Abstract: Current intent recognition algorithm has problems such as lack of evaluation of recognition results and incomplete domain knowledge, and there is an urgent need to study confidence level evaluation methods. The algorithm uses the spatiotemporal knowledge graph to uniformly represent entities containing spatiotemporal information, and abstractly represents the combat targets and their relationships as a triplet of spatiotemporal knowledge. Neural networks are trained with typical confrontation scene data, and the confidence levels of the entity and the knowledge graph are calculated and fused to obtain the final confidence level evaluation results. Simulation test results show that the method uses space-time and target model information to analyze the possibility that there is a certain combat intent in the combat targets, the confidence level of the intent recognition results is effectively evaluated, it is of great significant for the real "landing" of the intent recognition system.