AI Visualization for Analyzing Factors Influencing Tourists' Satisfaction Under Deep Learning Technology

Abstract

To analyze Chinese tourist satisfaction, this study proposed an advanced multimodal feature fusion deep learning framework integrating a robustly optimized bidirectional encoder representations from transformers pretraining approach (RoBERTa), a convolutional neural network (CNN), and a graph neural network (GNN). The model incorporated local sentiment pattern recognition, global review association modeling, and deep semantic understanding, thereby overcoming limitations of conventional unimodal models. RoBERTa enhanced the contextual interpretation of Chinese-language reviews through dynamic masking and extended training, identifying implicit sentiments. The CNN captured nuanced emotional cues using multi-scale convolutional kernels. The GNN constructed a semantic graph based on review similarity, unveiling widespread service deficiencies across scenic areas. These features were combined using a dynamic attention mechanism; this fusion enhanced sentiment classification reliability, especially for mixed expressions. The model achieved 92.1% accuracy and a 91.7% F1 score on a dataset of 125,000 tourism-related reviews. Further analysis employed data-driven documents and t-distributed stochastic neighbor embedding to create interactive geographic heatmaps and sentiment clustering diagrams, providing practical suggestions for tourism management. The framework can be integrated into scenic area management systems to enable real-time sentiment monitoring and facilitated data-driven policymaking by assisting local governments with optimizing tourism planning. The findings confirmed the technical advantages of multimodal fusion, while establishing a methodological foundation for the improved governance of tourism services and their cross-cultural adaptation.