摘要
网络文本中蕴含着大量的交通信息,对交通命名实体识别是地情变化监测、交通实体更新的重要前提。针对交通命名实体识别任务中缺少专业标注数据和有效识别方法导致识别效果无法满足需求的问题,文中提出一种融合深度学习模型的交通命名实体自动提取方法。该方法首先利用语言表征模型充分融合语境来提取文本特征,然后通过结合双向神经网络模型来学习上下文语境,最终由条件随机场模型对输出进行约束得到全局最优标记序列实现交通实体的识别。实验证明,该方法准确率可以达到90%以上,能够实现对交通实体的有效识别。
The network text contains a lot of traffic information,and the recognition of traffic named entities is an important prerequisite for monitoring changes in ground conditions and updating traffic entities.This paper proposes a method for automatic extraction of traffic named entities that incorporates deep learning models.The method first uses a linguistic representation model to fully integrate the context to extract textual features,then learns contextual information by combining a bidirectional neural network model,and finally constrains the output by a conditional random field model to obtain the global optimal token sequence to achieve the recognition of traffic entities.Experiments prove that this method can achieve an accuracy rate of over 90%and can achieve effective recognition of traffic entities.
作者
王树杰
马照亭
吴政
杨健男
WANG Shujie;MA Zhaoting;WU Zheng;YANG Jiannan(Chinese Academy of Surveying and Mapping,Beijing 100036,China)
出处
《测绘工程》
2024年第2期65-70,共6页
Engineering of Surveying and Mapping
基金
实景三维中国建设专项基金资助项目(121136000000210004)
国家重点研发计划(2018YFB2100702)。
关键词
交通命名实体
命名实体识别
双向神经网络
条件随机场
traffic named entity
named entity recognition
bidirectional neural network
conditional random fields