[1]游 兰,曾 晗,韩凡宇,等.基于 BERT-BiGRU 集成学习的情感语义识别[J].计算机技术与发展,2023,33(05):159-166.[doi:10. 3969 / j. issn. 1673-629X. 2023. 05. 024]
　YOU Lan,ZENG Han,HAN Fan-yu,et al.Sentiment Semantic Recognition Based on BERT-BiGRU Ensemble Learning[J].,2023,33(05):159-166.[doi:10. 3969 / j. issn. 1673-629X. 2023. 05. 024]

点击复制

基于 BERT-BiGRU 集成学习的情感语义识别()

分享到：

《计算机技术与发展》[ISSN:1006-6977/CN:61-1281/TN]

卷:

33

期数:

2023年05期

页码:

159-166

栏目:

人工智能

出版日期:

2023-05-10

文章信息/Info

Title:

Sentiment Semantic Recognition Based on BERT-BiGRU Ensemble Learning

文章编号:

1673-629X(2023)05-0159-08

作者:

游 兰¹; 2 ; 曾 晗¹; 2 ; 韩凡宇¹; 3 ; 金 红¹; 2; 3 ; 崔海波¹; 3; 4 ; 张家合¹; 4

1. 湖北大学 计算机与信息工程学院,湖北 武汉 430062;

2. 湖北省软件工程工程技术研究中心,湖北 武汉 430062;

3. 智慧政务与人工智能应用湖北省工程研究中心,湖北 武汉 430062;

4. 湖北省教育信息化工程技术研究中心,湖北 武汉 430062

Author(s):

YOU Lan1; 2 ; ZENG Han1; 2 ; HAN Fan-yu1; 3 ; JIN Hong1; 2; 3 ; CUI Hai-bo1; 3; 4 ; ZHANG Jia-he1; 4

1. School of Computer Science and Information Engineering,Hubei University,Wuhan 430062,China;

2. Engineering and Technical Research Center of Hubei Province in Software Engineering,Wuhan 430062,China;

3. Engineering Research Center of Hubei Province in Intelligent Government Affairs and Application of Artificial Intelligence,Wuhan 430062,China;

4. Engineering and Technical Research Center of Hubei Province in Educational Information,Wuhan 430062,China

关键词:

情感识别; BERT 预训练模型; 双向门控循环单元; 集成学习; 深层特征

Keywords:

sentiment recognition; BERT pre-training model; bidirectional GRU; ensemble learning; deep feature

分类号:

TP391

DOI:

10. 3969 / j. issn. 1673-629X. 2023. 05. 024

摘要:

如今,在社交网络上发表评论已成为公众对社会现象表达态度和立场的主要方式。 精准识别社交文本的情感倾向性对于舆情管控、社会维稳等有重要价值。 针对传统情感识别模型大多集中于评论的表层语义挖掘,存在分类效果不佳、泛化能力有限等问题,提出了一种基于 BERT-BiGRU 多模集成学习的深层情感语义识别方法。 首先,通过 BERT 预训练模型获取评论文本的上下文语义特征表示,再结合 BiGRU 提取深层非线性特征向量,实现单模型下的最优效果;接着,为了使模型效果稳定且多方面表现均衡,基于 BERT 系列预训练模型训练出表现优异且具有差异化的多个情感分类器;最后,利用数据扰动和投票策略的集成学习方法,实现各模型深层特征的充分融合。 实验结果显示:BERT-BiGRU 模型相较于其他传统模型,在两个公开数据集( COV19 和 ChnSenti)上具有更优的情感识别效果。

Abstract:

Nowadays,posting comments on social networks has become one of the main means for the public to express their attitudes andstandpoints on social events. Accurately identifying the sentiment orientation of social texts is of great value for public opinions control,social stability maintenance,etc. Since the traditional sentiment recognition models only focus on mining?

surface semantic on comments,there are some problems such as poor classification effect and limited generalization ability. Aiming at these problems,we propose a deepsentiment semantic recognition model based on BERT-BiGRU multi-model ensemble learning. Firstly,the contextual semantic featurerepresentation of the comment text is obtained through the BERT pre-training model,and then BiGRU is combined to extract the deepnonlinear feature vectors to achieve the optimal sentiment recognition results under the single model. Next,in order to make the effect ofthe model stable and achieve balanced performance in many aspects,several differentiated emotion classifiers with excellent performanceare trained based on the BERT series pre-training model. Finally,the ensemble learning method of data disturbance and voting strategy isused to achieve the full integration of the deep features of each model. The experimental results on two public datasets ( COV19 andChnSenti) show that the model proposed has better sentiment recognition effects than other traditional models.

常用功能