An efficient computational approach for automatic itinerary planning on web servers

Abstract

The automatic itinerary planning service requires to generate multiple-day schedules automatically under user-specified POIs and constraints. Knowing as an NP-Hard optimization problem, the task is commonly solved by (meta-)heuristic algorithms such as the genetic algorithms (GAs). However, considering the concurrent requests received by a web server in practice, the time efficiency of the existing itinerary planners can be rather unsatisfactory. To address the issue, this paper proposes a computational approach that hybridizing a GA with the reinforcement learning (RL) technology. The benefit is that we no longer need to re-execute the GA for each new request arrived. Instead, the approach keeps the historical solutions in track and maintains a RL agent to sequentially decide how to handle each new request. Experimental results show that the proposed approach is able to stably provide high-quality solutions, while greatly reducing the average time overhead of the web server.