Adaptive Trip Recommendation System: Balancing Travelers among POIs with MapReduce

Abstract

Travel recommendation systems provide suggestions to the users based on different information, such as user preferences, needs, or constraints. The recommendation may also take into account some characteristics of the points of interest (POIs) to be visited, such as the opening hours, or the peak hours. Although a number of studies have been proposed on the topic, most of them tailor the recommendation considering the user viewpoint, without evaluating the impact of the suggestions on the system as a whole. This may lead to oscillatory dynamics, where the choices made by the system generate new peak hours. This paper considers the trip planning problem that takes into account the balancing of users among the different POIs. To this aim, we consider the estimate of the level of crowding at POIs, including both the historical data and the effects of the recommendation. We formulate the problem as a multi-objective optimization problem, and we design a recommendation engine that explores the solution space in near real-time, through a distributed version of the Simulated Annealing approach. Through an experimental evaluation on a real dataset, we show that our solution is able to provide high quality recommendations, yet maintaining that the attractions are not overcrowded.