基于实体知识的石油炼化领域命名实体识别
Named Entity Recognition in Petroleum Refining Domain Based on Entity Knowledge
丁建新 1王晓伟 2温欣 1屈克将 1王建华 1赵艳红 1胡思颍2
作者信息
- 1. 昆仑数智科技有限责任公司 智慧油服事业部,北京 100071
- 2. 中国石油大学(北京) 信息科学与工程学院/人工智能学院,北京 102249
- 折叠
摘要
石油炼化领域中的命名实体识别任务存在着标注数据稀缺,以及现有的预训练语言模型不能很好识别领域组合和嵌套实体的问题.基于此,首先提出一种基于外部实体知识的数据增强方法(EEKR),通过引入外部实体知识库,将其与标注数据中的实体进行实体级别替换完成数据增强,有效解决标注数据稀缺的问题.之后提出了基于内部实体知识的命名实体识别模型(IIEKNER),通过获取标注样本中的内部实体嵌入,将内部实体知识融入预训练模型,从而可以更准确地识别出石油炼化领域中的嵌套和组合实体.实验结果显示,相比于其他模型,基于EEKR数据增强方法的IIEKNER模型的识别效果更优.
Abstract
Named entity recognition task in the petroleum refining domain suffers from the problems of scarcity of labeled data as well as the existing pre-trained language models cannot recognize domain combination and nested entities well.Based on this,a data augmentation method EEKR(External Entity Knowledge Replacement,EEKR)based on external entity knowledge is firstly proposed,which effectively solves the problem of scarcity of labeled data by introducing an external entity knowledge base and completing data augmentation by replacing it with entities in the labeled data at the entity level.After that,a named entity recognition model IIEKNER(Namd Entity Recognition Incorporating Internal Entity Knowledge,IIEKNER)is proposed,which incorporates internal entity knowledge into the pre-training model by obtaining internal entity embeddings in the labeled samples.Thus,nested and combined entities in the petroleum refining domain can be recognized more accurately.The experimental results show that compared to other models,the IIEKNER model based on EEKR data augmentation method has better recognition performance.
关键词
命名实体识别/石油炼化领域/数据增强/BERTKey words
named entity recognition/petroleum refining domain/data augmentation/BERT引用本文复制引用
基金项目
国家重点研发计划(2019YFC0312003)
出版年
2024
国家科技期刊平台
NSTL
万方数据