Ajrina, Navy Nurlyn (2026) KOMPARASI KINERJA MODEL DEEP LEARNING DAN WORD EMBEDDING DALAM MULTILABEL ASPECT-BASED SENTIMENT ANALYSIS PADA ULASAN HOTEL. Undergraduate thesis, UPN Veteran Jawa Timur.
|
Text (Cover)
22082010158-cover.pdf Download (960kB) | Preview |
|
|
Text (Bab 1)
22082010158-bab1.pdf Download (423kB) | Preview |
|
![[img]](https://repository.upnjatim.ac.id/style/images/fileicons/text.png) |
Text (Bab 2)
22082010158-bab2.pdf Restricted to Repository staff only until 25 May 2029. Download (693kB) |
|
|
Text (Bab 3)
22082010158-bab3.pdf Restricted to Repository staff only until 25 May 2029. Download (990kB) |
|
|
Text (Bab 4)
22082010158-bab4.pdf Restricted to Repository staff only until 25 May 2029. Download (8MB) |
|
|
Text (Bab 5)
22082010158-bab5.pdf Download (251kB) | Preview |
|
|
Text (Daftar Pustaka)
22082010158-daftarpustaka.pdf Download (225kB) | Preview |
|
|
Text (Lampiran)
22082010158-lampiran.pdf Restricted to Repository staff only until 25 May 2029. Download (206kB) |
Abstract
Visitor reviews on digital platforms such as Tripadvisor constitute an important source of information for the hospitality industry. However, the large volume and unstructured nature of these reviews make manual analysis difficult. This study aims to analyze sentiment toward the Room, Hotel, Location, and Service aspects using a Multilabel Aspect-Based Sentiment Analysis (ABSA) approach, as well as to compare the performance of deep learning architectures, word embeddings, and preprocessing techniques on hotel reviews from Surabaya, Sukabumi, and Bali. The dataset consisted of 7,821 hotel reviews written in Indonesian and English, where English-language reviews were translated into Indonesian to maintain linguistic consistency during analysis. The models were developed using 5-Fold Cross Validation, comprising one multilabel aspect classification model and four aspect-based sentiment classification models. This study compared combinations of deep learning architectures (CNN and BiLSTM), word embeddings (Word2Vec and FastText), and preprocessing techniques (with and without stemming). The results indicate that no single model combination consistently outperformed others across all classification tasks. BiLSTM with FastText and stemming achieved the best performance for multilabel aspect classification with an F1 Macro score of 0.9725. For sentiment classification, CNN tended to perform better, with the best-performing combinations being CNN + FastText + stemming for the Room aspect (F1 Macro 0.9882), BiLSTM + Word2Vec without stemming for Hotel (F1 Macro 0.9516), CNN + Word2Vec without stemming for Location (F1 Macro 0.9676), and CNN + Word2Vec + stemming for Service (F1 Macro 0.9725). These findings indicate that BiLSTM is more effective for multilabel aspect classification, whereas CNN performs better for sentiment classification. Word2Vec tended to provide better performance for sentiment classification on the Hotel, Location, and Service aspects, while the effect of stemming was inconsistent and dependent on the characteristics of each task. The five best-performing models were subsequently implemented in a web-based system to automatically analyze hotel reviews and provide interactive visualizations as a proof of concept for practical implementation
| Item Type: | Thesis (Undergraduate) | ||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Contributors: |
|
||||||||||||
| Subjects: | Q Science > Q Science (General) Q Science > QA Mathematics Q Science > QA Mathematics > QA76.6 Computer Programming Q Science > QA Mathematics > QA76.87 Neural computers T Technology > T Technology (General) T Technology > T Technology (General) > T58.6-58.62 Management Information Systems |
||||||||||||
| Divisions: | Faculty of Computer Science > Departemen of Information Systems | ||||||||||||
| Depositing User: | Navy Nurlyn Ajrina | ||||||||||||
| Date Deposited: | 26 May 2026 01:09 | ||||||||||||
| Last Modified: | 26 May 2026 01:22 | ||||||||||||
| URI: | https://repository.upnjatim.ac.id/id/eprint/52124 |
Actions (login required)
 |
View Item |