Abstract:

To address the problem that the current accuracy of relation extraction of civil aviation emergency domain ontology is low, this paper proposes a relation extraction model based on attention mechanism and bidirectional gated recurrent unit (BiGRU). Firstly, this paper queries the pre-trained word vector matrix and maps text words into vectors. Secondly, BiGRU is constructed to obtain the context semantic information of word sequence. Thirdly, attention mechanism is introduced at word level and sentence level repectively to allocate more weights to words and sentences that are more important for semantic representation. Finally, the model is trained and optimized. Experiments are conducted on the relation extraction of civil aviation emergency domain ontology, and the results show that this model has better accuracy of the relation extraction compared with traditional methods, which verifies the validity of the model and provides new method support for the automatic learning of relation extraction of civil aviation emergency domain ontology.

Key words: relation extraction, civil aviation emergency, attention mechanism, gated recurrent unit (GRU) model, domain ontology

摘要：

针对民航突发事件领域本体关系抽取准确率低的问题，提出了一种结合注意力机制与双向门控循环单元（BiGRU）的关系抽取模型。首先查询预先训练的词向量矩阵，将文本中每个词语映射为向量表示；其次构建BiGRU，得到词语序列的上下文语义信息；然后在词语层面和句子层面分别引入注意力机制，为表达语义关系更重要的词语和句子分配更大的权重；最后进行模型的训练与优化。将该模型应用在民航突发事件领域本体的关系提取中，实验结果表明该模型相较于其他方法具有更好的提取效果，验证了该模型的有效性，为民航突发事件领域本体关系的自动获取提供了新的方法支持。

关键词: 关系抽取, 民航突发事件, 注意力机制, 门控循环单元（GRU）模型, 领域本体