基于混合深度神经网络模型的网络评论情绪识别研究  

Emotion Recognition of Online Comments Based on Hybrid Deep Neural Network

作　　者：曹惠茹 钟麟 林延峰 许沛鑫 李小敏 胡嘉琪 Cao Huiru;Zhong Lin;Lin Yanfeng;Xu Peixin;Li Xiaomin;Hu Jiaqi(Department of Information Engineering,Guangzhou Institute of Technology,Guangzhou 510075,China;School of Electric-mechanical Engineering,Zhongkai University of Agricultural Engineering,Guangzhou 510550,China)

机构地区：[1]广州工程技术职业学院信息工程学院,广州510075 [2]仲恺农业工程学院机电工程学院,广州510550

出　　处：《机电工程技术》2024年第10期117-122,共6页Mechanical & Electrical Engineering Technology

基　　金：广东省哲学社会科学规划2023年度学科共建项目(GD23XXW13);2023年广东省广东大学生科技创新培育专项资金资助项目(pd-jh2023b0947);2022年度广州市高等教育教学质量与教学改革工程大学生创新创业训练计划项目(2022CXCYJH017)。

摘　　要：为了探究网络空间论坛中民众针对某热点事件的情绪状况,构建了一种基于混合深度神经网络的热点事件网络评论情绪识别方法。首先,以网络论坛为平台,建立了热点事件相关评论文本数据集;将情绪分为典型的5种类型,接着以CNN与LSTM传统深度学习框架为基础,设计了一种混合深度学习的网络论文情绪识别方法,然后以实际网络数据集对该方法进行了测试与性能对比。实验结果表明:所提出的混合深度学习模型对热点事件网络评论文本数据情绪识别效果良好,能够满足对于论坛中文本进行情感识别;与BiLSTM、CNN-BiGRU、CNN-GRU、CNN_LSTM、LSTM、SimpleRNN和textCNN模型进行比较,该模型情绪识别准确率分别提高了0.23%、0.35%、0.49%、0.47%、0.98%、3.5%和0.88%,所提出的方法具有较好的适应能力。In order to explore the emotional status of the public in cyberspace forums for a hotspot,a hybrid deep neural network method is constructed for emotion recognition of hotspots online comments.First,a dataset of commentaries related to a hot event is established on the platform of online forums.Emotion is divided into five typical types,and then based on the traditional deep learning framework of CNN and LSTM,a hybrid deep learning emotion recognition method for network papers is designed,and then the method is tested and compared with the actual network data set.The experimental results show that the proposed hybrid deep learning model has a good effect on the emotion classification number of the text data of the hotspots online comments,and can satisfy the emotion recognition of the text in the forum.Compared with BiLSTM,CNN-BiGRU,CNN-GRU,CNN_LSTM,LSTM,SimpleRNN and textCNN models,the emotion recognition accuracy of this model is increased by 0.23%,0.35%,0.49%,0.47%,0.98%,3.5%,0.88%,respectively.The proposed method has good adaptability.

关 键 词：情绪识别 热点事件 网络论坛 大数据 深度学习 

分 类 号：TP391.1[自动化与计算机技术—计算机应用技术]