基于深度学习的在线教师课程评论情感分析模型  

Sentiment analysis model of online teacher curriculum review based on deep learning

作　　者：肖海艳 XIAO Haiyan(Foreign Language Academy,Xianyang Normal University,Xianyang 712000,China)

机构地区：[1]咸阳师范学院外国语学院,陕西咸阳712000

出　　处：《电子设计工程》2023年第6期39-42,47,共5页Electronic Design Engineering

摘　　要：针对传统静态词向量存在语义表征弱以及循环序列模型训练效率低等问题,提出了基于MacBERT-BiSRU-AT的在线教师课程评论情感分析模型。通过预训练模型MacBERT获取评论文本符合上下文语义的动态向量表示,解决了静态词向量存在的一词多义问题,提升词向量语义表示质量。BiSRU模块用于提取评论文本高维情感特征,软注意力用于计算每个词对分类结果的影响程度大小,由分类层输出文本情感极性。在真实在线课程MOOC评论数据集进行实验,结果表明,MacBERT-BiSRU-AT模型分类F1值达到了91.33%,高于实验对比的其他模型,证明了模型的有效性。To address the problems of weak semantic representation of traditional static word vectors and low training efficiency of cyclic sequence model,an online teacher curriculum review emotion analysis model based on MacBERT-BiSRU-AT was proposed. Through the pre training model MacBERT,we can obtain the dynamic vector representation of the comment text that conforms to the context semantics,solve the problem of polysemy in the static word vector,and improve the semantic representation quality of the word vector. BiSRU module is used to extract high-dimensional emotional features of comment text,soft attention is used to calculate the impact of each word on the classification results,and the classification layer outputs the text emotional polarity. The experiment was conducted on the MOOC review dataset of real online courses. The results show that the classification F1 value of MacBERT BiSRU-AT model reaches 91.33%,higher than that of other models compared with the experiment,which proves the effectiveness of the model.

关 键 词：情感分析 MacBERT BiSRU 软注意力 

分 类 号：TN0[电子电信—物理电子学] TP391[自动化与计算机技术—计算机应用技术]