ASPECT-BASED SENTIMENT ANALYSIS IN INDONESIAN LANGUAGE USING GENERATIVE PRE-TRAINED LANGUAGE MODEL WITH MULTITASK LEARNING AND PROMPTING APPROACH

• Main Author: Zakya Suchrady, Randy
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/75263
• Institution: Institut Teknologi Bandung

Aspect-based sentiment analysis is a method in natural language processing aimed at identifying and understanding sentiments related to specific aspects of an entity. Aspects refer to words or phrases representing attributes or features of a particular entity. Previous research has utilized generative pre-trained language models to perform aspect-based sentiment analysis. LEGO-ABSA is one framework that successfully employs generative pre-trained language models in aspect-based sentiment analysis, particularly in English. LEGO-ABSA adopts a multitask learning and prompting approach to enhance the performance of the aspect-based sentiment analysis model. However, this approach has not been applied in the context of the Indonesian language. Therefore, this research aims to implement the multitask learning and prompting approach for aspect-based sentiment analysis in the Indonesian language using generative pre-trained language models. In this study, a model called Indo LEGO-ABSA is developed, which is an aspect-based sentiment analysis model utilizing generative pre-trained language models and trained through multitask learning and prompting. Indo LEGO-ABSA adopts the LEGO-ABSA framework with the mT5 model as its base and applies multitask learning to train all tasks within aspect-based sentiment analysis. The model is tested and compared with a previous aspect-based sentiment analysis model that adopted the GAS framework, named GAS-Indonesia. Indo LEGO-ABSA achieves f1-scores of 79.55%, 86.09%, 79.85%, 87.45%, and 88.09% for each task, namely Aspect Sentiment Triplet Extraction, Unified Aspect-based Sentiment Analysis, Aspect-Opinion Pair Extraction, Aspect Term Extraction, and Opinion Term Extraction. Meanwhile, GAS-Indonesia achieves f1-scores of 78.63%, 82.34%, 78.45%, 84.02%, and 87.74% for the same tasks.