Covid-19 information and disinformation: the role of Artificial Intelligence.

This article is intended to discuss Fake News in connection with Artificial Intelligence,  which  is one of the most important topics related to Communication and Social Networks.

The current situation due to Covid-19 pandemic brought many fears and uncertainties to our societies. In these last months, people from all over the world have relied on the media in order to search for the breaking news, checking the virus trend, preventive measures and treatments for it.

For that reason, we are witnessing an exponential increase of the volume of information that can be true, plausible and definitely fake. The so-called “fake news” have always been intentionally produced and widespread, and nowadays they have become a central issue because many web platforms and Social Networks such as Facebook, Instagram, Google, Twitter, Tik Tok and YouTube eliminate any kind of mediation by making inefficient all filters and controls of traditional publishing systems. Therefore, Social users can publish ideas or disseminate news through sharing, likes and retweets. In this context, while the urgent need to find an effective treatment for Covid-19 virus is still ongoing around the world, fake news proliferation, especially regarding healthcare, is constantly increasing and it constitutes a serious threat for public-health. Particularly in Social Networks, an automatic system which is able to detect and block fake news could represent a valid support to avoid the spread of viral and misleading information. 

AI has an important role in limiting the broadcast of fake news on social media, because it is capable of verifying a large number of posts and rapidly identifying those that are trying to overcome filters by inserting small changes in some contents already identified as false. Thus, AI task is to unmask false news, by classifying information through a “Truth-Scale”, which is composed by a high, medium and low level of truth. 

Deep Learning is used to analyze social media contents and to verify their authenticity. This last concept of authenticity includes many important signals such as: user’s features, the type of social network used and content polarity, along with users’ behavior, to whom a “truth-score” is assigned. In fact, fake news generators could show an unusual sharing behavior and they would tend to share also more extreme contents. All these characteristics, once gathered, contribute to give a more reliable estimate of authenticity.

Deep Learning algorithms responsible for detecting fake news usually employ a set of data (dataset), which is created thanks to data gathering from various social networks sources, such as Instagram, Facebook, Twitter. Hence, the activity of fake news recognition is formulated as a classification problem related to a single image, a specific text or a combination of a text and an image. Classification algorithms based on Deep Learning are used to solve this task of fake news recognition.

System of fake news recognition led by multimodal approach

In order to discriminate between true and false news, these algorithms require a first phase of training, in which a broad variety of data examples of both types of news is inserted (training set).

Thanks to this training set, the algorithm learns to recognize specific features belonging to fake news. That way, it is capable of distinguishing between true and false news. Therefore, the Deep Learning model consists of 3 components: networks for image classification, networks for text classification and total classification (text+image). When the trained algorithm has to analyze new news, it will be able to detect fake news according to what it had learnt in the training set. The used approach, based on neural networks, has the aim to reduce manual checks and, at the same time, to exploit these manual checks as forms of continuous learning.

In the future, these systems are going to reach such a high level of accuracy that they will be able to directly block and eliminate fake news on COVID-19 and on every general topic (violence, hate…).


Algorithm architecture for fake news recognition

The most advanced techniques of Deep Learning use neural networks on different layers that simulate biological neurons’ behavior. Deep Learning algorithm for fake news recognition has the purpose of extracting and classifying features automatically from multimedia contents (text/images/text+images) by exploiting the most famous social network. The system includes a learning phase aiming at collecting large datasets and learning from them. Nowadays, the possibility to do big data analytics requires advanced techniques and also computing power of computers working in parallel or on private clouds.