A multi‐agent system for itinerary suggestion in smart environments

Modern smart environments pose several challenges, among which the design of intelligent algorithms aimed to assist users. When a variety points interest are available, for instance, trajectory recommendations needed suggest users most suitable itineraries based on their interests and contextual constraints. Unfortunately, in many cases, these must be explicitly requested lack causes so-called cold-start problem. Moreover, lengthy travelling distances excessive crowdedness specific make itinerary planning more difficult. To address aspects, multi-agent suggestion system that aims at assisting an online collaborative way is proposed. A profiling agent responsible detection groups whose movements characterised by similar semantic, spatial temporal features; then, recommendation leverages information dynamically associates current user with clusters according Multi-Armed Bandit policy. Framing as reinforcement learning problem permits provide high-quality suggestions while avoiding both preference elicitation issues. The effectiveness approach demonstrated some deployments real-life scenarios, such campuses theme parks.