In today’s society, where a large part of the information passes through social platforms, the phenomenon of echo chambers is well known. These are homogeneous and closed discursive communities in which users are exposed only to news and opinions that align with their previous beliefs, excluding different views and interpretations.

However, one of the open problems regarding the phenomenon remains to find unbiased detection methods for the data.

A research group, in collaboration with the IMT School of Advanced Studies of Lucca, the Institute of Informatics and Telematics—CNR of Pisa, the Institute for the Applications of Computing “Mario Picone” (IAC), the Institute of Informatics and Telematics—CNR, and Enrico Research Center Fermi, has developed a method to identify echo chambers within political discussions on the X/Twitter platform.

The research, published on PNAS Nexus (https://doi.org/10.1093/pnasnexus/pgae177), took the debate on vaccination during Covid as a case study by analyzing a database of 1.87 million tweets in Italian collected between on 1st and 24th September 2021.

“Initially, we represented the system as a network in which the accounts are the nodes and the connections represent the interactions that occur between them” – explained Fabio Saracco of CREF. “For the analyses, we used an approach inspired by statistical physics. In practice, we build a virtual ensemble of networks that acts as a reference. In this “synthetic” set, the number of connections of the different nodes observed in the real system is maintained, but how the nodes are connected is completely random. Then we compare the real system with the ensemble, looking if any significant differences can reveal some non-trivial characteristics of the real system.”

The method allowed us to identify two groups: users exposed to the same news (news engagement communities of users-NEC) and groups linked by participation in a common discourse, the so-called discursive communities (DiCo).

Analyzing the COVID-19 vaccination database on Twitter/X, the researchers found that discourse communities were organized around the accounts of leading Italian politicians. In the case of NECs, the percentage of users showing significant similarity in the information diet was less than 1%. This means that these users tended to focus on very rare and niche sources of information and often linked to sites that spread conspiracy theories.

The study shows that echo chambers are formed when news engagement communities of users are part of the same discursive community and interact via retweets; this is because retweets are considered a form of approval of content created by others. Interestingly, the presence of users in the echo chambers is limited, around 0.35% of the total users. However, their impact on forming a common discourse is strong since users in echo chambers are responsible for almost a third of the retweet flow of the discursive community they belong to.

“The definition of echo chambers is independent of how verified the information shared within them is. This construction is more worrying when the quality of information is particularly low, being a mechanism that reinforces users’ already formed opinions and excludes any contrary opinion. Nonetheless, echo chambers are a worrying phenomenon for political communication because they limit the exchange of views and foment polarization,” concluded Saracco.

Contacts:

fabio.saracco@cref.it

anna.lopiano@cref.it (Ufficio stampa)

Title:

Entropy-based detection of Twitter echo chambers

Published on:

PNAS Nexus, Volume 3, Issue 5, May 2024, p. 177, https://doi.org/10.1093/pnasnexus/pgae177

Published: 25 April 2024

Authors:

Manuel Pratelli ( MT School For Advanced Studies Lucca, Istituto di Informatica e Telematica, CNR, Pisa)  

Fabio Saracco (“Enrico Fermi” Research Center, Rome, IMT School For Advanced Studies Lucca, Institute for Applied Computing “Mauro Picone”, CNR, Rome)

Marinella Petrocchi (Istituto di Informatica e Telematica, CNR, Pisa, IMT School For Advanced Studies Lucca)