BERT-Based Hybrid Deep Learning with Text Augmentation for Sentiment Analysis of Indonesian Hotel Reviews

Maxwell Thomson, Hendri Murfi, Gianinna Ardaneswari

2023

Abstract

Indonesia’s tourist industry plays a significant role in the country’s economic growth. Despite being impacted by COVID-19, the occupancy rate of hotels in June 2022 reached 50.28%, surpassing the previous record of 49.17% in January 2020. As hotel occupancy rates rise, it becomes increasingly important to analyze customer reviews of hotels through sentiment analysis to categorize the emotions expressed in the reviews. While a BERT-based hybrid deep learning model has been shown to perform well in sentiment analysis, the nature of class imbalance is often a problem. To address this, the text augmentation method provides a solution to increase the amount of minority class training data through existing data. This paper evaluates five word-level text augmentation methods for the BERT-based hybrid model on classifying sentiments in Indonesian hotel reviews. Our simulations show that the text augmentation methods can improve the model performance for all datasets dan unit measures. Moreover, the random swap method achieves the highest precision and specificity on two of three datasets.

Paper Citation

in Harvard Style

Thomson M., Murfi H. and Ardaneswari G. (2023). BERT-Based Hybrid Deep Learning with Text Augmentation for Sentiment Analysis of Indonesian Hotel Reviews. In Proceedings of the 12th International Conference on Data Science, Technology and Applications - Volume 1: DATA; ISBN 978-989-758-664-4, SciTePress, pages 468-473. DOI: 10.5220/0012127400003541

in Bibtex Style

@conference{data23,
author={Maxwell Thomson and Hendri Murfi and Gianinna Ardaneswari},
title={BERT-Based Hybrid Deep Learning with Text Augmentation for Sentiment Analysis of Indonesian Hotel Reviews},
booktitle={Proceedings of the 12th International Conference on Data Science, Technology and Applications - Volume 1: DATA},
year={2023},
pages={468-473},
publisher={SciTePress},
organization={INSTICC},
doi={10.5220/0012127400003541},
isbn={978-989-758-664-4},
}

in EndNote Style

TY - CONF

JO - Proceedings of the 12th International Conference on Data Science, Technology and Applications - Volume 1: DATA
TI - BERT-Based Hybrid Deep Learning with Text Augmentation for Sentiment Analysis of Indonesian Hotel Reviews
SN - 978-989-758-664-4
AU - Thomson M.
AU - Murfi H.
AU - Ardaneswari G.
PY - 2023
SP - 468
EP - 473
DO - 10.5220/0012127400003541
PB - SciTePress