A Corpus-based Statistical Study of Chinese Online Hate Speech

This study systematically analyzes Chinese online hate speech based on a large-scale corpus. Using the Log-Likelihood Ratio (LL), we extract keyword features across three domains: gender, race/nationality, and region. By integrating Keyword-in-Context (KWIC) analysis, the study examines the contextual characteristics and socio-cultural implications of hate speech. The findings reveal that gender-related hate speech reinforces gender antagonism through keywords such as zhinanai (misogynistic straight men) and lücha (cunning women). Racial hate speech exhibits a strong collectivist tendency, primarily addressing discrimination against Chinese people by foreign groups. Regional hate speech is shaped by economic, cultural, and political factors, with long-standing stereotypes influencing different patterns of expression. This study employs a keyword analysis approach based on statistical hypothesis test, overcoming the limitations of traditional subjective keyword identification, improving research reliability and consistency, and providing a scientific foundation for online hate speech regulation.