基于LDA主题模型的情感分析研究

doi:10.16180/j.cnki.issn1007-7820.2020.07.003

摘要/Abstract

摘要：

LDA主题模型在提取特征时缺乏对词语关联及相关词对的理解,这会影响情感极性分类的准确率。针对这一问题,文中提出一种在LDA主题模型中引入特征情感词对抽取方法的新模型,以改善特征情感词对的抽取效果。利用依存句法分析设计特征情感词对的识别方法,随后将识别方法作为约束条件引入LDA模型对特征情感词对进行抽取。通过吉布斯采样进行参数计算,给出了模型的生成过程。最后利用随机森林分类方法对文本进行情感极性分类。为验证文中模型的有效性,将其和另外两种模型一起进行实验,当主题个数为20时,文中所提模型分类的准确率、召回率、F值分别为81.54%、83.13%和82.33%,显著高于另外两种模型。

关键词: 产品评论, 情感分析, 依存句法, 特征抽取, LDA主题模型, 随机森林算法

Abstract:

LDA topic model lacks understanding of word association and related word pairs when extracting features, which affects the precision of emotional polarity classification. Aiming at this problem, this paper proposed a new model to introduce the feature-opinion pair extraction method in the LDA topic model to improve the extraction effect of the feature opinion pairs. Dependency parsing was used to design feature affective word pairs recognition methods of characteristic affective word pairs. Then the recognition method was introduced as a constraint condition into the LDA model to extract the feature sentiment word pairs. The parameters were calculated by Gibbs sampling, and the generation process of the model was proposed. Finally, the emotional polarity of the text was classified using the random forest classification method. In order to verify the validity of the proposed model, the experiment was carried out together with the other two models. When the number of subject was 20,the results showed that the precision, recall and F-Measure were 81.54%、83.13% and 82.33%, which were significantly higher than the other two models.

Key words: product reviews, sentiment analysis, dependency syntax, feature extraction, LDA topic model, random forest algorithm

中图分类号:

TP391.1

刘艳文,魏赟. 基于LDA主题模型的情感分析研究[J]. 电子科技, 2020, 33(7): 12-16.

LIU Yanwen,WEI Yun. Research of Emotional Analysis Based on LDA Topic Model[J]. Electronic Science and Technology, 2020, 33(7): 12-16.

图/表 14

图1

图2

表1

图3

图4

图5

图6

表2

表3

表4

表5

图7

图8

图9

参考文献 18

[1]	王洪伟, 宋媛, 杜战其, 等. 基于在线评论情感分析的快递服务质量评价[J]. 北京工业大学学报, 2017,43(3):403-412.
	Wang Hongwei, Song Yuan, Du Zhanqi, et al. Evaluation of service quality for express industry through sentiment analysis of online reviews[J]. Journal of Beijing University of Technology, 2017,43(3):403-412.
[2]	卿勇, 刘梦娟, 薛浩, 等. OPEN:一个基于评论的商品特征抽取及情感分析框架[J]. 计算机应用与软件, 2018,35(1):65-71.
	Qing Yong, Liu Mengjuan, Xue Hao, et al. OPEN:a framework for product feature extraction and sentiment analysis based on productcomments[J]. Computer Applications and Software, 2018,35(1):65-71.
[3]	马京苗. 网购用户评论中隐式评价对象的提取方法研究.[D]. 北京:北京交通大学, 2017.
	Ma Jingmiao. Research on extraction method of implicit evaluation objects in online user reviews[D]. Beijing:Beijing Jiaotong University, 2017.
[4]	Blei D M, Ng A Y, Jordan M I. Latent dirichlet allocation[J]. Journal of Machine Learning Research, 2003,3(3):993-1022.
[5]	李湘东, 廖香鹏, 黄莉. LDA模型下书目信息分类系统的研究与实现[J]. 现代图书情报技术, 2014,30(5):18-25.
	Li Xiaodong, Liao Xiangpeng, Huang Li. Research and implementation of bibliographicinformation classification classification system in LDA model[J]. New Technology of Library and Information Service, 2014,30(5):18-25.
[6]	Putri I R, Kusumaningrum R. Latent Dirichlet Allocation (LDA) for sentiment analysis toward tourism review in indonesia[J]. Journal of Physics:Conference Series, 2017,80(1):1-6
[7]	Wang W, Feng Y, Dai W. Topic analysis of online reviews for two competitive products using latent dirichlet allocation[J]. Electronic Commerce Research & Applications, 2018(7):29-33.
[8]	姬东鸿, 熊蜀峰. 面向产品评论分析的短文本情感主题模型[J]. 自动化学报, 2016,42(8):1227-1237.
	Ji Donghong, Xiong Shufeng. A short text sentiment-topic model for product review analysis[J]. ACTA Automatica Sinica, 2016,42(8):1227-1237.
[9]	Shams M, Baraani Dastjerdi A. Enriched LDA (ELDA): Combination of latent dirichlet allocation with word co-occurrence analysis for aspect extraction[J]. Expert Systems with Applications, 2017,80(6):136-146.
[10]	王伟, 周咏梅, 阳爱民, 等. 一种基于LDA主题模型的评论文本情感分类方法[J]. 数据采集与处理, 2017,32(3):629-635.
	Wang Wei, Zhou Yongmei, Yang Aimin, et al. Method of sentiment analysis for comment texts based on LDA[J]. Journal of Data Acquisition and Processing, 2017,32(3):629-635.
[11]	延丰, 杜腾飞, 毛建华, 等. 基于情感词典与LDA模型的股市文本情感分析[J]. 理论与算法, 2017,40(12):82-87.
	Yan Feng, Du Tengfei, Mao Jianhua, et al. Stock text sentiment analysis based on emotion dictionary and LDA model[J]. Electronic Measurement Technology, 2017,40(12):82-87.
[12]	孙悦, 袁健. 基于Spark的改进随机森林算法[J]. 电子科技, 2019,32(4):60-64.
	Sun Yue, Yuan Jian. Improved random forest algotithm based on Spark[J]. Electronic Science and Technology, 2019,32(4):60-64.
[13]	马勋, 周长胜, 吕学强, 等. 基于SAO结构的非分类关系抽取研究[J]. 计算机工程与应用, 2018,54(8):220-225.
	Ma Xun, Zhou Changsheng, Lü Xueqiang, et al. Extraction of non-taxonomic relation based on SAO structure[J]. Computer Engineering and Application, 2018,54(8):220-225.
[14]	杨荣根, 杨忠. 基于HMM中文词性标注研究[J]. 金陵科技学院学报, 2017,33(1):20-23.
	Yang Ronggen, Yang Zhong. Research of chinese part of speed tagging based on HMM[J]. Journal of Jinling Institute of Technology, 2017,33(1):20-23.
[15]	冯靖, 莫秀良, 王春东. 基于LDA改进的K-means算法在短文本聚类中的研究[J]. 天津理工大学学报, 2018,34(3):7-11.
	Feng Jing, Mo Xiuliang, Wang Chundong. Improved K-means algorithm based on Latent Dirichlet Allocation for short text clustering[J]. Journal of Tianjin University of Technology, 2018,34(3):7-11.
[16]	张志远, 杨宏敬, 赵越. 基于吉布斯采样结果的主题文本网络构建方法[J]. 计算机工程, 2017,43(6):150-157. doi: 10.3969/j.issn.1000-3428.2017.06.025
	Zhang Zhiyuan, Yang Hongjing, Zhao Yue. Topic text network construction method based on Gibbs sampling results[J]. Compute Engineering, 2017,43(6):150-157. doi: 10.3969/j.issn.1000-3428.2017.06.025
[17]	何明, 要凯升, 杨芃, 等. 基于标签信息特征相似性的协同过滤个性化推荐[J]. 计算机科学, 2018,45(6A):428-435.
	He Ming, Yao Kaisheng, Yang Peng, et al. Collaborative filtering personalized recommendation based on similarity of tag information feature[J]. Computer Science, 2018,45(S1):428-435.
[18]	Lin Chenghua, He Yulan. Joint sentiment topic model for sentiment analysis[C].New York:Proceedings of the 18^th ACM Conference on Information and Knowledge Management, 2009.

关系类型	缩写
主谓关系	SBV
动宾关系	VOB
定中关系	ATT
动补结构	CMP
状中关系	ADV

符号	说明
α	文档-主题分布的Dirichlet超参数
β	主题-词语分布的Dirichlet超参数
θ	文档-主题概率分布
ψ	主题-词语概率分布
z	词语的主题分布
γ	依存约束相关的Dirichlet超参数
w	词语
l	情感层标签
ε	情感-主题概率分分布
v	依存约束的转换参
ψ_p	依存约束的主题词语分布
S_p	依存约束关系集
N(d,s)	文档中的词语数
K	主题数
D	文档数
S	每个文档的句子数

符号	说明
TP	在文本中为正向, 分类结果为正向的数据
FP	在文本中为负向, 分类结果为正向的数据
FN	在文本中为正向, 分类结果为负向的数据
TN	在文本中为负向, 分类结果为负向的数据

模型	LDA	JST	本文模型
抽取对数	4 023	4 361	5 125
正确对数	2 723	3 102	4 037
抽取准确率	0.676	0.711	0.787