Predicting Bundles of Spatial Locations from Learning Revealed Preference Data

We propose the problem of predicting a bundle of goods, where the goods considered is a set of spatial locations that an agent wishes to visit. This typically arises in the tourism setting where attractions can often be bundled and sold as a package to visitors. While the problem of predicting future locations given the current and past trajectories is well-established, we take a radical approach by looking at it from an economic point of view. We view an agent's past trajectories as revealed preference (RP) data, where the choice of locations is a solution to an optimisation problem according to some unknown utility function and subject to the prevailing prices and budget constraint. We approximate the prices and budget constraint as the time costs to finish visiting the chosen locations. We leverage on a recent line of work that has established algorithms to efficiently learn from RP data (i.e., recover the utility functions) and make predictions of future purchasing behaviours. We adopt and adapt those work to our original setting while incorporating techniques from spatiotemporal analysis. We experiment with real-world trajectory data collected from a theme park. Our predictions show improved accuracies in comparison with the baseline methods by at least 20, one of which comes from the spatiotemporal analysis domain.