Research And Application Of Telecom Transaction Behavior Analysis And Risk Prediction Model Based On Big Data

With the development of telecom operators,the prevention and control of business risks has received attention.The common business such as phone bill recharge has the characteristics of small transaction amount and high demand frequency,which can easily be used by black and grey industries such as gambling for illegal operations.Therefore,it is very necessary to tap such black and gray industry users.However,traditional transaction systems still suffer from manual human processing and data lag,and their systematization has yet to be improved.The efficiency of handling transaction complaint information also varies depending on individuals,leading to a certain level of unpredictability.Furthermore,as data volume continues to grow,issues related to inefficient database storage and horizontal scalability have emerged.To address these problems,this thesis undertakes the following research and work:(1)Analyzing complaint data related to transactions from telecommunication operators,a complaint information text classification model is established.This model incorporates CNN(Convolutional Neural Networks)and RNN(Recurrent Neural Networks)for classifying transaction complaint information,enhancing the fault tolerance of the classification model.The combined decision probabilities of the two models are used for weighted discrimination,resulting in improved precision in complaint information text classification.The experimental results show that the average accuracy of discriminating three types of black involved in pornography,gambling and fraud reaches 98.02%.(2)Analyzing the payment transaction process and data of telecommunication operators,two transaction processes associated with underworld-related industries are identified,and corresponding models for underworld-related transaction behavior are established.A risk prediction model for telecommunication top-up transactions is developed based on payment and complaint record data,analyzing massive data to identify high-risk transaction features.Risk levels for transactions and user risks are predicted based on rule-based judgments and behavioral feature analysis,thereby further mitigating illegal transactions involving underworld-related activities.The experimental results show that the model has obvious advantages in discriminating underworld-related transactions and verifies the effectiveness and feasibility of the algorithm.(3)Designing and implementing a telecommunication transaction risk prevention and control system comprising the data access layer,data preprocessing layer,data storage layer,risk control model layer,data access layer,and application layer.The system utilizes actual transaction data flows to design a Hive data warehouse,partitioned at the hour level,and adopts columnar storage for transaction data.A hybrid architecture combining Mongo DB replication sets and sharding is employed to store and analyze highly aggregated data,providing vital support and security for data reporting and calculations related to telecommunication top-up transaction risk prediction model.