基于机器学习的《乙酉家乘》情绪分析模型构建

• 作者： 张燕 ¹
• 作者单位：
  1.  河北师范大学-文学院
• 发布时间：2025-02-20

摘要: 《乙酉家乘》作为日记体散文，具有时空意识明确、内容真实客观、文字简洁凝练、信息丰富多样的特点。该文不仅体现出黄庭坚“以史笔为文”的创作观念，还承载着作者或隐或现、复杂幽微的私人情绪。筛选出能够反映作者情绪的152篇有效日记，反复三次阅读文本得到建模者主观判断作者情绪的结果。统计分析可知，三次判断结果均一致的数量占比高达87.5%，验证了选择文本的有效性和主观判断情绪的稳定性。将三次判断一致的133篇日记作为标签集数据，同时选择特征集数据，通过机器学习训练构建出情绪判断模型。将模型判断情绪的结果与主观判断情绪的结果进行对比，准确率达87.9%。证明机器学习可以实现对建模者文本阅读能力的模拟和复现。

Abstract: The Yiyou Jia Sheng as a diary-style prose possesses distinct spatio-temporal awareness, objective and true content, concise and condensed language, and rich and diverse information. This text not only reflects Huang Tingjian's creative concept of "writing prose with the style of historical records", but also carries the author's complex and subtle private emotions, which are either implicit or explicit. By screening out 152 valid diaries that can reflect the author's emotions and reading the text three times, the subjective judgment of the author's emotions by the modeler was obtained. Statistical analysis shows that the proportion of consistent results in the three judgments is as high as 87.5%, verifying the effectiveness of the selected text and the stability of the subjective judgment of emotions. Taking the 133 diaries with consistent judgments in the three times as the labeled set data, and simultaneously selecting the feature set data, a model for emotion judgment was constructed through machine learning. By comparing the results of emotion judgment by the model with those of subjective judgment, the accuracy rate reached 87.9%. It is proved that machine learning can simulate and reproduce the text reading ability of the modeler.

• 来自： 张燕
• 分类： 综合学科 >> 综合学科
• 稿件状态： 已投稿期刊
• 引用： PSSXiv:202502.01088 (或此版本 PSSXiv:202502.01088V1)
  DOI:10.12451/202502.01088
  CSTR:32012.36.PSSXiv.202502.01088
• 推荐引用方式： 张燕.基于机器学习的《乙酉家乘》情绪分析模型构建.哲学社会科学预印本平台:https://zsyyb.cn/abs/202502.01088.[PSSXiv:202502.01088V1] (点此复制)