PSU Programming Students Create Fake News Identification Software
24.11.2020Fourth year students from the Institute of Foreign Languages and International Tourism at Pyatigorsk State University (majoring in “Intellectual systems in a humanitarian field”) has worked out a neural network model, which identifies fake news.
One day, being at informatics class of Associate Professor Olga Timchenko, Candidate of Economic Sciences, Pavel Posokhov, an author of Fake News Identifier project, shared his idea to search and detect fake news and asked for help. He told that he had already prepared a model that took 15th place among 1,041 participants at All-Russia Festival on Artificial Intellect and Algorithmic Programming “RuCode” (FakeNewsChallenge by Sberbank), but it needed more detailed elaboration.
The lecturer and groupmates supported the idea and started working on its realization. Pavel Posokhov, Stepan Skrylnikov, Valeriy Efanov, Stanislav Krapivin, Yuriy Kurochkin created a team, held several brainstorming sessions under the guidance of Olga Timchenko, and finalized the software.
After that, they decided to present their idea at the All-Russia Competition of Young Entrepreneurs, organized by the Russian Ministry of Science and Higher Education. The project of programming students on fake news identification made it to the final of the second All-Russia Competition of Young Entrepreneurs and got a big score after the online defense.
Pavel described his invention in fine and plain language: “The software is a neural network. It can learn and analyze using special mathematical transformations of information, “puts text into digits”. The bigger the network, the more factors it can count. Therefore, we used a model that we called BERT Large, and which has more than 340 million parameters. This allows it to be very precise. To determine the truthfulness of news, the model makes a comprehensive analysis of a text, considering the sense of a text and stylistics similar to what a person would do. Besides, the program can determine what type of news it is and specifies tags.”
Currently, there is no fully automatic solution for this global problem on the Russian market. So to fight fake news the students have worked out a functional model of a neural network, ready for release.
A more detailed interview of the project creators is available here.