Research On Accurate Hybrid Recommendation Method Based On Feature Engineering And Efficient Gradient Boosting Decision Tree Algorithm

With the rapid development of Internet and information technology,the amount of information on the Internet grows exponentially.In order to solve the problem of information overload and meet the information needs of users with different interests,areas of concern,behavior habits,consumption level and personal growth experience,collaborative filtering recommendation system has emerged.However,in the context of the exponential growth of Internet data volume,the data sparsity of collaborative filtering recommendation system is becoming more and more serious.In the high-dimensional item space,there are few items that users can access and grade,which also leads to the problem of extremely sparse scoring matrix.In order to solve the problem of sparse scoring matrix,this thesis proposes a hybrid recommendation method based on user behavior data,user and item attribute data from the perspective of Feature Engineering and recommendation methods.The application scenario in which telecommunication operators recommend package to users is discussed in detail.Firstly,the features from different sources are regularly combined to form a second combination feature.The word2 vec method is used to enhance the features based on the time series data of user behavior data and to build the relationship between users with the same consumption habits.Then,large-scale user data sets are modeled based on the efficient gradient boosting decision tree algorithm.In order to recommend personalized packages to users,feature importance and probability distribution are used to filter packages.Finally,based on the user multi-source data released by Unicom operators,a hybrid recommendation method is constructed to predict the user's interest package and complete the personalized package recommendation.The experimental results show that the data set based on feature combination and feature enhancement can significantly improve the accuracy of efficient gradient boosting decision tree model in predicting user selection packages.In addition,the hybrid recommendation method based on Feature Engineering and efficient gradient boosting decision tree model can effectively predict the categories of packages selected by large-scale users and the prediction accuracy is high.This proves the effectiveness and adaptability of the hybrid recommendation method based on Feature Engineering and efficient gradient boosting decision tree model.