Outdoor sport climbing is popular in Northern Italy due to its vast amount of rock climbing places (such as crags). New climbing crags appear yearly, creating an information overload problem for tourists who plan their sport climbing vacation. Recommender systems partly addressed this issue by suggesting climbing crags according to the most visited places or the number of suitable climbing routes. Unfortunately, these methods do not consider contextual information. However, in sport climbing, as in other outdoor activities, the possibility of visiting certain places depends on several contextual factors, for instance, a suitable season (winter/summer), parking space availability if traveling with a car, or the possibility of climbing with children if traveling with children. To address this limitation, we collected and analyzed the crag visits in Arco (Italy) from an online guidebook. We found that climbing contextual information, similar to users' content preferences, can be modeled by a correlation between recorded visits and crags features. Based on that, we developed and evaluated a novel context-aware climbing crags recommender system Visit & Climb, which consists of three stages as follows: (1) contextual information and content tastes are learned automatically from the users' logs by computing correlation between users' visits and crags' features; (2) those learned tastes are further made adjustable in a preference elicitation web interface; (3) the user receives recommendations on the map according to the number of visits made by a climber with similar learned tastes. To measure the quality of this system, we performed an offline evaluation (where we calculated Mean Average Precision, Recall, and Normalized Discounted Cumulative Gain for top-N), a formative study, and an online evaluation (in a within-subject design with experienced outdoor climbers N = 40, who tried three similar systems including Visit & Climb). Offline tests showed that the proposed system suggests crags to climbers accurately as the other classical models for top-N recommendations. Meanwhile, online tests indicated that the system provides a significantly higher level of information sufficiency than other systems in this domain. The overall results demonstrated that the developed system provides recommendations according to the users' requirements, and incorporating contextual information and crag characteristics into the climbing recommender system leads to increased information sufficiency caused by transparency, which improves satisfaction and use intention.
Read full abstract