A Joint Method for Combat Intent Recognition and Key Information Extraction

To alleviate the problems of poor quality and low efficiency in traditional combat plan making, we propose an intelligent combat plan generation method based on Bert pre-trained language model. First, we studied practical combat scenarios and military related websites, and constructed a military domain combat intent dataset that includes structured information such as combat categories, objects, and scenarios. Second, we utilize Bert pre-trained language model for semantic analysis of requirements, TextCNN (Convolutional Neural Network for Text) for combat intent recognition, and BiLSTM (Bidirectional Long Short-Term Memory) for key information extracting and entity normalization. Thus, based on the intent and key information, candidate schemes can be retrieved from the knowledge graph in the field of military operations in the future. Compared with traditional methods, the scheme quality and generation efficiency are significantly improved. This study provides an effective approach for intelligent decision support in the military field, and also offers references for intelligent scheme generation in other domains.