Competitive programming platforms have extensive archives of problems from past contests. The automatic tag system can help people use these archives effectively in training and practice.
We applied modern approaches to code analysis and natural language processing to predict tags using problem statements and submitted solutions.