基于强化学习和注意力机制的朝鲜语文本结构发现

doi:10.13229/j.cnki.jdxbgxb20200358

摘要/Abstract

摘要：

将注意力机制与深度强化学习相结合，利用标签信息研究如何自主学习出有效的朝鲜语文本结构化表示，提出了两种结构化表示模型：信息蒸馏注意力模型（ID-Attention）和层次结构注意力模型（HS-Attention）。ID-Attention选择与任务相关的重要单词，而HS-Attention在句中发现短语结构。两种表示模型中的结构发现是一个顺序决策问题，使用强化学习中的Policy Gradient实现。实验结果表明：ID-Attention能够识别朝鲜语重要单词；HS-Attention能够很好地提取出句子结构，在文本分类任务上有很好的性能表现，同时，两模型的结果对语料库的标注有很好的辅助作用。

关键词: 人工智能, 深度强化学习, 注意力机制, 文本结构发现, 朝鲜语自然语言处理

Abstract:

In this paper, attention mechanism is combined with deep reinforcement learning, and label information is used to study how to learn effective Korean language text structured representation independently. Two structured representation models are proposed, which are called Information Distilled Attention (IDA) and Hierarchically Structured Attention (HSA). IDA selects the important words related to the task, and HSA finds the phrase structure in the sentence. The structural discovery in both presentation models is a sequential decision problem that can be implemented using Policy Gradient (PG) in reinforcement learning. The experimental results show that the proposed IDA can recognize the important words of Korean, and HSA can extract the sentence structure well, and have good performance in the task of text classification. At the same time, the results of the two models have a good auxiliary effect on corpus tagging.

Key words: artificial intelligence, deep reinforcement learning, attention mechanism, text structure discovery, Korean natural language processing

中图分类号:

TP391.1

赵亚慧,杨飞扬,张振国,崔荣一. 基于强化学习和注意力机制的朝鲜语文本结构发现[J]. 吉林大学学报(工学版), 2021, 51(4): 1387-1395.

Ya-hui ZHAO,Fei-yang YANG,Zhen-guo ZHANG,Rong-yi CUI. Korean text structure discovery based on reinforcement learning and attention mechanism[J]. Journal of Jilin University(Engineering and Technology Edition), 2021, 51(4): 1387-1395.

图/表 9

图1

图2

图3

图4

表1

图5

图6

表2

表3

参考文献 23

1	Bengio Y, Courville A,Vincent P. Representation learning: a review and new perspectives[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence,2013,35(8):1798-1828.
2	Socher R, Pennington J, Huang E H. Semi-supervised recursive auto encoders for predicting sentiment distributions[C]∥The Conference on Empirical Methods in Natural Language Processing, Edinburgh, UK, 2011, 151-161.
3	毕玉德.朝鲜语自然语言处理研究管窥[J].中文信息学报, 2011,25(6):166-169, 182.
	Bi Yu-de. A research on Korean natural language processing[J]. Journal of Chinese Information Processing, 2011,25 (6): 166-169, 182.
4	Joulin A, Grave E, Bojanowski P. Bag of tricks for efficient text classification[C]∥The European Chapter of the ACL, Valencia, Spain, 2017:427-431.
5	Kim Y. Convolutional neural networks for sentence classification[C]∥The Conference on Empirical Methods in Natural Language Processing, Doha, Qatar,2014:1746-1751.
6	Zhu X, Sobihani P, Guo H. Long short-term memory over recursive structures[C]∥International Conference on Machine Learning, Miami, Florida, USA,2015:1604-1612.
7	Yang Z, Yang D, Dyer C. Hierarchical attention networks for document classification[C]∥Annual Conference of the North American Chapter of the Association for Computational Linguistics, San Diego,USA,2016:1480-1489.
8	Yogatama D, Blunsom P, Dyer C. Learning to compose words into sentences with reinforcement learning[C]∥ International Conference on Learning Representations,Toulon, France, 2017.
9	Chung J, Gulcehre C, Cho K. Empirical evaluation of gated recurrent neural networks on sequence modeling[C]∥NIPS, Montreal, Canada, 2014.
10	Zhang Tian-yang, Huang Min-lie, Zhao Li. Learning structured representation for text classification via reinforcement learning[C]∥The Association for the Advance of Artificial Intelligence, New Orleans,USA, 2018:6053-6060.
11	Williams R J. Simple statistical gradient-following algorithms for connectionist reinforcement learning[J]. Machine Learning, 1992,2(3/4):229-256.
12	Keneshloo Y, Ramakrishnan N, Reddy C K. Deep transfer reinforcement learning for text summarization[J]. Society for Industrial and Applied Mathematics, 2019, arxiv:.
13	Sutton R S, McAllester D A, Singh S P. Policy gradient methods for reinforcement learning with function approximation[C]∥NIPS, Denver, USA,2000:1057-1063.
14	Graves A, Schmidhuber Jürgen. Frame wise phoneme classification with bidirectional LSTM and other neural network architectures[J]. Neural Networks, 2005, 18(5/6):602-610.
15	Bahdanau D, Cho K, Bengio Y. Neural machine translation by jointly learning to align and translate[C]∥ International Conference on Machine Learning, Florida, USA,2015, arXiv:.
16	Tang D, Qin B, Liu T. Document modeling with gated recurrent neural network for sentiment classification[C]∥ Proceedings of the Conference on Empirical Methods in Natural Language Processing, Lisbon, Portugal, 2015:1422-1432.
17	Chung J, Ahn S, Bengio Y. Hierarchical multiscale recurrent neural networks[C]∥International Conference on Learning Representations, Toulon, France, 2017.
18	Pennington J, Socher R, Manning C D. Glove:Global vectors for word representation[C]∥Proceedings of the Conference on Empirical Methods in Natural Language Processing, Doha, Qatar, 2014:1532-1543.
19	Kingma D, Ba J. Adam: a method for stochastic optimization[C]∥International Conference on Learning Representations, Sandiego,USA,2015.
20	Mingjie T, Yahui Z, Rongyi C. Identifying word translations in scientific literature based on labeled bilingual topic model and co-occurrence features[C]∥The 17th China National Conference on Computational Linguistics, Changsha, China, 2018:79-92.
21	孟先艳,崔荣一,赵亚慧,等.基于双向长短时记忆单元和卷积神经网络的多语种文本分类方法[J].计算机应用研究, 2020,37(9):2669-2673.
	Meng Xian-yan, Cui Rong-yi, Zhao Ya-hui, et al. Multilingual text classification method based on bidirectional long-short memory unit and convolutional neural network[J]. Computer Application Research, 2020,37(9):2669-2673.
22	Vaswani A,Shazeer N, Parmar N. Attention is all you need[J]. Advances in Neural Information Processing Systems, 2017:5999-6009.
23	Lin Z, Feng M, Santos C N. A structured self-attentive sentence embedding[C]∥International Conference on Learning Representations, Toulon, France,2017.

相关文章 15

[1]	刘元宁,吴迪,朱晓冬,张齐贤,李双双,郭书君,王超. 基于YOLOv3改进的用户界面组件检测算法[J]. 吉林大学学报(工学版), 2021, 51(3): 1026-1033.
[2]	尚福华,曹茂俊,王才志. 基于人工智能技术的局部离群数据挖掘方法[J]. 吉林大学学报(工学版), 2021, 51(2): 692-696.
[3]	赵海英,周伟,侯小刚,张小利. 基于多任务学习的传统服饰图像双层标注[J]. 吉林大学学报(工学版), 2021, 51(1): 293-302.
[4]	赵宏伟,刘晓涵,张媛,范丽丽,龙曼丽,臧雪柏. 基于关键点注意力和通道注意力的服装分类算法[J]. 吉林大学学报(工学版), 2020, 50(5): 1765-1770.
[5]	车翔玖,董有政. 基于多尺度信息融合的图像识别改进算法[J]. 吉林大学学报(工学版), 2020, 50(5): 1747-1754.
[6]	欧阳丹彤,马骢,雷景佩,冯莎莎. 知识图谱嵌入中的自适应筛选[J]. 吉林大学学报(工学版), 2020, 50(2): 685-691.
[7]	李贻斌,郭佳旻,张勤. 人体步态识别方法与技术[J]. 吉林大学学报(工学版), 2020, 50(1): 1-18.
[8]	徐谦,李颖,王刚. 基于深度学习的行人和车辆检测[J]. 吉林大学学报(工学版), 2019, 49(5): 1661-1667.
[9]	高万夫,张平,胡亮. 基于已选特征动态变化的非线性特征选择方法[J]. 吉林大学学报(工学版), 2019, 49(4): 1293-1300.
[10]	杨顺,蒋渊德,吴坚,刘海贞. 基于多类型传感数据的自动驾驶深度强化学习方法[J]. 吉林大学学报(工学版), 2019, 49(4): 1026-1033.
[11]	欧阳丹彤,肖君,叶育鑫. 基于实体对弱约束的远监督关系抽取[J]. 吉林大学学报(工学版), 2019, 49(3): 912-919.
[12]	顾海军, 田雅倩, 崔莹. 基于行为语言的智能交互代理[J]. 吉林大学学报(工学版), 2018, 48(5): 1578-1585.
[13]	董飒, 刘大有, 欧阳若川, 朱允刚, 李丽娜. 引入二阶马尔可夫假设的逻辑回归异质性网络分类方法[J]. 吉林大学学报(工学版), 2018, 48(5): 1571-1577.
[14]	王旭, 欧阳继红, 陈桂芬. 基于垂直维序列动态时间规整方法的图相似度度量[J]. 吉林大学学报(工学版), 2018, 48(4): 1199-1205.
[15]	张浩, 占萌苹, 郭刘香, 李誌, 刘元宁, 张春鹤, 常浩武, 王志强. 基于高通量数据的人体外源性植物miRNA跨界调控建模[J]. 吉林大学学报(工学版), 2018, 48(4): 1206-1213.

Metrics

Viewed

Full text

Abstract

Cited

Shared

Discussed

类别	词条数	类别	词条数
动物	4582	植物	6172
微生物	5472	生物技术	1215
生物医学	2752	气候	708
海洋环境	810	地质	1735
海洋技术	819	材料工程	781
测控技术	1728	航天	4436
其他	1478

模型	准确率/%	模型	准确率/%
CNN	78.50	Self?Attention	82.91
LSTM	74.60	ID?LSTM	82.11
Bi?LSTM	78.14	HS?LSTM	83.65
MT	76.50	ID?Attention	84.84
T?BLSTM?CNN	81.68	HS?Attention	85.11

原始文本	??? ?? ??? ??? ?? ?? ?? ??? ?? ?? ?? ???? ?? ??? ?? ? ??? ?? ???? ??? ??? ??? ??（城市是受人类活动影响最大的地球表面，城市系统的碳循环在世界及地区的碳循环中具有重要的位置和作用。）
ID?Attention	??? ??~~???~~????????? ??? ?? ?? ?? ???? ?? ??????~~???~~??~~????~~??? ??? ?????
HS?Attention	??? \| ?? ??? ??? ?? ?? ?? \| ??? ?? ?? \| ?? ???? ?? ??? ?? ? ??? ?? ???? \| ??? ??? ??? ?? \|
原始文本	?? ????? ?? ??? ???? ?? ? ????? ?? ?? ???? ???? ??? ??? ?? ??? ? ??（从昆虫病原直接取样，检出及定量化可直接反映昆虫流行病学调查中病原丰度）
ID?Attention	?? ?????????? ???? ???????????? ????~~????~~??? ???????????
HS?Attention	?? ????? \| ?? ??? ???? ?? ? ????? \| ?? ?? ???? ???? ??? \| ??? ?? ??? ? ?? \|
原始文本	???? ?? ???? ? ?? ?? ??? ???? ????? ?? ?? ???? ???? ?? ?? ?? ?? ???? ???? ?? ??? ????? ??? ??（温室气体排放清单是目前计算城市碳排放量最常用的机理，并有助于在不同的产业或领域研究温室气体排放现状）
ID?Attention	~~????~~?????? ????? ??? ???? ?????????~~????~~???????????? ???? ???? ??~~???~~????? ??? ??
HS?Attention	???? ?? ???? ? \| ?? ?? ??? ???? \| ????? ?? ?? ???? ???? \| ?? ?? ?? ?? ???? \| ???? ?? ??? ????? \| ??? ??