The invention discloses a bipartite graph recommendation method based on a key user and a time context, which belongs to the technical field of personalized intelligent recommendation. The method comprises steps of acquisition of feedback data of a user to a goods, extraction of a key user collection, building of an interest preference neighbor collection of the user, material resource diffusion on a cut user-goods bipartite graph and final recommendation. By adopting the method, the key user group that plays a leading role in the recommendation system is mined, an interest nearest neighbor collection C for the target user is found out in the group, the bipartite graph is cut according to the collection C, nodes and edges in no relation or weak relation with the target user are removed, the calculation complexity is thus reduced, and the real-time performance of the recommendation algorithm is ensured. Besides, during the material diffusion process in the second step, a user score timedecay function is introduced, the different contribution degrees of different time scores to the recommendation results are reflected, and the algorithm recommendation accuracy is thus improved.