Optimizing Web Forms using A/B Testing and Reinforcement Learning
Final project for the course CSC2558 - Topics in Multidisciplinary Human-Computer Interaction at the University of Toronto.
Abstract
In this study, we have performed a series of experiments to test various approaches to optimize web forms by improving user experience. Our web forms are developed by incorporating Design Thinking principles. We then conduct a live Randomized A/B experiment and present a group of people with two versions of the form randomly. Furthermore, we simulate Epsilon Greedy and Thompson Sampling on the data gathered from our experiment to evaluate the likelihood of a user choosing to fill an Optimized Form over its Unoptimized alternative. We comprehensively compare the performance of these algorithms on our data and conclude that Beta Bernoulli Thompson Sampling performs the best, with least regret therefore establishing the superiority of probabilistic approach.