This application provides a method, apparatus, medium, and device for retrieving
multimodal data. The method includes: responding to receiving a query instruction indicating that input data belongs to the
multimodal data, encoding the input data to obtain a query vector corresponding to the query instruction; using the query vector to perform a vector search in a preset vector
database to obtain a set of candidate vectors; and based on each candidate vector in the set of candidate vectors, determining a preset entity corresponding to each candidate vector and associated data of the preset entity corresponding to each candidate vector from a preset
knowledge graph to obtain a set of candidate associated data, thereby determining the query result corresponding to the query instruction. This application solves the technical problem of low accuracy in
multimodal data retrieval in related technologies, achieving the effect of improving the accuracy of multimodal
data retrieval.