In order to address the challenges posed by multifaceted risk factors in air traffic control operations, a comprehensive analysis of unsafe operational incident reports was performed to extract risk-related information and identify underlying patterns. The latent Dirichlet allocation (LDA) model was utilized to uncover key risk topics and associated keywords, and the evolutionary relationships among different risk themes were systematically analyzed. A semantic network for the civil aviation air traffic control domain was constructed using the bidirectional encoder representation from Transformers(BERT) model to examine the interconnections and potential dependencies among risk topics. This network provides a theoretical foundation for quantifying the association between keywords. The findings indicate that the proposed approach enhances the digital representation of safety risks in air traffic control operations. It is concluded that the results offer valuable insights for advancing risk assessment and mitigation strategies in civil aviation air traffic control systems. The relevant research results can better mine air traffic control unsafe information and lay a foundation for accurately perceiving air traffic control operations risks.