This study focuses on path optimization and tourist behavior prediction in tourist scenes, and proposes a comprehensive solution based on big data. In view of the limitations of traditional tourism path planning methods that rely on static data and manual experience, and it is difficult to cope with the dynamic changes of scenic spots, and the shortcomings of existing tourist behavior prediction models in data fusion, dynamic adjustment and interpretability, this study integrates multi-source heterogeneous data such as tourist location data, social media check-in, real-time traffic data and weather information, and constructs a dynamic model of tourism scenes. In the aspect of path optimization, a dynamic path optimization algorithm based on graph neural network (GNN) is proposed. By dynamically adjusting the path weight, the shortest time or least congestion path planning is realized. The experimental results show that the algorithm is significantly superior to the traditional Dijkstra algorithm and A* algorithm in key indicators such as average transit time, congestion rate and path calculation delay. In the aspect of tourists’ behavior prediction, the LSTM-Transformer hybrid model integrating temporal and spatial characteristics is constructed, which can accurately predict tourists’ stay time, scenic spot preference and abnormal behavior. The model performs well in the average absolute error (MAE) of stay time and the accuracy of scenic spot preference, and has higher prediction accuracy and accuracy than the single LSTM model and ARIMA model. Through parameter sensitivity analysis, the optimal working range of the model under different environmental conditions is determined, which further improves the robustness and adaptability of the model. This study not only provides a new technical means for the path optimization of tourist attractions and the prediction of tourists’ behavior, but also provides a scientific basis for scenic spot managers to formulate resource allocation strategies and enhance tourists’ experience.