摘要
微博言论往往带有强烈的情感色彩,对微博言论的情感分析是获取用户观点态度的重要方法。许多学者都是将研究的重点集中在句子词性、情感符号以及情感语料库等方面,然而用户自身的情感倾向性并没有受到足够的重视,因此,提出了一种新的微博情感分类方法,其通过建模用户自身的情感标志得分来帮助识别语句的情感特征,具体地讲,将带有情感信息的微博语句词向量序列输入到长短期记忆网络(LSTM),并将LSTM输出的特征表示与用户情感得分进行结合作为全连接层的输入,并通过Softmax层实现了对微博文本的情感极性分类。实验表明,提出的方法UA-LSTM在情感分类任务上的表现超过的所有基准方法,并且比最优的基准方法MF-CNN在F1值上提升了3.4%,达到0.91。
Micro-blog's speech often has strong sentimental color, and the sentiment analysis of Micro-blog's speech is an important way to get users' opinions and attitudes. Many researchers conduct research via focusing on the parts of speech(POS), emotion symbol and emotion corpus. This paper proposes a novel method for Micro-blog sentiment analysis, which aims to identify the sentiment features of a text by modeling user sentiment tendency. Specifically, we construct a sentiment information embedded word embedding sequence, and input it into a long short term memory(LSTM) model to get a sentiment embedded output representation. Then we merge both the user sentiment tendency score and the output representation of LSTM, and use it as the input of a fully connected layer which is followed by a softmax layer to get the final sentiment classification result. The experiment shows that the performance of our proposed method UA-LSTM is better than all the baseline methods on the sentimental classification task, and it achieves the F1-score up to 0.91, with an improvement of 3.4% over the best baseline method MF-CNN.
作者
吴洁
朱小飞
张宜浩
龙建武
黄贤英
杨武
Jie WU;Xiao-fei ZHU;Yi-hao ZHANG;Jian-wu LONG;Xian-ying HUANG;Wu YANG(School of Computer Science and Engineering, Chongqing University of Technology, Chongqing 400054, China)
出处
《山东大学学报（理学版）》
CAS
CSCD
北大核心
2019年第3期46-55,共10页
Journal of Shandong University(Natural Science)
基金
国家自然科学基金资助项目(61702063
61502064
61502065)
国家社会科学基金资助项目(17XXW005)
重庆市基础科学与前沿技术研究项目(cstc2017jcyjBX0059
cstc2015jcyjBX0127
cstc2017jcyjAX0144
cstc2017jcyjAX0339
cstc2017jcyjAX0144)
重庆市教委人文社科重点研究项目(17SKG136)
关键词
情感分析
长短期记忆网络
用户情感倾向
sentiment analysis
long short term memory
user sentiment tendency
