Fake news or ‘Fake News’ and offensive content remain highly relevant topics for most media companies and society. The problem arises when too much information is shared at a high frequency over the Internet, so there is not enough time left for a manual review. As a result, and in the absence of solutions to monitor content, companies hire more and more employees to review and verify content and prevent inappropriate information from going viral.
However, this is not a long-term solution because as the volume of data grows rapidly, manual handling of misinformation and inappropriate content becomes very costly. Manual fact-checking also introduces biases into the selection process, as it is almost impossible to prevent the personal opinions and beliefs of team members from influencing your judgment. In addition to fake news, there is also the fight against offensive material, that is, content with violence and / or nudity, which is published on social networks and is manually flagged and reported by users.
And this is what a Deloitte study entitled ‘Cognitive Artificial Intelligence, the invisible invasion of the media’ tries to tell us, offering solutions and recommendations. The main claim of the consultancy is that “cognitive artificial intelligence (AI) can help ensure the accuracy of data, identify fake news and automate the censorship or removal of such content in an intelligent way. In the digital age, not only social media companies, but also traditional media companies can use cognitive technologies to detect inappropriate content, helping them to remain credible and trustworthy. “
The question then is how and where to start. Deloitte says that you can start by using natural language processing (NLP) techniques to extract information from a text, be it an article or a post on a social media platform. A classification model helps to identify the general topic (s) that a text is about and whether a text is clickbait or not. The latter would work in a similar way to how spam is identified. Other techniques for the extraction of information can be used, which basically consist of extracting structured data from a text.
There are many different techniques for performing text classification. A simple one would be unsupervised machine learning called k-means clustering , applied to a collection of sample texts. K-means groups those texts that are most similar in different groups. As this is a technique for unsupervised learning, the engineer would have to look at the resulting groups and assign names to them (eg appropriate and inappropriate). The resulting model can be used to classify new text. Of course, this is a very limited approach; there are much better, but more complex methods . An example would be using neural networks and a technique called ‘word2vec’.
In case a text comes with an image or a video, which is technically just a sequence of images, a model for image classification can be used to identify its content. Facebook, for example, uses neural networks to add tags about the content of a photo to each photo that is uploaded. If a tag is deemed problematic, the photo / video can simply be removed.
To build such a model, you would first have to create a database containing many tagged images containing the unwanted content. Using artificial intelligence and machine learning, these can be used to train a model (for example, a neural network) that displays every image posted on a website. The same principle applies to videos, but it requires a lot more data processing power.
In a next step, the information extracted from the content in the previous processing stages can be used to verify the facts, which could be done simply by comparing the information now structured with the information in the databases or by comparing the information with information extracted. from reliable sources.
In short, the use of cognitive AI enables media companies to combat fake news and violent content more effectively and with greater reliability. It can help editors and journalists make informed decisions about publishing a specific piece of content, as it has the power to cope with the vast amount of data and search the entire internet for evidence to verify a piece of content , by time that eliminates political and emotional judgment that human beings have on certain issues.
“The main benefits of cognitive artificial intelligence in the fight against fake news,” Delotte says, “enables considerable cost savings by eliminating time-consuming fact checks, so editors and journalists can focus on their main work and greater reliability ”.
