Towards Personalized Data-Driven Bundle Design with QoS Constraint
In this paper, we study the bundle design problem for offering personalized bundles of services using historical consumer redemption data. The problem studied here is for an operator managing multiple service providers, each responsible for an attraction, in a leisure park. Given the specific structure of interactions between service providers, consumers and the operator, a bundle of services is beneficial for the operator when the bundle is underutilized by service consumers. Such revenue structure is commonly seen in the cable television and leisure industries, creating strong incentives for the operator to design bundles containing lots of not-so-popular services. However, as customers might choose to bypass a bundle completely if it is not sufficiently attractive, we need to impose a quality of service (QoS) constraint on the lower bound of the perceived attractiveness. In this paper, we make two major contributions (1) recognizing the inherent differences in customer preferences, we propose an approach for detecting different user classes, and for each user class, make an appropriate bundle recommendation; and (2) in order to make the bundling scheme even more adaptive to unknown customer preferences, we propose a dynamic bundling strategy, which allows customers to “trade in” any number of undesirable services dynamically so that they can be replaced by an alternative set of services. A step to generate fixed or static bundles is also studied. The pros and cons of different bundling strategies are illustrated using a real-world dataset collected from a large leisure park operator in Asia that manages a large collection of attraction providers
