Named Entity Recognition Method for Sports Events based on Boundary Span Information

Named Entity Recognition (NER) is a research hotspot in the field of NLP, which is often used to identify entity boundaries and categories in text. In the field of sports events, due to the high complexity of the event name, the traditional named entity recognition method is difficult to accurately identify, resulting in low recognition accuracy. To address this problem, this paper proposes a named entity recognition method for sports events that integrates boundary span information. It mainly uses regression operations to refine the spatial location information of entities in sports field texts to improve the precision of recognition. This method predicts entity boundary words by adding boundary detection to improve the ability to capture boundary span information, and then realizes high-precision recognition of named entities of sports events. The experimental verification is carried out in the large-scale real corpus SGSum. The method achieves 97.91%, 98.22% and 97.39% F1 values in the identification of event name, team name and competition location name.