The overcrowding of scenic spots not only threatens tourists’ safety but also affects the travel experience. Traditional methods for addressing tourist overload have involved limited access and guided evacuation. While limited access has been effective, it often results in a diminished tourist experience. Moreover, the existing guided evacuation rarely considers the impact on tourists’ experience, resulting in a low willingness to cooperate and making it difficult to estimate evacuation effort efficiency. To solve these problems, this paper proposed a tourist evacuation route recommendation algorithm based on a graph neural network considering the similarity of tourism styles (PER-GCN) and designed a visualization system to simulate and analyse evacuation efficiency. First, the interaction matrix of tourists and scenic spots was constructed using graph mining to extract the high-order interaction information. In the output layer, the similarity between scenic spots and tourism styles was calculated to further improve the accuracy of scenic spot recommendations. Second, due to route complexity and the real-time carrying capacity of scenic spots, the researchers optimized the evacuation routes. Finally, taking the West Lake spot as the case study, the effectiveness of PER-GCN was verified. Additionally, a visualization system was designed to monitor tourist flow in real time and analyse tourist portraits according to the clustering results of scenic spot styles. In addition, the evacuation efficiency of scenic spots was analysed by adjusting the parameters of tourists’ willingness to cooperate, evacuation batch, and the weight of route complexity and scenic spot carrying capacity.