Detect fake news with Narrativa’s artificial intelligence

Detecting fake news is one of the great challenges in our society today. Social networks such as Twitter and Facebook have created tools to limit this type of content; Twitter specifically operates Birdwatch, a “community-based approach” to misinformation in which users identify information in tweets that they believe are misleading and write notes that provide informative context.

But this is clearly not enough, as we see more and more people become susceptible to misinformation. The consulting firm Gartner predicts that in 2022 more than half of the information that we will find on the Internet will be false.

This is a problem that crosses demographics. A recent study shows that young people are more likely to believe and share false information if they feel they have a common identity with the person who shared it in the first place.

At Narrativa we want to help the detection of fake news to build a more informed and transparent society and we have created an artificial intelligence model that we explain below.

Try our model at Hugging Face!

How to detect fake news?

To detect fake news we first check to see if the purpose of the content is to drum up controversy by determining if it’s in line with the majority of news already published on the subject. We also check to see if a source is cited or if the content is a screenshot, which can be easily fabricated. Our tool is capable of detecting news that is false with an amazing accuracy rate of 78% and it’s only getting better over time! The model currently works in Spanish and you can try by checking out this link.

How was the model created?

Narrativa’s fake news detection model is more efficient than other models created so far, and that’s *not* fake news. :) How can we prove it? Just this year, a competition was organized by IberLEF that sought to put NLP technologies to the test. IberLEF is a comparative evaluation campaign for Natural Language Processing systems and its goal is to encourage the research community to organize competitive text processing, understanding and generation tasks in order to define new research challenges and garner state-of-the-art results.

Just this year, a competition was organized by IberLEF that sought to put NLP technologies to the test

There were many proposals from all over the world and although Narrativa did not participate formally, we decided to challenge ourselves and managed to beat the winner of the actual competition! It seems we’re onto something!

While the best model had an F1-score of 0.76, ours secured a 0.78, which means an accuracy rate of 78%. We took both the headline and the body of the text into account (and used the RoBERTa model trained by the Barcelona Supercomputing Center on the BNE corpus). It is also important to clarify that our model detects fake news based on specific datasets (the dataset used was from January through July 2018).

These were the results. While the best model had an F1-score of 0.76, ours secured a 0.78

Who can use the fake news detection model?

Whoever wants to! At Narrativa we want to create artificial intelligence models that have a positive impact on society available to all people. In addition, this tool is made available to journalists and institutions so that they can use it and act to put an end to this new epidemic.

We are working to expand the range of analyzed news and in the future we will present a more complete and comprehensive model. The future of news detection undoubtedly passes through intelligence and at Narrativa we are prepared for it.

Do you want a solution for your media/company? Get in contact with us!