Year: 2025
Type: dissertation
Abstract: This study explores the application of Large Language Models (LLMs) for the classification of patient text messages written in Icelandic, which have been compiled and reviewed by physicians. The growing volume of clinical notes and their unstructured nature present significant challenges to medical professionals seeking efficient ways to process these documents. Although LLMs offer a promising solution, the scarcity of medical data presents a key obstacle. To address this, in addition to investigating the effectiveness of various LLMs, we also employ data augmentation techniques to artificially expand the training data, examining how this impacts the classification performance. Our findings show that Masked Language Models (MLMs), particularly Ice- BERT variants, achieved the highest performance in terms of accuracy and robustness, outperforming traditional models and Causal Language Models (CLMs). Data augmentation improved model performance in some cases, but highlighted the limitations of a small dataset a
Source: Skemman
Author: Kaustab Chandra Sahu
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W6982884178