Abstract
When visiting a busy urban area and its attractions, tourists often experience discrepancies between their expected and actual satisfaction. In fact, popular attractions tend to be crowded, forcing visitors to spend extra time in queues or cramped spaces, which negatively impacts their actual experience. Recommender systems that guide tourists to less crowded attractions could alleviate this problem. However, for the system to be effective, the recommendations must have sufficient salience to convince tourists to accept them at decision time (trip plan). Recommender systems may produce different levels of salience, and therefore will influence tourists in different ways. This raises the question of how the salience of recommendations can quantitatively affect tourists behaviour and satisfaction. In this paper, we propose a data-driven simulation protocol to model the impact of recommendation salience on the expected utility of the recommended items and ultimately on the users’ choices. It is based on the Plackett-Luce choice model, with a utility function that is decomposed in three components: popularity bias, topic preferences, and distance to attraction. By clearly separating these components of the decision utility, we can better model how recommendation induced salience affects user utility perception: by temporarily increasing the popularity of the recommended items. Using a case study in the city of Verona we simulate how a non personalised recommender system that promotes less crowded sites impacts on the users’ choices and ultimately on the actual experience of tourists. Our simulation approach provides insights for destination managers looking to implement similar systems on-site.