Date of Award

5-2026

Document Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

Department

School of Computing

Committee Chair/Advisor

Dr. Emma Dixon

Committee Member

Dr. Andrew Duchowski

Committee Member

Dr. Long Cheng

Committee Member

Dr. Guo Freeman

Abstract

This dissertation examines the accessibility, readability, emotional tone, and trustworthiness of digital, text‑based dementia information and investigates how older adults with and without dementia cognitively and affectively engage with such content. It further explores the potential of generative AI to improve dementia information delivery through the development of a domain‑specific retrieval‑augmented generation (RAG) large language model (LLM) chatbot designed to deliver accessible, stigma‑free information. The dissertation comprises four studies using complementary quantitative and qualitative methods. Study I conducts a large‑scale content analysis of 300 medical articles, 35 websites, and 50 blogs, applying readability metrics, LIWC‑based linguistic and sentiment analysis, and thematic analysis of intended audiences. Study II employs a mixed factorial design with webcam‑based eye‑tracking and facial expression analysis to examine visual attention, cognitive effort, and emotional responses among people with dementia (PWD) and older adults without dementia engaging with digital dementia information of varying complexity and credibility. Study III uses qualitative focus groups to investigate how PWD interact with and attempt to co‑customize commercially available generative AI chatbots. Study IV develops and evaluates two versions of a domain‑specific RAG‑LLM chatbot, comparing a medical‑only knowledge base with a hybrid knowledge base integrating medical evidence and lived‑experience sources. Findings show that most digital dementia information is too difficult to read and overly negative, despite targeting PWD. While PWD visually scan content similarly to other older adults, they experience higher cognitive load, less stable eye movements, and stronger emotional responses, particularly with complex medical content. Co‑customizing existing generative AI tools was found to be cognitively demanding and fatiguing. The hybrid RAG‑LLM chatbot outperformed the medical‑only version, delivering more readable, emotionally balanced, accurate responses with fewer hallucinations. Overall, this dissertation demonstrates a person‑centered approach to delivering accessible, trustworthy, and emotionally sensitive dementia information.

Author ORCID Identifier

0009-0002-5074-2013

Share

COinS
 
 

To view the content in your browser, please download Adobe Reader or, alternately,
you may Download the file to your hard drive.

NOTE: The latest versions of Adobe Reader do not support viewing PDF files within Firefox on Mac OS and if you are using a modern (Intel) Mac, there is no official plugin for viewing PDF files within the browser window.