1.College of Computer Science and Technology,Wuhan University of Science and Technology,Wuhan 430065,Hubei,China
2.Hubei Province Key Laboratory of Intelligent Information Processing and Real-Time Industrial System,Wuhan 430065,Hubei,China
3.Hubei Center for Data and Analysis,Hubei University of Economics,Wuhan 430205,Hubei,China
Show less
文章历史+
Received
Published
2021-03-30
2022-06-24
Issue Date
2026-07-23
PDF (764K)
摘要
针对标书文本重要信息的抽取需求,提出一种基于BERT(bidirectional encoder representations from transformers)的阅读理解式标书文本信息抽取方法。该方法将信息抽取任务转换为阅读理解任务,根据标书文本内容,生成对应问题,再抽取标书文本片段作为问题答案。利用BERT预训练模型,得到强健的语言模型,获取更深层次的上下文关联。相比传统的命名实体识别方法,基于阅读理解的信息抽取方法能够很好地同时处理非嵌套实体和嵌套实体的抽取,也能充分利用问题所包含的先验语义信息,区分出具有相似属性的信息。从中国政府采购网下载标书文本数据进行了实验,本文方法总体EM(exact match)值达到92.41%,F1值达到95.03%。实验结果表明本文提出的方法对标书文本的信息抽取是有效的。
Abstract
Aiming at the demand for extracting important information in the bidding text, a bidding text information extraction method based on BERT (Bidirectional Encoder Representations from Transformers) via reading comprehension is proposed. This method converts the information extraction task into a reading comprehension task, generates questions based on the content of the bidding text, and then extracts the text fragments as the answers to the questions. We use the BERT pre-trained model to obtain a robust language model and a deeper contextual association. Compared with traditional named entity recognition methods, the information extraction method via reading comprehension can handle the extraction of flat entities and nested entities at the same time, and can also make full use of the prior semantic information contained in the problem to distinguish information with similar attributes. Experiments were carried out by downloading the text data of the bid documents from the Chinese Government Procurement Network. The overall EM(exact match) value of the method in this paper reached 92.41%, and the F1 value reached 95.03%. The experimental results show that the method proposed in this paper is effective for extracting information from bidding texts.
信息抽取(information extraction)是从非结构化或半结构化的文本中对用户指定类型的实体、关系以及事件进行自动标识和分类,输出为结构化的信息[1]。由于其广泛应用,近年来,信息抽取成为自然语言处理(natural language processing)领域研究的热门课题。信息抽取主要包括命名实体识别(named entity recognition)、关系抽取(relation extraction)和事件抽取(event extraction)三个子任务[2],信息抽取的关键是命名实体的识别。早期常用的是基于规则的抽取方法[3],但该方法存在泛化能力差的问题。
深度学习的发展对自然语言处理领域产生了巨大的影响[4],在信息抽取中使用也相当普遍。在预训练模型发布之前,信息抽取主要模型以CNN和RNN[5]为主。Akbik等[6]通过动态存储每个词的所有上下文嵌入,并对这些嵌入进行池化操作以提取词的全局上下文嵌入,该方法显著提高了命名实体的识别效果。BERT(bidirectional encoder representations from transformers)模型的提出[7],使得多项自然语言处理任务取得了更好的效果,预训练模型被越来越多的学者使用。Xue等[8]提出针对联合实体和关系提取任务的集中注意力模型,该模型通过动态范围注意力机制将BERT模型集成到联合学习中,提高了共享参数层的特征表示能力。Qiu等[9]使用预训练模型,并以问答的方式,实现临床医疗文本的结构化。
F1:首先得到预测结果与标准答案的最长公共子串(longest common substring,LCS),根据LCS的长度和标准答案的长度计算召回率(R),根据LCS的长度和预测结果的长度计算精确率(P),再由R和P计算F1,如公式(3),然后同样取全部数据均值。F1用于计算预测结果与标准答案之间字符级别的匹配程度,反映了模糊匹配度。
LID M, ZHANGY, LID Y, et al. Review of entity relation extraction methods [J]. Journal of Computer Research and Development, 2020, 57(7): 1424-1448 (Ch). DOI: 10.7544/issn1000-1239.2020.20190358 .
[3]
GRISHMANR. Twenty-five years of information extraction [J]. Natural Language Engineering, 2019, 25(6): 677-692. DOI:10.1017/s1351324919000512 .
[4]
RAUL F. Extracting company names from text [C]// The Seventh IEEE Conference on Artificial Intelligence Application. New York: IEEE Press, 1991: 29-32. DOI:10.1109/CAIA.1991.120841 .
[5]
LAUZONF Q. An introduction to deep learning [C]//2012 11th International Conference on Information Science, Signal Processing and Their Applications (ISSPA). New York: IEEE Press, 2012: 1438-1439. DOI:10.1109/ISSPA.2012.6310529 .
[6]
YINW P, KANNK, YUM, et al. Comparative Study of CNN and RNN for Natural Language Processing [DB/OL].[2021-06-12].
[7]
AKBIKA, BERGMANNT, VOLLGRAFR. Pooled Contextualized Embeddings for Named Entity Recognition[DB/OL].[2021-06-12]. DOI:10.18653/v1/n19-1078 .
[8]
DEVLINJ, CHANGM W, LEEK, et al. BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding[EB/OL]. [2021-06-12].DOI: 10.18653/v1/n19-1423 .
[9]
XUEK, ZHOUY M, MAZ Y, et al. Fine-tuning BERT for joint entity and relation extraction in Chinese medical text [C]//2019 IEEE International Conference on Bioinformatics and Biomedicine (BIBM). New York: IEEE Press, 2019: 892-897. DOI:10.1109/BIBM47256.2019.8983370 .
[10]
QIUJ H, ZHOUY M, MAZ Y, et al. Question answering based clinical text structuring using pre-trained language model[C]//2019 IEEE International Conference on Bioinformatics and Biomedicine (BIBM). New York: IEEE Press, 2019: 1596-1600. DOI:10.1109/BIBM47256.2019.8983142 .
[11]
LEVYO, SEOM, CHOIE, et al. Zero-shot relation extraction via reading comprehension [C]//Proceedings of the 21st Conference on Computational Natural Language Learning (CoNLL 2017). Stroudsburg: Association for Computational Linguistics, 2017: 333-342. DOI:10.18653/v1/k17-1034 .
[12]
MCCANNB, KESKARN S, XIONGC M, et al. The Natural Language Decathlon: Multitask Learning as Question Answering[EB/OL]. [2021-08-06]. DOI: 10.48550/arXiv.1806.08730 .
[13]
LIX Y, YINF, SUNZ J, et al. Entity-relation extraction as multi-turn question answering [C]//Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics. Stroudsburg: Association for Computational Linguistics, 2019: 1340-1350. DOI:10.18653/v1/p19-1129 .
[14]
QIUL, ZHOUH, QUY R, et al. QA4IE: A Question Answering Based Framework for Information Extraction[DB/OL].[2021-06-12]. DOI: 10.1007/978-3-030-00671-6_12 .
[15]
LIX Y, FENGJ R, MENGY X, et al. A unified MRC framework for named entity recognition[C]//Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics. Stroudsburg: Association for Computational Linguistics, 2020: 5849-5859. DOI:10.18653/v1/2020.acl-main.519 .
[16]
RAJPURKARP, ZHANGJ, LOPYREVK, et al. SQuAD: 100,000+ Questions for Machine Comprehension of Text [DB/OL].[2021-06-12]. DOI:10.18653/v1/d16-1264 .
SEOM, KEMBHAVIA, FARHADIA, et al. Bidirectional Attention Flow for Machine Comprehension [EB/OL]. [2021-06-12]. DOI: 10.1109/cvpr.2017.571 .
[19]
RAJPURKARP, JIAR, LIANGP. Know what you don’t know: Unanswerable questions for SQuAD [C]//Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics (Volume 2: Short Papers). Stroudsburg: Association for Computational Linguistics, 2018: 784-789. DOI:10.18653/v1/p18-2124 .
LINB Y, XUF, LUOZ Y, et al. Multi-channel BiLSTM-CRF model for emerging named entity recognition in social media [C]//Proceedings of the 3rd Workshop on Noisy User-Generated Text. Stroudsburg: Association for Computational Linguistics, 2017: 160-165. DOI:10.18653/v1/w17-4421 .
[22]
CUIY M, CHEW X, LIUT, et al. Pre-Training with Whole Word Masking for Chinese BERT [EB/OL]. [2021-06-12]. DOI: 10.1109/taslp.2021.3124365 .