computer science, and various datasets of high-dimensional representations of large word lists are now publicly available. We apply machine learning methods to word vector data to develop predictions of EPA ratings in an existing questionnaire dataset. The success of our final models range from explaining 76% of variance in ratings of the evaluation dimension to 57% for the activity dimension. We use the occasion of this paper to specify predictions and an analysis plan for the next step of validating our model using new questionnaire data we will collect on new concepts.

Authors

• Austin van Loon, Stanford

• Jeremy Freese, Stanford University