Identifying social norms and their violation is a challenge facing several projects in computational
science. This paper presents a novel approach to identifying social norm violations.
We used GPT-3,
zero-shot classification, and automatic rule discovery to develop simple predictive models grounded in
psychological knowledge.
Tested on two massive datasets, the models present significant predictive performance and show that even complex social situations can be functionally analyzed through modern computational tools.
This paper presents a novel approach to identifying social norm violations. The models developed in this
paper rely on state-of-the-art language models such as GPT37,8 NLI-based Zero Shot Text Classification9,10 and
automatic rule discovery11, but are also grounded in deep theoretical understanding of human psychology and
social emotions. It must be explained in a nutshell that we
- (1) use GPT-3 to support the identification of top-level categories of social norm violation in a bottom-up manner,
- (2) use NLI-based zero shot text classification, as it is an approach relevant for situations where big datasets are not available for training a model, and
- (3) use automatic rule discovery because we seek to identify the simplest possible models that can be understood by social scientists while avoiding the black-box complexification of some models.
