Description:

Why do people feel different emotions toward the same event? And what roles do emotions play when people talk about each other? Answering these questions automatically from digital footprints will help us better understand human behavior, emotions, and biases. This talk presents our recent work tackling these two questions, harnessing data from social media. First, we investigate the emotional reactions and triggers in the context of global crises, exemplified by the COVID-19 pandemic. We craft new datasets and tasks, with which we will not only be able to train models that predict perceived emotion, but also to identify the specific experiences that trigger an emotion in a long social media post (in the form of a short summary). Next, we investigate interpersonal emotions and their impact on intergroup bias manifested in language. We argue that the language we produce is biased in one way or another, whether we intend to or not. We present data and models to learn to characterize this "generalized intergroup bias", where we find fine-grained interpersonal emotions to be a key indicator.