Kalman-Based Adaptive Moment Estimation Optimisation Algorithm to Enhance GPT in LLMs for Medical Sentiment Analysis...

Area of Science:

• Natural Language Processing (NLP)
• Machine Learning
• Healthcare Informatics

Background:

• Large Language Models (LLMs) have advanced medical sentiment analysis from patient feedback.
• High training costs and data requirements limit LLM application in real-world healthcare.

Purpose of the Study:

• To introduce KAdam-EnGPT4LLM, an optimization algorithm for efficient LLM training.
• To develop GPT4ALL-MediSentAly-KAdam for improved medical sentiment analysis.

Main Methods:

• Proposed KAdam-EnGPT4LLM, combining Kalman filters and Adaptive Moment Estimation.
• Applied KAdam-EnGPT4LLM to train GPT4ALL for medical sentiment analysis, creating GPT4ALL-MediSentAly-KAdam.
• Evaluated performance on medical question-and-answer datasets.

Main Results:

• GPT4ALL-MediSentAly-KAdam demonstrated faster convergence and more stable training.
• Achieved superior Accuracy, Recall, F1-score, and Runtime compared to traditional LLMs.
• Outperformed baseline models including GPT4ALL, Ada, Babbage, Curie, and Duvinci.

Conclusions:

• KAdam-EnGPT4LLM significantly enhances LLM training efficiency and reduces costs for medical sentiment analysis.
• GPT4ALL-MediSentAly-KAdam offers a more effective solution for analyzing patient feedback in healthcare.