基于BERT-BiLSTM-CRF模型的地质领域实体识别研究

Named Entity Recognition in Geological Field Based on BERT-BiLSTM-CRF

针对地质文本进行地质实体识别,对地质专业研究人员挖掘和分析数据具有重大作用,同时这项技术也是构建地质领域知识图谱以及很多上层应用的基础。目前,地质领域实体识别的研究还在发展中,应用较少,而地质专业数据量却呈爆发式增长,所以数据处理技术显得尤为重要。基于此,论文提出了一种基于BERT-BiLSTM-CRF模型并且融合约束规则的命名实体识别技术,用以辅助地质专业人员处理地质数据。首先由BERT层处理输入文本序列,将其转换为包含上下文特征的字向量,然后将字向量输入到BiLSTM层中对上下文特征进行学习,输出单个汉字的得分,CRF层将BiLSTM的得分和自身学习到的隐含规则进行整合,输出最后综合得分,从而选择最佳标签。实验结果表明,对比传统方法以及现在流行的深度学习方法,此方法的准确率、召回率、F1值均为最高值,分别为92.05%、94.82%、93.41%。

Recognition of geological entities based on geological texts plays a major role in mining and analyzing data for geo-logical researchers.This technology is also the basis for building knowledge graph in the geological field and lots of upper-level ap-plications.At present,the research of entity recognition in the field of geology is still under development,with fewer applications,but the amount of geological data is increasing exponentially,so data processing technology is particularly important.Therefore this paper proposes a named entity recognition technology based on the BERT-BiLSTM-CRF model and constraint rules to assist geologi-cal professionals in processing geological data.First,the BERT layer receives the input text sequences,and converts them into word vectors with contextual features.Next the word vectors are input into the BiLSTM layer to learn the contextual features,and the BiL-STM layer outputs the scores of every single chinese character.After this,the CRF layer integrates the scores from the BiLSTM layer and the implicit rules which are learned by itself,then the final comprehensive scores are output so as to select the best label.The experimental results show that compared with the traditional method and the popular deep learning method,the precision,recall,and F1 value of this method are all the highest values,which are 92.05%,94.82%,and 93.41%,respectively.