Research On Recommendation System Based On Mixed Model

With the continuous development of Internet technology,information on the Internet is growing at an explosive speed,although people can get more and more the way of information,but also have to face the problem of information overload in vast amounts of information to find out the valuable information more and more difficult to solve this problem,the problem of information overload,recommendation system personalized came into being and has become a major way to solve the problem of information overload.Therefore,it is of great social value for the research of the recommendation system and the improvement of the performance of the recommended system.This paper first introduces the common algorithm of recommendation system,focusing on the matrix decomposition algorithm is studied,and the matrix decomposition algorithm is improved at the same time,and put forward the scheme of fusion fusion through the improved matrix decomposition algorithm and classification algorithm in machine learning,finally through the analysis and comparison of experimental verification scheme is helpful to recommend the effect of improving.Specifically,the main work of this paper includes the following three aspects:1)the matrix decomposition algorithm in the recommendation system is studied,which is usually influenced by the social factors of their friends in addition to the factors affected by their own interests when they are recommended in practice,because the matrix decomposition algorithm is usually considered only from the two aspects of the user and the project.Instead of considering the influence of social factors on scoring,Therefore,in this paper,we improve the SVD++ algorithm of the matrix decomposition algorithm,and add the neighborhood social factors into the model.Comparison experiments show that compared with the traditional matrix decomposition algorithm,the improved algorithm helps to improve the effectiveness of the recommendation and also alleviates the sparsity of the recommended system.2)compared with single algorithm recommendation,fusion of models can improve the recommended performance better.Therefore,this paper proposes a multi model fusion scheme for classification algorithms to produce the recommended results.The classificationalgorithms are the logical regression algorithm(LR),the Gradient Boosting Decision Tree algorithm(GBDT)and the random forest algorithm(RF).Through Gradient Boosting Decision Tree algorithm,the combination feature is trained as the feature of LR.After that,the model is fused through Stacking mode.The contrast experiment shows that the fusion of models can significantly improve the recommended performance.3)selected two data sets to carry out experimental comparison and analysis,the first experiment is to verify the matrix decomposition improvement algorithm,mainly through the data set of social relations have Epinions data set to carry out the experiment,by comparing the improved algorithm and the original matrix decomposition algorithm to compare the decrease of RMSE.The second experiment is a contrast experiment based on the fusion of classification algorithms in machine learning.The data set used is a Tmall data set,and the features are extracted by using the method of Feature Engineering for the data provided,and the model fusion is carried out in the way of Stacking.F1 and AUC are used as the evaluation index,and the experimental verification shows the phase.Compared with a single model,the F1 and AUC of multi model fusion is worth upgrading.