For the purpose of constructing a knowledge graph for aviation unsafe events, information extraction technologies are focused on for knowledge graph. A dataset for entity relationship extraction and a domain dictionary have been constructed by analyzing and cleaning aviation unsafe events raw data. The major challenges in entity relationship extraction for knowledge graph are determining entity boundaries and improving relationship classification accuracy. In view of this, a semantic information augmentation method that integrated domain dictionary embedding was proposed for domain entity and relationship joint extraction. The proposed knowledge graph construction method was validated for the scenario of numerous unsafe event recordings in airline safety information reports. Experiments demonstrate the effectiveness of the knowledge graph construction method by analyzing the precision and recall metrics of the proposed entity relationship extraction algorithms.