用于文本情感极性分析的动态卷积神经网络超限学习算法

贾熹滨; 李宁; 靳亚

doi:10.11936/bjutxb2016040093

用于文本情感极性分析的动态卷积神经网络超限学习算法

doi: 10.11936/bjutxb2016040093

北京工业大学计算机学院, 北京 100124

基金项目: 国家自然科学基金资助项目(61175115,61370113)

详细信息

作者简介:
作者简介: 贾熹滨(1969—), 女, 副教授, 主要从事视觉图像理解方面的研究, E-mail:jiaxibin@bjut.edu.cn

中图分类号: TP391
计量
- 文章访问数: 97
- HTML全文浏览量: 52
- PDF下载量: 0
- 被引次数: 0
出版历程
- 收稿日期: 2016-04-27
- 网络出版日期: 2022-09-09
- 刊出日期: 2017-01-01

Dynamic Convolutional Neural Network Extreme Learning Machine for Text Sentiment Classification

College of Computer Science, Beijing University of Technology, Beijing 100124, China

摘要

摘要: 为改善动态卷积神经网络在文本情感分类方法中的泛化能力,提出了一种动态卷积超限学习算法. 对动态卷积神经网络的输出层加以改进,使用浅层随机神经网络替代全连接层,利用参数随机生成的扰动性能,降低分类端对训练样本的依赖,避免过拟合,提升分类性能. 在公共数据集上的实验证明:相对改进前的动态卷积学习算法以及超限学习机,所提出的方法在准确率、F₁测度等多个标准指标上均体现了更优的分类性能.
- 文本情感分类 /
- 动态卷积 /
- 超限学习机 /
- 动态卷积超限学习算法
Abstract:
Aim ing at improving the generalization performance of the dynamic convolutional neural network on text sentiment classification, a dynamic convolutional extreme learning machine algorithm was proposed. This algorithm modified the output layer of dynamic convolutional neural network by replacing the fully connection layer with the shallow random neural network. By utilizing the perturbation ability of the random generation of parameters, it is prone to mitigate the dependence on training samples and avoid over-fitting to improve the classification performance. Experiments on several public data sets show that this approach outperforms the dynamic convolutional neural network and extreme learning machine under the evaluation metrics including accuracy rate, F₁-measure, etc.
- text sentiment classification /
- dynamic convolutional /
- extreme learning machine /
- dynamic convolutional extreme learning machine
The authors have declared that no competing interests exist.

HTML全文

图 1 以词数为7的输入语句为样例的CNN结构

Figure 1. CNN architecture for the seven word input sentences

下载: 全尺寸图片幻灯片

图 2 以词数为7的输入语句为样例的DCNN结构

Figure 2. DCNN architecture for the seven word input sentences

下载: 全尺寸图片幻灯片

图 3 ELM算法结构

Figure 3. Architecture of ELM

下载: 全尺寸图片幻灯片

图 4 以词数为7的输入语句为样例的DCELM结构图

Figure 4. DCELM architecture for the seven word input sentences

下载: 全尺寸图片幻灯片

图 5 4种算法在不同数据集上的精确率

Figure 5. Precision of four algorithms on different data sets

下载: 全尺寸图片幻灯片

图 6 4种算法在不同数据集上的召回率

Figure 6. Recall of four algorithms on different data sets

下载: 全尺寸图片幻灯片

图 7 4种算法在不同数据集上的F₁测度

Figure 7. F₁-measure of four algorithms on different data sets

下载: 全尺寸图片幻灯片

图 8 ELM-f的准确率随隐层节点改变的变化情况

Figure 8. Accuracy of ELM-f with the change of hidden nodes

下载: 全尺寸图片幻灯片

图 9 ELM-v的准确率随隐层节点改变的变化情况

Figure 9. Accuracy of ELM-v with the change of hidden nodes

下载: 全尺寸图片幻灯片

图 10 DCELM的准确率随隐层节点改变的变化情况

Figure 10. Accuracy of DCELM with the change of hidden nodes

下载: 全尺寸图片幻灯片

表 1 3种数据集概述

Table 1. Description of three data sets

数据集	l	N	\|V\|	Test
ChnSentiCorp	33	3009	4870	500
MioChnCorp-2	26	120000	13560	20000
SST-2	19	9613	16185	1821

下载: 导出CSV

表 2 4种算法在不同数据集上的准确率

Table 2. Accuracy on four data sets%

数据集	ELM-f	ELM-v	DCNN	DCELM
ChnSentiCorp	83.80	75.00	89.00	90.00
MioChnCorp-2	87.34	81.54	90.05	91.49
SST-2	77.27	62.84	81.71	82.70

下载: 导出CSV

参考文献(17)

[1]	DAVE K, LAWRENCE S, PENNOCK D M.Mining the peanut gallery: opinion extraction and semantic classification of product reviews[C]//Proceedings of the 12th international conference on World Wide Web. New York: ACM, 2003: 519-528.
[2]	KIM S M, HOVY E.Determining the sentiment of opinions[C]//Proceedings of the 20th international conference on Computational Linguistics. Stroudsburg: ACL, 2004: 1367-1373
[3]	YANG D, YANG A M.Classification approach of Chinese texts sentiment based on semantic lexicon and naive Bayesian[J]. Application Research of Computers, 2010, 27(10): 3737-3739. (in chinese)
[4]	JI J Z, ZHANG L L, WU C S, et al.Semantic weight-based naive Bayesian algorithm for text sentiment classification[J]. Journal of Beijing University of Technology, 2014, 40(12): 1884-1890. (in chinese)
[5]	PANG B, LEE L, VAITHYANATHAN S.Thumbs up?: sentiment classification using machine learning techniques[C] ∥Proceedings of the ACL-02 Conference on Empirical Methods in Natural Language Processing. Stroudsburg: ACL, 2002: 79-86.
[6]	SOCHERRPENNINGTONJHUANG EHet al.Semi-supervised recursive autoencoders for predicting sentiment distributions Proceedings of the Conference on Empirical Methods in Natural Language Processing. Stroudsburg: ACL2011151161 SOCHER R, PENNINGTON J, HUANG E H, et al.Semi-supervised recursive autoencoders for predicting sentiment distributions[C]// Proceedings of the Conference on Empirical Methods in Natural Language Processing. Stroudsburg: ACL, 2011: 151-161.
[7]	SOCHERRPERELYGINAWUJet al.Recursive deep models for semantic compositionality over a sentiment treebank Proceedings of the 2013 Conference on Empirical Methods in Natural Language Processing. Stroudsburg: ACL201316311642 SOCHER R, PERELYGIN A, WU J, 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. Stroudsburg: ACL, 2013: 1631-1642.
[8]	JURGOVSKY J, GRANITZER M.Comparing recursive autoencoder and convolutional network for phrase-level sentiment polarity classification[C]// International Conference on Applications of Natural Language to Information Systems. Berlin: Springer, 2015: 160-166.
[9]	KALCHBRENNER N, GREFENSTETTE E, BLUNSOM P.A convolutional neural network for modelling sentences[C] ∥Proceedings of the 52nd Annual Meeting of the Association for Computational Linguistics. Baltimore: ACL, 2014: 655-665.
[10]	KIM Y.Convolutional neural networks for sentence classification[C]//Proceedings of the 2014 Conference on Empirical Methods in Natural Language Processing. Doha: ACL, 2014: 1746-1751.
[11]	HERMANN K M, BLUNSOM P.The role of syntax in vector space models of compositional semantics[C]//Proceedings of the 51st Annual Meeting of the Association for Computational Linguistics. Sofia: ACL, 2013: 894-904.
[12]	SOCHERRHUVALBMANNING CDet al.Semantic compositionality through recursive matrix-vector spaces Proceedings of the 2012 Joint Conference on Empirical Methods in Natural Language Processing and Computational Natural Language Learning. Stroudsburg: ACL201212011211 SOCHER R, HUVAL B, MANNING C D, et al.Semantic compositionality through recursive matrix-vector spaces[C]// Proceedings of the 2012 Joint Conference on Empirical Methods in Natural Language Processing and Computational Natural Language Learning. Stroudsburg: ACL, 2012: 1201-1211.
[13]	HUANG G B, ZHU Q Y, SIEW C K.Extreme learning machine: theory and applications[J]. Neurocomputing, 2006, 70(1/2/3): 489-501.
[14]	HUANG GBZHU QYSIEW CK.Extreme learning machine: a new learning scheme of feedforward neural networks 2004 Proceedings 2004 IEEE International Joint Conference on Neural Networks. Piscataway: IEEE20042985990 HUANG G B, ZHU Q Y, SIEW C K.Extreme learning machine: a new learning scheme of feedforward neural networks[C]// 2004 Proceedings 2004 IEEE International Joint Conference on Neural Networks. Piscataway: IEEE, 2004, 2: 985-990.
[15]	MIKOLOV T, YIH W, ZWEIG G.Linguistic regularities in continuous spaceword representations[C]//Proceedings of the 2013 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies. Atlanta: ACL, 2013: 746-751.
[16]	TAN S, ZHANG J.An empirical study of sentiment analysis for Chinese documents[J]. Expert Systems with Applications, 2008, 34(4): 2622-2629.
[17]	LIN Y, LEI H, WU J, et al.An empirical study on sentiment classification of chinese review using word embedding[C]// 29th Pacific Asia Conference on Language, Information and Computation. Shanghai: Computer Science, 2015: 258-266.