G-T-ERNIE: Multi-Label Classifier with Text–Label Joint Modeling for Tourism Texts

With the rapid growth of user reviews on tourism platforms and the rising demand for intelligent services, multi-label classification has become essential for review retrieval and service optimization. Traditional methods struggle with the complex semantics and multi-label nature of reviews, leading to weak feature representation and lower accuracy. To address this issue, this study proposes a G-T-ERNIE-based multi-label text classification model. ERNIE is used to capture contextual semantics, TextCNN extracts local n-gram features, and a label co-occurrence graph with GCN models dependencies among labels. In addition, label semantic vectors are integrated to enhance label representation and enable joint modeling of text and labels. Experiments on a self-constructed Jingdezhen tourism review dataset show that the model outperforms existing methods in Micro-F1 and Hamming Loss. To further verify generalization, additional tests were conducted on the CAIL2019 Marriage and Family Elements dataset, confirming the model’s effectiveness and robustness. Moreover, three sets of ablation experiments on the Jingdezhen dataset were designed to examine the contribution of the label structure and semantic fusion module, and the results demonstrated the effectiveness of these components. Overall, the findings provide useful insights for intelligent review analysis and service optimization, supporting the digital and intelligent development of the tourism industry.