• echo chamber
• emotion analysis
• NLP
• polarization
• sentiment analysis
• SNA
• social network analysis
• topic modeling

Social network sites (SNSs) have reshaped how information is spread, in favour of a faster way of sharing ideas and participating in public discussions. Despite the impressive number of benefits, SNSs bring with them an equivalent amount of polluting phenomena that cannot be ignored. Among these issues, we include echo chambers, i.e. polarized systems in which information, ideologies, and beliefs are amplified as the only truthful view of reality, without contemplating rebuttal or openness to different ideas. Although many efforts were made to study echo chambers, current research lacks a rigorous analytical framework to analyze EC's diachronic evolution. Moreover, scarce to non-existent attention was given to characterizing the individual behaviours of the users therein. In this thesis is presented a study on echo chamber detection on Reddit discussion boards revolving around the first two years of Trump's presidency. At first, for each chosen topic, we model the interaction network of the users via node-attributed graphs, so that each user is characterized by their own political leaning. Then, we extract communities and assess the risk of them being echo chambers by looking at the topological cohesion and the ideology homogeneity. Afterwards, we focus on their temporal evolution to gain new insights into their stability. The second part of the work is about users and the different ways of acting depending on whether they are trapped or not inside an echo chamber. We analyze their linguistic productions, looking at text-specific features, the emotion and sentiment vehicolated through their words, and finally, the topics they talked about. This analysis is then enhanced by relating the temporal dimension to the topic and sentiment, to have a fine-grained overview of the users. This provides insights into the stability of ECs through time and the tendency of their members to focus on a single controversial topic.