Abstract
This article focuses on sport climbing and on the design of innova- tive tools for supporting climbers to browse and search routes to climb. The difficulty of a route, its grade, is normally assessed by expert climbers, named route setters. A regular climber, after trying a route, may perceive it more or less difficult than the route setter. It is important to estimate this climber’s perceived difficulty of the routes in order to suggest the routes that have a target perceived dif- ficulty as expected by the climber. We develop a knowledge-based approach that uses domain-specific features to train a predictive model. Additionally, the problem is modeled as a rating prediction task in a recommender system, using a matrix factorization ap- proach with a custom normalization solution. The knowledge-based approach enables us to develop a grade prediction explanation func- tionality. In off-line experiments, we demonstrate improvements over a baseline. Moreover, we show how the proposed techniques can be exploited in an app developed by a major company offering information services to the sport climbing market.