Dense retrieval method based on query expansion and delayed interaction

The invention discloses a dense retrieval method based on query expansion and delayed interaction, and belongs to the technical field of information retrieval. Comprising the following steps: data preprocessing: constructing a data set for query expansion; using a transform structure introduced with evaluation sorting, and combining the comparison loss to expand the query statement; generating a final representation of the query based on the query extension statement, and performing sentence-level division on the document to generate a multi-view representation; and based on a context delay interaction framework, performing fine-grained interaction on queries and documents, and sorting the documents according to the correlation scores. According to the dense retrieval method, an end-to-end query expansion mode is introduced, multi-view representation is provided for queries and documents, and a delay interaction strategy is adopted to calculate correlation scores of the queries and the documents. According to