The Research Of Hybrid Recommendation Algorithm Based On Matrix Decomposition

In the context of the "big data era",the recommendation system extract the user's historical preference data through analysis,infer the items that the target user may like and recommend them to the user in combination with the preference relationship between users and the similarity between projects.Because of its excellent scalability and easy implementation,matrix decomposition model is the most common and popular technology in model-based collaborative filtering recommendation algorithm.Single recommendation algorithms have many limitations,such as data sparsity and cold start.Therefore,this thesis studies a hybrid recommendation algorithm combining several basic recommendation algorithms.The main work of this thesis is as follows:Summarize the existing recommendation algorithms and relevant theories of matrix decomposition model.Aiming at the problem of sparse user scoring matrix,improve the model through K-means algorithm,cluster user data with K-means algorithm,and then reconstruct the scoring matrix according to the clustering results.Through this optimization,the dimension and sparsity of the matrix can be significantly reduced,The improved matrix decomposition model is obtained.Comparative experiments are carried out on Movie Lens dataset,and the performance of the algorithm in accuracy,recall,F-measure are analyzed.A label based matrix decomposition recommendation algorithm is proposed.The algorithm adds label features to the traditional implicit factorization model(LFM),decomposes the user item scoring matrix,and obtains the user matrix and item matrix.Then,according to the feature classification of the user's historical scoring data of items,the type of label is added as a new feature value.After experimental analysis,the recommendation accuracy can be effectively improved by using label data.In terms of alleviating the cold start problem and algorithm optimization,a hybrid recommendation algorithm integrating user and project content information and matrix decomposition is proposed,which integrates the user and project deviation into the matrix decomposition model,so as to reduce the number of learning parameters in the matrix decomposition model,improve the calculation efficiency,and further alleviate the cold start problem of users and projects.Through design experiments,it is verified that the performance of the hybrid recommendation algorithm is better than the traditional matrix decomposition model and the single recommendation algorithm.