Abstract: Hierarchical label text classification has always been one of the research hotspots in the field of natural language processing. Because of the hierarchical structure between labels of text data, it brings some challenges to traditional text classification methods. When the text is divided into multiple labels with hierarchical structure, because the relevant background knowledge is often needed, and the conventional classification method can not model the hierarchical relationship between the labels, a hierarchical label generation method is proposed. In this method, the relevant knowledge in the corpus was divided into multiple semantic clusters by constructing cluster clusters. The relevant semantic clusters were queried for the current text to enhance semantic information, and the hierarchical labels were generated successively by using the generation method. The experimental results on three datasets of WOS, BGC and AAPD show that the hierarchical label generation method based on fusion clusters can effectively improve the classification performance of hierarchical labels, and achieves the best result in the comparison experiment with the representative models selected in the past five years.