- CUSTOMER: Spanish center for cybersecurity (INCIBE).
- INDUSTRY: Public administration
- SERVICES: Misinformation classification engine
- TECHNOLOGY: Natural Language Processing, Deep Learning
Addressing misinformation at scale and speed
Malicious content such as fake news is part of our information ecosystem. Even genuine content should sometimes be considered malicious if its purpose is to harm someone else’s image. A recent example of this is the 26 March article in The Guardian criticising Spain’s management of the COVID-19 pandemic when the British government initially tried to ignore the situation too ( “How did Spain get its coronavirus response so wrong?” ). This single article has already been shared over 15000 times.
Malicious content affects trust towards institutions and countries, has negative effects on decision-making by individuals and in general has devastating effects on the functioning of society.
There are many examples of fake news interfering with politics and the democratic process, as well as those having direct and indirect effects in the economy. Additional examples of fake news affecting Spain include the hoax that the Government is about to implement a basic income for all; that Spain is an irresponsible country neglecting the benefit of vaccination against coronavirus ; that 5G technology is helping spread coronavirus in countries like Spain, Italy and France ; how conspiracy theories are using COVID-19 to spread ; and that the Spanish government is using the coronavirus pandemic as an excuse to engage in undemocratic practices . Most importantly, the Oxford Internet Institute has reported that negative and harmful propaganda is being produced by nations such as China, Russia and Turkey to damage the reputation of Western governments while positioning themselves as world leaders in the response to the coronavirus pandemic .
 Tremlett G (2020) How did Spain get its coronavirus response so wrong? The Guardian, 26 March 2020 : http://tiny.cc/piu7lz
 The Economist (2020) How 5G conspiracy theories used covid-19 to go viral. Date: 8/4/2020. URL: http://tiny.cc/6p9vmz
 Oxford Internet Institute (2020) State-backed media in Russia, China, Iran and Turkey successful in sharing misleading stories on COVID-19. Date: 9 April 2020. URL: http://tiny.cc/w1mrmz
The project main goals were: (1) Identify in near real-time malicious tweets against the interests of Spain, (2) prioritise the most offending messages and their authors in order to orchestrate a counter-response.
We validated our solution for the identification of malicious campaigns against Spain by focusing on the current coronavirus situation in the country with a specific focus on the social network Twitter.
We harvested 160 million tweets in English containing hashtags related to coronavirus and posted between January 1st and June 20th, 2020.We made use of our cyberdefense platform, Sherlock, to identify Twitter content that is harmful to the interests of Spain in the context of the coronavirus pandemic.
Near real time misinformation detection
The identification of harmful content and the corresponding offenders proceeds in three stages: (i) Broad Search, where we mine all tweets that meet specific search criteria; (ii) Misinformation Profiling, where we profile the impact of specific messages; and (iii) Attacker Profiling, where we estimate the potential impact a given offender might have when trying to spread their message (Figure 1).
AI-assisted decision making processes
We believe that tackling the problem of online misinformation calls for the combination of artificial intelligence-based technologies and human based decision making.
The outstanding levels of performance recently achieved by artificial intelligence (AI) enables the partial automation of current human based processes deployed by the defense and security forces to identify online misinformation.
We believe that empowering security and defence agents with AI-based technologies is key to mitigate online misinformation in near real time at scale.