Adversarial Behaviour in Moral Value Expression: A Statistical and Frame-Semantic Analysis of Social Media

Author

Prevedello, Giulio and Verheyen, Lara and Brugnoli, Emanuele and Lo Sardo, D Ruggiero and Van Trijp, Remi

Abstract

Adversarial behaviour in online spaces has become a threat to user well-being and societal unity, so it is important to understand why users may interact with each other in hostile ways. This paper contributes to this understanding by exploring how adversarial behaviour is grounded in moral value expression by analyzing English texts from the Moral Foundations Twitter Corpus (enriched with toxicity scores). First, through statistical analysis, we confirmed the assumption of earlier studies that the relation between a tweet’s toxicity level and its moral value expression is topic-specific. Next, we performed a more precision-based analysis by extracting semantic frames (e.g. identifying who is inflicting harm on whom). We found that automatic frame extraction matches human annotations with low sensitivity but high specificity. Importantly, replacing human annotations with semantic frames does not impact the toxicity score distribution of tweets, which shows that semantic frame extraction is a promising method for detecting which (moral) stance a user takes on a particular issue. As such, identification of moral value expression may potentially become an important component for developing value-aware technologies that can greatly contribute to safer and more positive social media experiences.