New research-based tools battle disinformation in social media
Fake or fact? Social media is flooded with disinformation, and it’s hard to know who’s telling the truth. A European team led by DSV researchers has created tools to expose ”fake news”.
When false information is presented as facts, people are misled and democracy may be at risk. This was the starting point of a three-year long EU-funded project which has now come to an end.
Nine partners from seven countries joined forces to battle disinformation. The project was led by eGovlab, a centre of excellence in e-Government studies. The centre is affiliated with the Department of Computer and Systems Sciences at Stockholm University.
Mattias Svahn is head of research at eGovlab and responsible for coordinating the project.
“We have developed two anti-disinformation tools that can be used in social media. The idea is to combine human and artificial intelligence to identify false information”, says Mattias Svahn.
The hybrid solution is necessary, explains Svahn. Artificial intelligence is needed to handle the enormous amounts of data, but it’s not smart enough to understand nuances in the material. To do this, we need human intelligence.
The project, called Co-inform, ran from April 2018 until July 2021. Stockholm University/eGovlab is the project owner and Swedish representative. They have worked closely together with European universities, research institutes and SMEs.
“Co-inform is a developed prototype with close to a 75 percent accuracy in a laboratory environment. We have created brand new algorithms and made important contributions to research in this area. Before the project started, we didn’t know it was possible to build a hybrid model where human fact-checkers cooperate with machine learning in social media”, says Mattias Svahn.
Co-creation and transparency have led the work, which has involved journalists, policymakers, and other citizens. A series of workshops has been organized, in Sweden, Austria, and Greece, around migration and the covid-19 pandemic.
Plug-in for Twitter
The project culminated in an online conference on September 29–30, 2021, which was open to the public. A demonstration of Co-inform was given and further questions regarding disinformation in society were discussed. How can we navigate in the new, digital reality?
“We showed the different parts of the tools and talked about how society can benefit from them. Public policymakers are an important target group”, says Svahn.
“One of the functions is a plug-in for Twitter. It shows both known and potential disinformation. The assessment is made by AI, in cooperation with human fact-checkers around the world. As an individual Twitter user, you can also provide input which is checked by the human team. This loop increases the precision of the system and makes it unique”, says Mattias Svahn.
This function uses so called nudging. If you are trying to retweet a post that has been identified as disinformation, the system will ask: “Are you really sure you want to retweet this?” It doesn’t forbid you to do it, but it makes you think your decision through.
“The system also explains how the assessment was made. This increases its credibility.”
Troll farm activities made visible
By aggregating information from Twitter, Co-inform can also function as a weather report for disinformation, Svahn explains. A dashboard shows journalists and policymakers which subjects are attracting a lot of disinformation.
“The tool reveals increased activities from troll farms, and warns you if storms are rising. This information is of value for agencies responsible fopr keeping the public informed, or when important decisions are being made.”
Co-inform is an open platform which developers and entrepreneurs can continue to work on.
“The research part of the project is over. All material is published as open source material on Github, a platform where software developers and companies can use it”, says Mattias Svahn.
More about the research
Co-inform is an EU funded project within Horizon 2020
Nine partners from seven countries cooperated in the project:
eGovlab at Stockholm University, Sweden
Cyprus University of Technology, Cyprus
Open University, United Kingdom
University of Stuttgart, Germany
FactCheckNI, United Kingdom
Expert System Iberia, Spain
International Hellenic University, Greece
International Institute för Applied Systems Analysis, Austria
Last updated: February 14, 2022
Source: Department of Computer and Systems Sciences, DSV