| 引用本文: | 赵晓芳,金志刚.融合表情符号与短文本的微博多维情感分类[J].哈尔滨工业大学学报,2020,52(5):113.DOI:10.11918/201907004 |
| ZHAO Xiaofang,JIN Zhigang.Multi-dimensional sentiment classification of microblog based on Emoticons and short texts[J].Journal of Harbin Institute of Technology,2020,52(5):113.DOI:10.11918/201907004 |
|
| 摘要: |
| 表情符号已成为网络语言重要组成部分,是分析社交媒体情感的主要特征之一.目前分析社交媒体情感符号的方法多针对Emoji,对颜文字的情感倾向没有相应分析.为获取中文媒体的多维度情感并分析热点话题的群体情感走向,本文以微博为例提出一种新的融合表情符号与短文本的多维情感分类方法.在该框架中,采用深度学习模型分析文本与Emoji组合部分、颜文字部分,分别计算两部分的7种情感强度,挖掘各部分与情感标签的深层次关联,并设计计算模型来反映语句包含的多维情感属性,实现对语句多维情感强度的检测.实验选择NLPCC2014数据集和爬取的带有颜文字的微博数据集进行验证,实验证明当文本与Emoji组合、颜文字占比分别为0.6和0.4时情感分类效果最好,且含颜文字的语句情感分类性能指标始终高于不含颜文字的语句,这表明融合表情符号和短文本的形式有效提高了情感检测精度.该方法为研究群体情感趋势提供了更细粒度的分析,为中文社交媒体的情感分析提供了新思路. |
| 关键词: 情感分类 Emoji 颜文字 深度学习 |
| DOI：10.11918/201907004 |
| 分类号:TP391 |
| 文献标识码:A |
| 基金项目:国家自然科学基金(71502125) |
|
| Multi-dimensional sentiment classification of microblog based on Emoticons and short texts |
|
ZHAO Xiaofang,JIN Zhigang
|
|
(School of Electrical and Information Engineering, Tianjin University, Tianjin 300072, China)
|
| Abstract: |
| Emoticons have become an important component of network language and is one of the main characteristics of the analysis of social media sentiment. The current social media sentiment analysis methods most focus on Emoji, while there is no study on the sentiment trend of kinesics. In order to obtain the multi-dimensional sentiment polarity of Chinese social media and analyze the group sentiment trend on hot topics, this paper proposes a new multi-dimensional sentiment classification method based on deep learning, which combines Emoticons with short texts. In this framework, the text and Emoji combination and the kinesics in microblog sentences were analyzed using deep learning model, and seven sentiment intensities of the two parts were obtained to explore the correlation between each part and sentiment labels. Then, a computational model was designed to reflect the multi-dimensional sentiment polarity contained in microblog sentences, which can realize the detection of the multi-dimensional sentiment intensity of sentences. The experiment utilized the NLPCC2014 dataset and the crawled microblog dataset containing kinesics for verification. Results show that when the proportion of the text and Emoji combination and the kinesics were 0.6 and 0.4, the effect of sentiment classification was the best. The sentiment classification performance indicator of the sentences containing kinesics was always higher than that without kinesics, which indicates that the combination of Emoticon and short texts can effectively improve the accuracy of microblog sentiment detection. The experiment provides a more fine-grained analysis for group sentiment trend and a new idea for Chinese social media sentiment analysis. |
| Key words: sentiment classification Emoji kinesics deep learning |
