Date of Award
5-2025
Degree Type
Thesis
Degree Name
Master of Science (MS)
Department
Interdisciplinary Informatics
First Advisor
Dr. Ada Rhodes Wish
Second Advisor
Dr. Joel Elson
Third Advisor
Dr. Jorge Fandinno Garcia
Abstract
This thesis combines two complementary research efforts examining how artificial intelligence, specifically OpenAI’s ChatGPT, interacts with and processes human emotions. The first paper explores ChatGPT’s emotional adaptability, investigating whether the model’s responses vary in tone, style, or empathy when exposed to emotionally charged prompts. The study uses sentiment analysis tools, readability scoring, and emotional congruence matching to assess how the model interprets emotional cues embedded in the text. Drawing from theories in affective computing, psychology, and language modeling, this research aims to understand whether ChatGPT can simulate emotional awareness and adapt its communication style based on perceived human sentiment. These evaluations are grounded in a conversational context, allowing for detailed measurement of response modulation across different emotions such as happiness, sadness, fear, and anger.
The second paper extends this work by directly comparing human emotional perception with AI-generated emotion responses using Plutchik’s Wheel of Emotions. We designed a structured rating tool to capture emotional intensity ratings from human volunteers and ChatGPT across eight core emotions grouped into opposite pairs: joy-sadness, trust- disgust, fear-anger, and anticipation-surprise. Instead of using fixed labels, both sets of respondents provided ratings on a bipolar sliding scale, reflecting emotions' direction and intensity. This parallel data collection approach created a unified format for side-by-side comparison between natural and synthetic emotional expression. The study emphasizes the probabilistic nature of emotions and uses Plutchik’s visual model to interpret emotional gradients, oppositions, and blended responses across the dataset.
These papers offer a layered investigation into how large language models detect, interpret, and simulate emotional content in human-like ways. The combined body of work explores ChatGPT’s linguistic output in emotional contexts and builds a framework to measure and visualize its performance alongside human ratings. The approach centers on staying clear and straightforward. It avoids complicated tools and focuses instead on capturing the real depth and variety of emotions. These ideas add meaningful value to the evolving field of emotional AI. They offer a solid starting point for future studies, especially those aiming to build systems that respond to people in thoughtful, human-centered ways.
Recommended Citation
Pirisingula, Sridivya, "EVALUATING EMOTIONAL RESPONSIVENESS IN LARGE LANGUAGE MODELS: PLUTCHIK’S EMOTIONS AS A FRAMEWORK" (2025). Interdisciplinary Informatics Theses, Dissertations, and Student Creative Activity. 3.
https://digitalcommons.unomaha.edu/interdiscipinformaticsstudent/3
Comments
The author holds the copyright to this work and any reuse or permission must be obtained from them directly.