网络舆论中隐性词汇的情感意涵研究

Sentimental Connotations of Implicit Words in Public Opinion Online

网络舆论研究大多关注显性情感,对隐性情感的分析较为缺乏。本研究使用加权词向量方法和双向长短时记忆网络(Bi-LSTM)方法进行情感多分类研究,进一步挖掘网络舆论中隐性词汇的情感意涵。研究首先以词频-逆文档频率(TF-IDF)理论思想为基础计算词向量的情感权重,然后对词向量进行情感加权,并将加权后的词向量输入Bi-LSTM模型进行特征计算,最后使用Softmax函数完成情感的多分类预测。同时通过回溯的方式,对情感权重值较高的代表性词汇(隐性词汇)进行情感意涵分析。结果表明,该方法不仅能够有效提升多类情感判别的准确率,还可以通过这些代表性词汇加深对隐性词汇中情感意涵的认识和把握,这对于我国这一具有自我审查文化传统的社会及其舆论的把握具有重要意义。

On the research of public opinion,most studies focus on the explicit emotion,while little attention has been paid to the sentimental analysis of implicit words.The current study attempts to explore the various sentimental connotations of implicit words through the weighted word vector method and LSTM method.Firstly,the emotion weight of the word vector was calculated based on TF-IDF theory,then the emotion weight was carried out on the word vector.Secondly,the weighted word vector was input into the Bi-LSTM model for feature calculation.Finally,softmax function was used to complete the multi-classification prediction of emotion.Meanwhile,the sentimental meanings are illustrated through dating back to the implicit“representative”words with high emotional weight.The results show that the method proposed in this study can effectively improve the accuracy of multiemotion prediction.More importantly,these“representative”words are beneficial to deepen the understanding and grasp of the emotional meanings of implicit words,which is of great significance to the public opinion in the context of self-censorship in China.