基于多尺度语义协同网络的高校网络舆论情感分类研究  

Research on Sentiment Classification of University Network Public Opinion based on Multi-scale Semantic Collaborative Network

作　　者：张舜标[1] ZHANG Shunbiao(Guangdong Agriculture Industry Business Polytechnic,Guangzhou 510507,China)

机构地区：[1]广东农工商职业技术学院,广东广州510507

出　　处：《广东农工商职业技术学院学报》2023年第3期62-67,共6页Journal of Guangdong Agriculture Industry Business Polytechnic

摘　　要：针对基础深度学习模型特征提取能力不足,循环网络训练效率低等问题,将高校社交网络平台评论文本数据作为研究对象,提出了基于多尺度语义协同网络的高校网络舆论情感分类模型。预训练模型ALBERT(A Lite BERT)通过结合当前词的具体上下文动态调整向量表示,提升词向量语义表征质量。多尺度语义协同网络捕捉评论文本不同尺度下的多通道融合情感特征,软注意力机制计算每个特征对情感分类结果的影响权重大小,加权求和后得到情感分类特征表示,线性层输出分布概率并得到具体情感倾向。在真实高校图书馆社交网络平台用户评论数据集进行实验,结果表明该模型F1分数达到了97.46%,优于近期表现优秀的实验对比模型,且通过消融实验证明了各个功能模块的有效性。To address the problems of insufficient feature extraction ability of basic deep learning model and low training efficiency of recurrent network,this paper takes the comment text data of university social network platform as the research object,and proposes a multi-scale semantic collaborative network based sentiment classification model of university network public opinion.The pre training model ALBERT(A Lite BERT)improves the semantic representation quality of word vectors by dynamically adjusting the vector representation in combination with the specific context of the current word.The multi-scale semantic collaborative network captures the multi-channel fusion sentiment features of comment texts at different scales.The soft attention mechanism calculates the influence weight of each feature on the emotional classification results.After weighted summation,the emotional classification feature representation is obtained.The linear layer outputs the distribution probability and obtains the specific emotional tendency.The experiment was conducted on the user comment data set of the social network platform of the real university library,and the result showed that the F1 score of the model reached 97.46%,which was better than the experimental comparison model with excellent performance in recent years,and the ablation experiment proved the effectiveness of each functional module.

关 键 词：情感分析 ALBERT 多尺度语义协同网络 时序卷积网络 软注意力 

分 类 号：G717[文化科学—职业技术教育学]