Abstract

In this study, we explore the possibility of studying emotions expressed in Korean literature by using emotion data based on Korean popular narratives. Accordingly, we introduce literature and content evaluation data collected by Chung-Ang University's Humanities Contents Research Institute, and utilize it to develop long short-term memory (LSTM) and Bidirectional Encoder Representations from Transformers (BERT) model trained to perform binary classification (positive/negative) for each data sample of several categories including “Literature,” “Content,” and ”Literature+Content.” A three-class LSTM 3 classification model was also constructed to distinguish anger, sadness, and joy. The two-class LSTM and BERT models showed an accuracy in predicting emotions of 80-87%, and predicted negative emotions more accurately than positive emotions. Based on its learning method, the BERT model exhibited higher prediction accuracy than the LSTM. For each emotion data sample, the “Literature+Content” model showed high predictive accuracy. These results are identified as related to the the total amount of data learned from the perspective of information engineering. However, from the perspective of the humanities, the results are identified as originating from the differences in emotional aspects according to the characteristics of the basic data, that is, the differences between novels and dramas. Subsequently, using the emotional deep learning model, the data for judging the emotions of The Cloud Dream of the Nine and the manual emotion input data of The Cloud Dream of the Nine were compared and verified. Thus, this work shows how an emotional deep learning model can be used in literary research.