Analisis Sentimen Berbasis Aspek Menggunakan Modified CABILSTM-IndoBERT pada Sistem Komparasi Produk Kecantikan

Valentina, Clara (2026) Analisis Sentimen Berbasis Aspek Menggunakan Modified CABILSTM-IndoBERT pada Sistem Komparasi Produk Kecantikan. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5027221016-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (3MB) | Request a copy

Abstract

Pesatnya pertumbuhan industri kosmetik menghasilkan volume ulasan pengguna yang sangat besar pada platform digital seperti Female Daily Network. Fenomena information overload dan subjektivitas ulasan menyulitkan konsumen dalam melakukan komparasi produk bibir secara objektif. Penelitian ini mengembangkan sistem Aspect-Based Sentiment Analysis (ABSA) berbasis website bernama LipSense yang mengimplementasikan arsitektur CABiLSTM-IndoBERT untuk memfasilitasi komparasi produk bibir secara real-time. Penelitian ini membandingkan kinerja model Baseline IndoBERT, CABiLSTM-IndoBERT dengan Cross Entropy Loss, dan model usulan dengan Focal Loss dan Class Weighting melalui 19 eksperimen pada format klasifikasi 3-kelas dan 4-kelas. Hasil pengujian menunjukkan model CABiLSTM-IndoBERT dengan IndoBERT Frozen dan Cross Entropy Loss tanpa pembobotan mencapai Macro F1-score terbaik sebesar 0,6679, mengungguli model Baseline (0,5199). Berbeda dari hipotesis awal, Focal Loss secara konsisten menurunkan performa, sementara Class Weighting dengan skema Sqrt-normalized terbukti lebih efektif memitigasi ketidakseimbangan kelas. Model terbaik diintegrasikan ke dalam sistem yang menghitung Adjusted Net Sentiment Score (NSS) dan memvisualisasikannya melalui Radar chart sebagai alat bantu pengambilan keputusan konsumen. Hasil User Acceptance Testing terhadap 15 responden menunjukkan tingkat penerimaan tinggi dengan grand mean 4,61 dari skala 5,00.
==================================================================================================================================
The rapid growth of the cosmetics industry generates massive volumes of user reviews on digital platforms such as Female Daily Network. Information overload and review subjectivity complicate objective product comparisons for lip products. This research develops a web-based Aspect-Based Sentiment Analysis (ABSA) system called LipSense implementing a CABiLSTM-IndoBERT architecture to facilitate real-time lip product comparisons. This research compares the performance of Baseline IndoBERT, CABiLSTM-IndoBERT with Cross Entropy Loss, and the proposed model with Focal Loss and Class Weighting through 19 experiments across 3-class and 4-class classification formats. Testing results show that CABiLSTM-IndoBERT with Frozen IndoBERT and Cross Entropy Loss without Weighting achieves the best Macro F1-Score of 0.6679, outperforming the Baseline model (0.5199).
Contrary to the initial hypothesis, Focal Loss consistently decreases performance, while Class Weighting with a Sqrt-normalized scheme proves more effective in mitigating class imbalance. The best model is integrated into a system that calculates Adjusted Net Sentiment Score (NSS) and visualizes it through Radar charts as a consumer decision-support tool. User Acceptance Testing results from 15 respondents show a high acceptance rate with a grand mean of 4.61 out of a 5.00 scale.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Analisis Sentimen Berbasis Aspek, IndoBERT, CABiLSTM, Produk Bibir, Female Daily, Focal Loss, Class Imbalance.
Subjects:	T Technology > T Technology (General) > T57.5 Data Processing
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Information Technology > 59201-(S1) Undergraduate Thesis
Depositing User:	Clara Valentina
Date Deposited:	16 Jul 2026 07:51
Last Modified:	16 Jul 2026 07:51
URI:	http://repository.its.ac.id/id/eprint/135221

Actions (login required)

View Item