This paper focuses on the data sparsity, performance and scalability problems in e-commerce personalized recommendation algorithms. This paper proposes a KNN-SVM algorithm and a collaborative chaotic filtering algorithm based on the above two improvements, taking into account the traditional similarity of collaborative filtering algorithms In the case of sparse data, the calculation is inaccurate. The structural similarity and traditional similarity are organically combined, and a combined similarity method is proposed, which better compensates for the inaccurate calculation of similarity in the case of sparse data. Based on the basic matrix factorization algorithm and the deviation-based matrix factorization algorithm, the user nearest neighbor model in the collaborative filtering algorithm is introduced into the matrix factorization model.