基于多输入模型及句法结构的中文评论情感分析方法

doi:10.11959/j.issn.2096-0271.2021059

大数据 ›› 2021, Vol. 7 ›› Issue (6): 41-52.doi: 10.11959/j.issn.2096-0271.2021059

• 专题：大数据支撑的智能应用 • 上一篇下一篇

基于多输入模型及句法结构的中文评论情感分析方法

张宝华¹, 张华平¹, 厉铁帅², 商建云¹

¹ 北京理工大学计算机学院，北京 100081
² 中央军事委员会政法委员会，北京 100120

出版日期:2021-11-15 发布日期:2021-11-01
作者简介:张宝华（1996- ），男，北京理工大学计算机学院硕士生，主要研究方向为自然语言处理、情感分析
张华平（1978- ），男，博士，北京理工大学计算机学院副研究员，主要研究方向为大数据搜索与挖掘、自然语言处理、社交网络
厉铁帅（1975- ），男，中央军事委员会政法委员会高级工程师，主要研究方向为计算机应用
商建云（1965- ），女，博士，北京理工大学计算机学院高级工程师，主要研究方向为自然语言处理、数据挖掘
基金资助:
国家自然科学基金资助项目(61772075);北京市自然科学基金资助项目(4212026)

Chinese comment sentiment analysis method based on multi-input model and syntactic structure

Baohua ZHANG¹, Huaping ZHANG¹, Tieshuai LI², Jianyun SHANG¹

¹ School of Computer Science ＆Technology, Beijing Institute of Technology, Beijing 100081, China
² Politics and Law Commission of Central Military Commission of the People’s Republic of China, Beijing 100120, China

Online:2021-11-15 Published:2021-11-01
Supported by:
The National Natural Science Foundation of China(61772075);Beijing Municipal Natural Science Foundation(4212026)

摘要/Abstract

摘要：

海量的网络文本给情感分析任务带来了巨大的机遇和挑战，传统基于规则的方法已经很难胜任这类文本的分析工作，现有的深度学习方法存在一些不足，一方面模型的输入只包括文本嵌入矩阵，缺乏其他特征的使用；另一方面，词嵌入算法会导致文本结构信息缺失，进而影响分析效果。在对基于规则的情感分析方法中的句法规则进行研究的基础上，提出了一种结合MCNN、LSTM和全连接神经网络的多输入模型。同时在深度学习模型中构建了句法特征提取器来提取句法特征。在3个公开数据集上进行了实验，结果表明，构建的模型较其他模型拥有更好的分类性能，且句法规则特征的引入对模型的分类效果有一定的提升。

关键词: 情感分析, 句法规则, 多输入模型

Abstract:

Massive network texts have brought huge opportunities and challenges to sentiment analysis tasks.Traditional rule-based methods have been difficult to analyze such texts.Existing deep learning methods have some shortcomings.On the one hand, the inputs of the model only include the text embedding matrix, lack the use of other features.On the other hand, the algorithm of word embedding will lead to the lack of text structure information, then impact the result.Based on the research of syntactic rule in the rule-based sentiment analysis methods, a multi-input model combined with MCNN, LSTM and fully connected neural network was proposed.Meanwhile, a syntactic feature extractor to combine the syntactic features was constructed in the deep learning model.Experiments on three public data sets were conducted.The results show that the model constructed in this article has better classification performance than other models, and the introduction of syntactic rule features has a little improvement in the classification effect of the model.

Key words: sentiment analysis, syntactic rule, multi-input model

中图分类号:

TP183

张宝华, 张华平, 厉铁帅, 商建云. 基于多输入模型及句法结构的中文评论情感分析方法[J]. 大数据, 2021, 7(6): 41-52.

Baohua ZHANG, Huaping ZHANG, Tieshuai LI, Jianyun SHANG. Chinese comment sentiment analysis method based on multi-input model and syntactic structure[J]. Big Data Research, 2021, 7(6): 41-52.

图/表 6

图1

图2

图3

表1

表2

表3

参考文献 23

[1]	COLLOBERT R , WESTON J , BOTTOU L ,et al. Natural language processing (almost) from scratch[J]. Journal of Machine Learning Research, 2011,12: 2493-2537.
[2]	KIM Y . Convolutional neural networks for sentence classification[J]. arXiv preprint,2014,arXiv:1408.5882.
[3]	KALCHBRENNER N , GREFENSTETTE E , BLUNSOM P . A convolutional neural network for modelling sentences[J]. arXiv preprint,2014,arXiv:1404.2188.
[4]	ZHANG Y , WALLACE B . A sensitivity analysis of (and practitioners’ guide to) convolutional neural networks for sentence classification[J]. arXiv preprint,2015,arXiv:1510.03820.
[5]	GAO J , PANTEL P , GAMON M ,et al. Modeling interestingness with deep neural networks[C]// Proceedings of the 2014 Conference on Empirical Methods in Natural Language Processing.[S.l.:s.n.], 2014: 2-13.
[6]	SHEN Y L , HE X D , GAO J F ,et al. A latent semantic model with convolutional-pooling structure for information retrieval[C]// Proceedings of the 23rd ACM International Conference on Conference on Information and Knowledge Management. New York:ACM Press, 2014: 101-110.
[7]	ZHANG R , LEE H , RADEV D R . Dependency sensitive convolutional neural networks for modeling sentences and documents[J]. arXiv preprint,2016,arXiv:1611.02361.
[8]	CHO K , VAN MERRI?NBOER B , GULCEHRE C ,et al. Learning phrase representations using RNN encoder-decoder for statistical machine translation[J]. arXiv preprint,2014,arXiv:1406.1078.
[9]	SOCHER R , PERELYGIN A , WU J Y ,et al. Recursive deep models for semantic compositionality over a sentiment treebank[C]// Proceedings of the 2013 Conference on Empirical Methods in Natural Language Processing.[S.l.:s.n.], 2013: 1631-1642.
[10]	TRAN K , BISAZZA A , MONZ C . Recurrent memory network for language modeling[J]. arXiv preprint,2016,arXiv:1601.01272.
[11]	CHEN P , SUN Z Q , BING L D ,et al. Recurrent attention network on memory for aspect sentiment analysis[C]// Proceedings of the 2017 Conference on Empirical Methods in Natural Language Processing. Stroudsburg:Association for Computational Linguistics, 2017: 452-461.
[12]	宋婷, 陈战伟, 杨海峰 . 基于分层注意力网络的方面情感分析[J]. 大数据, 2020,6(5): 82-91.
	SONG T , CHEN Z W , YANG H F . Aspect sentiment analysis based on a hierarchical attention network[J]. Big Data Research, 2020,6(5): 82-91.
[13]	WANG Y Q , HUANG M L , ZHU X Y ,et al. Attention-based LSTM for aspect-level sentiment classification[C]// Proceedings of the 2016 Conference on Empirical Methods in Natural Language Processing. Stroudsburg:Association for Computational Linguistics, 2016: 606-615.
[14]	CHENG Y , YAO L B , XIANG G X ,et al. Text sentiment orientation analysis based on multi-channel CNN and bidirectional GRU with attention mechanism[J]. IEEE Access, 2020,8: 134964-134975.
[15]	YANG L , LI Y , WANG J ,et al. Sentiment analysis for e-commerce product reviews in Chinese based on sentiment lexicon and deep learning[J]. IEEE Access, 2020,8: 23522-23530.
[16]	LI W , ZHU L Y , SHI Y ,et al. User reviews:sentiment analysis using lexicon integrated two-channel CNN-LSTM family models[J]. Applied Soft Computing, 2020,94: 106435
[17]	LI W J , QI F , TANG M ,et al. Bidirectional LSTM with self-attention mechanism and multi-channel features for sentiment classification[J]. Neurocomputing, 2020,387: 63-77.
[18]	USAMA M , AHMAD B , SONG E M ,et al. Attention-based sentiment analysis using convolutional and recurrent neural network[J]. Future Generation Computer Systems, 2020,113: 571-578.
[19]	BASIRI M E , NEMATI S , ABDAR M ,et al. ABCDM:an attention-based bidirectional CNN-RNN deep model for sentiment analysis[J]. Future Generation Computer Systems, 2021,115: 279-294.
[20]	JIN N , WU J X , MA X ,et al. Multi-task learning model based on multi-scale CNN and LSTM for sentiment classification[J]. IEEE Access, 2020,8: 77060-77072.
[21]	BASIRI M E , NEMATI S , ABDAR M ,et al. ABCDM:an attention-based bidirectional CNN-RNN deep model for sentiment analysis[J]. Future Generation Computer Systems, 2021,115: 279-294.
[22]	USAMA M , AHMAD B , SONG E M ,et al. Attention-based sentiment analysis using convolutional and recurrent neural network[J]. Future Generation Computer Systems, 2020,113: 571-578.
[23]	张宝华, 李奀林, 张华平 ,等. 基于层次结构的情感单元表示方法[J]. 计算机工程与科学, 2021:已录用.
	ZHANG B H , LI E H , ZHANG H P ,et al. Representation of sentiment unit based on hierarchical structure[J]. Computer Engineering and Science, 2021:accepted.

数据集	总评论数/个	句子最大长度/字符	句子最小长度/字符	平均包含单句数量/个	包含关联词的句子数量/个
Douban	100 000	99	11	4.32	56 998
Hotel	10 000	1 985	4	8.74	8 276
NLPCC2014	10 000	1 004	3	5.48	6 419

模型	Hotel	Douban	NLPCC2014
MCNN_S_LSTM_NN	91.72%	86.42%	76.65%
MCNN_LSMT_NN	91.59%	84.55%	75.49%
MCNN_S_LSTM	91.45%	84.61%	75.21%
MCNN_LSTM	91.01%	84.29%	74.20%
MCNN_NN	90.54%	84.52%	72.54%
S_LSTM_NN	90.51%	83.94%	72.90%
MCNN	90.35%	83.55%	71.25%
S_LSTM	89.90%	83.70%	73.50%
LSTM	88.40%	83.64%	71.55%

模型	Hotel				Douban				NLPCC2014
模型	P	R	F1	ACC	P	R	F1	ACC	P	R	F1	ACC
CNN	88.18%	89.40%	88.78%	88.61%	82.61%	86.17%	84.36%	83.94%	78.32%	67.67%	72.57%	73.56%
LSTM	87.79%	89.70%	88.74%	88.40%	86.01%	80.54%	83.23%	83.64%	75.92%	68.45%	72.00%	72.65%
MCNNALSTM	90.61%	90.79%	90.70%	90.52%	85.38%	84.16%	84.77%	84.16%	74.36%	75.50%	74.93%	73.80%
SLCABG	91.35%	88.91%	90.12%	90.15%	86.13%	83.89%	85.00%	85.11%	79.08%	71.46%	75.08%	75.40%
ATTConv-RNN rand	92.54%	88.42%	90.43%	90.55%	86.19%	84.85%	85.51%	85.55%	81.70%	67.60%	73.98%	75.35%
ABCDM	89.48%	92.67%	91.05%	90.80%	84.20%	86.91%	85.53%	85.22%	78.59%	73.96%	76.20%	76.05%
SAMF-BiLSTM	93.25%	87.52%	90.30%	90.50%	84.72%	85.49%	85.10%	84.96%	74.82%	78.20%	76.47%	75.05%
CNN-BiLSTM	92.32%	88.12%	90.17%	90.30%	86.13%	83.89%	84.77%	84.79%	78.32%	71.07%	74.52%	74.80%
MC-AttCNN-AttBiGRU	91.35%	90.99%	91.17%	91.10%	85.53%	86.80%	86.16%	85.99%	75.99%	78.11%	77.03%	75.85%
MCNN_S_LSTM_NN	90.74%	93.08%	91.87%	91.70%	88.27%	84.17%	86.17%	86.42%	77.56%	77.34%	77.45%	76.65%

基于多输入模型及句法结构的中文评论情感分析方法

Chinese comment sentiment analysis method based on multi-input model and syntactic structure

在线阅读

PDF下载

可视化

摘要/Abstract

引用本文

使用本文

图/表 6

参考文献 23

相关文章 6

Metrics

推荐阅读 0

[1]	张伶俐, 褚琦凯, 王桂娟, 张巍瀚, 蒲慧, 宋振金, 吴亚东. 文本情感可视分析技术及其在人文领域的应用[J]. 大数据, 2022, 8(6): 56-73.
[2]	李爱黎, 张子帅, 林荫, 王秋菊, 杨建安, 孟炜程, 张岩峰. 基于社交网络大数据的民众情感监测研究[J]. 大数据, 2022, 8(6): 105-126.
[3]	徐康庭, 宋威. 结合语言知识和深度学习的中文文本情感分析方法[J]. 大数据, 2022, 8(3): 115-127.
[4]	宋云生. 一种情感判别分析体系在汽车品牌舆情管理中的应用[J]. 大数据, 2017, 3(6): 55-64.
[5]	范珈瑜. 基于文本挖掘的游客对古镇旅游态度的分析[J]. 大数据, 2017, 3(6): 93-101.
[6]	郭鹏. 收益管理中基于大数据仓库的需求无约束估计：框架与挑战[J]. 大数据, 2017, 3(3): 68-83.