Graduation Semester and Year




Document Type


Degree Name

Doctor of Philosophy in Computer Science


Computer Science and Engineering

First Advisor

Manfred Huber

Second Advisor

David Levine


In today’s fast-paced and globally connected world, businesses are creating products with more significance to user personalization and customization. This has amplified the importance of capturing and learning user preferences as more information from users can lead to the designing and development of products that will improve user engagement and performance. Numerous algorithms based on collaborative filtering and recommender systems have been used to learn user preferences, but almost all of them require big datasets to train on. This creates a dependency on collecting more and more user information which might lead to ethical considerations and privacy concerns. To solve this dependency, we are proposing a novel architecture that aims to predict a user’s preference with minimal interactions and that integrates it into an overall system to optimize user performance. The concept of collaborative filtering in an embedding space is used where based on the information from previous users, the architecture predicts the preference of the new user, with minimal feedback interactions. This is then integrated with a second learning agent that aims at optimizing task settings to optimize a user performance measure while considering the predicted user preferences. In this dissertation, we have experimented with the structure of the architecture as well as the tools and algorithms that are used to create the architecture. We have then used parts of the architecture in two different domains to experiment and understand the positive impact and usefulness of the research. In the end, we used the complete architecture in a gaming domain and recorded its performance. Applications in various domains have shown promising results where it has effectively learned on-line about user preferences with minimal interaction with the real user and shown the ability to optimize task contexts for the specific user without the need for large amount of negative user experiences during learning.


Generative AI, User preference prediction, Collaborative filtering, Recommender system, Siamese network, Conditional generative adversarial network, Reinforcement learning


Computer Sciences | Physical Sciences and Mathematics


Degree granted by The University of Texas at Arlington