一种基于BiLSTM-CRF的命名实体识别方法研究

A Method of Building Entity Knowledge Base Based on Bilstm-crf

针对目前从开源数据中挖掘结构化数据存在的数据繁复,新词多等问题,提出了一种命名实体识别的方法。该方法将文本进行实体规范、去停用词和加入词典特征等数据预处理后形成的信息作为输入,通过BiLSTM-CRF算法实现命名实体的识别。最后通过三种不同的语料对该方法进行了验证,实验结果表明了融合领域特征的命名实体识别达到了91.63%。

Aiming at the problems of complex data and many new words in mining structured data from open source data,a method of named entity recognition is proposed.In this method,the information formed after data preprocessing,such as entity specification,removing stop words and adding dictionary features,is used as input,and named entity recognition is realized by BiLSTM-CRF algorithm.Finally,three different corpora are used to verify the method.The experimental results show that the named entity recognition of fusion domain features reaches 91.63%,and the automatic expansion of weapon equipment entity database is realized.