The invention discloses a recurrent neural network method for improving the time sequence data mining capability of a recommendation model. The method includes the following steps that: step 1, time gate-based network transformation is performed on a conventional neural network; step 2, interaction interval time-added time sequence data are inputted into a model; step 3, the prediction values of the model on each item of the sequence are calculated; step 4, the loss value of the model is calculated, if the loss value is lower than a preset value and tends to be stable, step 6 is executed, otherwise, step 5 is executed; step 5, the gradient of each parameter is calculated according to the loss value, the parameters are updated, the method returns to the step 3; and step 6, the interest of auser is predicted based on the current model. With the method of the invention adopted, the time information mining capacity of the neural network in the recommendation field can be improved, and themodel can process long-term general features and short-term temporary features contained in long-term data more easily, and can perform even better in a personalized recommendation system.