Analisis Sentimen Berbasis Aspek dengan Kategorisasi Aspek dengan teknik Transfer Learning (studi kasus : ulasan hotel di Provinsi Bali)

Chang, Hong Kwang (2024) Analisis Sentimen Berbasis Aspek dengan Kategorisasi Aspek dengan teknik Transfer Learning (studi kasus : ulasan hotel di Provinsi Bali). Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5026201127-Undergraduate_Thesis.pdf Restricted to Repository staff only until 1 October 2026. Download (8MB) | Request a copy

Abstract

Pariwisata merupakan sektor yang sedang dikembangkan oleh pemerintah karena mempunyai peran vital terhadap pembangunan nasional dan pendapatan negara. Menurut data dari Kemenparekraf, kunjungan wisatawan mancanegara ke Indonesia pada Januari-Juli 2023 meningkat sebesar 196,85% dari bulan Januari-Juli 2022. Ini merupakan indikator besarnya potensi pariwisata di Indonesia.
Akomodasi merupakan hal yang penting dalam sektor pariwisata. Seringkali wisatawan akan mencari akomodasi yang cocok sesuai preferensi masing masing berdasarkan teks ulasan di platform-platform seperti Google Maps, Traveloka, Trip Advisor, PegiPegi. Ulasan-ulasan yang banyak ini menjadi kesulitan tersendiri untuk mengelola umpan balik wisatawan. Hal ini dapat diselesaikan dengan analisis sentimen berbasis aspek. Akan tetapi, pada analisis berbasis aspek sering kali akan muncul aspek yang sangat banyak dan mengakibatkan kesulitan menganalisa data. Oleh karena itu, pada task ABSA, dikenalkan ASQE (Aspect Sentiment Quadruple Extraction) yang mengekstrak 4 elemen yaitu aspek, opini, kategori aspek, dan polaritas dalam satu kalimat. Penelitian ini akan menghasilkan model ASQE yang bermanfaat untuk menganalisis teks ulasan berbahasa Inggris hotel di Bali.
Berdasarkan hasil penelitian, model pre-trained berbasis T5 mampu memiliki performa yang cukup baik untuk melakukan ekstraksi quadruple pada domain teks berbahasa Inggris. Performa yang dicapai model yaitu mencapai nilai F1 sebesar 0.557 yang mendekati model baseline berdasarkan referensi. Secara kuantitatif, model dapat mengekstrak quadruple dengan cukup baik dan model dapat memahami teks dengan baik. Selain itu, model juga terbukti dapat diimplementasikan pada teks ulasan yang banyak dan menarik insight yang komprehensif ketimbang membaca teks ulasan secara manual, seperti melihat perbandingan sentimen secara keseluruhan, sebaran aspek setiap kategori, perbandingan sentimen setiap kategori, perbandingan sentimen setiap sub-kategori, serta analisis kelebihan dan kekurangan hotel.
Kata kunci: ASQE, pariwisata, sentimen analisis berbasis aspek,

================================================================
The government is presently focusing on the development of the tourism sector due to its crucial role in national progress and government revenue. As per information from the Ministry of Tourism and Creative Economy (Kemenparekraf), the number of foreign tourists visiting Indonesia between January and July 2023 has surged by 196.85% in comparison to the corresponding period in 2022, indicating the substantial potential of the tourism industry in Indonesia.
A fundamental aspect of the tourism sector is accommodation. Travelers frequently look for appropriate lodging based on their preferences, relying on reviews from platforms like Google Maps, Traveloka, Trip Advisor, and PegiPegi. The sheer volume of these reviews presents a challenge in effectively managing tourist feedback. To address this, aspect-based sentiment analysis is employed. However, such analysis often results in a multitude of aspects, complicating the data analysis process. Consequently, in the Aspect-Based Sentiment Analysis (ABSA) task, ASQE (Aspect Sentiment Quadruple Extraction) is introduced to extract four key elements – aspect, opinion, aspect category, and polarity – from a single sentence. Thus, this study will utilize the ASQE model to evaluate English-language hotel reviews in Bali, generating four distinct components
Based on the research results, the pre-trained model based on T5 demonstrated quite good performance in extracting quadruples in the domain of English text. The model achieved an F1 score of 0.557, which is close to the baseline model according to the reference. Quantitatively, the model can extract quadruples well and understand the text effectively. Additionally, the model has been proven to be implementable on large review texts, drawing comprehensive insights compared to manually reading review texts, such as overall sentiment comparison, aspect distribution in each category, sentiment comparison in each category, sentiment comparison in each sub-category, as well as analysis of the strengths and weaknesses of hotels
Keywords: ASQE, tourism, aspect based sentiment analysis

Item Type:	Thesis (Other)
Uncontrolled Keywords:	ASQE, pariwisata, sentimen analisis berbasis aspek,ASQE, tourism, aspect based sentiment analysis
Subjects:	T Technology > T Technology (General) > T59.7 Human-machine systems.
Divisions:	Faculty of Information Technology > Information System > 57201-(S1) Undergraduate Thesis
Depositing User:	Chang Hong Kwang
Date Deposited:	09 Aug 2024 08:35
Last Modified:	24 Sep 2024 06:54
URI:	http://repository.its.ac.id/id/eprint/112568

Actions (login required)

View Item