Research And Implementation Of User Portrait Algorithms Based On Personal Data

With the development of the Internet,the scale of users and the data generated by the Internet have increased exponentially,and the phenomenon of "information overload" has also emerged.How to extract useful information from a large amount of data and recommend content of interest to users has become a problem faced by major Internet companies.Recommendation system can solve these problems very well.User portrait is the basis of recommendation system.Through comprehensive analysis of user attributes,behavior and other information,users can be classified,which can provide better services for users and better prevent users from losing.A good user profile can improve the performance of recommendation system.The main task of this paper is to portray the user's attributes according to the search terms of the user's history of one month.In view of the shortcomings of the traditional user portrait model in generalization and accuracy,this paper proposes an improved user portrait model.In this paper,a two-level user portrait model based on fusion algorithm is proposed.The first-level model mainly realizes the function of the relationship between users and search terms at different levels.The second-level model combines the relationship between user attributes and search terms at different levels by using fusion algorithm to construct the final mapping model between search terms and users.The main research work of this paper is as follows:(1)A new SVM_EM algorithm is proposed.Aiming at the problem of scarcity of manual labeling data sets,this paper combines SVM with EM,uses a small number of manual labeled data sets and some unlabeled data sets to train classifiers,and reduces the cost of manual labeling.Experiments show that the classification accuracy of the combined algorithm is higher than that of the traditional SVM algorithm.(2)A new two-level user portrait model is proposed.In the first level model,SVM_EM features are combined to learn the differences between users' words,Doc2 Vec is used to learn the relationship between semantics,and deep neural network is used to learn the deep relationship between semantics.In the second level model,fusion algorithm is used to mine the relationship between tags in depth,which improves the prediction accuracy of user attribute tags and the generalization ability of the model.Power.(3)Based on the user portrait model proposed in this paper,a user portrait prototype system is implemented and tested.The system performs well in throughput and request success rate.