Joint Extraction of Organizations and Relations for Emergency Response Plans with Rich Semantic Information Based on Multi-head Attention Mechanism

At present, deep learning-based joint entity-relation extraction models are gradually gaining the capability to accomplish complex tasks. However, research progress in specific fields is relatively slow. Compared with other areas, emergency plan text possesses unique characteristics such as high entity density, extensive text, and numerous professional terms. These features challenge some general models, which struggle to handle the semantic information of emergency plan text effectively. In response to this, the paper addresses the complex semantics of emergency plan text. It proposes a joint extraction model for emergency plan organization and relationship, based on the multi-head attention mechanism (MA-JE). This model enriches semantic information by obtaining contextual information from various perspectives and different levels. The aim is to deeply mine and utilize sentence semantic information through extensive feature extraction of emergency plan text. The proposed model and the baseline model were separately tested on the Chinese emergency response plan dataset. The results indicate that the proposed approach surpasses existing baseline models in the joint extraction of entities and their relations. Furthermore, ablation experiments were conducted to verify the effectiveness of each module within the model.