The invention belongs to the technical field of
data management and knowledge mapping, and particularly relates to a
metadata modeling management method for table data,
JSON (
JavaScript Object Notation), graph data, text, picture and video data, which specifically comprises the following steps of: respectively constructing nodes and directed edges to generate initial sub-graphs for each of the table data
JSON, the graph data, the text, the picture and the video data; according to the embedding similarity, utilizing a large
language model to merge nodes and edges representing the same entity or relationship to realize clustering, and generating a
metadata graph; receiving a
natural language query of a user, analyzing the query into a target node or edge through a
language model, and generating a corresponding embedded vector; calculating the embedding similarity between the embedding vector and nodes and edges in the graph according to the embedding vector, and screening related nodes and edges based on the embedding similarity to form a candidate set; and setting budget limitation, scoring candidate nodes and edges, selecting a group with the highest
score to form a sub-graph, and returning
metadata and relationships related to query in the sub-graph.