摘要
命名实体识别是从农产品信息文本数据中有效抽取信息的关键一步,旨在从非结构化文本中确定与农产品特性相关的命名实体。农业领域的命名实体识别研究大多集中在农业病虫害领域,关于农产品信息文本的实体识别研究较少,通过采用BMES标注的方式对爬虫获取的农产品信息文本数据进行标注,提出融合BERT的BiLSTM-CRF深度学习模型对该文本数据进行实体抽取。将该模型与多种神经网络模型的实验结果比较发现,融合BERT的BiLSTM-CRF模型对农作物、地区、富含营养成分等3种命名实体识别的准确率和召回率分别为82.25%和84.54%,明显优于IDCNN-CRF等神经网络模型,说明该方法能有效识别抽取农产品信息文本数据命名实体。基于此,命名实体识别作为中文文本信息抽取的关键技术,在农产品信息推荐系统、智能问答等方面将会有广泛的应用前景。
Named entity recognition is a key step to effectively extract information from agricultural products information text data,which aims to determine named entities related to agricultural products characteristics from unstructured text.Most of the researches on named entity recognition in the field of agriculture focus on the field of agricultural diseases and pests,and there is less research on entity recognition of agricultural products information text.BMES annotation was used to label the agricultural products information text data obtained by crawlers,and a BiLSTM-CRF deep learning model integrating BERT was proposed to extract the entity from the text data.Comparing the experimental results of this model with a variety of neural network models,it was found that the accuracy and recall rate of BiLSTM-CRF model fused with BERT for the recognition of three named entities such as crops,regions and rich nutrients are 82.25%and 84.54%respectively,which is significantly better than IDCNN-CRF and other neural network models,the results showed that this method can effectively recognize and extract the named entities of agricultural products information text data.Based on this,as the key technology of Chinese text information extraction,named entity recognition will have a wide application prospect in agricultural products information recommendation system,intelligent question answering and so on.
作者
贺子康
杨勇
杨国峰
张欣钰
He Zikang;Yang Yong;Yang Guofeng;Zhang Xinyu(Agricultural Information Institute,CAAS,Beijing 100081;College of Biosystems Engineering and Food Science,Zhejiang University,Hangzhou 310058,Zhejiang)
出处
《农业展望》
2022年第5期105-111,共7页
Agricultural Outlook
基金
中国农业科学院科技创新工程项目(CAAS-ASTIP-201X-AII)。
关键词
农产品信息
命名实体识别
深度学习模型
信息抽取
agricultural products information
named entity recognition
deep learning model
information extraction
