Automated evaluation of sport and leisure accident reports using natural language processing

This paper describes the automated evaluation of sport and leisure accident report data. Two approaches were taken to analyze the provided data. The first was about exploring the data and possible accident prevention measures. For this, GPT-3 and BERTopic are used. In the second approach, GPT-3 is used as well. This approach aims to find out, for specific predefined measures, how many accident cases could have been prevented using this measure. The results of these analyses are displayed using charts and can be used to find which measures would help prevent many accidents.