The application discloses a few-sample multi-
modal knowledge graph completion
system and method based on a large
language model, the completion
system comprises a data preprocessing module, a relationship prior acquisition module, a relationship topology calibration module, a relationship
level structure reasoning module, a multi-
modal information regulation and fusion module, an entity structure reasoning module and a prediction output module, and the completion method comprises the following steps: S1, data preprocessing; S2, constructing a prompt word and guiding LLM prediction; S3, relationship graph construction and calibration; S4, relationship
level structure reasoning; S5, multi-
modal regulation and fusion; S6, entity
level structure reasoning; S7, prediction output; and S8, model pre-training and reasoning. The application solves the problems of relationship context information scarcity and multi-modal
noise interference in multi-modal few-sample connection prediction, efficiently utilizes multi-modal LLM prior knowledge, reasonably performs multi-modal information and structure information interaction, and simultaneously realizes accurate few-sample connection prediction.