Joint Extraction of Organizations and Relations for Emergency Response Plans With Rich Semantic Information Based On Multi-Head Attention Mechanism

Abstract

At present, deep learning-based joint entity-relation extraction models are gradually able to accomplish complex tasks, but the research progress in specific fields is relatively slow. Compared with other fields, emergency plan text has the characteristics of high entity density, long text, and many professional terms, which make some general models unable to handle the semantic information of emergency plan text well. Therefore, this paper addresses the problem of complex semantics of emergency plan text, and proposes a joint extraction model of emergency plan organization and relationship based on multi-Head Attention Mechanism (MA-JE) to enrich semantic information, starting from multiple perspectives and different levels to obtain contextual information, aiming to deeply mine and use sentence semantic information through deep feature extraction of emergency plan text. The proposed model and the baseline model are experimented separately on the Chinese emergency response plan dataset, and the results show that the proposed approach outperforms existing baseline models for joint extraction of entity and their relations. In addition, ablation experiments were performed to verify the validity of each module in the model.