Enhancing Multimodal Tourism Review Sentiment Analysis Through Advanced Feature Association Techniques DOI Creative Commons
Peng Chen, Lingmei Fu

International Journal of Information Systems in the Service Sector, Год журнала: 2024, Номер 15(1), С. 1 - 21

Опубликована: Июль 17, 2024

The development of tourism services presents significant opportunities for extracting and analyzing customer sentiment. However, with the advent multimodality, travel reviews have brought new challenges. Early methods detecting such merely combined text image features, resulting in poor feature correlation. To address this issue, our study proposes a novel multimodal review sentiment analysis method enhanced by relevant features. Initially, we employ fusion model that combines BERT Text-CNN extraction. This approach strengthens semantic relationships filters noise effectively. Subsequently, utilize ResNet-51 extraction, leveraging its ability to learn complex visual representations. Additionally, integrating an attention mechanism further enhances modality correlation, thereby improving effectiveness. On Multi-ZOL dataset, achieves accuracy 90.7% F1 score 90.8%. Similarly, on Ctrip it attains 83.6% 84.1%.

Язык: Английский

Enhancing Multimodal Tourism Review Sentiment Analysis Through Advanced Feature Association Techniques DOI Creative Commons
Peng Chen, Lingmei Fu

International Journal of Information Systems in the Service Sector, Год журнала: 2024, Номер 15(1), С. 1 - 21

Опубликована: Июль 17, 2024

The development of tourism services presents significant opportunities for extracting and analyzing customer sentiment. However, with the advent multimodality, travel reviews have brought new challenges. Early methods detecting such merely combined text image features, resulting in poor feature correlation. To address this issue, our study proposes a novel multimodal review sentiment analysis method enhanced by relevant features. Initially, we employ fusion model that combines BERT Text-CNN extraction. This approach strengthens semantic relationships filters noise effectively. Subsequently, utilize ResNet-51 extraction, leveraging its ability to learn complex visual representations. Additionally, integrating an attention mechanism further enhances modality correlation, thereby improving effectiveness. On Multi-ZOL dataset, achieves accuracy 90.7% F1 score 90.8%. Similarly, on Ctrip it attains 83.6% 84.1%.

Язык: Английский

Процитировано

0