Research

Main Topics

Riccardo Cantini's research activities are centered around Deep Learning and Big Data analysis, with a specific focus on data generated on major social media platforms. The primary objective is to extract a wide range of information about users to outline their perception of real-world facts and events, providing a data-driven approach for an in-depth understanding of socio-political phenomena. In this context, cutting-edge machine learning and deep learning models and techniques are employed, with a particular emphasis on the use of Large Language Models, along with frameworks for parallel and distributed computing. Key applications include identifying discussion topics in online conversations, estimating the political polarization of social media users, and exploring the connection between user polarization and emotions expressed in posted content. Another aspect addressed is the dynamism of online conversations and the impact of false information on user discussions. To this end, topic-oriented techniques for false information detection in social content are developed, as well as time-adaptive models capable of effectively handling dynamically evolving scenarios. The research activities also focus on sustainable artificial intelligence, with a focus on ensuring an efficient, fair, and trustworthy use of Large Language Models. In this context, techniques for efficient fine-tuning, energy-aware models, bias elicitation techniques, and advanced retrieval-augmented generation systems are currently explored, as well as the use of curriculum learning and knowledge distillation for learning effectively from limited amounts of data and scarce computing resources.