Two-sided Facility Location

Recent years have witnessed the rise of many successful e-commerce marketplace platforms like the Amazon marketplace, AirBnB, Uber/Lyft, and Upwork, where a central platform mediates economic transactions between buyers and sellers. Motivated by these platforms, we formulate a set of facility location problems that we term Two-sided Facility location. In our model, agents arrive at nodes in an underlying metric space, where the metric distance between any buyer and seller captures the quality of the corresponding match. The platform posts prices and wages at the nodes, and opens a set of ”virtual markets” or facilities to route the agents to. The agents at any facility are assumed to be matched. The platform ensures high match quality by imposing a distance constraint between a node and the facilities it is routed to. It ensures high service availability by ensuring flow to the facility is at least a pre-specified lower bound. Subject to these constraints, the goal of the platform is to maximize the social surplus subject to weak budget balance, i.e., profit being non-negative. We present an approximation algorithm for this problem that yields a (1+ ǫ) approximation to surplus for any constant ǫ > 0, while relaxing the match quality (i.e., maximum distance of any match) by a constant factor. We use an LP rounding framework that easily extends to other objectives such as maximizing volume of trade or profit. We justify our models by considering a dynamic marketplace setting where agents arrive according to a stochastic process and have finite patience (or deadlines) for being matched. We perform queueing analysis to show that for policies that route agents to virtual markets and match them, ensuring a low abandonment probability of agents reduces to ensuring sufficient flow arrives at each virtual market. Such an analysis also helps us posit facility location variants that capture settings where the platform elicits deadlines truthfully by posting lotteries over different prices and wages for different deadlines. ∗Department of Computer Science, Duke University, Durham, NC 27708-0129. Email: [email protected]. Supported by NSF grant CCF-1408784. †School of Operations Research and Information Engineering, Cornell University, Ithaca, NY 14853. Email: [email protected] ‡Google Research, Mountain View, CA 94043. Email: [email protected] §Google Research, Mountain View, CA 94043. Email: [email protected] ¶Department of Computer Science, Duke University, Durham, NC 27708-0129. Email: [email protected]. Supported by NSF grants CCF-1408784, CCF-1637397, and IIS-1447554.