A page resource caching method and apparatus, a device, and a medium, relating to the technical field of artificial intelligence. The method comprises: obtaining a page loading completion signal for an i-th accessed page, the page loading completion signal carrying a user identifier to be predicted and an i-th accessed page address real value (S1); in response to the page loading completion signal, obtaining a target user portrait according to the user identifier to be predicted (S2); using a first preset page prediction model to predict a (i+1)-th accessed page address according to the i-th accessed page address real value and the target user portrait, to obtain a (i+1)-th accessed page address predicted value (S3); searching in a local cache according to the (i+1)-th accessed page address predicted value (S4); when a cache result is not found, obtaining a page resource from a server according to the (i+1)-th accessed page address predicted value, to obtain a page resource to be cached (S5); and storing the page resource to be cached in the local cache (S6). Therefore, a page resource that is likely to be accessed next time by a user is automatically cached to a local cache on the premise of no perception by the user, thereby increasing the speed of loading of an unopened page.