logo SBA

ETD

Archivio digitale delle tesi discusse presso l’Università di Pisa

Tesi etd-05102021-194726


Tipo di tesi
Tesi di dottorato di ricerca
Autore
NIZZOLI, LEONARDO
URN
etd-05102021-194726
Titolo
Leveraging Social Media and AI to foster Secure Societies against Online and Offline Threats
Settore scientifico disciplinare
ING-INF/05
Corso di studi
INGEGNERIA DELL'INFORMAZIONE
Relatori
tutor Prof. Avvenuti, Marco
tutor Dott. Tesconi, Maurizio
tutor Dott. Cresci, Stefano
commissario Prof. Di Pietro, Roberto
commissario Prof. Quattrociocchi, Walter
commissario Prof. Vecchio, Alessio
commissario Prof. Cimino, Mario Giovanni Cosimo Antonio
commissario Prof. Sharma, Rajesh
Parole chiave
  • artificial intelligence
  • coordinated behavior
  • cryptocurrency manipulation
  • deep learning
  • Discord
  • geoparsing
  • machine learning
  • social media
  • social sensing
  • Telegram
  • Twitter
Data inizio appello
20/05/2021
Consultabilità
Completa
Riassunto
This thesis pursues the objective of leveraging social media data and Artificial Intelligence (AI) techniques to foster secure societies against online and offline threats. Thus, it provides contributions in two directions: (i) uncovering threats in online ecosystems, and (ii) grounding the online information on the unfolding offline threats. In both cases, on the one hand, we investigated how to improve AI techniques that enable essential applications; on the other, we devised and applied comprehensive approaches to target specific threats.

Focusing on uncovering online threats, our first contribution consists of a general deep learning framework for automatically detecting extremist propaganda contents on OSM when approaching realistic conditions. In the second contribution, we mapped and characterized cryptocurrency manipulations within a large online ecosystem, focusing on the mechanisms adopted by scammers to recruit participants. Finally, in the third contribution, we devised a new network-based framework for uncovering and characterizing coordinated behavior on social media in the context of the recent UK General Election.

Moving to the problem of grounding online information on unfolding offline threats, we proposed a new technique for solving the geoparsing problem, which consists of identifying location mentions in text and linking them to the corresponding geographic coordinates. Finally, our last contribution is an AI-powered system for enriching available social crisis data in the aftermath of mass disasters with solicited information from on-the-ground witnesses.
File