In the last decade, online Social Networking Sites have become a fundamental part of our everyday life. Billions of individuals worldwide participate in such virtual communities, sharing and discussing messages, photos, videos, and other user-generated content. News consumption habits have also changed, and more and more individuals consume online news on social platforms such as Facebook and Twitter rather than traditional media such as newspapers and TV. However, online social media also expose us and make us vulnerable to a variety of false and misleading information which erodes public trust towards institutions, with severe backlashes in the real world. One example is the ongoing COVID-19 pandemic, which has been accompanied by waves of potentially unreliable information which undermine medical intervention and governmental efforts to circumvent the spread of the disease. In this work, we leverage a network and computer science approach to tackle the problem of disinformation -- a term we use hereby as a shorthand to indicate all sorts of misleading, false and potentially harmful information -- spreading in online social networks. Focusing on Twitter and Facebook, we study the mechanisms and the actors involved in the spread of false information and other malicious content during relevant events such as political elections and the ongoing COVID-19 pandemic, when the need of reliable information for the public is higher. We carry out a systematic comparison of reliable information, published by mainstream and traditional news websites, versus unreliable information conveyed by websites that have been repetitively flagged for sharing disinformation, misinformation, hoaxes, fake news and hyper-partisan propaganda. We provide evidence of superspreaders of disinformation, i.e., influential users which are responsible for most of the disinformation shared online, and we unveil links with far-right communities, which oftentimes exploit fabricated information to push their agenda. At the same time, we show that reliable information accounts for the majority of news stories circulating online and that disinformation has a small yet non-negligible online prevalence which can still influence individuals’ opinions and feelings. We further investigate the interplay between vaccine-related disinformation shared on Twitter and the vaccine hesitancy and uptake rates measured across U.S. regions, following the roll-out of the COVID-19 vaccination program. Building a regression model which takes into account demographics, socio-economic and political factors, we find a significant association between online disinformation and vaccine outcomes. Finally, drawing on the results of aforementioned analyses, we deploy a methodology to accurately classify news articles based on the interactions between users that naturally take place on Twitter. Following the intuition that users shape different diffusion patterns depending on the content they share, we train and test off-the-shelf machine learning classifiers that can classify the veracity of a news article, without the need of looking at its content. All in all, our results contribute to a better understanding of the issue of disinformation spreading in online social media, and highlight the need for intervention by platforms and governments to address this issue in a timely fashion.
Negli ultimi anni, i Social Network hanno acquisito un ruolo importante nella nostra vita quotidiana. Miliardi di persone si organizzano in comunità virtuali su scala globale, condividendo e discutendo messaggi, foto, video e altri tipi di contenuto. Anche la fruizione di notizie è cambiata, e sempre più individui si rivolgono a piattaforme come Facebook e Twitter per informarsi, abbandonando media tradizionali come giornali e televisione. Tuttavia, i "social" ci espongono e ci rendono vulnerabili a una varietà di informazioni false e ingannevoli, e contribuiscono a erodere la fiducia nelle istituzioni con gravi conseguenze nel mondo reale. Un esempio è la pandemia di COVID-19 attualmente in corso, accompagnata da un'infodemia di notizie inattendibili che minano gli sforzi a contenere il virus. In questo lavoro, utilizziamo un approccio a metà tra scienza delle reti e informatica per affrontare il problema della disinformazione -- un termine che utilizziamo come cappello per indicare i vari tipi di informazione inattendibile, falsa e potenzialmente dannosa -- che circola sui Social Network online. Focalizzando la nostra attenzione su Twitter e Facebook, studiamo i meccanismi e gli attori coinvolti nella diffusione di disinformazione durante eventi rilevanti quali elezioni politiche e la pandemia in corso, quando la possibilità di avere accesso a informazioni attendibili è cruciale. Effettuiamo una comparazione sistematica dell'informazione attendibile, prodotta da siti di notizie tradizionali e "mainstream", in contrapposizione all'informazione inattendibile prodotta da siti che sono stati ripetutamente richiamati per aver condiviso disinformazione, bufale, "fake news" e propaganda faziosa. Forniamo prove della presenza di super diffusori di disinformazioni, i.e., utenti influenti che sono responsabili per la maggior parte dei contenuti di disinformazione condivisi online, e riveliamo collegamenti tra le comunità di estrema destra, che spesso fanno ricorso a notizie contraffatte per promuovere la loro ideologia. Allo stesso tempo, mostriamo che la maggioranza delle notizie che circolano online proviene da siti attendibili, e che la disinformazione ha una presenza limitata ma non trascurabile che può influenzare le opinioni e i sentimenti degli utenti online. Investighiamo l'influenza della disinformazione relativa ai vaccini che si diffonde su Twitter sulla campagna di vaccinazione degli Stati Uniti. Utilizzando un modello di regressione lineare multipla, che tiene conto di fattori demografici, socio-economici e politici, troviamo un'associazione significativa tra la disinformazione online e le vaccinazioni. In conclusione, basandoci sui risultati delle analisi sovra-citate, costruiamo una metodologia per classificare accuratamente le notizie sulla base delle interazioni sociali tra utenti che avvengono su Twitter. Seguendo l'intuizione che gli utenti danno vita a diversi "pattern" di diffusione a seconda del contenuto condiviso, alleniamo e testiamo dei classificatori ad apprendimento automatico che possono verificare la veracità di un articolo senza aver bisogno di guardare al contenuto. I nostri risultati contribuiscono ad una maggiore comprensione del problema della disinformazione che circola sulle piattaforme social, e sottolineano l'urgenza di interventi da parte di piattaforme e governi per contrastare il fenomeno.
Characterization and detection of disinformation spreading in online social networks
Pierri, Francesco
2021/2022
Abstract
In the last decade, online Social Networking Sites have become a fundamental part of our everyday life. Billions of individuals worldwide participate in such virtual communities, sharing and discussing messages, photos, videos, and other user-generated content. News consumption habits have also changed, and more and more individuals consume online news on social platforms such as Facebook and Twitter rather than traditional media such as newspapers and TV. However, online social media also expose us and make us vulnerable to a variety of false and misleading information which erodes public trust towards institutions, with severe backlashes in the real world. One example is the ongoing COVID-19 pandemic, which has been accompanied by waves of potentially unreliable information which undermine medical intervention and governmental efforts to circumvent the spread of the disease. In this work, we leverage a network and computer science approach to tackle the problem of disinformation -- a term we use hereby as a shorthand to indicate all sorts of misleading, false and potentially harmful information -- spreading in online social networks. Focusing on Twitter and Facebook, we study the mechanisms and the actors involved in the spread of false information and other malicious content during relevant events such as political elections and the ongoing COVID-19 pandemic, when the need of reliable information for the public is higher. We carry out a systematic comparison of reliable information, published by mainstream and traditional news websites, versus unreliable information conveyed by websites that have been repetitively flagged for sharing disinformation, misinformation, hoaxes, fake news and hyper-partisan propaganda. We provide evidence of superspreaders of disinformation, i.e., influential users which are responsible for most of the disinformation shared online, and we unveil links with far-right communities, which oftentimes exploit fabricated information to push their agenda. At the same time, we show that reliable information accounts for the majority of news stories circulating online and that disinformation has a small yet non-negligible online prevalence which can still influence individuals’ opinions and feelings. We further investigate the interplay between vaccine-related disinformation shared on Twitter and the vaccine hesitancy and uptake rates measured across U.S. regions, following the roll-out of the COVID-19 vaccination program. Building a regression model which takes into account demographics, socio-economic and political factors, we find a significant association between online disinformation and vaccine outcomes. Finally, drawing on the results of aforementioned analyses, we deploy a methodology to accurately classify news articles based on the interactions between users that naturally take place on Twitter. Following the intuition that users shape different diffusion patterns depending on the content they share, we train and test off-the-shelf machine learning classifiers that can classify the veracity of a news article, without the need of looking at its content. All in all, our results contribute to a better understanding of the issue of disinformation spreading in online social media, and highlight the need for intervention by platforms and governments to address this issue in a timely fashion.| File | Dimensione | Formato | |
|---|---|---|---|
|
PHD_thesis-6.pdf
accessibile in internet per tutti
Dimensione
19.12 MB
Formato
Adobe PDF
|
19.12 MB | Adobe PDF | Visualizza/Apri |
I documenti in POLITesi sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/10589/188756