Measuring Cultural Relativity of Emotional Valence and Arousal using Semantic Clustering and Twitter
This work addresses the debate on cultural relativity of emotions for researchers in psychology and computational linguistics by providing evidence from large-scale text data, though it is incremental as it applies existing techniques to a new domain.
The study tackled the problem of measuring cultural differences in emotions by analyzing valence and arousal of emotion keywords on Twitter across Europe, Asia, and North America, finding significant regional variations such as Europeans appearing more positive and aroused, North Americans more negative, and Asians more positive but less aroused.
Researchers since at least Darwin have debated whether and to what extent emotions are universal or culture-dependent. However, previous studies have primarily focused on facial expressions and on a limited set of emotions. Given that emotions have a substantial impact on human lives, evidence for cultural emotional relativity might be derived by applying distributional semantics techniques to a text corpus of self-reported behaviour. Here, we explore this idea by measuring the valence and arousal of the twelve most popular emotion keywords expressed on the micro-blogging site Twitter. We do this in three geographical regions: Europe, Asia and North America. We demonstrate that in our sample, the valence and arousal levels of the same emotion keywords differ significantly with respect to these geographical regions --- Europeans are, or at least present themselves as more positive and aroused, North Americans are more negative and Asians appear to be more positive but less aroused when compared to global valence and arousal levels of the same emotion keywords. Our work is the first in kind to programatically map large text corpora to a dimensional model of affect.