Representation Learning And Applications Based On Mobile User Visiting Sequences

With the rapid development of mobile devices and wireless technologies,mobile Internet plays an essential role for delivering networked services in our daily life.Thus,understanding of mobile applications and user behavior is important for better networked services provisioning.In this paper,we propose a general knowledge mining method based on the real mobile network traffic.It uses the representation learning that based on neural network to learn the representation vectors of domains and users.And we apply the representation learning result to many different machine learning scenarios.The main contributions of this paper are as follows:Based on large-scale mobile DPI traffic,we use distributed computing technology to extract the user visiting sequences.The vector representation learnings of domains and users is based on three-layer neural network that is to predict the surrounding domains,and apply it to various machine learning tasks,such as clustering and classification.For the domain vector that we learned from the DPI data,we apply domain vectors to multi-class classification task of mobile servers,and validate the performance of the method.The accuracy of classify different companies' domain can measure up 93%,different business category's domain can reach 85%.Furthermore,we analyze the business interest relations between domains by measuring the similarity of domain vectors.For the user vectors that we learned from the DPI data,we apply user vectors to user clustering,and we analyzed the user clusters' business interest results in detail based on the artificial knowledge rules of.In addition,we extracted some feature based on the user representation vectors,then we use the isolated forest algorithm to detect the anomaly user.The experiments validated the effectiveness of our method.