结合主题模型的中国古代诗人大五人格预测  

Combining Topic Model for Ancient Chinese Poets Big Five Personality Traits Analysis

作　　者：闫滢钰 汶东震 张冬瑜[1] 林鸿飞[1] YAN Yingyu;WEN Dongzhen;ZHANG Dongyu;LIN Hongfei(Computer Science and Technology,Dalian University of Technology,Dalian 116024,China)

机构地区：[1]大连理工大学计算机科学与技术学院,辽宁大连116024

出　　处：《山西大学学报（自然科学版）》2023年第3期546-556,共11页Journal of Shanxi University(Natural Science Edition)

基　　金：国家自然科学基金(62076046,62076051)。

摘　　要：诗歌作为一种重要的文学体裁体现着古代诗人的人格特征。目前研究主要基于现代语言的数据集进行人格分析,缺少对于古人的相关分析任务,影响数字人文领域研究进程和古代诗人的画像构建。因此,本文开展对于古代诗人进行人格特征分析的研究,并以大五人格理论为标注标准,构建了一个针对古代诗人的大五人格数据集,该数据集包括了581位唐宋两代著名诗人,参考现有的文学评论,对其宜人性、外倾性、神经质、开放性和尽责性进行标注。在此数据集上将语言模型与深度学习模型相融合,基于交叉熵损失函数进行人格特征等级的学习,提出了基于主题增强的大五人格特征预测模型。实验结果表明,准确率达到了0.71,证明所提出的数据集和模型对古代诗人人格特征分析和建模研究有着良好的促进效果。As an important literary genre,poetry embodies the personality characteristics of ancient poets.At present,the research is mainly based on the data set of modern language for personality analysis.The lack of relevant analysis tasks for the ancients has affected the research process in the field of digital humanities and hindered the construction of portraits of ancient poets.This paper proposes the task of analyzing the personality characteristics of ancient poets.Based on the Big Five personality theory,we constructed a Big Five personality dataset for ancient poets,including 581 famous poets of Tang and Song dynasties,and labeled their agreeableness,extraversion,neuroticism,openness and conscientiousness.In order to test the validity of the dataset,a variety of common machine learning and deep learning models were tested on the dataset,and a Big Five personality trait prediction model was proposed combining the theme model.The experimental results showed that the accuracy of the dataset in the final model reaches 0.71,which proved that the proposed dataset has a good promotion effect on the analysis and modeling of the personality characteristics of ancient poets.

关 键 词：大五人格 古代诗人 多任务学习 隐式狄利克雷分布 

分 类 号：O436[机械工程—光学工程]