Tripadvisor Hotel Review Text Mining Based on RoBERTa and LDA Models

In the hotel industry, the development trend of online platforms is booming, and the massive online review data generated provides an important basis for business operators to carry out public opinion analysis and commercial marketing. Effective analysis and mining of hotel online review information can help hoteliers understand the needs of customers, so as to take effective measures to improve service quality and enhance customer satisfaction. Based on the hotel review data on Tripadvisor, this paper proposes a method based on Robustly Optimized BERT Pretraining Approach (RoBERTa) pre-training model and Latent Dirichlet Allocation (LDA) topic analysis model to conduct emotion classification and topic mining for hotel reviews, so as to enable hotel managers to more accurately understand the emotional tendencies and concerns expressed by customers. The experimental results show that the RoBERTa model is used to classify the emotion of the review text with high accuracy. Meanwhile, the topic analysis of the positive and negative emotion comments after classification based on the LDA model can help hoteliers find the advantages and disadvantages of services and products, and propose improvement measures to address the shortcomings, so as to improve the competitiveness of hotels.