Sentiment analysis of noisy malay text using a large language model

Due to the informality of social media, Malay user-generated content sentiment analysis is dificult. Existing methods struggle to capture cultural and contextual details. This study proposes publishing an open-source annotated dataset, fine-tuning an open-source large language model (LLM), and using an open-source chatbot interface to createa robust sentiment analysis model for noisy Malay text. The research addresses three main issues: lack of labelled Malay social media data, in suficient generic Malay language models, and lack of practical sentiment analysis tools. Its three goals are to create a diverse dataset with accurate sentiment labels, parameter-efficiently fine-tune an LLM, and export the model for an interactive chatbot. The process involves collecting social media data using Contextual Lexical Adaptation, pre-processing and analysing it, fine-tuning the TinyLlama LLM using LoRA, and comparingit to traditional models. Real-world applications, such as sentiment analysis of Malaysian tweets, will be shown using a locally deployed chatbot interface for fine-tuned model inference. This study lays the groundwork for practical sentiment analysis, benefiting businesses, researchers, and politicians seeking data-driven insights. This research aims to revolutionise open-source Malay sentiment analysis by addressing current limitations through an integrated approach.