ETD

• community detection
• natural language processing
• polarization
• sentiment analysis
• social network analysis

In recent years, the rapid growth of social media platforms has profoundly transformed online communication, enabling large-scale interaction, opinion sharing, and content dissemination. This has also made it possible to study user behavior through computational and data-driven methods, providing valuable insights into the dynamics of digital communities and online discourse. Within this context, the present thesis investigates patterns of user interaction and sentiment propagation across social networks.

Social media users interact with images, comments, and shared content posted by members of their network. They express both positive and negative attitudes, showing support or opposition toward other users or influencers.
While scrolling through social media, we can perceive an increasing tendency among users to share negative comments. More and more influencers are complaining about feeling attacked by users. In fact, by paying attention to this kind of opinion expression, we can observe a growing climate of hostility in online debates. Usually, this kind of behavior can be observed in discussions about sensitive topics such as politics, religion, human rights, and others.
In contrast, to investigate this growing phenomenon in non-sensitive contexts, this thesis starts by collecting data from two social media platforms — Instagram and YouTube — focusing on neutral topics, such as food and travel, at two different levels of analysis: national and international. Regarding users, by applying classical inferential statistical methods, we observe a correlation between negative comments under posts and the likelihood of receiving further negative comments.
Then, various networks of user interactions have been constructed, based on the categorization of users as positive or negative. Community discovery algorithms have been applied to analyze the resulting factions. This investigation reveals that negativity operates as a contagious force in online discourse: initial negative comments significantly amplify subsequent hostile interactions, creating cascading effects most pronounced in direct user mention networks. Interestingly, national communities exhibit homogeneous negative patterns, while international audiences display more heterogeneous patters, reflecting the complex social structures of the users discourses around any kind of topic.