摘要：AbstractIt is important to identify the personality of ancient people, which may help us understand more about culture change. The traditional method, self-report, is difficult to conduct research on ancient Chinese historical individuals. With the development of artificial intelligence, Transfer Learning provides a new way to identify the personality of ancient people. In this paper, we proposed Unsupervised Entropy Regression Domain Adaptation(UERDA) and Semi-supervised Entropy Regression Domain Adaptation (SERDA) based on Transfer Learning. We used the micro-blog data on Sina Weibo and autobiography data of ancient Chinese individuals for distribution adaptation, to train the personality prediction model and predicted the big-five personality traits of ancient individuals. For joint distribution adaptation, UERDA uses information entropy to realize unsupervised domain adaptation based on the regression task. SERDA is a semi-supervised method based on UERDA to improve the accuracy of the Transfer Learning model in the regression task. Compared with traditional algorithm PCA, RMSE of UERDA has decreased by 2.23 on average, and RMSE of SERDA has decreased by 3.16 on average. The experimental results showed that Transfer Learning achieved good performance in terms of the prediction of the big five personality traits of ancient people, which provides a new way for research on ancient people and culture change.
关键词：KeywordsCognitive controlpsychobiographybig five personalitytransfer learningdomain adaptationunsupervised learningsemi-supervised learninginformation entropy
