区级人大报告中的地名自动标注研究  

Research on automatic labeling of place names in the report of the National People′s Congress Report

作　　者：喻航 李红莲[1] 吕学强[2] YU Hang;LI Honglian;Lü Xueqiang(School of Information and Communication Engineering,Beijing Information Science&Technology University,Beijing 100101,China;Beijing Key Laboratory of Internet Culture and Digital Dissemination Research,Beijing Information Science&Technology University,Beijing 100101,China)

机构地区：[1]北京信息科技大学信息与通信工程学院,北京100101 [2]北京信息科技大学网络文化与数字传播北京市重点实验室,北京100101

出　　处：《北京信息科技大学学报（自然科学版）》2020年第6期31-36,共6页Journal of Beijing Information Science and Technology University

基　　金：国家自然科学基金项目(61671070);北京成像技术高精尖创新中心项目(BAICIT-2016003);国家社会科学基金重大项目(14@ZH0363);国家语委重点项目(ZDI135-53)。

摘　　要：针对区级人大报告中地名的自动标注问题,采用BERT(bidirectional encoder representations from transformers)预训练语言模型、条件随机场模型加改进双向长短时记忆神经网络的方法来识别地名,并利用屏蔽敏感词的算法实现对识别到的地名的自动标注。BERT模型具备更好的捕获上下文信息特征的能力;改进的双向长短时记忆网络缩短了模型的训练时间,提升了模型的准确率。实验结果表明,该方法提升了标注结果的F值。For the automatic labeling of place names in district-level people′s congress reports,the BERT(bidirectional encoder representations from transformers)pre-trained language model,the conditional random field model and improved two-way long and short-term memory neural network method were used to recognize place names,and the algorithm of shielding sensitive words were used to realize the automatic labeling of recognized place names.The BERT model has a better ability to capture contextual information features;the improved two-way long and short-term memory network shortens the training time and improves the accuracy of the model.The experimental results show that this method improves the F value of the labeling result.

关 键 词：人大报告 自动标注 BERT 神经网络 

分 类 号：TP391.1[自动化与计算机技术—计算机应用技术]