地质灾害文本实体关系联合抽取研究  

Research on Joint Extraction of Entity and Relation in Geological Hazard Text

作　　者：周榆婷 陈平华[1] 陈建平[2] 

机构地区：[1]广东工业大学计算机学院,广东 广州 [2]肇庆学院计算机科学与软件学院,广东 肇庆

出　　处：《计算机科学与应用》2021年第11期2672-2681,共10页Computer Science and Application

摘　　要：针对传统实体关系联合抽取方法存在效率低下、错误传播、实体冗余等问题,提出基于双向长短时记忆神经网络和条件随机场并融合注意力机制的地质灾害实体与关系联合抽取方法。使用一种新标注方案,将地质灾害文本实体关系联合抽取问题转化为序列标注问题。用字符级嵌入进行作文本向量化表示,使用BiLSTM-Attention-CRF模型实现地质灾害文本实体关系联合抽取。实验结果表明:在地质灾害语料集上,实体识别的F-score值达到了85.4%,关系抽取的F-score达到了63.6%,证明了该方法的优越性和有效性。In view of the problems of low efficiency, error propagation, and entity redundancy in traditional entities and relations extraction method, this article proposes a joint geological hazard entities and relations extraction method based on bi-directional long short-term memory and conditional random fields with attention mechanism. This method employs a new tagging scheme which represents both entity and relation information by the tags and converts the joint extraction task to a tagging task. This method applies character embedding as input, and extracts geological hazard entities and relations with BiLSTM-Attention-CRF model. The results shows that, on geological hazard corpus, the method achieves 85.4% F-score for named entity recognition and 63.6% F-score for relations extraction which proves the superiority and effectiveness of the method.

关 键 词：命名实体识别 关系抽取 联合抽取 标注模式 

分 类 号：P69[天文地球—地质学]