Korean twitter emotion classification and application

Abstract

Twitter is one of the most popular social network services analyzed in the field of computer science. Especially, it is an interesting source to investigate people’s opinions in the context of a variety of topics and situations. In particular, many recent studies try to investigate social media during the crisis situations that range from natural disasters to man-made conflicts. In this dissertation, we present an emotion analysis method that classifies fine-grained emotions in Korean Twitter posts. Using the proposed method, we investigate people's emotional responses expressed on Twitter during the 2015 Middle East Respiratory Syndrome (MERS) outbreak in South Korea. In comparison to active research going on with tweets written in English, emotions expressed in Korean Twitter messages have not been studied in depth. Consequently, it is difficult to find related Korean microblogging studies as well as datasets or resources that are publically available. In addition, elaborate emotion lexicons are important to distinguish fine-grained emotions, but existing lexicon sets are mostly in English whereas building such lexicons is known to be extremely labor-intensive or resource-intensive. In this research, we constructed Korean Twitter Emotion Analysis (KTEA) dataset annotated by emotions as well as various resources that are particularly attuned to Korean Twitter domain. Moreover, we present a method that automatically builds fine-grained emotion lexicon sets from the annotated dataset. Based on the constructed dataset and resources, we suggest Korean-specific and Twitter-specific features as well as an effective classification method using machine learning algorithms that classify Korean Twitter messages into fine-grained emotion categories. Moreover, we conduct a case study of how Korean Twitter users responded to MERS outbreak using our proposed emotion analysis method. Analysis results on tweets related to MERS outbreak help to understand the behaviors of humans and the characteristics of sociocultural system. We believe our method can be further harnessed by the media to automatically investigate public opinions as well as the authorities to gain insights for quickly deciding the assistance policies.