摘要
针对传统情感分析方法对微博短文本应用效果不佳的问题,提出将文本情感特征与深度学习模型融合的微博情感分析新机制.通过词向量计算文本的语义特征,结合基于表情字符的情感特征,利用卷积神经网络挖掘特征集合与情感标签间的深层次关联,训练情感分类器.实验结果表明,相比基于词典的机器学习模型,该机制将情感分析准确率与F度量分别相对提升21.29%和19.20%.该机制结合语义和情感特征,利用卷积神经网络的抽象特征提取能力,改善微博短文本的情感分析精度.
Since traditional sentiment analysis for Weibo had poor performance,A novel mechanism was proposed based on convolutional neural networks with sentiment features.This novel mechanism combined semantic features from word vectors with sentiment features from emoticons and utilized convolutional neural networks to mine the deep correlation between features and labels.Experiment results showed that accuracy and F measure are relatively improved by 21.29% and 19.20% respectively compared with machine learning model based on lexicon.The performance of Weibo sentiment analysis is improved through integrating sentiment features and utilizing the features extraction ability of convolutional neural networks.
作者
金志刚
胡博宏
张瑞
Jin Zhigang;Hu Bohong;Zhang Rui(School of Electronic bifomiation Engineering,Tianjin University,Tianjin 300072,China;Tianjin International Engineering Institute,Tianjin University,Tianjin 300072,China)
出处
《南开大学学报(自然科学版)》
CAS
CSCD
北大核心
2020年第5期77-81,86,共6页
Acta Scientiarum Naturalium Universitatis Nankaiensis
基金
国家自然科学基金(61571318)
青海省科技项目(2015-ZJ-904)
海南省科技项目(ZDYF2016153)。
关键词
情感分析
卷积神经网络
微博短文本
表情字符
sentiment analysis
convolutional neural networks
Weibo short text
emoticons