Creating structured and FAIR marathon data

Abstract

Marathon running poses distinct challenges that demand personalized, data- driven solutions; however, the domain has witnessed relatively little technological innovation to meet those needs. This work concentrates on creating structured and FAIR datasets specifically for marathon-oriented dialogue system. Using role-playing frameworks, the research creates natural and context-specific dialogues to fill gaps in race selection, training guidance, and marathon preparation. Furthermore, large language models are leveraged for generating a synthetically dialogical seed dataset with domain-specific expertise validation and the extraction of real-world scenarios using crowdsourced professional athletes and coaches, for adding authenticity. Efforts here are targeted to come up with something that could be scalable to offer higher-quality resources which TODS will subsequently utilize to yield action-worthy insights and recommendation values customized to runners’ profiles. This work there- fore underlines the transformative potential of Target-oriented Dialogue System (TODS) in delivering tailored help for specific niched domains, with the methodologies underlying having broader applicability across endurance sports.