Frankie

Detecting disinformation based on trusted fact-checkers.

Comment 5

The Problem

Disinformation is spreading like wildfire in Europe, for political or commercial manipulation, hampering efforts to contain the pandemic, and ripping apart our societies' social fabric. Lives literally depend on all of us listening to health authorities and our ability to access, rely and trust reliable information. Yet more and more European citizens find themselves unable to do so.

Many technical approaches against the spread of fake news either rely entirely on human fact-checks or solely on AI. With Frankie, our system, we want to combine the accuracy and pluralism of fact-checks from trusted journalists with the scalability of machine-based validation.

The Solution

Our system, "Frankie", detects disinformation about COVID-19 by checking if similar texts were already debunked on a website of independent fact-checkers and high-quality publishers. If that is the case, we provide a link to the fact-check and thus not only inform the user of a disinformation but also help to disseminate the work of these organizations.

At the moment our primary user interface is a chrome extension to "scan" visited content on misinformation about COVID-19. We also have an experimental text field on our website, which can be used to try out Frankie's capabilities. With our API we will add many more UIs in the near future and offer other organizations to integrate the service into their apps and websites, as well. This will make fact-checking for our end-user as easy as possible.

While designing the system we also keep in mind ethical values and privacy issues. First and foremost, we use only pre-audited checkers – by organizations independent from us - to minimize conflict of interest. We are also committed to respecting the GDPR and relevant intellectual property law in Member States .

The Impact

Frankie helps readers of COVID-19 related articles or social media posts quickly determine if the article they are reading has content, similar to that which has already been debunked by a journalist as misinformation. With our approach they are not only warned but also enabled to spread the source of trusted high-quality journalism.

We also see Frankie as a tool to showcase the excellent work already done by fact-checkers and to raise citizens' awareness of the importance of media pluralism and the unprecedented existential challenge COVID represents for the news sector.

Accomplishments during this Hackathon

The team and idea for Frankie launched at this Hackathon. We are especially proud that - despite the small time-frame - we completed the following prototypes:

  • An initial version of the back-end
  • A chrome extension, parsing text and communicating with the back-end (see video and github)
  • A crawler capable of scraping more than 3000 IFCN reviewed fact-checks
  • The initial version of our project website, including a feature to give users the opportunity to try out our API

We also layed out the foundation to make Frankie a success story in the near future by:

What's next for Frankie

To further develop and disseminate Frankie, we will tap into the inspiring diversity of skills and contacts of our team. Together we combine advanced AI and developer skills with and in-depth knowledge of the European civil society and regulations which, we are confident, will allow us to strive for maximum accuracy while observing the highest standards of social responsibility and human rights.

We are incredibly proud of how far we already have gone in developing Frankie over the last few days, but these efforts risk to go to waste unless we receive financial support soon. The good news is that we don't need much:

A 15K pilot grant will allow us to confidently participate in competitive calls for projects and market our product within a few weeks:

With the money we will pay for:

  • The usage and maintenance of Cloud Infrastructure
  • Small stipends for design and marketing team
  • Small stipends technology team building, supporting and managing an Open Source Community
  • Small stipends to draft & submit high-quality proposals and Licensing content

This, in turn, will:

  • Make Frankie more stable and teach her to match text pieces far more accurate and efficient than the current model
  • Help us promote local fact-checkers in different regions with support in languages other than English
  • Integrate Frankie into many more UIs for example a mobile app, voice-features or an interactive chat-bot, which could be used for the education on disinformation and media pluralism
  • Make us compliant with the relevant EU laws and ethical standards.

How do we see ourselves working?

We will leverage our existing contacts to partner with fact-checkers, traditional media, European civil society and policymakers for our product dissemination and continuous improvement. We are also committed to transparency and a human rights-based approach – including with regards to the recognition of the intrinsic limitations of the technology that we offer.

Our Ethics

With Frankie we will be careful not to overstate our results. Frankie is a machine – which means that s/he is vulnerable to inaccuracy, bias – and that can have consequences for freedom of expression if it is used to streamline content moderation by social media platforms. To avoid these harms, we will only license Frankie to social media providers who provide their users with meaningful and rapid access to remedy. Frankie’s code will be open and auditable and we will rely on the Open Source Community for its continuous improvement. Developers on our end will sign a strict code of conduct and under no scenario will Frankie source outside the fact-checks community already vetted by the IFCN

Our Team

Esther Martinez is a political scientist from Spain, she has helped Human Rights Defenders outside Europe to withstand the pressures of state-sponsored defamation campaigns and currently works to protect media pluralism in the EU. During the hackathon, she has supported the development of the business strategy, reviewed issues of legal compliance (with regards copyright and GDPR as well as the ethics of automated detention of false information). Moving forward, she will help with partnership building with civil society, dissemination of the project amongst key EU stakeholders and fundraising. Linkedin.

Myriam Barnés is a mathematician from Spain. She’s currently working as a Data Analyst at K Fund Venture Capital and is co-founder of helloleia.ai a non-commercial exploration of tech and design. She’s using NLP methods to compare speeches related to climate change and sustainability for her final math degree project. During the hackathon she’s been in the algorithmic part of Frankie working with BERT pre-trained models and testing different distances to fine-tune the implementation. On next steps she’ll work on the matching sentences problem to improve the accuracy and develop specific strategies depending on the type of text and context input (keywords, length of text, multilingual,...). Linkedin.

Jan Klatte is a Software Engineer and Machine Learning specialist from Germany. He currently develops software for autonomous driving at Continental in Frankfurt. There he focuses on perception and object prediction algorithms, which oftentimes rely heavily on machine learning methods. He also gained experience in web development at previous positions, which he applied during the hackathon to develop the chrome extension. As a team lead he also worked on finding the right mentors and on the final pitch of the project. His next contributions for Frankie will be the improvement of the extensions UI, as well as the accuracy of the matching algorithm. Linkedin.

Elsa Delmas is a French Graphic Designer living in England. She has experience in brand design and has also worked on creating infographics for the University of the West of England Science Communication Department. Her challenge during this hackathon was to give Frankie a visual identity that would reflect the purpose of the project: providing users with an empowering fact-checking tool which aims to restore trust in the news sector. She also stretched her skill to UI and UX, and learnt to work with front-end developers to implement the design on the website and chrome extension. Following this hackathon, she hopes to continue learning about user-journeys and UI implementations to provide Frankie with an ever-more efficient communication. Linkedin.

Maurits De Roover is a Belgian data scientist/developer. He studied business engineering at the Catholic University of Leuven, worked at Deloitte as an auditor and it currently studying an additional data science master. In the meanwhile, he works as a data analyst at a fintech company. His challenge during this challenge was to build the website, the backend infrastructure, the API connection. He also built the quick matching functionality, using fuzzywuzzy, that our tool is currently built around whilst our algorithm is being finetuned. In the next steps, he will integrate the new algorithm and help stabilize the chrome extension. Linkedin.

Danielle Paes Barretto de Arruda Camara is a Data Scientist with MSc. and PhD in Electrical engineering and Master in Data Science and Entrepreneurship. She is Brazilian and lives in the Netherlands since 2012 having both Brazilian and Dutch nationality. She has been working as a freelance data scientist being involved in projects in different domains. During the hackathon, she has been responsible to find the most adequate data source, retrieve data and put it in the necessary format to be processed. She also supported the group finding actual statistical information about the impact of misinformation related to COVID-19. In the future, she will continue working on subjects related to data and will support improvements of the algorithm. Linkedin.

+ 7 more
Share this project:

Updates