In Natural Language Processing (NLP) and Information Extraction, Named EntityRecognition (NER) presents a significant challenge – autonomously identifying entities like Person, Location, and Organization from text. Despite NER research flourishing in English and Chinese, Turkish remains underrepresented, particularly in specific domains. The convergence of sports and technology has revolutionized sports management, enhancing performance and engaging fans. As global sports investments rise, the impact spans finance, marketing, and psychology. Qualitative insights from textual data offer a deeper understanding of athlete-team-supporter dynamics.
The integration of deep learning techniques in addressing Turkish NER, compared to conventional methods, remains underexplored. Furthermore, more research is needed to investigate interpretability and explainability within transformer-based models.
This study introduces domain-specific Turkish NER data sets and evaluates transformer-based models, shedding light on their interpretability. Notably, we contribute annotated sports data sets, compare models, analyze annotation formats’ impacts, and explore named entity distribution effects through cross-validation. Additionally, interpretability was employed to elucidate the rationale and mechanisms behind the predictions generated by the models. The findings bridge performance and understanding in Turkish NER, paving the way for enriched sports research and management practices.