Exploiting Similar Behavior of Users in a Cooperative Optimization Approach for Distributing Service Points in Mobility Applications

In this contribution we address scaling issues of our previously proposed cooperative optimization approach (COA) for distributing service points for mobility applications in a geographical area. COA is an iterative algorithm that solves the problem by combining an optimization component with user interaction on a large scale and a machine learning component that provides the objective function for the optimization. In each iteration candidate solutions are generated, suggested to the future potential users for evaluation, the machine learning component is trained on the basis of the collected feedback, and the optimization is used to find a new solution fitting the needs of the users as good as possible. While the former concept study showed promising results for small instances, the number of users that could be considered was quite limited and each user had to evaluate a relatively large number of candidate solutions. Here we deviate from this previous approach by using matrix factorization as central machine learning component in order to identify and exploit similar needs of many users. Furthermore, instead of the black-box optimization we are now able to apply mixed integer linear programming to obtain a best solution in each iteration. While being still a conceptual study, experimental simulation results clearly indicate that the approach works in the intended way and scales better to more users.