Research
Hold on to Your People: A Resource for Office Managers, Principal Dentists and Dental Spouses
75-Minute Interactive Discussion / Lecture · Ontario Dental Association Annual Spring Meeting
Toronto, ON · May 8-10, 2025
Jeremy Barr
Robust Cross-Validation for Small-Sample NLP: Transformer Fine-Tuning with Clinical Text
Seminar · MCSC Seminar (Modelling and Computational Science)
SHA361, Ontario Tech University · September 24, 2025
Jeremy Barr
Small-sample text classification presents fundamental challenges for reliable model training, particularly in domains where labels may be unreliable. Using alexithymia detection as a test case, we demonstrate that transformer fine-tuning can achieve clinical-grade performance (89% accuracy, κ=0.77)…
Fine-Tuning Large Language Models for Psychological Assessment: Lessons from Research on Alexithymia and Emotional Intelligence
12-Minute Talk · Canadian Psychological Association Annual National Convention
St. John's, NL · June 12-14, 2025
Jeremy Barr, Samantha M. van Rens, Colin T. Henning, N. Jacob Persi, Alexander McIntosh, James D.A. Parker
Traditional psychological research is constrained by small sample sizes due to data collection limitations and privacy concerns. Using alexithymia and Emotional Intelligence as test cases, we demonstrate novel approaches to fine-tuning large language models that can help address these fundamental…
No Words for Feelings: Developing Large Language Model Representations of Alexithymia
Poster · Canadian Psychological Association Annual National Convention
St. John's, NL · June 12-14, 2025
Jeremy Barr, Samantha M. van Rens, Colin T. Henning, Alexander McIntosh, James D.A. Parker
Alexithymia, characterized by difficulties identifying and describing emotions, presents unique challenges for psychological assessment due to its self-report paradox. Traditional measures like the Toronto Alexithymia Scale (TAS-20) rely on individuals' ability to recognize their emotional…
Large Language Model-based Classification for Emotional and Social Intelligence
Poster · Canadian Psychological Association Annual National Convention
Ottawa, ON · June 21-23, 2024
Jeremy Barr, Jacob Persi, Colin Henning, James D.A. Parker
This pilot study investigates the feasibility of fine-tuning a large pre-trained language model, specifically DistilBERT, to classify levels of emotional and social Intelligence (EI and SI) using a modified constructed-response Situational Judgment Test (SJT). The main aim of this study was to…