Code annotation generation method based on dense retrieval

The invention discloses a code annotation generation method based on dense retrieval and relation modeling, which uses dense retrieval to retrieve related annotations as additional inputs, and extracts the relation between codes and the additional annotation inputs, thereby improving the annotation generation quality. According to the method, a retriever is mainly used for retrieving related annotations of input codes in a retrieval set, the codes and the retrieved annotations are spliced and then input into a discriminator, a fusion representation vector, a relation representation vector and relation classification are obtained, the fusion representation vector and the relation representation vector are input into a second multi-layer decoder, and code annotations are generated. According to the method, the relation between the additional input and the original input is considered for the first time, the generator can be effectively prevented from being misled, and then the generation effect is improved. 本发明