한국 사회 실시간 감정지도

Modern society has achieved high economic growth and quantitative expansion. However, in the process, many conflicts have arisen among regions, classes, genders, and age groups. The imbalance in the distribution of resources and opportunities and the weakening of social trust due to intensified competition have caused various social problems such as political polarization, excessive competition in private education, suicide, low birthrate, and hate crimes.

The 'Research Group for Social Emotions and Potential Conflicts' of the Center for Computational Social Science conducted a convergence research of computer science and social science that uses big data to collect, discriminate, visualize, and analyze potential social anxiety and social conflict factors expressed in the form of various emotions. As a result, it established the ‘Sentimap’(http://sentimap.hanyang.ac.kr), a spatio-temporal emotion distribution analysis and visualization system that reveals the emotions in Korean society, and opened related services. In the Emotion Map, seven emotions, namely: anger, hate, fear, joy, surprise, and sadness, centering on anxiety - can be seen through the depth of each color.

The Emotion Map analyzes the distribution of various emotions of members of local communities across the country in depth by time and region, and visualizes the results in space and time. The deep learning technology of computer science was used instead of the limited methods such as surveys mainly used in previous demographic studies. The research team judged that artificial intelligence technology which can do an in-depth analysis of large-scale data would provide an optimal solution for analyzing complex social problems from various angles, and devised a method to use social media to analyze the emotional distribution of members of society in space and time. We selected Twitter, which has the most popular user base and at the same time includes both time and spatial information, and collected tweet messages posted across the country in real time.

In particular, we created emotion classification technology using the latest deep learning language model BERT, which overcomes the limitations of the classical probability-based method (Naïve Bayesian classification). In addition, we improved emotion classification performance by utilizing the Attention Mechanism that reflects word association information, and developed an emotion index calculation model to effectively quantify the distribution of emotions.

The contents of the research were recognized for their excellence both internally and externally, by being published in Social Science Computer Review, an SSCI/SCIE journal in the top 5%, and receiving the top prize in the Communication and Technology (CAT) category at the ICA academic conference, the most prestigious conference in the field of communication.

The significance of this study is that it successfully developed an emotion classification model and real-time emotion distribution visualization tool through convergence research, and demonstrated that it can be useful for the study of conflict and anxiety of actual members of society by applying to the interpretation of various social problems. For example, in 2021, it has successfully proved the correlation between the COVID-19 pandemic and the resulting distribution of anxiety among members of society. By effectively analyzing various social issues and problems inherent in each time and region, and the potential psychological state of members of society, it is expected to be practically used for policymaking, decision-making, and analyses of social phenomena.