Facing the fact that the existing data-driven machine learning landslide susceptibility assessment methods fail to fully express the spatio-temporal correlation between landslides and disaster-prone environments in complex geological environments, the problems of sample imbalance and insufficient feature expression remain prominent. This study focuses on the assessment of landslide susceptibility in graph neural networks under the constraint of geographical environment consistency. By constructing geographical nodes and weighted graph models, the precise assessment of landslide susceptibility has been achieved. Finally, a graph neural network model integrating the attention mechanism was further proposed.