A fine-tuned large language model for domain-specific with reinforcement learning

Ismail, Amelia Ritahani and Aminuddin, Amira Shazleen and Nurul, Afiqa and Zakaria, Noor Azura (2024) A fine-tuned large language model for domain-specific with reinforcement learning. In: 2024 3rd International Conference on Creative Communication and Innovative Technology (ICCIT), 7 Agust 2024, IIUM, Kuala Lumpur.

PDF - Published Version Restricted to Repository staff only Download (553kB) | Request a copy

Abstract

Large Language Models (LLMs) like GPT-3 and BERT have significantly shown advancement in natural language processing by providing robust tools for understanding and generating human languages. However, their broad but shallow knowledge across many domains often leads to less effective performance in domain-specific tasks, where detailed and special- ized knowledge is needed. To address this limitation, this paper investigates the effectiveness of fine-tuning LLMs for specific domains. The approach incorporates reinforcement learning to integrate user feedback, allowing the model to dynamically adjust and refine its responses. This will ensure the model adapts iteratively, improving communication and interaction with users. The fine-tuned model’s performance is evaluated using two domain-specific datasets—medical and dental. Evaluation metrics such as Levenshtein distance and cosine similarity are used to assess the textual accuracy and semantic relevance of the fine-tuned model. The results from the dental the medical datasets indicate a low level of textual differences and strong semantic alignment, respectively. These suggest that the fine- tuned model effectively processes and preserves the integrity of domain-specific content with the potential of fine-tuning LLMs to enhance their applicability in specific domains.

Item Type:	Proceeding Paper (Plenary Papers)
Uncontrolled Keywords:	Large Language Model, Reinforcement Learning, Domain-Specific, Fine-Tuning
Subjects:	Q Science > QA Mathematics > QA75 Electronic computers. Computer science
Kulliyyahs/Centres/Divisions/Institutes (Can select more than one option. Press CONTROL button):	Kulliyyah of Information and Communication Technology > Department of Computer Science Kulliyyah of Information and Communication Technology > Department of Computer Science
Depositing User:	Amelia Ritahani Ismail
Date Deposited:	18 Dec 2024 10:21
Last Modified:	18 Dec 2024 10:21
URI:	http://irep.iium.edu.my/id/eprint/116732

Actions (login required)

View Item