基于细粒度信息交互注意力的情绪分类方法  

An Emotion Classification Method Based on Interactive Attention Mechanism with Fine-Grained Information

作　　者：胡慧君[1,2,3] 易洋 施琦 唐东昕 刘茂福 HU Hujun;YI Yang;SHI Qi;TANG Dongxin;LIU Maofu(School of Computer Science and Technology,Wuhan University of Science and Technology,Wuhan 430065,Hubei,China;Hubei Province Key Laboratory of Intelligent Information Processing and Real-Time Industrial System,Wuhan 430065,Hubei,China;Big Data Science and Engineering Research Institute,Wuhan University of Science and Technology,Wuhan 430065,Hubei,China;The First Affiliated Hospital of Guizhou University of Traditional Chinese Medicine,Guiyang 550002,Guizhou,China)

机构地区：[1]武汉科技大学计算机科学与技术学院,湖北武汉430065 [2]智能信息处理与实时工业系统湖北省重点实验室,湖北武汉430065 [3]武汉科技大学大数据科学与工程研究院,湖北武汉430065 [4]贵州中医药大学第一附属医院,贵州贵阳550002

出　　处：《武汉大学学报（理学版）》2023年第3期400-408,共9页Journal of Wuhan University:Natural Science Edition

基　　金：国家重点研发计划(2019YFC1712500);贵州省科技计划项目(黔科合支撑[2021]一般095)。

摘　　要：现有细粒度分析方法未能充分利用细粒度情绪信息来增强上下文与评价目标间的语义关联性,且对多词构成的评价目标仅平均化处理,损失了词间内容与关系信息,导致分类不精准。针对上述问题,本文提出了一种基于细粒度信息交互注意力(interactive attention with fine-grained information,FGIA)的情绪分类方法,通过采用更加细粒度的注意力机制来实现评价目标与上下文之间的充分交互,同时得到目标对上下文以及上下文对目标的交互注意力表示,进而辅助完成情绪分类。在本文构建的COVID-19网络舆情中文数据集上进行了实验验证,结果表明,FGIA能够有效地提升网络舆情数据情绪分类的准确性,相比于主流的分类方法,在各项评价指标上均取得了较高的提升。The fine-grained information in the text cannot be fully used by the existing fine-grained emotion analysis methods,leading to the weak relevance of the contextual semantic representation of the target.Moreover,the target is generally composed of multiple words,and average pooling will lose the content and relationship information inside,resulting in inaccurate classification.This paper proposes a novel classification method based on an interactive attention mechanism with fine-grained information(FGIA).Our FGIA method can ensure that the interaction between target and context can be effectively completed,and simultaneously calculate the attention vectors of context and target.Finally,emotion classification can be obtained with its guidance.Extensive experimental results on the Chinese Dataset of COVID-19 Online Public Opinion demonstrate that our FGIA can effectively complete emotion classification.Compared with several state-of-the-art methods,our FGIA can yield superior performance according to the evaluation metrics.

关 键 词：情绪分类 细粒度信息 交互注意力 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]