TY - GEN
T1 - Aspect-Based Sentiment Analysis of Hotel Reviews Using Multiscale Convolutional Neural Network
AU - Awantina, Rachma
AU - Fithriasari, Kartika
AU - Widhianingsih, Tintrim Dwi Ary
N1 - Publisher Copyright:
© 2025 IEEE.
PY - 2025
Y1 - 2025
N2 - Aspect-Based Sentiment Analysis (ABSA) is an essential research topic in Natural Language Processing (NLP), enabling a deeper understanding of customer opinions across multiple service dimensions. Data-driven approaches, particularly those based on deep learning, have shown strong potential for improving sentiment prediction performance. However, capturing sentiment variations within specific aspects remains challenging, especially when the dataset is imbalanced, with more positive reviews than negative ones in real scenarios. This study employs a Multiscale Convolutional Neural Network (MCNN) to analyze hotel reviews covering five aspects: Cleanliness, Comfort, Service, Food, and Location. Aspect identification was conducted using a predefined term list. The optimal MCNN configuration - with a vector size of 200, filter sizes [3, 4, 5], and 100 filters - achieved a test accuracy of 96.36%. Among the aspects, Comfort achieved the overall highest performance, while Location showed slightly lower metrics, demonstrating that the MCNN consistently captures sentiment across all five evaluation aspects of hotel reviews.
AB - Aspect-Based Sentiment Analysis (ABSA) is an essential research topic in Natural Language Processing (NLP), enabling a deeper understanding of customer opinions across multiple service dimensions. Data-driven approaches, particularly those based on deep learning, have shown strong potential for improving sentiment prediction performance. However, capturing sentiment variations within specific aspects remains challenging, especially when the dataset is imbalanced, with more positive reviews than negative ones in real scenarios. This study employs a Multiscale Convolutional Neural Network (MCNN) to analyze hotel reviews covering five aspects: Cleanliness, Comfort, Service, Food, and Location. Aspect identification was conducted using a predefined term list. The optimal MCNN configuration - with a vector size of 200, filter sizes [3, 4, 5], and 100 filters - achieved a test accuracy of 96.36%. Among the aspects, Comfort achieved the overall highest performance, while Location showed slightly lower metrics, demonstrating that the MCNN consistently captures sentiment across all five evaluation aspects of hotel reviews.
KW - aspect-based sentiment analysis
KW - hotel reviews
KW - multiscale convolutional neural network
UR - https://www.scopus.com/pages/publications/105035997395
U2 - 10.1109/BTS-I2C67944.2025.11399408
DO - 10.1109/BTS-I2C67944.2025.11399408
M3 - Conference contribution
AN - SCOPUS:105035997395
T3 - Beyond Technology Summit on Informatics International Conference, BTS-I2C 2025
SP - 335
EP - 339
BT - Beyond Technology Summit on Informatics International Conference, BTS-I2C 2025
A2 - Wibowo, Ferry Wahyu
A2 - Kurniawati, Lintang Setyo
A2 - Al Faruq, Habibatul Azizah
A2 - Dasuki, Moh.
A2 - Kurniawan, Isman
PB - Institute of Electrical and Electronics Engineers Inc.
T2 - 2nd Beyond Technology Summit on Informatics International Conference, BTS-I2C 2025
Y2 - 18 December 2025
ER -