The invention discloses an active content caching method based on federated learning, and belongs to the technical field of wireless communication. The method comprises steps that firstly, in each round of communication, a user downloads a global model, training is conducted locally through a stacked automatic encoder, and a local model and implicit features of the user and a file are obtained; secondly, in each round of communication, the user updates the model and sends the model to a server, and all the local models are aggregated to generate a global model; thirdly, after training is finished, the user sends the implicit features of the user and the file to the server, the server firstly calculates the user similarity and the file similarity, then a certain user is randomly selected, and a decoder of the stacked automatic encoder is used for recovering pseudo score matrixes of the user and the file; and finally, scores of the group of users on all the files are calculated by usingcollaborative filtering, and the file with the highest average score is selected for caching. On the premise of ensuring the cache hit rate, a problem of data sharing between neighbor users is effectively avoided, so private data of the user is safer.