|Di GIOVANNI MARCO||Cycle: XXXIII |
Section: Computer Science and Engineering
Tutor: GATTI NICOLA
Advisor: BRAMBILLA MARCO Major Research topic
:Exploring and Challenging the Limits of Language Models and their Application to Human-generated Textual ContentAbstract:
Social networks are an enormous source of human-generated content.
Hidden somewhere, useful information lies, challenging to find, collect and encode in meaningful features.
In my thesis, I apply state-of-the-art language models (LMs) to analyze publicly shared textual content.
In the field of NLP, attention-based LMs gained popularity due to their recent success in many heterogeneous NLP benchmark datasets.
I use classic and novel approaches to compare and classify users based on the text they share.
Tasks include extraction of emerging knowledge, represented by users similar to a given set of accounts, community detection, user classification (e.g. political inclination) and controversy detection in online discussions.
Finally, I investigate how those models behave when they are fed with text, generated with an evolutionary adversarial algorithm (EFSG), and how to increase their robustness to obtain better models, able to deal with noisy data.