Abstract
Research on understanding and processing human emotions using artificial intelligence technology is active. Understanding emotions is a more challenging task than understanding language, and in the case of text, it is crucial to identify emotions based on context. In this regard, the latest language models learn not only emotional vocabulary but also the relationships with surrounding words, so the learning patterns may vary depending on what the emotional dataset addresses.
This study performs text mining based on Named-Entity Recognition on emotion datasets for AI training and analyzes which entity types and vocabularies appear in emotional expressions. It examines the types of entities that appear in different styles of text, such as empathetic conversations, literary works, and social media comments, considering the implications for the utilization and construction of emotion datasets. This study provides analytical data of AI training datasets from the perspective of affective computing. Additionally, I expect that the Named Entity Recognition analysis methods used in this study will provide methodological assistance to text mining researchers. |