This invention discloses a
product diversity complementary recommendation method in the field of e-commerce. The method includes: inputting the product category
label of the target product into a pre-trained large
language model to determine the complementary relationship between the target product and candidate products; constructing a product complementarity graph by treating products as nodes and the complementary relationship between the target product and candidate products as edge information; establishing neighbor relationships for the target product based on the product complementarity graph, combining the feature vectors and edge information of neighbor nodes; learning the product embedding representation through a graph neural network based on the neighbor relationships to determine the product embedding vector; determining the set of neighbor nodes for the target product based on the similarity between product embedding vectors; weighting intra-class complementarity and inter-class complementarity to select multiple product neighbors from the target product's set of neighbor nodes; and performing a self-balancing re-
ranking of the comprehensive scores of the product neighbors through a constructed relevance-diversity weight mechanism to obtain the final recommendation result.