As the most widely used recommendation technology in the recommendation system, collaborative filtering faces severe problems such as data sparsity, real-time, scalability and other issues, affecting the final recommendation effect. If only the K-means clustering is used to improve the recommendation, there will be no obvious optimization effect due to defects such as local optimality. To solve the corresponding problem, this paper firstly uses a new metaheuristic algorithm called Cuckoo Search (CS algorithm) to optimize the K-means algorithm, and then clusters the users and items in the data set with the optimized algorithm as well as weighted fusion. Finally, get a list of recommendations. The experimental results show that the recommendation process proposed in this paper effectively improves the data sparsity and scalability, moreover, the recommendation efficiency and quality are significantly improved.