融入主题特征的中越跨语言情感分类模型  

A Chinese and Vietnamese Cross-lingual Sentiment Classification Model Incorporating Topic Features

作　　者：施忆雪 余正涛[1,2] 相艳[1,2] 张亚飞 SHI Yixue;YU Zhengtao;XIANG Yan;ZHANG Yafei(Faculty of Information Engineering and Automation,Kunming University of Science and Technology,Kunming,Yunnan 650500,China;Yunnan Key Laboratory of Artificial Intelligence,Kunming University of Science and Technology,Kunming,Yunnan 650500,China)

机构地区：[1]昆明理工大学信息工程与自动化学院,云南昆明650500 [2]昆明理工大学云南省人工智能重点实验室,云南昆明650500

出　　处：《中文信息学报》2022年第3期36-44,共9页Journal of Chinese Information Processing

基　　金：云南省重大科技专项计划项目(202002AD080001);国家重点研发计划(2018YFC0830105,2018YFC0830100);国家自然科学基金(61762056,61472168,61972186)。

摘　　要：越南语网络评论的情感分类是越南语事件观点分析的基础。越南语资源匮乏,标注困难,可借助中文标注语料进行跨语言情感分类,实现越南语评论的情感极性预测。但现有的跨语言情感分类模型忽略了主题信息对加强情感表征学习、减小语言差异的作用。为此,该文提出了一种融入主题特征的中越跨语言情感分类模型。将中文①和越南语的主题词分布作为外部知识引入模型,利用门控机制将主题表征与语义表征进行融合编码,并通过对抗过程使模型学习到语言分布差异最小的表征,最终完成情感分类任务。实验结果表明,该模型能更快拟合出语言分布差异,其宏F_(1)值较多个基线模型均有明显提高。Sentiment classification of Vietnamese online comments is the viald for the opinion analysis of Vietnamese event.As Vietnamese is a low-resource language,the cross-lingual sentiment classification can be performed with the help of Chinese annotated corpus to help the sentiment polarity prediction of Vietnamese.In this paper,a cross-lingual sentiment classification model of Chinese and Vietnamese comments incorporating confrontational topic features is proposed.The topic distributions of Chinese and Vietnamese are introduced into the model as external knowledge,and a gate layer is used to encode representation from topic representations with semantic representations.The model is optimized to learn the representations with the smallest differences in language distributions through the adversarial learning to finally complete the sentiment classification task.The experimental results show that the proposed model can significantly improve marco F1 values compared with several baseline models.

关 键 词：跨语言情感分析 主题模型 社交媒体评论 对抗学习 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]