摘要
在玻璃文物领域中,有大量的文物数据没有被充分研究并利用.提出基于字向量的BiLSTM-CRF模型与玻璃文物特征字典相结合的命名实体识别方法,首先基于玻璃文物特征字典BIO标注文本并构建玻璃文物语料库,接着使用BiLSTM训练语料库并输出结果,最后CRF层结合训练结果对实体进行筛选并分类,抽取出正确的实体.实验表明,该方法在玻璃文物语料库中准确率达到91.56%,召回率达到88.25%,综合评价指标达到89.87%.
In the field of glass cultural relics,a large amount of cultural relics data has not been fully studied and utilized.A method of NER based on the combination of the word-based vector BiLSTM-CRF model and the glass cultural relic feature dictionary is proposed.Firstly,the BIO-labeled text based on glass cultural relic feature dictionary is constructed into a glass cultural relic corpus,and then BiLSTM is utilized to train the corpus and output the results.Finally,CRF combines the training results to filter and classify the entities,and extract the correct entities.Experiments show that this method has an accuracy rate of 91.56%,a recall rate of 88.25%,and a harmonic mean value of 89.87%in the glass cultural relics corpus.
作者
杨云
宋清漪
云馨雨
史雯倩
尚梦丹
YANG Yun;SONG Qing-yi;YUN Xin-yu;SHI Wen-qian;SHANG Meng-dan(School of Electronic Information and Artificial Intelligence, Shaanxi University of Science & Technology, Xi′an 710021, China;School of Information Engineering, Minzu University of China, Beijing 100081, China)
出处
《陕西科技大学学报》
北大核心
2022年第3期179-184,共6页
Journal of Shaanxi University of Science & Technology
基金
国家重点研发计划项目(2019YFC1520200)。
关键词
玻璃文物
玻璃文物特征字典
命名实体识别
双向长短时记忆网络
条件随机场
glass cultural relics
glass cultural relics feature dictionary
named entity recognition
bidirectional long short term memory
conditional random field