Knowledge graph technology can understand the fragmented massive text data and deeply excavate the hidden military knowledge. It is one of the effective technical ways to process the massive battlefield information at present. Operational documents are an important part of battlefield data, which have the characteristics of complex and diverse naming, standardized structure, and a large number of overlapping entity relationships. However, most of the existing knowledge extraction methods cannot solve the overlapped multi-relation extraction problem, which means one or two entities are shared among multiple relational triples contained in a sentence of combat document. In order to solve these problems, a joint entity and relationship extraction model based on sequence generation model and position attention mechanism is proposed via improving training word vectors in military field. Finally, overlapping entity relationships is extracted by the proposed model according to the characteristics of combat documents. By adding entity label and location information to guide the generation of attention pointer aim at combat documents, the experimental results show that our proposed model improves the accuracy of identifying overlapping entity relationships compared with other knowledge extraction models, the F1 value is improved by 4.8%–12.1%. In addition, the extraction effect of this model is also significantly improved compared with the baselines.