Recommender systems have become a regular feature in our daily lives. They influence the books we read, the movies we watch and the content we consume on social media. There is opportunity to apply recommender systems to more complex domains, such as exercise, and in this paper we consider how such systems can play a role in supporting runners as they train for a marathon. However, making recommendations for more complex domains introduces additional challenges such as how to provide varied recommendations and how to evaluate these suggestions. In this work we address both of these issues using a stratified case-based recommendation approach and the use of so-called pseudo-cases for evaluation. The stratified approach allows for different recommendations to be generated for each runner based on whether they would like to continue along their current training trajectory, or target a more ambitious or a more conservative goal. We further describe how to evaluate these recommendations in terms of their feasibility, plausibility, effectiveness and safety using a large-scale, real-world dataset of more than 130,000 runners and their marathon training experiences.
© Copyright 2024 Artificial Intelligence XLI. Alle Rechte vorbehalten.