Commodity recommendation scene-oriented user personalized model construction method

The invention discloses a commodity recommendation scene-oriented user personalized model construction method, which comprises the following steps of: (1) collecting model parameters of all users by a server, and searching K nearest neighbor users for each user by utilizing a K nearest neighbor (KNN) algorithm so as to realize rough matching of a personalized model; (2) each user carries out more detailed matching according to a rough matching result of the server in combination with local data and model parameters of the user so as to obtain a user model parameter set which is more matched with the features of the user; and (3) each user constructs a personalized sub-model by aggregating the fine-grained model parameters, so that the problem of model overfitting caused by insufficient local data volume when the user uses local data to train the model is effectively solved. According to the method, the problem of insufficient data volume when a user locally trains a personalized model is solved, and the accur