Analisis Sentimen Berbasis Aspek pada Ulasan Pondok Wisata dengan Unified Generative Framework

Mufti, Amri (2024) Analisis Sentimen Berbasis Aspek pada Ulasan Pondok Wisata dengan Unified Generative Framework. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5026201032-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only until 1 October 2026. Download (9MB) | Request a copy

Abstract

Pariwisata di Indonesia merupakan salah satu kontributor penting bagi perekonomian negara dan kesejahteraan masyarakat, dengan Pulau Bali sebagai salah satu destinasi wisata utamanya. Dalam konteks kesejahteraan tersebut, program Community Based Tourism (CBT) dikembangkan, yang dinilai mampu memberi peluang bagi masyarakat lokal untuk menjadi agen pariwisata yang berdaya dan memajukan daerah wisata. Sebagai bentuk partisipasi masyarakat dalam CBT, pondok wisata (homestay), suatu pilihan utama bagi wisatawan yang mencari pengalaman otentik dengan nuansa budaya setempat, dibangun dan dikelola oleh masyarakat Bali, terutama di kawasan Ubud. Seiring dengan adanya program CBT dan meningkatnya permintaan akan pondok wisata, pengelola perlu mengetahui aspek penting dalam menyediakan pelayanan yang memuaskan tamu. Informasi tersebut dapat diperoleh melalui situs Electronic Word of Mouth (E-WOM), seperti TripAdvisor. Namun, pemilik pondok wisata acapkali harus membaca ulasan yang tidak relevan dan tidak bermanfaat, membuang waktu, terutama dengan peningkatan jumlah ulasan. Meskipun teknik analisis sentimen terhadap pondok wisata telah dilakukan, sentimen hanya dianalisis pada level dokumen dan tidak mengidentifikasi aspek secara spesifik. Untuk mengatasi hal tersebut, teknik analisis sentimen pada level aspek bernama Aspect-Based Sentiment Analysis (ABSA) digunakan, yang dapat mengidentifikasi aspek-aspek tertentu yang dinilai dalam ulasan pondok wisata dan mengekstrak sentimen terkait. Penelitian ABSA sebelumnya cenderung berfokus pada subset tertentu dari tugas ABSA saja sehingga memerlukan model yang rumit dan sulit untuk dipecahkan dalam sebuah kerangka kerja yang terpadu. Oleh sebab itu, penelitian ini menggunakan pendekatan Unified Generative Framework (UGF) dengan model pre-trained Bidirectional Autoregressive Transformer (BART) untuk menyelesaikan tugas ABSA pada ulasan pondok wisata berbahasa Inggris di Bali. Hasil penelitian menunjukkan bahwa aspek yang paling kerap di bahas pada ulasan pondok wisata di Bali sejak 2018 hingga 2023 adalah room, staff, breakfast, place, dan family. Model mencapai skor F1 tertinggi, sebesar 75,15, dengan menyetel hiperparameter; learning rate sebesar 6e-5, batch size sebesar 16, num beams sebesar 4, jumlah epoch sebesar 50, dengan tipe decoder avg_score, dan length penalty sebesar 1. Model tersebut mempunyai kinerja baik saat memprediksi ulasan dengan panjang teks kurang dari 250, jumlah aspek kurang dari 25 di dalam satu ulasan, memiliki sentimen positif, serta aspek dan opini yang terdiri dari satu kata.
================================================================================================================================
Tourism in Indonesia is a significant contributor to the country's economy and the well-being of its people, with Bali being one of its main tourist destinations. In the context of well-being, the Community Based Tourism (CBT) program has been developed, which is believed to provide opportunities for local communities to become empowered tourism agents and advance the tourism areas. As a form of community participation in CBT, homestays, which are the preferred choice for travelers seeking an authentic experience with local cultural nuances, are built and managed by the local community in Bali, particularly in the Ubud area. With the implementation of the CBT program and the increasing demand for homestays, it is important for the operators to understand the essential aspects of providing satisfactory services to guests. Such information can be obtained through Electronic Word of Mouth (E-WOM) platforms, such as TripAdvisor. However, homestay owners often have to read irrelevant and unhelpful reviews, which wastes time, especially with the growing number of reviews. Although sentiment analysis techniques for homestays have been conducted, sentiments are only analyzed at the document level and do not specifically identify aspects. To address this issue, Aspect-Based Sentiment Analysis (ABSA) is employed, which can identify specific aspects evaluated in homestay reviews and extract related sentiments. Previous ABSA research has tended to focus on specific subsets of ABSA tasks, requiring complex and challenging models to be solved within an integrated framework. This study utilizes a Unified Generative Framework (UGF) approach, using a pre-trained Bidirectional Autoregressive Transformer (BART) model to accomplish ABSA tasks in English-language homestay reviews in Bali. The results show that the most frequently discussed aspects in the reviews of homestays in Bali from 2018 to 2023 are room, staff, breakfast, place, and family. The model achieves the highest F1 score, 75.15, by tuning the hyperparameters; learning rate of 6e-5, batch size of 16, num beams of 4, number of epochs of 50, with decoder type of avg_score, and length penalty of 1. The model performs well when predicting reviews with text length less than 250, number of aspects less than 25 in one review, having positive sentiment and single-word aspects and opinions.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	ABSA, BART, E-WOM, Unified Generative Framework
Subjects:	G Geography. Anthropology. Recreation > G Geography (General) > G155 Tourism Q Science > Q Science (General) > Q325.5 Machine learning. Support vector machines.
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Information System > 57201-(S1) Undergraduate Thesis
Depositing User:	Amri Mufti
Date Deposited:	15 Jul 2024 01:17
Last Modified:	15 Jul 2024 01:17
URI:	http://repository.its.ac.id/id/eprint/108289

Actions (login required)

View Item