Analisis Sentimen Berbasis Aspek pada Ulasan Desa Wisata Berkonsep Sustainable Tourism Menggunakan IndoBERT dengan Pendekatan Hierarchical Multi-Task Learning

Ho, Jason (2026) Analisis Sentimen Berbasis Aspek pada Ulasan Desa Wisata Berkonsep Sustainable Tourism Menggunakan IndoBERT dengan Pendekatan Hierarchical Multi-Task Learning. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5026221005-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (8MB) | Request a copy

Abstract

Pariwisata berkelanjutan menjadi salah satu fokus penting dalam pengembangan sektor pariwisata Indonesia, sejalan dengan meningkatnya kebutuhan memahami persepsi wisatawan melalui ulasan daring. Pendekatan pipeline dan single-task pada Aspect-Based Sentiment Analysis (ABSA) rentan mengalami error propagation ketika kesalahan pada tahap ekstraksi aspek memengaruhi klasifikasi sentimen pada tahap berikutnya. Penelitian ini mengembangkan model IndoBERT berbasis Hierarchical Multi-Task Learning (H-MTL) dengan mekanisme cross-attention antar-subtugas, kemudian membandingkannya dengan baseline Single-Task IndoBERT+CRF serta referensi EMC-GCN. Metodologi mencakup pemanfaatan 2.511 ulasan desa wisata, preprocessing menuju skema sequence labeling BIEOS 13 kelas, pelatihan dua-fase dengan joint optimization, serta evaluasi pada level entitas melalui Precision, Recall, dan F1-Score. Hasil eksperimen menunjukkan kedua arsitektur secara konsisten melampaui referensi EMC-GCN (F1 = 0,6761), dengan H-MTL mencapai F1 = 0,7578 pada data original dan Single-Task mencapai F1 tertinggi sebesar 0,7600 pada data augmentasi. Keunggulan H-MTL atas baseline bersifat tipis namun lebih stabil lintas seed (std-dev 0,0067 vs 0,0166), dengan gain +0,0714 poin pada kasus aspek-opini berjarak ≥6 token yang membuktikan manfaat cross-attention secara kondisional. Penelitian ini memberikan dekomposisi empiris kontribusi tiap komponen arsitektur ABSA-MTL untuk Bahasa Indonesia, dan model diterapkan pada 13.278 ulasan dari 38 desa wisata di 34 provinsi dalam bentuk dashboard analitik interaktif.
===================================================================================================================================
Sustainable tourism has become a major focus in Indonesia's tourism development, alongside the growing need to understand visitor perceptions derived from increasingly massive and diverse online reviews. The main challenge in analyzing these reviews lies in the limitations of pipeline and single-task approaches in Aspect-Based Sentiment Analysis (ABSA), which are prone to error propagation when inaccuracies in aspect extraction negatively affect sentiment classification. This research developed an IndoBERT-based Hierarchical Multi-Task Learning (H-MTL) model with a cross-attention mechanism between subtasks, and compared it against a Single-Task IndoBERT+CRF baseline and the EMC-GCN reference. The methodology included utilizing 2,511 tourism village reviews, preprocessing the data into a BIEOS sequence-labeling scheme of 13 classes, two-phase training with joint optimization, and entity-level evaluation through Precision, Recall, and F1-Score. Experimental results show that both architectures consistently outperformed the EMC-GCN reference (F1 = 0.6761), with H-MTL achieving F1 = 0.7578 on original data and Single-Task achieving the highest F1 of 0.7600 on augmented data. The advantage of H-MTL over the baseline is marginal but more stable across seeds (std-dev 0.0067 vs 0.0166), with a +0.0714 gain on aspect-opinion pairs separated by ≥6 tokens, demonstrating the conditional benefit of cross-attention. This research provides an empirical decomposition of the contribution of each component in MTL-based ABSA for the Indonesian language, and the model is deployed on 13,278 reviews from 38 tourism villages across 34 provinces as an interactive analytical dashboard for village managers.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Aspect Based Sentiment Analysis, IndoBERT, Hierarchical Multi-Task Learning, Ulasan Pariwisata, Sustainable Tourism, Online Reviews
Subjects:	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:	Jason Ho
Date Deposited:	15 Jul 2026 08:54
Last Modified:	15 Jul 2026 08:54
URI:	http://repository.its.ac.id/id/eprint/135022

Actions (login required)

View Item