基于特征融合的层次结构微博情感分类

doi:10.11959/j.issn.1000-0801.2016182

摘要/Abstract

摘要：

情感分类是观点挖掘的热点研究之一，微博文本情感分类具有很高的应用价值。鉴于传统特征选择方法存在语义缺陷，采用神经网络语言模型，提出了基于概率模型的对词向量进行权重分配的深层特征表示方法，构建文本语义向量。将文本深层特征与浅层特征融合，构建融合语义信息的特征向量，弥补传统特征选择方法语义的缺陷。采用SVM层次结构分类模型，实现多种情感分类。实验结果表明，采用特征融合的层次结构情感分类方法，能有效提高微博情感分类的准确率。

关键词: 情感分类, 词向量, 深层特征, 特征融合, 层次结构分类模型

Abstract:

Sentiment classification is an important issue of opinion mining.It has a high application value to classify sentiment in micro-blogs.As traditional feature selection method has semantic gap,a neural network language model was used to propose a deep feature representation method based on probability model to distribute weight to the word vector.Using this method,text semantic vector could be built.In order to avoid the semantic gap,the deep features and shallow features of text were integrated and feature vector that contained semantic information was constructed.With SVM hierarchical classification model,a variety of sentiments could be classified.Experimental results show that the hierarchical sentiment classification method based on feature fusion can improve the accuracy of sentiment classification in micro-blogs.

Key words: sentiment classification, word vector, deep feature, eature fusion, hierarchical classification model

朱宪莹,刘箴,金炜,刘婷婷,刘翠娟,柴艳杰. 基于特征融合的层次结构微博情感分类[J]. 电信科学, 2016, 32(7): 106-114.

Xianying ZHU,Zhen LIU,Wei JIN,Tingting LIU,Cuijuan LIU,Yanjie CHAI. Hierarchical micro-blog sentiment classification based on feature fusion[J]. Telecommunications Science, 2016, 32(7): 106-114.

图/表 13

图1

表1

表2

图2

图3

图4

表3

表4

表5

图5

图6

表6

表7

参考文献 15

[1]	PANG B , LEE L , VAITHYANATHAN S . Thumbs up sentiment classification using machine learning techniques[J]. Computer Science, 2009(10): 79-86.
[2]	PANG B , LEE L . Seeing stars: exploiting class relationships for sentiment categorization with respect to rating scales[J]. Arxiv Cornell University Library, 2004: 115-124,arXiv:cs/0506075v1.
[3]	ALAM C O , ROTH D , SPROAT R . Emotions from text:machine learning for text-based emotion prediction[C]// Conference on Human Language Technology and Empirical Methods in Natural Language Processing, October 6-8, 2005, Vancouver,British Columbia,Canada. New York: ACM Press, 2005: 579-586.
[4]	ECKMAN P . Universal and cultural differences in facial expression of emotion[EB/OL]. [2016-05-02]. .
[5]	GHAZI D , INKPEN D , SZPAKOWICZ S . Hierarchical versus flat classification of emotions in text[C]// NAACL HLT 2010 Workshop on Computational Approaches to Analysis and Generation of Emotion in Text, June 5, 2010, Los Angeles,USA. New York: ACM Press, 2010: 140-146.
[6]	GHAZI D , INKPEN D , SZPAKOWICZ S . Hierarchical approach to emotion recognition and classification in texts[M]. Berlin:Springer, 2010: 40-50.
[7]	HUANG S , PENG W , LI J , et al. Sentiment and topic analysis on social media: a multi-task multi-label classification approach[C]// The 5th Annual ACM Web Science Conference, May 2-4, 2013, Paris,France. New York: ACM Press, 2013: 172-181.
[8]	LIU S M , CHEN J H . A multi-label classification based approach for sentiment classification[J]. Expert Systems with Applications, 2015,42(3): 1083-1093.
[9]	XU H , YANG W , WANG J . Hierarchical emotion classification and emotion component analysis on chinese micro-blog posts[J]. Expert Systems with Applications, 2015,42(11): 8745-8752.
[10]	CHO S H , KANG H B . Text sentiment classification for SNS-based marketing using domain sentiment dictionary[C]// 2012 IEEE International Conference on Consumer Electronics (ICCE), Jan 13-16, 2012, Las Vegas,NV,USA. New Jersey: IEEE Press, 2012: 717-718.
[11]	SUN X , LI C , YE J . Chinese microblogging emotion classification based on support vector machine[C]// 2014 International Conference on Computing,Communication and Networking Technologies (ICCCNT), July 11-13, 2014, Hefei,China. New Jersey: IEEE Press, 2014: 1-5.
[12]	刘翠娟，刘箴，柴艳杰，等. 基于微博文本数据分析的社会群体情感可视计算方法研究[J]. 北京大学学报(自然科学版), 2016,52(1): 178-186. LIU C J , LIU Z , CHAI Y J , et al. Visual study on calculation method of social groups emotional based on the micro-blog post analysis[J]. Journal of Peking University(Natural Science Edition), 2016,52(1): 178-186.
[13]	HINTON G E . Learning distributed representations of concepts[EB/OL]. [2002-08-01].
[14]	MIKOLOV T , CHEN K , CORRADO G , et al. Efficient estimation of word representations in vector space[J]. Computer Science, 2013(9),arXiv: 1301.3781v3.
[15]	MIKOLOV T , SUTSKEVER I , Chen K , et al. Distributed representations of words and phrases and their compositionality[J]. Advances in Neural Information Processing Systems, 2013(26): 3111-3119.

情感类型	微博数目/条
无情感	7 490
开心	1 310
喜好	1 879
厌恶	1 239
伤心	1 018
生气	595
吃惊	324
害怕	145

文本类别	实际属于c_i类的文本数	实际不属于c_i类的文本数
分类器判别为c_i的文本数	A	B
分类器判别不为c_i的文本数	C	D

特征选择方法		有情感			无情感
特征选择方法	P	R	F1	P	R	F1
CHI	0.746	0.759	0.753	0.709	0.716	0.716
CHI+WWBP	0.777	0.783	0.780	0.754	0.747	0.750
IG	0.744	0.772	0.758	0.732	0.701	0.716
IG+WWBP	0.770	0.785	0.777	0.753	0.736	0.744
MI	0.744	0.754	0.734	0.704	0.661	0.682
MI+WWBP	0.752	0.765	0.759	0.731	0.716	0.723
WWBP	0.762	0.767	0.765	0.750	0.736	0.733

特征选择方法		积极			消极
特征选择方法	P	R	F1	P	R	F1
CHI	0.822	0.752	0.786	0.794	0.805	0.799
CHI+WWBP	0.822	0.794	0.808	0.812	0.838	0.825
IG	0.847	0.785	0.815	0.810	0.866	0.837
IG+WWBP	0.848	0.794	0.820	0.817	0.866	0.841
MI	0.806	0.751	0.778	0.780	0.830	0.804
MI+WWBP	0.816	0.776	0.795	0.798	0.834	0.816
WWBP	0.809	0.801	0.805	0.814	0.822	0.818

特征选择方法		厌恶			伤心			生气			害怕
特征选择方法	P	R	F1	P	R	F1	P	R	F1	P	R	F1
CHI	0.551	0.771	0.643	0.590	0.722	0.649	0.632	0.326	0.430	0.533	0.133	0.213
CHI+WWBP	0.564	0.739	0.640	0.606	0.765	0.676	0.612	0.353	0.447	0.530	0.150	0.234
IG	0.516	0.788	0.623	0.580	0.742	0.651	0.663	0.236	0.349	1.000	0.033	0.065
IG+WWBP	0.540	0.773	0.636	0.590	0.780	0.676	0.655	0.295	0.406	0.857	0.100	0.179
MI	0.448	0.877	0.593	0.631	0.537	0.580	0.693	0.202	0.312	1.000	0.017	0.033
MI+WWBP	0.515	0.829	0.636	0.623	0.752	0.681	0.703	0.248	0.367	1.000	0.033	0.065
WWBP	0.532	0.752	0.623	0.625	0.750	0.681	0.554	0.260	0.354	0.000	0.000	0.000