融合自注意力机制和语义词典的危机情绪分类研究  

Crisis Emotion Classification Combining Self-Attention and Semantic Dictionary

作　　者：由丽萍[1] 刘越 王世兴 You Liping

机构地区：[1]山西大学经济与管理学院,山西太原030006 [2]山西医科大学汾阳学院,山西汾阳032200

出　　处：《情报理论与实践》2022年第5期189-195,共7页Information Studies:Theory & Application

基　　金：国家自然科学基金面上项目“面向汉语篇章语义分析的框架推理技术研究”的成果,项目编号:61772324。

摘　　要：[目的/意义]在社会危机事件发生后,及时、有效地对社交网络上的文本内容进行情绪分类,有助于准确掌握公众情绪状态、优化社会危机事件管控方案。[方法/过程]基于框架语义理论构建危机情绪分类词典和危机情绪类别体系,采用融合自注意力机制的LSTM神经网络分类模型,实现优化的、细粒度的危机情绪分类。[结果/结论]以微博危机事件评论数据为例,通过不同参数的组合以及模型对比实验,获得了较高的准确率,验证了模型的可行性和有效性。为社交网络文本危机情绪分类提供优化的理论模型和方法支持,同时为相关领域的研究提供语义资源。[Purpose/significance]After the occurrence of social crisis events,timely and effective emotion classification of text content on social networks is helpful to accurately grasp the public emotional state and optimize the control scheme of the social crisis events.[Method/process]The crisis emotion classification dictionary and the crisis emotion category system are constructed based on the Frame Semantic theory,and the classification model combining LSTM neural network with self-attention is adopted to realize the optimized and fine-grained crisis emotion classification.[Result/conclusion]Taking Weibo crisis comment data as an example,this paper captured the data of Weibo comments and obtained high accuracy through the combination of different parameters and model comparison,which verified the feasibility and effectiveness of the model.This paper provides optimized theoretical model and method support for crisis emotion classification in social network texts,and meanwhile provides a semantic resource for crisis emotion research in related fields.

关 键 词：危机情绪 长短期记忆网络 自注意力机制 语义词典 情绪分类 

分 类 号：B842[哲学宗教—基础心理学] TP183[哲学宗教—心理学] TP391.1[自动化与计算机技术—控制理论与控制工程]