基于分层注意力网络的方面情感分析

doi:10.11959/j.issn.2096-0271.2020045

摘要/Abstract

摘要：

基于深度学习的方面情感分析是自然语言处理的热点之一。针对方面情感，提出基于方面情感分析的深度分层注意力网络模型。该模型通过区域卷积神经网络保留文本局部特征和不同句子时序关系，利用改进的分层长短期记忆网络（LSTM）获取句子内部和句子间的情感特征。其中，针对LSTM添加了特定方面信息，并设计了一个动态控制链，改进了传统的LSTM。在SemEval 2014的两个数据集和Twitter数据集上进行对比实验得出，相比传统模型，提出的模型的情感分类准确率提高了3%左右。

关键词: 深度学习, 方面情感, 区域卷积神经网络, 分层长短期记忆网络, 注意力机制, 动态控制链

Abstract:

Aspect sentiment analysis based on deep learning is one of the hot spots in natural language processing.Aiming at aspect sentiment,a deep hierarchical attention network model based on aspect sentiment analysis was proposed.The local features of the text and the temporal relationship of different sentences were retained in model through the convolutional neural network,and the emotional features within and between sentences were obtained by using the layered long shortterm memory network (LSTM).Among them,specific aspects of information were added to LSTM and a dynamic control chain was designed to improve the traditional LSTM.A comparative experiment is conducted on the two data sets in SemEval 2014 and the Twitter data set.Compared with the traditional model,the accuracy of sentiment classification of the proposed model increases by about 3%.

Key words: deep learning, aspect sentiment, regional convolutional neural network, layered long short-term memory, attention mechanism, dynamic control chain

中图分类号:

TP391

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.

Ting SONG, Zhanwei CHEN, Haifeng YANG. Aspect sentiment analysis based on a hierarchical attention network[J]. Big Data Research, 2020, 6(5): 82-91.

图/表 6

图1

图2

表1

表2

图3

图4

参考文献 22

[1]	PANG B , LEE L . Opinion mining and sentiment analysis[J]. Foundations and Trends in Information Retrieval, 2008,2(1-2): 1-135.
[2]	KIRITCHENKO S , ZHU X D , CHERRY C ,et al. NRC-Canada-2014:detecting aspects and sentiment in customer reviews[C]// The 8th International Workshop on Semantic Evaluation.[S.l.:s.n]. 2014: 437-442.
[3]	SU J S , ZENG J L , XIONG D Y ,et al. A hierarchy-to-sequence attentional neural machine translation model[J]. IEEE/ACM Transactions on Audio Speech ＆Language Processing, 2018(99): 1-1.
[4]	YIN W P , SCHUTZE H , XIANG B ,et al. Attention-based convolutional neural network for modeling sentence pairs[J]. Transaction of the Association for Computational Linguistics, 2016,4(11): 259-272.
[5]	AL-RIFAIE M M , BISHOP J M . Swarmic sketches and attention mechanism[C]// International Conference on Evolutionary and Biologically Inspired Music and Art. Heidelberg:Springer, 2013.
[6]	LUMER E D . Selective attention to perceptual groups:the phase tracking mechanism[J]. International Journal of Neural Systems, 1992,3(1): 1-17.
[7]	Z H ANG Z , ROBINSON D , TEPPER J . Detecting hate speech on Twitter using a convolution-GRU based deep neural network[C]// ESWC 2018. Cham:Springer, 2018.
[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, 2018: 2514-2523.
[9]	XIAO R , CUI X H , ZHOU P P ,et al. LSTM based on the classification of emotion about user evaluation on shopping site[C]// 2016 International Conference on Identification,Information and Knowledge in the Internet of things. Piscataway:IEEE Press, 2016.
[10]	梁斌, 刘全, 徐进 ,等. 基于多注意力卷积神经网络的特定目标情感分析[J]. 计算机研究与发展, 2017,54(8): 1724-1735.
	LIANG B , LIU Q , XU J ,et al. Emotional score of specific target based on multi-attention convolutional neural network[J]. Computer Research and Development, 2017,54(8): 1724-1735.
[11]	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.[S.l.:s.n]. 2016: 606-615
[12]	ZHU Y H , GAO X , ZHANG W L ,et al. A bi-directional LSTM-CNN model with attention for aspect-level text classification[J]. Future Internet, 2018,10(12): 116-127.
[13]	MA Y K , PENG H Y , CAMBRIA E . Targeted aspect-based sentiment analysis via embedding commonsense knowledge into an attentive LSTM[C]// The 32nd AAAI Conference on Artificial Intelligence. Menlo Park:AAAI Press, 2018: 2576-5883.
[14]	TANG D U , QIN B , FENG X C ,et al. Effective LSTMs for target dependent sentiment classification[C]// The 26th International Conference on Computational Linguistics.[S.l.:s.n]. 2016: 3298-3307.
[15]	VASWANI A , SHAZEER N , PARMAR N ,et al. Attention is all you need[C]// The 31st Conference on Neural Information Processing Systems. Cambridge:MIT Press, 2017: 6000-6010.
[16]	RUDER S , GHAFFARI P , BRESLIN J G . A hierarchical model of reviews for aspectbased sentiment analysis[J]. arXiv preprint, 2016,arXiv:1609.02745.
[17]	LI D , WEI F R , TAN C Q ,et al. Adaptive recursive neural network for target-dependent Twitter sentiment classification[C]// The 52nd Annual Meeting of the Association for Computational Linguistics.[S.l:s.n]. 2014: 49-54.
[18]	KIM Y , . Convolutional neural networks for sentence classification[C]// The 2014 Conference on Empirical Methods in Natural Language Processing.[S.l.:s.n]. 2014: 1746-1751.
[19]	LIU J M , ZHANG Y . Attention modeling for targeted sentiment[C]// The 15th Conference of the European Chapter of the Association for Computational Linguistics.[S.l.:s.n]. 2017: 572-577.
[20]	WANG X , LIU Y C , SUN C J ,et al. Predicting polarities of Tweets by composing word embeddings with long short-term memory[C]// Meeting of the Association for Computational Linguistics＆ the International Joint Conference on Natural Language Processing.[S.l.:s.n]. 2015: 1343-1353.
[21]	MA D H , LI S J , ZHANG X D ,et al. Interactive attention networks for aspectlevel sentiment classification[C]// The 26th International Joint Conference on Artificial Intelligence.[S.l.:s.n]. 2017.
[22]	TAY Y , LUU A T , HUI S C . Learning to attend via word-aspect associative fusion for aspect-based sentiment analysis[C]// National Conference on Artificial Intelligence.[S.l.:s.n]. 2018: 5956-5963.

情感极性	Restaurant		Laptop		Twitter
情感极性	训练集/个	测试集/个	训练集/个	测试集/个	训练集/个	测试集/个
积极	2 164	728	994	341	1 561	173
消极	807	196	870	128	1 560	173
中立	637	196	464	169	3 127	346
总计	3 608	1 120	2 328	638	6 248	692

模型名称		数据集
模型名称	Restaurant	Laptop	Twitter
CNN	67.23%	59.32%	70.03%
LSTM	74.30%	66.50%	66.50%
ATT-CNN	68.19%	62.11%	72.18%
ATAE-LSTM	78.50%	69.27%	69.88%
LSTM-R	75.98%	72.59%	—
IAN	78.60%	72.10%	—
AF-LSTM	75.44%	68.81%	—
SVM	80.00%	70.49%	—
HANMAS	79.85%	75.44%	76.39%