Managing math anxiety with AI: Innovative approach of AI-assisted learning
Loading...
Supplementary material
Other Title
Authors
Kaur, Rajwinder
Author ORCID Profiles (clickable)
Degree
Master of Applied Technologies (Computing)
Grantor
Unitec, Te Pūkenga – New Zealand Institute of Skills and Technology
Date
2025
Supervisors
Bell, Jamie
Ardekani, Iman
Ardekani, Iman
Type
Masters Thesis
Ngā Upoko Tukutuku (Māori subject headings)
Keyword
math anxiety
chatbots
prompt engineering
natural language processing
human-computer interaction
AI in education
educational technology
chatbots
prompt engineering
natural language processing
human-computer interaction
AI in education
educational technology
ANZSRC Field of Research Code (2020)
Citation
Kaur, R. (2025). Managing math anxiety with AI: Innovative approach of AI-assisted learning (Unpublished document submitted in partial fulfilment of the requirements for the degree of Master of Applied Technologies (Computing)). Unitec, Te Pūkenga - New Zealand Institute of Skills and Technology
https://hdl.handle.net/10652/6950
Abstract
RESEARCH QUESTIONS
1. How much assistance did fine-tuning provide to help the chatbot?
2. Does prompt engineering offer a substitute for fine-tuning?
3. How did the pre-trained model perform in solving fraction problems?
ABSTRACT
Math anxiety is a psychological condition characterized by feelings of tension, fear, apprehension, or helplessness that interfere with a person's ability to engage in and perform mathematical tasks. It often manifests when individuals are required to solve math problems, attend math classes, or take math-related tests. Addressing math anxiety involves both emotional and cognitive support, including encouraging a growth mindset, offering patient guidance, and using supportive learning tools such as AI-powered tutoring systems or chatbots that provide personalized and anxiety-sensitive assistance. The goal of the research presented here was to assist learners with math anxiety by producing an AI-assisted chatbot that provides emotional support in ways that reduces anxiety. A chatbot was developed but it was decided that it would be unethical to test it with people with math anxiety because of the potential for the chatbot to produce responses that could be harmful. Then the question became, how could the chatbot be developed and tested so that there was increased confidence in positive outcomes before testing with people. Specifically, the first research question became: what prompting techniques can be used to influence chatbot responses to align better with previously determined supportive examples in the dataset? The second research question was: does finetuning a chatbot on a training dataset provide significant improvements in response accuracy and alignment compared to prompt engineering? To address these questions, it was decided to develop a dataset of user inputs and chatbot responses that could serve three purposes. The primary purpose was to test that the chatbot gives reasonable emotional supportive responses, appropriate for people with math anxiety. The second purpose was guiding the responses of the chatbot using context optimisation via prompt engineering, inspired by cache augmented generation. The final purpose of the dataset was to fine-tune the model of the chatbot to give responses that were closely aligned to the dataset. Using BLEU and ROUGE metrics, showed that the fine-tuned model gave the best performance on an evaluation split of the dataset. The contributions of this research are the dataset, the chatbot and the experimental results from finetuning and context optimization by prompt engineering.
Publisher
Permanent link
Link to ePress publication
DOI
Copyright holder
Author
Copyright notice
All rights reserved