摘要
基于微博文本的情感分析已经成为近些年的研究热点.为了更有效分析微博文本的情感极性,实现将表情符的情感表现加入到文本分类的任务中,提出了一种将深度学习与情感符号相结合的学习方法.该方法利用双向门控循环(BiGRU)神经网络对文本进行特征提取,然后利用自注意力机制对文本和表情符号的向量表示进行特征提取,最后通过softmax识别出最终的情感极性.本文采集了三种微博语料集,实验结果表明,与融合表情符的自注意力机制和BiLSTM模型和与仅输入纯文本的自注意力机制和BiGRU模型相比,融合表情符号的自注意力机制和BiGRU网络模型有效地提高了情感分类的准确率.
Recently,sentiment analysis of Microblog has become a hot research topic.In order to analyze the sentiment polarity of the microblog text more effectively,and realize the task of adding the sentiment polarity of emoji to sentiment classification,this paper proposed a learning method combining deep learning with emotional symbols.This method uses the bidirectional Gated Recurrent Unity(BiGRU)neural network to extract the features of plain text,and then uses the self-attention mechanism to obtains a new feature representation,finally recognize the final sentiment polarity of the microblogs text by softmax.The experiments result on three microblog corpuses collected show that compared with the self-attention and BiLSTMmodels and the self-attention and BiGRU model that only inputs plain text,the self-attention and BiGRU models effectively improves the classification accuracy.
作者
陈亚茹
陈世平
CHEN Ya-ru;CHEN Shi-ping(School of Optical-Electrical and Computer Engineering,University of Shanghai for Science and Technology,Shanghai 200093,China)
出处
《小型微型计算机系统》
CSCD
北大核心
2020年第8期1590-1595,共6页
Journal of Chinese Computer Systems
基金
国家自然科学基金项目(61472256,61170277)资助
上海市一流学科建设项目(S1201YLXK)资助
上海理工大学科技发展基金项目(16KJFZ035,2017KJFZ033)资助
沪江基金项目(A14006)资助。
关键词
情感分析
表情符
自注意力机制
双向门控循环神经网络
sentiment analysis
emoji
self-attention mechanism
bidirectional Gated Recurrent Unity(BiGRU)
