Research On Recommendation Algorithm Based On Gradient Boosting Decision Tree And Deep Belief Network

With the continuous development of science and technology and the Internet,the total amount of global information is explosive,people enter the information age,mass data continue to produce,information overload problem is more and more serious.The number of these data information increases the difficulty of people to obtain the necessary information,how to make full use of these massive data information,screen out the "garbage" information,analysis and excavation of the information people need to become a hot topic,The emergence of the algorithm effectively addresses this situation.Recommended algorithm through the analysis of massive sparse data processing,learning to the user's hobbies and behavior habits,dig out the user's possible information needs,the resulting prediction results recommended to the user.For the recommended algorithm,the greater the amount of data it processes,the more likely it is to learn the potential links in the data,collect information that is of interest to the user,and more accurately recommend the information of the user's needs,so the recommended algorithm It is a hot topic for people to study in the information age with rapid data growth.Recommended algorithm research,the massive information extracted in the effective information recommended to the needs of such information users to meet the information needs of different users,saving users the time required to search information to improve the utilization of information.Recommended algorithm is widely used,one of the most economic value of the application areas is the field of e-commerce,the subject of research in the e-commerce scenarios recommended algorithm.In the age of big data,e-commerce scenarios,the amount of data is huge,the data attribute dimension is high,the traditional recommendation algorithm can not effectively deal with such data,the recommended effect is limited.The algorithm of gradient lifting decision tree can effectively deal with the feature data of large number and high attribute dimension,and it has better application in search sorting and advertising click rate.The data in this paper is also applicable to the gradient promotion decision tree algorithm.The gradient decision tree algorithm is a popular machine learning algorithm,which can deal with a large amount of sparse data,learn the potential links in the data,and generate higher accuracy recommendation results than the traditional recommendation algorithm.However,in the practical application,when the data feature is too large,the learning effect of the decision tree is improved.We introduce the deep belief network to solve the problem that the data feature is too big.The deep belief network is the main algorithm in the field of deep learning,which can be used to identify the characteristics and generate the characteristic data.By using the advantages of processing the data feature,the feature data which is useless to the recommendation result is screened out in the feature selection work,Introducing the gradient to improve the decision tree model learning,can effectively improve the recommendation accuracy rate.By using different algorithm models,we can use the single algorithm to deal with the complementary advantages of different data of different problems,and solve the complicated problems in large-scale data volume and high-dimensional feature selection process.In this paper,a recommendation algorithm based on gradient enhancement decision tree and deep belief network is researched.The main work is as follows:1.A feature set construction algorithm based on deep belief network is proposed.First,in order to prevent the original data in the training process has appeared in the fitting or less fit to do the data preparation,including data description,data analysis and data preprocessing.Secondly,according to the results of data preprocessing and statistical analysis,combined with the factors influencing the recommended results in real life,the feature set of basic category and cross combination category is generated from the twelve major angles.Then,in the process of selecting high-dimensional data collection feature selection,the deep belief network model of deep learning domain is introduced to solve the feature selection and high attribute dimension problem of feature set.By training the weights between neurons in the model,the whole model network generates the training feature data according to the maximum probability,thus reducing the computational complexity of the training algorithm.This method of deep belief network is introduced in the construction phase of the feature model,which plays an important role in improving the recommendation accuracy and recommendation effect of the recommendation system.2.The decision tree algorithm based on the idea of gradient lifting is proposed.Firstly,starting from the core algorithm of the recommended algorithm,the decision tree is used as the basic learner,and the idea of the gradient is introduced to train a large number of basic learning models through iterative way.We will integrate these trained basic learners to realize the weak learning The combination becomes a strong learner,which improves the generalization ability and recommendation effect of the recommended algorithm.By using the deep belief network model,the original data is combined with the actual rules to construct the feature set,and these feature sets are used to train the optimization gradient decision tree to construct the final recommendation algorithm model.3.Recommendation algorithm based on gradient enhancement decision tree and deep belief network.Firstly,the primary algorithm model of deep belief network is constructed.The feature set is extracted and the final feature set is selected by using the deep belief network model.Secondly,according to the actual business scenarios,generate all the relevant rules that users and commodities may exist,take the time axis as the baseline and refer to the time forgetting rule,divide the training set,test set and verification set.Finally,the feature set based on the deep belief network model is used to fully train the gradient lifting decision tree.Combining with the verification set of the model,the parameters of the model are validated and the parameters of the two models are adjusted to realize the fusion of the algorithm model to generate the final recommendation model.Based on the real user-commodity behavior data of Alibaba Mobile e-commerce platform,a large number of experiments were carried out.By implementing the gradient enhancement decision tree and the deep belief network,the stochastic forest algorithm with the single recommendation algorithm,the gradient lifting decision tree algorithm And the logistic regression algorithm,it is concluded that the gradient lifting decision tree can quickly learn the potential links in the data and generate the recommended results in the data processing a large number of sparse data and characteristic attributes.At the same time,in order to filter a large number of feature data lifting algorithms in the data set Efficiency,to avoid the impact of too much feature data on the recommended effect,introduce the deep belief network algorithm,through its training to reduce the number of features.Through the fusion of these two recommended algorithms and experimental comparison verification,summed up the convergence of the recommended algorithm in the e-commerce scene has a high recommended effect,and achieved good performance.