Air fare sentiment via Backtranslation-CNN-BiLSTM and BERTopic

Abstract Civil aviation passengers’ comments about airlines or airports on social media are the key to improving service quality. In order to make effective use of these data, in-depth analysis is needed to provide solid support for service improvement of airlines and airports. Due to its uniqueness, accurate modeling and analysis are required. First, the data are accurately collected from various network platforms and reprocessed. In this process, transfer learning, artificial data annotation, and term frequency–inverse document frequency (TF-IDF) analysis technology are innovatively integrated to ensure data quality and analysis depth. Then, according to the characteristics of the review data, the civil aviation domain-specific word vector based on Word2Vec was customized and developed, and the backtranslation – convolutional neural networks – bi-directional long short-term memory (Backtranslation-CNN-BiLSTM) model was constructed for sentiment analysis. The model is verified by multi-dimensional evaluation indicators, which shows excellent performance indicators and ensures reasonable efficiency. Finally, the cutting-edge BERTopic modeling technology was used to deeply mine the passenger comment topics to reveal the focus and potential needs of passengers. This study successfully constructed the technical system of civil aviation passenger comment sentiment analysis, which provided technical support for industry service optimization.