摘要
互联网中存在大量的与地理信息相关的信息,其中文本信息是这些信息的重要组成部分。针对构建地理实体库过程中依赖人工制定规则、信息提取不充分等问题,本文通过利用爬虫技术获取百度百科文本信息并构建地理实体语料库,端到端的深度学习模型BERT-BiLSTM-CRF模型对自建的地理实体语料库进行了命名实体识别(NER),模型在传统的BiLSTM-CRF模型上加入了BERT预训练模型,使得模型可以更好地结合文本上下文及语义信息。结果表明,该模型相比于BiLSTM-CRF模型和BiLSTM模型在地理实体命名实体识别中取得了更好的结果,且对进一步构建地理实体知识图谱、知识库等具有重要意义。
There is a large amount of information related to geographic information in the internet,among which text information is an important part.To address the problems of relying on manual rule-making and inadequate information extraction in the process of building a geographic entity corpus,the text information of Baidu Encyclopedia was obtained in this paper obtains and was explored to a geographic entity corpus by using crawler technology,and based on the BERT-BiLSTM-CRF model,the self-built geographic entity.The model added a BERT pre-training model to the traditional BiLSTM-CRF model,so that it can better combine textual context and semantic information.The results showed that the model achieved better results than the BiLSTM-CRF model and the BiLSTM model in geographic entity named entity recognition,and was important for further construction of geographic entity knowledge graphs and knowledge bases.
作者
汤洁仪
李大军
刘波
TANG Jieyi;LI Dajun;LIU Bo(College of Surveying and Mapping Engineering,East China University of Technology,Nanchang Jiangxi 330032,China)
出处
《北京测绘》
2023年第2期143-147,共5页
Beijing Surveying and Mapping
基金
国家自然科学基金(42161064)
江西省自然科学基金(20212BAB204003)。