• Acrophily
• Climate change
• Opinion polarization
• Social media
• Social network

This thesis explores the phenomenon of acrophily, the tendency of individuals to connect with those who express more extreme opinions. While traditional studies have focused on homophily, in which people connect based on similarity, acrophily goes further, fostering the development of dangerous behaviours such as radicalization. The following thesis project aims to develop a measure to capture this phenomenon. Then, other related measures were implemented: "extremophily," which considers not just neighbours in the same faction but also those in the opposing faction, and "exposure," which defines how a node is influenced by extreme views as it traverses the graph. In this field, defining whether an opinion is extreme or not is important. To distinguish moderate from extreme users, we devised a methodology based on a definition of thresholds. We first tested the developed measures in a political context, a field in which acrophily has already been studied, using data from a recent study of discussions on Reddit during the first two and a half years of Trump's presidency. The main factions in these discussions are "pro-Trump" and "anti-Trump" users. Next, the metrics will be tested in the climate change field, a context where acrophily has never been detected. Ad hoc networks have been built to capture interactions in the online climate change debate coming up on Reddit in 2022, and each user is categorised by the use of the BERT model. The opinion distribution for the considered periods indicated that both political and climate change networks show bimodal distributions of opinions, thus confirming the presence of extreme users. Nevertheless, climate change debates captured more moderate opinions than political debates, which showed more extreme opinions. A significant result is the presence of acrophily in both contexts, with higher density peaks in the political context. Extremophily showed users' openness to different opinions. In particular, in political debates, there is a more constructive dialogue between different ideologies. In contrast, climate change debates showed a more closed view, with a preference to interact with moderates or extremes. In the last metrics results, it has been observed that users participating in political debates are exposed to both moderate and extreme individuals equally. It was different on climate change, where some individuals are more exposed to radicals but the majority to moderate, with at most one extreme user. We conducted more detailed analyses. In particular, the correlation matrix and regressions demonstrated the independence of the used measures from other variables, and the Markov chain allowed for a detailed study of the evolution of various opinions over time, demonstrating that our labelling procedure was reasonable.