Enhancing tourism sentiment analysis with deep learning: a comprehensive study on social media data

Sentiment analysis plays a crucial role in extracting valuable insights from tourists reviews. Existing sentiment analysis techniques designed for general text often fall short when applied to tourism-related comments. This paper proposes an innovative deep learning-based approach for analysing comments in tourism. We utilise a BERT-based word embedding technique to capture vector representations of each preprocessed tweet. We use three deep learning methods, namely BERT, LSTM, and BiLSTM, which are commonly employed in various text processing tasks, for classifying sentiment in tourism tweets. Therefore, the proposed sentiment analysis methods include BERT and two combined methods: BERT-LSTM and BERT-BiLSTM. We tune the hyperparameters of the proposed deep learning methods and assess their performance by evaluating precision, recall, accuracy, and F1-score. Our research findings reveal that the proposed BERT model outperforms other methods on the benchmark dataset, achieving an impressive precision rate of 96.17%, a recall of 96%, an accuracy of 96%, and an F1-score of 96%. Besides, the other proposed deep learning methods outperform previous methods on the mentioned criteria. The results demonstrate the effectiveness of the proposed methods in accurately classifying sentiments in tourism-related tweets.