Researchers from Jordan University of Science and Technology and Duquesne University created algorithms that could detect bots in languages other than English, thus opening the way for stopping fake news in languages other than English.
Social networks are increasingly used to actively spread misinformation. In fact, the numbers are astounding: more than one-fifth of stories shared on Tweeter during the 2016 US presidential elections were fake. Similarly, when the Catalan independence wave hit in 2017, bots were used to inundate audience with fake or biased content about the demonstrations. The spread of fake news has even been used to manipulate stock value!
It is critical that we learn how to deal with this challenge, especially in the new era of the new Coronavirus, when fake news can cause people to throw caution to the wind and risk their lives. Unfortunately, some misinformation is being spread by fake profiles and autonomous bots – even if those are remotely controlled by human beings.
As the use of bots and fake profiles – also known as sockpuppets – increases, more humans are exposed to dealing with them. It is therefore of the utmost importance to determine whether any piece of content has been written by actual human beings or by bots masquerading as such. This is the task of author profiling, in which pieces of text are studied to uncover the characteristics of the author: gender, age, native language and other characteristics.
There are currently many AI engines designed to do exactly that, but they are mainly trained and focused on the English language, making them largely useless for other languages.
Researchers from the Jordan University of Science and Technology in Jordan and Duquesne University in Pennsylvania, USA, recently came up with a novel method for detecting bots based on text alone – in Spanish no less. They trained their machine-learning algorithm on a large dataset that contained tweets in Spanish, in the hope of developing an optimized algorithm for the detection of bots.
The algorithm developed this way outperformed the state-of-the-art systems for detecting bots in Spanish, with a success rate of whopping 94%, proving that the researchers method of machine learning was capable of such impressive feats.
It can only be hoped that the same machine learning algorithms could be applied with the same rate of success for other languages as well. This development would mean that misinformation and fake news might actually be eradicated – or at least minimized – in languages other than English.
Then again, considering the situation with fake news in English at the moment – author profiling algorithms or not – perhaps we shouldn’t hold our breath. Fake news, it seems, are here to stay.
The research was conducted by Mohammed N. AlRashdan, Malak Abdullah and Mahmoud Al-Ayyoub from Jordan University of Science and Technology in Jordan, and by Yaser Jaraweh from Duquesne University from Pennsylvania, USA.
Original content by Nawartna