The invention discloses an adaptive customized recommendation method based on users and articles. The method comprises two stages of training and customized recommendation. For the training stage, firstly, data including the user personal information, user behavior characteristics and object evaluation of the users is acquired through a platform; similar users are clustered according to the user data, a mean difference matrix of the object evaluation of the users is calculated, a prediction model based on user clustering is established, and an evaluation prediction error of the model for all the objects is calculated; similarities among objects are calculated according to attributes of the objects, mean object evaluation difference of the users is calculated, a prediction model is established, an adaptive prediction model based on the users and the objects is formed. For the customized recommendation stage, firstly, user attribute clustering is determined, the adaptive prediction model integrated with the users and the objects is utilized, evaluation of the users for the objects is predicted, and the objects with high prediction evaluation are recommended to the users. The method is advantaged in that the method has adaptive capability and higher accuracy compared with a traditional customized recommendation method.