Along with the North American elections or the current coronavirus pandemic in which we are still immersed, disinformation campaigns , with the dissemination of hoaxes and fake news, have increased notably in recent years. But this is not something new. We can think, for example, of propaganda in times of war, when this type of campaign was used in order to influence public opinion against a specific enemy.
But, as we have already seen, experts now face a palpable and latent problem, especially if we take into account the use of the Internet and social networks when disseminating this type of campaign. For example, the spread of disinformation through social media has the power to strengthen conspiracy theories, encourage denial campaigns against the coronavirus or COVID-19 vaccines, or even change the final outcome of elections.
Researchers from the MIT Lincoln Laboratory’s Artificial Intelligence Algorithms and Software Architectures Group have been part of a team of specialists who set out to better understand and understand these types of campaigns, for which they launched the Reconnaissance of Influence Operations (RIO) program. . Its objective? Create a system capable of automatically detecting narratives created to misinform , as well as identifying those individuals who shared them and participated in their dissemination on social networks.
After several years of work, they managed to make this system capable of combining multiple analysis techniques with which to create a comprehensive vision of where and how disinformation campaigns are spreading. Thus, they developed a statistical approach to help determine not only whether a specific social media account was spreading the campaign, but also to what extent that particular account was causing the network as a whole to change, thereby amplifying the message.
On the other hand, the researchers also applied a new approach to machine learning , useful to allow RIO to classify these accounts by analyzing data related to behaviors, such as whether the said account interacts with foreign media and, in addition, what languages it uses. In this way, the system was able to detect hostile accounts that are only activated in different campaigns, ranging from the French presidential elections of 2017 to the spread of information related to COVID-19.
One of the main advantages of this system is that it is able to detect and quantify the impact of social media accounts operated by both humans and bots . What’s more, it would help predict how different countermeasures could stop a particular disinformation campaign from spreading .
Reference: Steven T. Smith et al, Automatic detection of influential actors in disinformation networks, Proceedings of the National Academy of Sciences (2021). DOI: 10.1073 / pnas.2011216118
