Research On Hybrid Recommendation Algorithm Based On XGBoost And SVD

With the rapid development of the information society in the new era,the explosive growth of information has brought new problems to people.Recommendation algorithm is one of the technologies that can effectively solve information overload.Traditional serial recommendation algorithms can no longer handle the current data scale.Matrix decomposition algorithm is one of the more popular recommendation algorithms.It can decompose high-dimensional matrices into low-dimensional matrices to improve the accuracy of recommendation results.There are also many shortcomings,such as single implicit feedback information,and inaccurate correlation between users and products.This paper analyzes the common problems and solutions of collaborative filtering algorithms in detail.Based on the problem of data sparsity and inaccurate recommendation accuracy,two optimized collaborative filtering recommendation hybrid models are proposed for research:(1)Design the first model is a hybrid recommendation model based on the improved clustering method and trust factor.First,the K-means clustering and Canopy clustering are mixed.The optimized clustering method can reduce the computational cost and solve the problem.The random k value affects the stability of clustering.In view of the problem of data sparseness,SVD technology is proposed to optimize the original matrix.After the matrix dimension is reduced,the implicit feedback eigenvector is used as a new parameter value,and the trust factor is introduced to optimize the similarity calculation method,which can improve the Recommended accuracy.The experiments are carried out on the Taobao Double Eleven data set and the Book-crossings data set.Through experimental comparison,the performance of the optimized model proposed in this paper is higher than other high-performance models,which confirms that the hybrid model proposed in this paper has high accuracy and effectiveness.(2)The second model is designed to be a collaborative filtering personalized recommendation model based on XGBoost and SVD.First,the characteristics of the original matrix optimized by SVD and the characteristics of XGBoost regular learning are combined to obtain significantly better recommendation performance than using a single algorithm.It can solve the problem of sparse data of the original matrix and the problem of low recommendation accuracy.For the user data set of tens of millions of levels,the introduction of the hybrid clustering method can better solve the problems of memory limitation and high computational cost.The experiments were carried out on the Taobao Double Eleven data set.Through experimental comparison,the performance of the optimization model proposed in this paper is higher than other high-performance models,which confirms that the recommendation model proposed in this paper has high accuracy and effectiveness.