Risk Co-De model: a machine learning approach to monitor the risk construction in social media

Socio-psychological processes such as denial, moral disengagement, and psychological distance can hinder the willingness to act in front of risks. In order to promote an effective risk communication, it is thus fundamental to identify these mechanisms and intervene timely. In this contribution, we will present the Consciousness to Denial (Co-De) model and discuss a derived machine learning tool. The model combines the different nuances of denial, disengagement, and distance into four macro and nine micro categories. These categories are at the core of a tool dedicated at analysing psychosocial processes expressed in natural language in social media communication. A total of 2813 tweets in Italian, related to major events (Covid-19 pandemic, climate change), were collected and manually annotated by four experts according to the nine micro-categories. We then identified the linguistic features (i.e. expressions in the language) of the model’s categories and created a training set to train machine learning algorithms aimed at the automatic classification of texts. Finally, we tested the effectiveness of the model. Prodigy annotation tool was used. Results showed a satisfactory accuracy in predicting the model categories. The effectiveness of the model and of resulting tools for implementing effective risk communication will be discussed, focusing on the potential of studying psycho-social processes through machine learning.