基于多模态联合注意力机制的网民情感分析研究  

Sentiment Analysis of Online Users Based on Multimodal Co-attention

作　　者：范涛 吴鹏[2] 王昊[1] 凌晨[2] Fan Tao;Wu Peng;Wang Hao;Ling Chen(School of Information Management,Nanjing University,Nanjing 210023;School of Economics and Management,Nanjing University of Science and Technology,Nanjing 210094)

机构地区：[1]南京大学信息管理学院,南京210023 [2]南京理工大学经济管理学院,南京210094

出　　处：《情报学报》2021年第6期656-665,共10页Journal of the China Society for Scientific and Technical Information

基　　金：国家自然科学基金项目“突发事件网民负面情感的模型检测研究”(71774084),“社会化影响下个体信息认知处理中的扭曲与偏见机制研究”(71471089)。

摘　　要：网民情感对网络舆情的演化和发展具有重要的推动作用,如何精准分析网民情感对于网络舆情管理有着重大的现实意义。现有的网民情感研究主要是基于文本进行分析,鲜有研究从多模态视角出发,结合文本和图片对网民情感进行分析。在多模态情感分析中,现有研究多结合不同模态的全面特征、进行高维融合,容易造成信息冗余、引入噪声,忽略了不同模态之间的交互性和相关性。本文提出基于多模态联合注意力机制的情感分析模型,通过联合词引导的注意力机制和图引导的注意力机制,结合文本和图片分析网民情感,并且在"新冠肺炎疫情"等网络舆情数据中进行实证研究,与相关优异的基线模型进行了对比分析。实验结果表明,本文提出的基于多模态联合注意力机制模型具有一定的优越性,能够有效捕捉不同模态间的交互和相关关系。The sentiment of online users significantly influences the evolution and development of online public opinion events.Accurately recognizing the sentiment of online users has practical implications for managing online public opinion events.Extant studies about the sentiment analysis of online users are mostly based on texts,lacking in research on the combination of texts and images.In multimodal sentiment analysis,existing research typically unifies the comprehensive unimodal features and performs high-dimensional fusion.This can easily cause information redundancy and introduce the noise,ignoring the interaction and correlation between different modalities.Thus,we propose a sentiment analysis model based on multimodal co-attention to analyze the sentiment of online users,unifying word-guided and image-guided attention mechanisms.To this end,we conduct empirical experiments on multimodal datasets,such as"COVID-19."The results show that the proposed model based on multimodal co-attention is superior to past models and can capture the interaction and relationship between different modalities.

关 键 词：网络舆情 情感分析 多模态融合 神经网络 注意力机制 

分 类 号：G252[文化科学—图书馆学]