Please wait a minute...
Advanced Search
数据分析与知识发现  2021, Vol. 5 Issue (8): 1-12     https://doi.org/10.11925/infotech.2096-3467.2021.0181
  研究论文 本期目录 | 过刊浏览 | 高级检索 |
融合BERT与关系位置特征的军事领域关系抽取方法*
马江微1,吕学强1,游新冬1(),肖刚2,韩君妹2
1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
2复杂系统仿真总体重点实验室 北京 100101
Extracting Relationship Among Military Domains with BERT and Relation Position Features
Ma Jiangwei1,Lv Xueqiang1,You Xindong1(),Xiao Gang2,Han Junmei2
1Beijing Key Laboratory of Internet Culture and Digital Dissemination Research, Beijing Information Science and Technology University, Beijing 100101, China
2National Key Laboratory of Complex System Simulation, Beijing 100101, China
全文: PDF (1209 KB)   HTML ( 38
输出: BibTeX | EndNote (RIS)      
摘要 

【目的】 解决军事文本中实体关系重叠引起的关系抽取困难问题,改善军事文本关系抽取效果。【方法】 使用BERT模型作为输入文本的编码器,采用分层强化学习方法分别进行关系与其对应实体的解码,并在实体解码过程中融合关系位置特征,构建军事领域关系抽取模型。【结果】 在军事武器装备数据集上F1值达到82.2%,相较其他方法提升约8个百分点。在公开的NYT10、NYT10-sub数据集上F1值分别达到71.8%和69.0%,相较其他方法提升约7个百分点与9个百分点。【局限】 在人工标注数据集上抽取效果较好,在存在噪声的远程监督数据集上效果有待提升。【结论】 所提方法较目前主流方法在军事领域的关系抽取中效果更好,同时具有一定的泛化能力。

服务
把本文推荐给朋友
加入引用管理器
E-mail Alert
RSS
作者相关文章
马江微
吕学强
游新冬
肖刚
韩君妹
关键词 关系抽取BERT关系位置特征强化学习    
Abstract

[Objective] This article addresses the difficulties of relationship extraction due to overlapping entity relationship in military texts. [Methods] We used the BERT model as the encoder for the input texts, and used the hierarchical reinforcement learning approach to decode relationship and their corresponding entities. Then, we merged the relational position features in the entity decoding process to construct a relationship extraction model for military domains. [Results] The F1 value reached 82.2% on the military weapon and equipment dataset, which was about 8% higher than other methods. Using the publicly available NYT10 and NYT10-sub datasets, the F1 values reached 71.8% and 69.0%, which was about 7% and 9% higher than other methods. [Limitations] The new method’s extraction performance is better on manually annotated datasets. More research is needed to improve it performance on remotely supervised datasets with much noise. [Conclusions] The HBP method could effectively extract relationship among the military domains, and has some generalization potentiality.

Key wordsRelation Extraction    BERT    Relation Position Feature    Reinforcement Learning
收稿日期: 2021-02-24      出版日期: 2021-09-15
ZTFLH:  TP391  
基金资助:*北京市自然科学基金项目(4212020);国防科技重点实验室基金项目(6412006200404);北京信息科技大学“勤信人才”培育计划项目(QXTCP B201908)
通讯作者: 游新冬 ORCID:0000-0002-3351-4599     E-mail: youxindong@bistu.edu.cn
引用本文:   
马江微, 吕学强, 游新冬, 肖刚, 韩君妹. 融合BERT与关系位置特征的军事领域关系抽取方法*[J]. 数据分析与知识发现, 2021, 5(8): 1-12.
Ma Jiangwei, Lv Xueqiang, You Xindong, Xiao Gang, Han Junmei. Extracting Relationship Among Military Domains with BERT and Relation Position Features. Data Analysis and Knowledge Discovery, 2021, 5(8): 1-12.
链接本文:  
https://manu44.magtech.com.cn/Jwk_infotech_wk3/CN/10.11925/infotech.2096-3467.2021.0181      或      https://manu44.magtech.com.cn/Jwk_infotech_wk3/CN/Y2021/V5/I8/1
Fig.1  HBP模型结构
Fig.2  HBP的分层强化学习过程
Fig.3  高层关系识别结构
Fig.4  低层实体抽取结构
项目 军事武器装备数据集 NYT10
关系类型 16 29
训练集 2 532 66 816
验证集 133 3 516
测试集 297 4 006
Table 1  数据集统计
数据样本 “贝劳伍德”号航空母舰1942年12月6日在纽约海军造船厂下水,1943年3月31日服役。
下水时间 “/H_B 贝/H_I 劳/H_I 伍/H_I 德/H_I ”/H_I 号/H_I 航/H_I 空/H_I 母/H_I 舰/H_I
1/T_B 9/T_I 4/T_I 2/T_I 年/T_I 1/T_I 2/T_I 月/T_I 6/T_I 日/T_I 在/N
纽/N 约/N 海/N 军/N 造/N 船/N 厂/N 下/N 水/N ,/N 1/N
9/N 4/N 3/N 年/N 3/N 月/N 3/N 1/N 日/N 服/N 役/N
服役时间 “/H_B 贝/H_I 劳/H_I 伍/H_I 德/H_I ”/H_I 号/H_I 航/H_I 空/H_I 母/H_I 舰/H_I
1/N 9/N 4/N 2/N 年/N 1/N 2/N 月/N 6/N 日/N 在/N
纽/N 约/N 海/N 军/N 造/N 船/N 厂/N 下/N 水/N ,/N 1/T_B
9/T_I 4/ T_I 3/ T_I 年/ T_I 3/ T_I 月/ T_I 3/ T_I 1/ T_I 日/ T_I 服/N 役/N
Table 2  数据标注示例
关系类型 关系类型
舰船-装备-武器 舰船-满载排水量-重量
舰船-搭载-飞机 舰船-舰宽-宽度
舰船-属于-国家、舰队 舰船-舰长-长度
舰船-继承-舰船 舰船-造价-价格
舰船-级下号-舰船 舰船-服役时间-时间
舰船-代号-字符 舰船-下水时间-时间
舰船-舷号-字符 舰船-航速-速度
舰船-标准排水量-重量 舰船-吃水-深度
Table 3  军事武器装备数据集关系类型说明
参数 参数值
状态向量大小 768
隐藏层向量大小 768
位置向量大小 30
Batch Size 16
学习率 4e-5
β 0.90
γ 0.95
Table 4  实验参数设置
Fig.5  军事武器装备数据集实例
方法 准确率 召回率 F 1
NovelTagging 0.468 0.158 0.237
CopyR 0.336 0.206 0.256
GraphRel 0.438 0.305 0.360
HRL 0.538 0.336 0.414
BL 0.960 0.605 0.742
HBP 0.838 0.807 0.822
Table 5  军事武器装备数据集对比实验结果
方法 NYT10 NYT10-sub
准确率 召回率 F 1 准确率 召回率 F 1
SPTree 0.492 0.557 0.522 0.272 0.315 0.292
NovelTagging 0.593 0.381 0.464 0.256 0.237 0.246
CopyR 0.569 0.452 0.504 0.392 0.263 0.315
GraphRel 0.639 0.600 0.619
HRL 0.714 0.586 0.644 0.815 0.475 0.600
HBP 0.749 0.689 0.718 0.842 0.585 0.690
Table 6  NYT10数据集对比实验结果
数据集 方法 准确率 召回率 F 1
NYT10 HBP(+LSTM+POS) 0.707 0.589 0.643
HBP(+BERT) 0.710 0.661 0.685
HBP(+BERT+POS) 0.749 0.689 0.718
NYT10-sub HBP(+LSTM+POS) 0.803 0.485 0.605
HBP(+BERT) 0.816 0.580 0.678
HBP(+BERT+POS) 0.842 0.585 0.690
军事武器装备
数据集
HBP(+LSTM+POS) 0.540 0.537 0.538
HBP(+BERT) 0.804 0.761 0.782
HBP(+BERT+POS) 0.838 0.807 0.822
Table 7  消融实验结果
[1] 林旺群, 汪淼, 王伟, 等. 知识图谱研究现状及军事应用[J]. 中文信息学报, 2020, 34(12):9-16.
[1] ( Lin Wangqun, Wang Miao, Wang Wei, et al. A Survey to Knowledge Graph and Its Military Application[J]. Journal of Chinese Information Processing, 2020, 34(12):9-16.)
[2] 鄂海红, 张文静, 肖思琪, 等. 深度学习实体关系抽取研究综述[J]. 软件学报, 2019, 30(6):1793-1818.
[2] ( E Haihong, Zhang Wenjing, Xiao Siqi, et al. Survey of Entity Relationship Extraction Based on Deep Learning[J]. Journal of Software, 2019, 30(6):1793-1818.)
[3] Mintz M, Bills S, Snow R, et al. Distant Supervision for Relation Extraction without Labeled Data [C]// Proceedings of the Joint Conference of the 47th Annual Meeting of the ACL and the 4th International Joint Conference on Natural Language Processing of the AFNLP. 2009: 1003-1011.
[4] Gormley M R, Yu M, Dredze M. Improved Relation Extraction with Feature-Rich Compositional Embedding Models [C]// Proceedings of the 2015 Conference on Empirical Methods in Natural Language Processing. 2015: 1774-1784.
[5] Zheng S C, Wang F, Bao H Y, et al. Joint Extraction of Entities and Relations Based on a Novel Tagging Scheme [C]// Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics. 2017: 1227-1236.
[6] Takanobu R, Zhang T Y, Liu J X, et al. A Hierarchical Framework for Relation Extraction with Reinforcement Learning [C]// Proceedings of the AAAI Conference on Artificial Intelligence. 2019: 7072-7079.
[7] Devlin J, Chang M W, Lee K, et al. BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding [C]// Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies. 2019: 4171-4186.
[8] Lample G, Ballesteros M, Subramanian S, et al. Neural Architectures for Named Entity Recognition [C]// Proceedings of the 2016 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies. 2016: 260-270.
[9] Santos C N D, Xiang B, Zhou B W. Classifying Relations by Ranking with Convolutional Neural Networks [C]// Proceedings of the 53rd Annual Meeting of the Association for Computational Linguistics and the 7th International Joint Conference on Natural Language Processing. 2015: 626-634.
[10] 张琴, 郭红梅, 张智雄. 融合词嵌入表示特征的实体关系抽取方法研究[J]. 数据分析与知识发现, 2017, 1(9):8-15.
[10] ( Zhang Qin, Guo Hongmei, Zhang Zhixiong. Extracting Entity Relationship with Word Embedding Representation Features[J]. Data Analysis and Knowledge Discovery, 2017, 1(9):8-15.)
[11] 张东东, 彭敦陆. ENT-BERT:结合BERT和实体信息的实体关系分类模型[J]. 小型微型计算机系统, 2020, 41(12):2557-2562.
[11] ( Zhang Dongdong, Peng Dunlu. ENT-BERT: Entity Relation Classification Model Combining BERT and Entity Information[J]. Journal of Chinese Computer Systems, 2020, 41(12):2557-2562.)
[12] 李卫疆, 李涛, 漆芳. 基于多特征自注意力BLSTM的中文实体关系抽取[J]. 中文信息学报, 2019, 33(10):47-56.
[12] ( Li Weijiang, Li Tao, Qi Fang. Chinese Entity Relation Extraction Based on Multi-Features Self-Attention Bi-LSTM[J]. Journal of Chinese Information Processing, 2019, 33(10):47-56.)
[13] 吴粤敏, 丁港归, 胡滨. 基于注意力机制的农业金融文本关系抽取研究[J]. 数据分析与知识发现, 2019, 3(5):86-92.
[13] ( Wu Yuemin, Ding Ganggui, Hu Bin. Extracting Relationship of Agricultural Financial Texts with Attention Mechanism[J]. Data Analysis and Knowledge Discovery, 2019, 3(5):86-92.)
[14] Li Q, Ji H. Incremental Joint Extraction of Entity Mentions and Relations [C]// Proceedings of the 52nd Annual Meeting of the Association for Computational Linguistics. 2014: 402-412.
[15] Ren X, Wu Z Q, He W Q, et al. CoType: Joint Extraction of Typed Entities and Relations with Knowledge Bases [C]// Proceedings of the 26th International Conference on World Wide Web. 2017: 1015-1024.
[16] Miwa M, Bansal M. End-to-End Relation Extraction Using LSTMs on Sequences and Tree Structures [C]// Proceedings of the 54th Annual Meeting of the Association for Computational Linguistics. 2016: 1105-1116.
[17] Bekoulis G, Deleu J, Demeester T, et al. Joint Entity Recognition and Relation Extraction as a Multi-head Selection Problem[J]. Expert Systems with Applications, 2018, 114:34-45.
doi: 10.1016/j.eswa.2018.07.032
[18] Li X Y, Yin F, Sun Z J, et al. Entity-Relation Extraction as Multi-Turn Question Answering [C]// Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics. 2019: 1340-1350.
[19] Wei Z P, Su J L, Wang Y, et al. A Novel Cascade Binary Tagging Framework for Relational Triple Extraction [C]// Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics. 2020: 1476-1488.
[20] Zeng D J, Liu K, Lai S, et al. Relation Classification via Convolutional Deep Neural Network [C]// Proceedings of the 25th International Conference on Computational Linguistics. 2014: 2335-2344.
[21] Ji G, Liu K, He S W, et al. Distant Supervision for Relation Extraction with Sentence-level Attention and Entity Descriptions [C]// Proceedings of the 31st AAAI Conference on Artificial Intelligence. 2017: 3060-3066.
[22] 李芊芊, 张克亮. 基于依存分析的军事领域英文实体关系抽取研究[J]. 情报工程, 2019, 5(1):98-112.
[22] ( Li Qianqian, Zhang Keliang. Entity Relation Extraction Based on Dependency Parsing in Military Field[J]. Technology Intelligence Engineering, 2019, 5(1):98-112.)
[23] 田佳来, 吕学强, 游新冬, 等. 基于分层序列标注的实体关系联合抽取方法[J]. 北京大学学报(自然科学版), 2021, 57(1):53-60.
[23] ( Tian Jialai, Lü Xueqiang, You Xindong, et al. Joint Extraction of Entities and Relations Based on Hierarchical Sequence Labeling[J]. Acta Scientiarum Naturalium Universitatis Pekinensis, 2021, 57(1):53-60.)
[24] Vaswani A, Shazeer N, Parmar N, et al. Attention is All You Need[J]. Advances in Neural Information Processing Systems, 2017, 30:5998-6008.
[25] 李培芸, 李茂西, 裘白莲, 等. 融合BERT语境词向量的译文质量估计方法研究[J]. 中文信息学报, 2020, 34(3):56-63.
[25] ( Li Peiyun, Li Maoxi, Qiu Bailian, et al. Integrating BERT Word Embedding into Quality Estimation of Machine Translation[J]. Journal of Chinese Information Processing, 2020, 34(3):56-63.)
[26] Sutton R S, Barto A G. Reinforcement Learning: An Introduction[M]. MIT Press, 1998.
[27] Bishop C M. Pattern Recognition and Machine Learning[M]. Springer, 2006.
[28] van Rijsbergen C J. Information Retrieval[M]. Butterworths, 1975.
[29] Sutton R S, McAllester D A, Singh S P, et al. Policy Gradient Methods for Reinforcement Learning with Function Approximation[A]// Advances in Neural Information Processing Systems[M]. MIT Press, 2000: 1057-1063.
[30] Riedel S, Yao L M, McCallum A. Modeling Relations and Their Mentions without Labeled Text [C]// Proceedings of Joint European Conference on Machine Learning and Knowledge Discovery in Databases. Springer, Berlin, Heidelberg, 2010: 148-163.
[31] Zeng X R, Zeng D J, He S Z, et al. Extracting Relational Facts by an End-to-end Neural Model with Copy Mechanism [C]// Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics. 2018: 506-514.
[32] Fu T J, Li P H, Ma W Y. GraphRel: Modeling Text as Relational Graphs for Joint Entity and Relation Extraction [C]// Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics. 2019: 1409-1418.
[33] Su J L. Hybrid Structure of Pointer and Tagging for Relation Extraction: A Baseline[EB/OL].(2019-06-03).[2021-02-24]. https://github.com/bojone/kg-2019-baseline.
[1] 陈杰,马静,李晓峰. 融合预训练模型文本特征的短文本分类方法*[J]. 数据分析与知识发现, 2021, 5(9): 21-30.
[2] 周泽聿,王昊,赵梓博,李跃艳,张小琴. 融合关联信息的GCN文本分类模型构建及其应用研究*[J]. 数据分析与知识发现, 2021, 5(9): 31-41.
[3] 王一钒,李博,史话,苗威,姜斌. 古汉语实体关系联合抽取的标注方法*[J]. 数据分析与知识发现, 2021, 5(9): 63-74.
[4] 李文娜, 张智雄. 基于联合语义表示的不同知识库中的实体对齐方法研究*[J]. 数据分析与知识发现, 2021, 5(7): 1-9.
[5] 王昊, 林克柔, 孟镇, 李心蕾. 文本表示及其特征生成对法律判决书中多类型实体识别的影响分析[J]. 数据分析与知识发现, 2021, 5(7): 10-25.
[6] 喻雪寒, 何琳, 徐健. 基于RoBERTa-CRF的古文历史事件抽取方法研究*[J]. 数据分析与知识发现, 2021, 5(7): 26-35.
[7] 张乐, 冷基栋, 吕学强, 崔卓, 王磊, 游新冬. RLCPAR:一种基于强化学习的中文专利摘要改写模型*[J]. 数据分析与知识发现, 2021, 5(7): 59-69.
[8] 刘文斌, 何彦青, 吴振峰, 董诚. 基于BERT和多相似度融合的句子对齐方法研究*[J]. 数据分析与知识发现, 2021, 5(7): 48-58.
[9] 陆泉, 何超, 陈静, 田敏, 刘婷. 基于两阶段迁移学习的多标签分类模型研究*[J]. 数据分析与知识发现, 2021, 5(7): 91-100.
[10] 阮小芸,廖健斌,李祥,杨阳,李岱峰. 基于人才知识图谱推理的强化学习可解释推荐研究*[J]. 数据分析与知识发现, 2021, 5(6): 36-50.
[11] 尹鹏博,潘伟民,张海军,陈德刚. 基于BERT-BiGA模型的标题党新闻识别研究*[J]. 数据分析与知识发现, 2021, 5(6): 126-134.
[12] 宋若璇,钱力,杜宇. 基于科技论文中未来工作句集的学术创新构想话题自动生成方法研究*[J]. 数据分析与知识发现, 2021, 5(5): 10-20.
[13] 胡昊天,吉晋锋,王东波,邓三鸿. 基于深度学习的食品安全事件实体一体化呈现平台构建*[J]. 数据分析与知识发现, 2021, 5(3): 12-24.
[14] 王倩,王东波,李斌,许超. 面向海量典籍文本的深度学习自动断句与标点平台构建研究*[J]. 数据分析与知识发现, 2021, 5(3): 25-34.
[15] 常城扬,王晓东,张胜磊. 基于深度学习方法对特定群体推特的动态政治情感极性分析*[J]. 数据分析与知识发现, 2021, 5(3): 121-131.
Viewed
Full text


Abstract

Cited

  Shared   
  Discussed   
版权所有 © 2015 《数据分析与知识发现》编辑部
地址:北京市海淀区中关村北四环西路33号 邮编:100190
电话/传真:(010)82626611-6626,82624938
E-mail:jishu@mail.las.ac.cn