Date of Award
1-1-2023
Language
English
Document Type
Dissertation
Degree Name
Doctor of Philosophy (PhD)
College/School/Department
Department of Computer Science
Content Description
1 online resource (x, 182 pages) : illustrations (some color)
Dissertation/Thesis Chair
Shaghayegh Sahebi
Committee Members
Petko Bogdanov, Chinwe Ekenna, Peter Brusilovsky
Keywords
Human-computer interaction, User interfaces (Computer systems), Machine learning, Computer users
Subject Categories
Computer Sciences
Abstract
A tremendous amount of user data is collected daily due to technological progress that enables us to understand users better. The availability of this data also advances machine learning-based technologies, which aim to learn generic global patterns of user behavior from large volumes of data. User modeling is the process of building user profiles and finding the inherent representation of the user. Precise user modeling is critical for predicting users' future behavior and providing personalized services or products to individuals. Many machine learning models have been explored for user modeling to meet the increasing demand for user-centric technologies. Machine learning for user modeling is challenging because a user's inherent representation, such as user preference and cognitive state, is typically not observable and should be estimated or reasoned from the collected (observed) data, such as the user's demographic data and user-item response data. Also, many well-established user modeling methods are built based on the assumption that the user's inherent representation is static over time and that the user's sequential activities or events are independent. However, temporal dynamics naturally exist in user behavior, and the inherent representation of users may change over time in many real scenarios. This thesis aims to develop and improve sequential user modeling (SUM) and recommendation in applications where users' inherent representations are dynamic and unobserved but could be approximated from partially observed data. Our focus will be on educational and mental health domains, in which student knowledge states or user mental states are dynamic and not observable.
Recommended Citation
Wang, Chunpai, "Sequential user modeling and recommendation under partially observable environment" (2023). Legacy Theses & Dissertations (2009 - 2024). 3265.
https://scholarsarchive.library.albany.edu/legacy-etd/3265
