Commodity recommendation model training method

The embodiment of the invention provides a commodity recommendation model training method. The method comprises the following steps: associating attributes of a user and a recommended commodity to obtain a sparse vector; performing embedded coding on category variables in the sparse vectors to obtain embedded vectors of the features; calculating an attention score of each pair of feature interaction in the mapping embedded vector; wherein the attention score comprises the attention score of pairwise interaction between each pair of features and the attention score of each pair of features to all feature domains; and performing full connection on the representation of the feature domain, predicting to obtain a target recommended commodity of the user, and training to complete a commodity recommendation model after the loss function converges. According to the embodiment of the invention, the calculation of the attention score is divided into two parts, one part is the pairwise interaction between the features,