2025 Theses Doctoral

Platform Operations: From Models to Methods

Kumar, Akshit

Online platforms – from on-demand home-services and e-commerce fulfillment networks to content streamers – thrive on their ability to match demand with supply. Doing so well requires algorithms that cope with uncertainty, high-dimensional type spaces, mis-aligned and multiple objectives, and informational frictions. This dissertation develops models and methods that illuminate these challenges and propose simple, and often near-optimal solutions. The dissertation is organized into five self-contained but thematically linked chapters – the first three chapters are related to problems in online matching, dynamic resource allocation and multi-objective optimization in order fulfillment problems, while, the last two chapters deal with design and optimization of recommendation systems.

First, we tackle algorithm design questions that arise in online matching and dynamic resource allocation problems. How should a centralized matching platform dynamically match heterogeneous service providers with heterogeneous customers? What are the fundamental drivers of algorithmic performance in these dynamic resource allocation platforms? Is there a unifying algorithmic principle which can solve large class of dynamic resource allocation problems? How to make resource allocation decisions with multiple and often competing objectives?

In Chapter 1, we study dynamic two-sided matching with heterogeneous demand and supply modeled as highly dimensional weight and feature vectors respectively. We show that simple myopic policies such as Greedy which are practically prevalent can impact be highly sub-optimal. We develop a forward-looking supply aware policy dubbed Simulate-Optimize-Assign-Repeat (SOAR). We prove that SOAR achieves the optimal regret scaling under different assumptions on the demand and supply distributions. In Chapter 2, we broaden the scope to general online resource allocation problems. We identify a novel driver of algorithmic performance – the spatial distribution of demand types. We develop a unifying algorithm dubbed Repeatedly Act using Multiple Simulations (RAMS) which is a generalization of SOAR studied in Chapter 1. In Chapter 3, we turn to multi-objective optimization in the context of order fulfillment problems. We develop a principled framework for weight generation to enable the weighted objective approach.

Next, we take the viewpoint of a system designer and tackle platform design questions in the context of recommendation systems. By optimizing for measurable proxies, are recommendation systems at risk of significantly under-delivering on utility? If so, how can one improve utility which is seldom measured? Different information provisioning tools exists, such as public rankings and personalized recommendations, but when do these tools work and when do they not? What is the role of the market setting in the driving the efficacy of these different information provisioning tools?

In Chapter 4, we study a stylized model of repeated user consumption. We demonstrate that optimizing for measurable proxies like engagement can lead to significant utility losses. Instead, we propose a utility-aware policy that initially recommends diverse set of options. As the platform becomes more forward-looking, our utility-aware policy achieves the best of both worlds: near-optimal utility and near-optimal engagement simultaneously. Our study elucidates an important feature of recommendation systems; given the ability to suggest multiple items, one can perform significant exploration without incurring significant reductions in engagement. By recommending high-risk, high-reward items alongside popular items, systems can enhance discovery of high utility items without significantly affecting engagement.

In Chapter 5, we ask when personalized recommendations are worth their added complexity relative to public rankings. In unconstrained supply settings, both public rankings and personalized recommendations improve welfare, with their relative value determined by the degree of preference heterogeneity. In contrast, in supply-constrained settings, revealing just the common term of the utility, as done by public rankings, provides limited benefit since the total common value available is limited by capacity constraints, whereas personalized recommendations, by revealing both common and idiosyncratic terms, significantly enhance welfare by enabling agents to match with items they idiosyncratically value highly.