First real-time fact-checking tool to fight against the fake news and disinformation
1.03.2019 – 31.08.2019
Project budget and EU funding
€71’429 / €50’000
Newtral has recently finished an SME Instrument Phase 1 project, which was funded by the European Commission (€50’000 grant with a 70% reimbursement rate). Project objectives were:
Technical and Practical feasibility (SO1): We will review the state of art of involved technologies to define an initial technical roadmap for an end-to-end solution for automated fact-checking.
Commercial and Business feasibility (SO2): Analysis of target markets, main stakeholders, customers and competitors in potential markets to identify potential partners, mostly active fact-checkers and reference customers or partners.
Financial feasibility (SO3): Adjustment and fine-tuning of business model. Adapting financial plan to target market segments, prices and operational costs. Revenue projections. Capital requirements, cost-benefit analysis, financing options and ROI. Elaboration of new business plan based on results.
Project context and societal challenge addressed
People consume news because they need to be informed about the state-of-the-world and events around them. As consumers do not have access to the original information it is difficult for them to assess the credibility and veracity of news. They need to trust on news producers and intermediaries. Winning consumers trust is becoming the central issue of our times as businesses compete for attention in a digital world, but consumer trust has been declining in the news media, being currently at its lowest point.
The uprising of the “Fake News” phenomenon has had a deep effect on the news media industry and our societies, provoking that only 42% of the population shows some trust on news.
Algorithm-driven news distribution platforms have reduced market entry costs and widened the market reach for news publishers and readers. At the same time, content-curation algorithms have been designed to maximize traffic and advertising revenue not news accuracy. This situation weakens the role of the traditional gatekeepers (media industry) as quality intermediaries and facilitates the distribution of false and fake news content.
Currently, fact-checking is, essentially, a manual process. At Newtral we envision a hybrid human-machine system where some of fact-checking tasks (and eventually most of the process in the future) are totally automated.
In the long run, human intervention will be only needed for the most complex fact-checks although journalists will always supervise decisions made by machine learning systems.
Fact-checking comprises 4 phases:
PHASE I: Monitoring sources – relevant sources (social media, tv shows, newspapers, YouTube channels…) are daily checked to discover relevant conversations
PHASE II: Spot relevant facts – journalists review the content and highlight the most relevant facts to be checked.
PHASE III: Data verification – journalist must compare data and statements behind each relevant fact against official sources, review its context, contact experts to obtain more information and submit a final evaluation about the claim. A database of ‘verified information’ is built as output of this process.
PHASE IV: Exploitation – Verified statements are published and distributed to the audience
Recent advances in the fields of Natural Language Understanding (NLU) and machine learning have generated the appropriate conditions to automate the most cumbersome fact-checking tasks. Newtral’s goal is to develop the first real-time automated fact-checking tool to fight against the fake news and disinformation. We plan to combine human intelligence and computer intelligence to increase the reach and speed of fact-checkers. While computers are good at monitoring and processing massive amounts of data, journalists are good at understanding context and reading behind the lines of human language. By mixing both worlds our automated tool help journalists to massively scale up fact-checking by applying AI technologies in ther day-to-day operations.
This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 855556