Nowadays, Gender-Based Violence (GBV) has undergone a normalization process, whereby violent behaviors, by being justified as normal, have become subtle and difficult to recognize. In NLP, GBV has been investigated within the broad topic of Hate Speech detection, distinguishing between the different targets of hateful contents. Considering the pervasiveness of GBV and its media representation in our society, the main goal of our research is to explore people’s reactions to femicide events, considered the most brutal expression of GBV. In particular, we collected 932 YouTube comments in response to the news regarding Giulia Cecchettin’s femicide and we proposed an annotation task through a fine-grained annotation schema that builds upon Ferrando et al. with some modifications. The qualitative analysis of the annotated comments revealed some differences from the GBV-Maltesi dataset, especially regarding misogyny, aggressiveness and responsibility attribution. We tested different LLMs, investigating their ability to recognize the presence of aggressiveness and responsibility in both Maltesi and Cecchettin datasets and to indicate their target, using different prompts.
Analyzing Femicide Reactions in YouTube Comments: a Comparative Study of Giulia Cecchettin and Carol Maltesi
Sveva Silvia Pasini;Chiara Zanchi;
2025-01-01
Abstract
Nowadays, Gender-Based Violence (GBV) has undergone a normalization process, whereby violent behaviors, by being justified as normal, have become subtle and difficult to recognize. In NLP, GBV has been investigated within the broad topic of Hate Speech detection, distinguishing between the different targets of hateful contents. Considering the pervasiveness of GBV and its media representation in our society, the main goal of our research is to explore people’s reactions to femicide events, considered the most brutal expression of GBV. In particular, we collected 932 YouTube comments in response to the news regarding Giulia Cecchettin’s femicide and we proposed an annotation task through a fine-grained annotation schema that builds upon Ferrando et al. with some modifications. The qualitative analysis of the annotated comments revealed some differences from the GBV-Maltesi dataset, especially regarding misogyny, aggressiveness and responsibility attribution. We tested different LLMs, investigating their ability to recognize the presence of aggressiveness and responsibility in both Maltesi and Cecchettin datasets and to indicate their target, using different prompts.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
