SORS: Computational affective science: from digital traces to generative agents

Abstract

The wealth of text data generated by social media has enabled new kinds of analysis of emotions with language models. These models are often trained on small and costly datasets of text annotations produced by readers who guess the emotions expressed by others in social media posts. This affects the quality of emotion identification methods due to training data size limitations and noise in the production of labels used in model development. I present LEIA, a model for emotion identification in text that has been trained on a dataset of more than 6 million posts with self-annotated emotion labels, achieving state of the art performance. Beyond that, LEIA outperforms humans at identifying emotions in social media, opening the door to new developments in Affective AI. Building on LEIA, we want to understand social and collective factors of emotional experiences. To do so, Generative Agent-Based Modelling (GABM) combines Large Language Models with Analytical Sociology in simulations of social media with an unprecedented level of resolution and accuracy. Our recent results on the Collective Turing Test show the face validity of GABM for simulating online discussions, bearing promise to simulate policies and interventions in online platforms.