摘要： 为提高井下安全管理效率，实现煤矿安全生产，根据煤矿行业相关标准规范，并结合井下不安全行为领域知识，采用BIO标注策略构建了一个包含8类实体类别、2359条样本的煤矿井下不安全行为语料库。针对煤矿井下不安全行为命名实体识别任务中存在的语义信息利用不足、实体分布不均衡、实体边界模糊的问题，提出一种基于Global Pointer和对抗训练的煤矿井下不安全行为命名实体识别模型。首先，采用改进的分层RoBERTa模型充分利用多层语义信息增强井下不安全行为文本向量化，并结合对抗训练对词嵌入层进行扰动，缓解数据不平衡问题，增强模型的鲁棒性；其次，在特征提取层采用BiGRU可以更有效的捕获语料的上下文语义特征，加强文本语义关联；最后，在解码层构造Global Pointer，获得更准确的实体边界识别结果。为了验证本方法提出的模型的有效性，在自建的小样本煤矿井下不安全行为数据集上进行了实验。结果表明，本文模型的精确率、召回率和 值分别是78.77%、78.20%、78.48%，相比于BERT-Global Pointer模型提高了2.27％、0.63％、1.45％，为构建井下不安全行为知识图谱提供基础。

Abstract: In order to improve the efficiency of underground safety management and realize coal mine safety production, according to the relevant standards and norms of the coal mine industry, combined with the knowledge of the field of underground unsafe behavior, a coal mine unsafe behavior corpus containing 8 entity categories and 2359 samples was constructed by using BIO labeling strategy. Aiming at the problems of insufficient semantic information utilization, unbalanced entity distribution and fuzzy entity boundary in the named entity recognition task of unsafe behavior in coal mine, a named entity recognition model of unsafe behavior in coal mine based on Global Pointer and adversarial training is proposed. Firstly, the improved hierarchical RoBERTa model is used to make full use of multi-layer semantic information to enhance the text vectorization of underground unsafe behavior, and the word embedding layer is disturbed by adversarial training to alleviate the problem of data imbalance and enhance the robustness of the model. Secondly, the use of BiGRU in the feature extraction layer can more effectively capture the contextual semantic features of the corpus and enhance the semantic association of the text. Finally, Global Pointer is constructed in the decoding layer to obtain more accurate entity boundary recognition results. In order to verify the effectiveness of the model proposed by this method, experiments were carried out on the self-built small sample coal mine underground unsafe behavior data set. The results show that the accuracy, recall and value of the model in this paper are 78.77 %, 78.20 % and 78.48 %, respectively, which are 2.27 %, 0.63 % and 1.45 % higher than the BERT-Global Pointer model. It provides a basis for constructing a knowledge graph of unsafe behavior in underground mines.