In the past few years, the growth of online social networks has led to an increasingly serious problem of cyberbullying, making its detection a pressing issue. In this study, we propose the Knowledge Enhancement Network for Cyberbullying Detection (KNCD), an approach that utilizes the Large Language Model (LLM) to extract global information from the labeled data to enhance the knowledge of individual samples. Specifically, KNCD extracts essential information from representative samples with the LLM. After text encoding, the K-Nearest Neighbor (KNN) approach is applied to align this extracted knowledge with the samples. By combining the global knowledge with the samples, the semantic information of the samples is expanded. Experimental results on tweet data show that our KNCD model achieves promising performance in cyberbullying detection.