Presenter: Giordano Paoletti
Thursday, April 10th, 2025, 5:00 PM
Location: Sala Grande Covivio, Corso Ferrucci 112
ABSTRACT
Telegram has emerged as a pivotal platform for political discourse, particularly regarding high-stakes events such as the 2024 US elections. In this ongoing project, we leverage network analysis and advanced NLP techniques to analyze a large-scale dataset consisting of hundreds of millions of Telegram messages across thousands of groups. Preliminary topic modeling has identified discussions around critical events, including the Trump assassination attempt and Biden’s withdrawal from the race. However, when combined with a community detection analysis, we found that approximately 80% of the content in the dataset originated from a small, tightly knit community of groups and channels primarily discussing cryptocurrencies.
This raises significant concerns about the methods used to collect the dataset and suggests the need for improved data collection practices in the future. Currently, we are exploring summarization methods using Llama 3.1 and refining semantic encodings via LLM2Vec library combined with Llama 3 to pinpoint nuanced topics within political discussions. This talk will present our methodological approach, initial findings, and future directions, highlighting the challenges and opportunities in capturing dynamic political conversations on Telegram.
BIOGRAPHY
Giordano is a 2nd year Ph.D. student at DAUIN, with a background as a complex systems physicist. He previously spent one year as a researcher at ISI Foundation, where he broadened his expertise in computational social science. His research passionately investigates the dynamics of social networks, with a particular emphasis on political discourse analysis using advanced network analysis, natural language processing, and large language models.