BERT for Sentiment Analysis in Thai Hotel Reviews

The rapid growth of the tourism and hospitality industry has resulted in a significant increase in customer reviews shared online. These reviews help tourists discover new accommodations with a favorable atmosphere and reasonable prices; the volume of reviews makes it challenging for travelers to choose the right option. Negative reviews, in particular, can influence booking decisions and impact a hotel’s image. Sentiment analysis that categorizes comments has become an important tool for analyzing customer feedback automatically. Moreover, the Thai language has unique characteristics, such as its diverse writing styles, punctuation, and multiple meanings of a single word, which pose language barriers for sentiment analysis. Our method, employing the Bidirectional Encoder Representations from Transformers (BERT) model, analyzes hotel reviews in Thai, classifying sentiments into three categories: positive, neutral, and negative. This study uses a dataset of 37,011 hotel reviews. Our experiment results show that the BERT model has an accuracy of 89.31% and an F1 score of 89.43%, outperforming prior research. The findings contribute insights to a deeper understanding of customer reviews for the hospitality industry, enabling hotel operators to respond to customer feedback more effectively and improve their services. Analyzing and distilling reviews from customer feedback may assist tourists and others in making quicker choices. Finally, the results of this study show that using BERT for sentiment analysis can help businesses grow and become more competitive in the quickly changing tourism market.