The detection of aggression and toxicity in user texts has become an increasingly popular topic in computer science and linguistics. At the hackathon, participants were tasked with training an AI algorithm to identify negative messages, delete them, or rewrite them to make them sound more respectful – all without losing a user’s main point, if possible.
“It took us 24 hours to solve this problem. We took the ready-made translator No Language Left Behind and converted it from Russian to the Tatar language. Then, we translated quite a few pairs of negative and benign messages into Tatar and used it to train our model,” shares Daniil Antonov, an ITMO student and a member of the Sota team.
The solution developed by the students is based on Russian-trained AI models that were modified for the less widespread Tatar language.
The competition was organized by the Tatarstan Academy of Sciences and AIRI Institute and brought together 368 students across the country.
