The invention discloses a patent recommendation method based on a knowledge graph and Transformer. The method comprises the following steps: crawling a patent resource library to construct the knowledge graph; mining a knowledge graph through a hybrid model of a graph convolution network and an attention mechanism to obtain content feature representation vectors of users and patents; mining user information through a Transformer model to obtain a sequence feature vector of historical preferences of the user; cascading and combining the content feature representation vector and the sequence feature vector, and inputting into a Softmax layer of a Transformer model for calculation to obtain recommended probability values of a plurality of candidate patents; and carrying out Top-k sorting on the plurality of probability values to obtain Top-k patents as recommendation results of the target user. According to the method, knowledge graph rich feature representation is adopted, Transformer isadopted to mine behavior sequence features, and the accuracy and interpretability of recommendation results are improved.