Data-Driven Models with Applications in Crowdsourcing, Retail, and Peer-to-Peer Lending

Abstract

Making predictions based on historical data has a wide range of applications. Nowadays with the advent of more and more online platforms, we are able to design various machine learning models and test them with online data. In this thesis, there are three main chapters, each focusing on a specific application of data-driven modeling problems. The first chapter is devoted to the development of rank aggregation algorithms in the setting of crowdsourcing. The basic question we want to answer is how to efficiently collect a large amount of high-quality pairwise comparisons for the ranking purpose. The second chapter focuses on designing and comparing different demand prediction models for online retail with promotions. The data obtained from online retail platforms has a distinctive hierarchical structure and our goal is to utilize such a structure to help us achieve higher prediction accuracy. Experimental evaluations on both synthetic and real data are available at the end of the chapter. Finally, the last chapter focuses on the online peer-to-peer lending market. In particular, we want to infer the impact of IPOs on firms operational behaviors. We use one firm (public) as treatment and the other one (private) as control to conduct causal inference.