基于胶囊网络的方面级情感分类研究

doi:10.11959/j.issn.2096-6652.202031

智能科学与技术学报 ›› 2020, Vol. 2 ›› Issue (3): 284-292.doi: 10.11959/j.issn.2096-6652.202031

基于胶囊网络的方面级情感分类研究

徐志栋¹,陈炳阳²,王晓^3,⁴,张卫山²()

¹ 中国人民解放军国防大学国家安全学院，北京100091
² 中国石油大学（华东）计算机科学与技术学院，山东青岛 266580
³ 青岛智能产业技术研究院，山东青岛 266109
⁴ 中国科学院自动化研究所复杂系统管理与控制国家重点实验室，北京100190

修回日期:2020-09-11 出版日期:2020-09-20 发布日期:2020-10-20
作者简介:徐志栋（1981- ），男，博士，中国人民解放军国防大学国家安全学院研究员，主要研究方向为自然语言处理、社会计算|陈炳阳（1993- ），男，中国石油大学（华东）计算机科学与技术学院博士生，主要研究方向为数据挖掘、自然语言处理、情感识别|王晓（1987- ），女，博士，青岛智能产业技术研究院常务副院长，中国科学院自动化研究所复杂系统管理与控制国家重点实验室副研究员，主要研究方向为社会计算、智能交通与车联网、人工智能等|张卫山（1970- ），男，博士，中国石油大学（华东）计算机科学与技术学院教授、博士生导师，黄岛区“智能大数据处理创新人才团队”负责人，中国石油大学“石油大数据处理”团队负责人，主要研究方向为大数据智能处理、人工智能
基金资助:
国家重点研发计划基金资助项目(2018YFE0116700);山东省自然科学基金资助项目(ZR2019MF049)

Research on capsule network-based for aspect-level sentiment classification

Zhidong XU¹,Bingyang CHEN²,Xiao WANG^3,⁴,Weishan ZHANG²()

¹ School of Security,China People’s Liberation Army National Defence University,Beijing 100091,China
² College of Computer Science and Technology,China University of Petroleum,Qingdao 266580,China
³ Qingdao Academy of Intelligent Industries,Qingdao 266109,China
⁴ The State Key Laboratory for Management and Control of Complex Systems,Institute of Automation,Chinese Academy of Sciences,Beijing 100190,China

Revised:2020-09-11 Online:2020-09-20 Published:2020-10-20
Supported by:
The National Key Research and Development Program of China(2018YFE0116700);The Natural Science Foundation of Shandong Province(ZR2019MF049)

摘要/Abstract

摘要：

由于文本中多种情感极性混合而难以判断，方面级情感分析成为当前研究的热点。考虑到多面句表达时会在一定程度上对不同目标的多重情感造成特征重叠，进而影响文本情感分类效果，提出一种基于胶囊网络的方面级情感分类模型（SCACaps）。模型使用序列卷积分别提取上下文和方面词的特征，同时引入交互注意力机制，减少二者对彼此的影响，并对文本特征表示进行重构后传入胶囊网络。胶囊层间通过引入高层胶囊系数对路由算法进行优化，整个迭代更新过程的参数全局共享，以保存较完整的文本特征信息。通过与多个模型进行对比实验发现，SCACaps的分类效果最佳，同时，在小样本学习中SCACaps也有较好的表现。

关键词: 胶囊网络, 序列卷积, 交互注意力, 方面级, 情感分类

Abstract:

Considering the difficulty of judging the mixed multiple sentimental polarities in a text,aspect-level sentiment analysis has become a hot research topic.Multiple sentiments of different targets when expressing multi-faceted in a sentence,it will cause problems such as feature overlap,which will have a negative impact on text sentiment classification.A capsule network-based model for aspect-level sentiment classification (SCACaps) was proposed.Sequential convolution was used to extract the features of context and aspect words separately,and an interactive attention mechanism was introduced to reduce the mutual influence on each other,and then the text feature representation was transmitted into the capsule network after reconstruction.The routing algorithm was optimized by introducing high-level capsule coefficients between the capsule layers,and the global parameters were shared in the entire iterative update process to save relatively complete text feature information.By comparing with multiple models,the SCACaps model has the best classification effect,and the SCACaps model also performs better in small sample learning.

Key words: capsule network, sequential convolution, interactive attention, aspect-level, sentiment classification

中图分类号:

TP301

徐志栋, 陈炳阳, 王晓, 等. 基于胶囊网络的方面级情感分类研究[J]. 智能科学与技术学报, 2020, 2(3): 284-292.

Zhidong XU, Bingyang CHEN, Xiao WANG, et al. Research on capsule network-based for aspect-level sentiment classification[J]. Chinese Journal of Intelligent Science and Technology, 2020, 2(3): 284-292.

图/表 11

图1

图2

图3

图4

图5

表1

表2

表3

胶囊网络参数"

胶囊网络	可调参数	参数值
初始胶囊	胶囊维度	8
	层数	1
	参数共享	是
高层胶囊	胶囊维度	16
	胶囊个数	2
	迭代次数	3
	层数	1
	参数共享	是
	激活函数	$\frac{{‖ s_{j} ‖}^{2} + ε}{0.5 + {‖ s_{j} ‖}^{2}}$

表3

表4

表5

表6

参考文献 22

[1]	赵传君, 王素格, 李德玉 . 跨领域文本情感分类研究进展[J]. 软件学报, 2020,31(6): 1723-1746.
	ZHAO C J , WANG S G , LI D Y . Research progress on cross-domain text sentiment classification[J]. Journal of Software, 2020,31(6): 1723-1746.
[2]	LIU B . Sentiment analysis and opinion mining[J]. Synthesis lectures on Human Language Technologies, 2012,5(1): 1-167.
[3]	AUGENSTEIN I , ROCKT?SCHEL T , VLACHOS A ,et al. Stance detection with bidirectional conditional encoding[C]// The 2016 Conference on Empirical Methods in Natural Language Processing. Stroudsburg:ACL Press, 2016: 876-885.
[4]	WANG Y Q , HUANG M L , ZHU X Y ,et al. Attention-based LSTM for aspect-level sentiment classification[C]// The 2016 Conference on Empirical Methods in Natural Language Processing. Stroudsburg:ACL Press, 2016: 606-615.
[5]	MA D H , LI S J , ZHANG X D ,et al. Interactive attention networks for aspect-level sentiment classification[C]// The 26th International Joint Conference on Artificial Intelligence. CA:AAAI, 2017: 4068-4074.
[6]	陈虹, 杨燕, 杜圣东 . 用户评论方面级情感分析研究[J]. 计算机科学与探索, 2020:已录用.
	CHEN H , YANG Y , DU S D . Research on aspect-level sentiment analysis of user reviews[J]. Journal of Frontiers of Computer Science and Technology, 2020:accepted.
[7]	KIM Y . Convolutional neural networks for sentence classification[J]. arXiv preprint, 2014,arXiv:1408.5882.
[8]	XUE W , LI T . Aspect based sentiment analysis with gated convolutional networks[C]// The 56th Annual Meeting of the Association for Computational Linguistics. Stroudsburg:ACL Press, 2018: 2514-2523.
[9]	韩建胜, 陈杰, 陈鹏 ,等. 基于双向时间深度卷积网络的中文文本情感分类[J]. 计算机应用与软件, 2019,36(12): 225-231.
	HAN J S , CHEN J , CHEN P ,et al. Chinese text sentiment classification based on bidirectional temporal deep convolutional network[J]. Computer Applications and Software, 2019,36(12): 225-231.
[10]	高巍, 孙盼盼, 李大舟 ,等. 基于序列卷积网络的Twitter情感分类[J]. 计算机工程与设计, 2020,41(5): 1314-1319.
	GAO W , SUN P P , LI D Z ,et al. Twitter sentiment classification based on sequential convolution network[J]. Computer Engineering and Design, 2020,41(5): 1314-1319.
[11]	FAN F F , FENG Y S , ZHAO D Y . Multi-grained attention network for aspect-level sentiment classification[C]// The 2018 Conference on Empirical Methods in Natural Language Processing.[S.l.:s.n]. 2018.
[12]	曾锋, 曾碧卿, 韩旭丽 ,等. 基于双层注意力循环神经网络的方面级情感分析[J]. 中文信息学报, 2019,33(6): 108-115.
	ZENG F , ZENG B Q , HAN X L ,et al. Double attention neural network for aspect-based sentiment analysis[J]. Journal of Chinese Information Processing, 2019,33(6): 108-115.
[13]	曾义夫, 蓝天, 吴祖峰 ,等. 基于双记忆注意力的方面级别情感分类模型[J]. 计算机学报, 2019,42(8): 1845-1857.
	ZENG Y F , LAN T , WU Z F ,et al. Bi-memory based attention model for aspect level sentiment classification[J]. Chinese Journal of Computers, 2019,42(8): 1845-1857.
[14]	HINTON G E , KRIZHEVSKY A , WANG S D . Transforming auto-encoders[C]// International Conference on Artificial Neural Networks. Heidelberg:Springer, 2012: 44-51.
[15]	HINTON G E , SABOUR S , FROSST N . Matrix capsules with EM routing[C]// International Conference on Learning Representations.[S.l.:s.n]. 2018.
[16]	CHEN Z , QIAN T Y . Transfer capsule network for aspect level sentiment classification[C]// The 57th Annual Meeting of the Association for Computational Linguistics. Stroudsburg:ACL Press, 2019: 547-556.
[17]	XIAO L Q , ZHANG H L , CHEN W Q ,et al. Mcapsnet:capsule network for text with multi-task learning[C]// The 2018 Conference on Empirical Methods in Natural Language Processing.[S.l.:s.n]. 2018.
[18]	ZHAO W , YE J B , YANG M ,et al. Investigating capsule networks with dynamic routing for text classification[C]// The 2018 Conference on Empirical Methods in Natural Language Processing. Stroudsburg:ACL Press, 2018: 3110-3119.
[19]	杨云龙, 孙建强, 宋国超 . 基于 GRU 和胶囊特征融合的文本情感分析[J]. 计算机应用, 2020:已录用.
	YANG Y L , SUN J Q , SONG G C . Text sentiment analysis based on feature fusion of gated recurrent unit and capsule[J]. Journal of Computer Applications, 2020:accepted.
[20]	王敏蕊, 高曙, 袁自勇 ,等. 基于动态路由序列生成模型的多标签文本分类方法[J]. 计算机应用, 2020:已录用.
	WANG M R , GAO S , YUAN Z Y ,et al. Sequence to sequence model with dynamic routing for multilabel text classification[J]. Journal of Computer Applications, 2020:accepted.
[21]	林悦, 钱铁云 . 基于胶囊网络的跨领域情感分类方法[J]. 南京信息工程大学学报, 2019,11(3): 286-294.
	LIN Y , QIAN T Y . Cross-domain sentiment classification by capsule network[J]. Journal of Nanjing University of Information Science ＆Technology, 2019,11(3): 286-294.
[22]	ZHANG B W , XU X F , YANG M ,et al. Cross-domain sentiment classification by capsule network with semantic rules[J]. IEEE Access, 2018,6: 58284-58294.

数据集	积极情感		中性情感		消极情感
数据集	训练集/个	测试集/个	训练集/个	测试集/个	训练集/个	测试集/个
Restaurant	2 164	728	637	196	807	196
Laptop	994	341	464	169	870	128

参数	参数值
词向量维度	300
因果空洞卷积层数	5
卷积核大小	5
空洞因子	2n
残差模块数	4
激活函数	ReLU
优化函数	Adam
dropout	0.2

模型	Restaurant	Laptop
BiLSTM	77.13%	68.76%
Seq-CNN	77.19%	68.78%
ATAE-LSTM	78.48%	69.87%
IAN	78.61%	72.18%
MGAN	81.24%	75.38%
Bi-IAN	79.38%	72.24%
SCACaps	81.64%	76.26%

模型	Restaurant	Laptop
BiLSTM	63.74%	62.42%
Seq-CNN	63.79%	62.48%
ATAE-LSTM	63.38%	62.37%
IAN	68.13%	65.82%
MGAN	71.87%	72.38%
Bi-IAN	69.94%	66.74%
SCACaps	72.34%	72.43%

模型	accuracy	F1分数
BiLSTM	63.76%	—
Seq-CNN	64.03%	—
ATAE-LSTM	65.42%	—
IAN	65.57%	—
MGAN	71.84%	65.27%
Bi-IAN	66.32%	62.69%
SCACaps	73.28%	67.14%

基于胶囊网络的方面级情感分类研究

Research on capsule network-based for aspect-level sentiment classification

在线阅读

PDF下载

可视化

摘要/Abstract

引用本文

使用本文

图/表 11

参考文献 22

相关文章 11

Metrics

推荐阅读 0

[1]	王晓, 杨林瑶, 胡斌, 侯家琛. 平行推理：一种基于ACP方法的虚实互动的知识协同框架[J]. 智能科学与技术学报, 2023, 5(1): 69-82.
[2]	张欣, 詹志辉. 智能优化算法在供应链网络中的应用[J]. 智能科学与技术学报, 2022, 4(2): 174-183.
[3]	李亚玲, 杨林瑶, 葛俊, 覃缘琪, 王晓. 博弈5.0：基于平行系统和机器博弈的社会认知平行博弈[J]. 智能科学与技术学报, 2021, 3(4): 507-520.
[4]	秦蕊, 李娟娟, 王晓, 朱静, 袁勇, 王飞跃. NFT：基于区块链的非同质化通证及其应用[J]. 智能科学与技术学报, 2021, 3(2): 234-242.
[5]	王飞跃, 王艳芬, 陈薏竹, 田永林, 齐红威, 王晓, 张卫山, 张俊, 袁勇. 联邦生态：从联邦数据到联邦智能[J]. 智能科学与技术学报, 2020, 2(4): 305-313.
[6]	李涛, 魏庆来. 基于深度强化学习的智能暖气温度控制系统[J]. 智能科学与技术学报, 2020, 2(4): 348-353.
[7]	刘莹莹, 王占山. 异构多智能体系统的输出同步：一个基于数据的强化学习方法[J]. 智能科学与技术学报, 2020, 2(4): 394-400.
[8]	王飞跃,曹东璞,魏庆来. 强化学习：迈向知行合一的智能机制与算法[J]. 智能科学与技术学报, 2020, 2(2): 101-106.
[9]	师强,杨明. 基于能量函数的智能车辆路径曲率平滑方法研究[J]. 智能科学与技术学报, 2020, 2(2): 161-168.
[10]	刘远平,宋昱锴,张小燕,刘贤强. 基于深度学习的髋关节应力分布算法研究[J]. 智能科学与技术学报, 2019, 1(3): 260-268.
[11]	陈妮亚,阮佳阳,黄金苗,杨伟. 结合深度学习与生物特征识别在冷链拣选中的算法研究[J]. 智能科学与技术学报, 2019, 1(1): 88-95.