Analisis Sentimen Berbasis Aspek (ABSA) Menggunakan LDA, BERTopic, dan IndoBERT-GRU Terhadap Ulasan Pelanggan pada Aplikasi Coffee Purveyor Indonesia

Alyadrus, Diva Ardelia (2026) Analisis Sentimen Berbasis Aspek (ABSA) Menggunakan LDA, BERTopic, dan IndoBERT-GRU Terhadap Ulasan Pelanggan pada Aplikasi Coffee Purveyor Indonesia. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5026221029-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (9MB) | Request a copy

Abstract

Sektor makanan dan minuman (F&B) di Indonesia berkembang pesat sehingga perusahaan kopi lokal seperti Kopi Kenangan dan Fore Coffee semakin mengandalkan layanan digital yang menghasilkan ribuan ulasan pengguna sebagai sumber electronic word-of-mouth. Namun, analisis sentimen konvensional belum mampu mengidentifikasi sentimen pada berbagai aspek dalam satu ulasan, sementara pengembangan model deep learning untuk Aspect-Based Sentiment Analysis (ABSA) masih terkendala keterbatasan data berlabel. Penelitian ini bertujuan membangun kerangka kerja ABSA menggunakan pendekatan hibrida dengan membandingkan Latent Dirichlet Allocation (LDA) dan BERTopic untuk ekstraksi aspek, serta memanfaatkan Distant Supervision berbasis Large Language Model (Qwen2.5-7B dengan 3× self-consistency) untuk membangun silver dataset yang digunakan dalam pelatihan model IndoBERT-GRU. Hasil penelitian menunjukkan bahwa LDA (Exp3, K=6) memperoleh nilai coherence 0,5989, sedangkan BERTopic (MiniLM, min_cluster_size=1000) menghasilkan topik yang lebih bersih tanpa noise, namun dengan nilai coherence yang lebih rendah (0,5452). Kedua metode secara konsisten mengidentifikasi aspek RASA, LAYANAN, dan PROMO, sedangkan aspek HARGA ditemukan bersifat cross-cutting, sehingga penelitian menetapkan lima aspek akhir yang digunakan pada tahap klasifikasi sentimen. Evaluasi model IndoBERT-GRU mencapai nilai macro-F1 sebesar 0,7624, dengan performa terbaik pada aspek APLIKASI (0,8636). Secara keseluruhan, penelitian ini menunjukkan bahwa kombinasi topic modeling, distant supervision, dan model berbasis IndoBERT efektif digunakan untuk membangun sistem ABSA dengan kebutuhan anotasi manual yang minimal.
=======================================================================================================================================
Indonesia's Food and Beverage (F&B) industry has experienced rapid growth, prompting local coffee companies such as Kopi Kenangan and Fore Coffee to increasingly rely on digital services that generate thousands of user reviews as a valuable source of electronic word-of-mouth. However, conventional sentiment analysis is unable to identify sentiments toward multiple aspects within a single review, while the development of deep learning models for Aspect-Based Sentiment Analysis (ABSA) remains constrained by the limited availability of labeled data. This study aims to develop a hybrid ABSA framework by comparing Latent Dirichlet Allocation (LDA) and BERTopic for aspect extraction and employing Large Language Model-based Distant Supervision (Qwen2.5-7B with 3× self-consistency) to construct a silver dataset for training an IndoBERT-GRU model. The results show that LDA (Exp3, K=6) achieved a coherence score of 0.5989, whereas BERTopic (MiniLM, min_cluster_size = 1000) produced cleaner topics without noise but obtained a lower coherence score (0.5452). Both methods consistently identified the TASTE, SERVICE, and PROMOTION aspects, while the PRICE aspect was found to be cross-cutting, leading to the definition of five final aspects for the sentiment classification stage. The IndoBERT-GRU model achieved a macro-F1 score of 0.7624, with the best performance on the APPLICATION aspect (0.8636). Overall, this study demonstrates that the combination of topic modeling, distant supervision, and an IndoBERT-based model is effective for developing an ABSA system while requiring minimal manual annotation.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	ABSA, IndoBERT, BERTopic, LDA, Distant Supervision
Subjects:	T Technology > T Technology (General) T Technology > T Technology (General) > T57.5 Data Processing
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Information System > 57201-(S1) Undergraduate Thesis
Depositing User:	Diva Ardelia Alyadrus
Date Deposited:	28 Jul 2026 07:35
Last Modified:	28 Jul 2026 07:35
URI:	http://repository.its.ac.id/id/eprint/138928

Actions (login required)

View Item