Deconstructing "Japanese-ness": Analyzing Perceptions of Japan's Ethnic Minorities on Twitter Using Natural Language Processing

Abstract

This study analyzes Japanese tweets via natural language processing techniques, in order to provide insight into current perceptions of ethnic minorities in Japan. While existing work has studied different topics on Japanese Twitter or researched online racism in other countries, most studies rely on manual labeling and purely qualitative analysis, and none have studied online racial discrimination in a Japanese context. Drawing on the existing sociological literature, this research fills this gap by introducing a rigorous quantitative component to the study of Japan’s minorities, applying Word2Vec semantic vector similarity analysis and Latent Dirichlet Allocation (LDA) analysis to sixteen total tweet datasets. By studying changes over time in common topics and keywords associated with different minority groups, this work compares and contrasts discourse pertaining to Zainichi Koreans, native Ainu and Okinawan populations, Hāfu, and Southeast Asian migrants, in order to develop a broadened understanding of racial discrimination and attitudes toward minorities in Japan today. The data suggest that the conversation surrounding a minority group generally reflects a level of awareness of social issues that corresponds to the group’s history of advocacy efforts and established presence in Japan.