Aspect Extraction on Indonesia Beauty Review Product using Pre-trained Language Model - Dalam bentuk buku karya ilmiah

CLARISA HASYA YUTIKA

Informasi Umum

Kode

24.05.594

Klasifikasi

006.35 - Natural Language Processing, Computer Science

Jenis

Karya Ilmiah - Thesis (S2) - Reference

Subjek

Natural Language Processing

Dilihat

96 kali

Informasi Lainnya

Abstraksi

Currently, the development of the beauty industry from 2022 to mid-2023 has grown by 21.9%. With the increase in the beauty industry, there are currently many forums to discuss and review these beauty products. One of the forums is Female Daily. The results of these reviews can convince consumers to buy these products. Based on Brightlocal data in consumer reviews, as many as 82% of respondents will delay buying a product if there is a negative review. So, the reviews greatly affect product quality. One way to get a conclusion of reviews is with aspect-based sentiment analysis. So, this research performs aspect extraction with token-level classification on beauty product reviews using pre-trained language models, namely IndoBERT and mBERT. The best performances are achieved by IndoBERT of 69% F1-score, while mBERT is 68% F1-score. This is because IndoBERT is better at handling Indonesian contexts due to its training on a large and specific Indonesian dataset, while mBERT is more flexible for multilingual data but less optimized for Indonesian-specific tasks. In addition, the performance for each aspect, the highest performances are obtained by aspect price of 75% F1-score at B-Price and 71% F1-score at I-Price, compared to the aspects of packaging and texture. This is because the price aspect is easier to recognize as it tends to have more specific terminology, such as cheap, expensive, etc.<br />