Literary Emotion Classification Using a Emotion Lexicon

Recently, emotion analysis technology has been developed to classify emotions felt in texts such as sentences and documents, and it is being used in a variety of research, including literature. Because emotions are an important foundation for understanding literature, it is important to apply emotion analysis technology in literary research, and efforts are needed to improve problems that arise during the emotion analysis process for the development of technology. This study explains the problems that arise from existing labeling methods when constructing learning data, and introduces ‘emotion lexicon-based emotion labeling’ to improve these problems. To prove the effectiveness of emotion lexicon-based labeling, we constructed two learning data with different emotion labeling methods, entered them into a emotion analysis model, and compared the results. As a result, the accuracy of the model using emotion lexicon-based labeling was 91.95%, showing approximately 20% higher performance than the model using existing labeling. In addition, as a result of going beyond quantitative evaluation and directly checking the analysis results by inputting literary passages from the CSAT into the model, the model using emotion lexicon-based labeling showed better predictive power.