Research On Personalized Recommendation Based On Tree Model

With the rapid development of the Internet,the scale of information resources is growing rapidly,and information overload has become a major problem in society.Since the user has become more difficult to extract useful information from the mass information,the intelligent recommendation technology for individuals has emerged as the times require.In view of the advantages of personalized recommendation system,such as active service,high degree of personalization and user driven,recommendation technology has become a hot topic in academia and industry.Tree model is a prediction model based on rules to classify samples,with the further expansion of user volume and information resources,this thesis has carried out in-depth and careful research in view of the problems of feature dimension curse,positive and negative interaction sample imbalance and model overfitting,and the main achievements are as follows:(1)Based on the historical data of goods,a hierarchical commodity prediction model is proposed in this thesis.In order to realize the effective representation of the commodity data,a commodity characterization method based on temporal and spatial characteristics is proposed.To alleviate the problem of dimensionality curse,we construct a Bagging based hierarchical ensemble learning model,which solve the high deviation problem and improve the performance of the prediction model,the experimental results verify the validity of the proposed model in this thesis;(2)In view of the imbalance of positive and negative interaction samples in commodity data,a personalized debiased DART integration model is proposed in this thesis.In order to describe the personalized portrait of the user,the user’s personalized characterization is obtained by using the sparse coding method of the decision tree.Further,we build multiple base DART decision tree models by using down sampling,and generate strong learner by integrating method,avoiding overfitting problem and improving the prediction performance of the model;(3)For the Dropout method in the DART model,the importance of the decision tree is not fully taken into account,a DART model based on importance ranking is proposed in this thesis.By using the error rate of decision tree as the importance ranking rule,the direction of fitting residuals is improved,and the generalization ability of the model is improved.The prediction performance of the proposed model is verified by experiments.