Design And Implementation Of Intelligent Anti-fraud Decision-making Platform For Telecommunication Data

The rapid growth of information has detonated the arrival of the big data era.With the continuous development of the communications industry,more and more telecom fraud has appeared in the daily lives.Telecom fraud has become the main form of fraud that affects people's daily lives,and the current anti-fraud methods are relatively passive and cumbersome,and cannot meet the needs of efficient anti-fraud.Therefore,the research on telecom fraud is imminent.Based on the signaling data and call text data in the telecom anti-fraud scenario,this paper respectively proposes anti-fraud comprehensive decision-making recognition algorithms,fraud pattern discovery and trend analysis algorithms,which can efficiently carry out corresponding telecom anti-fraud recognition and decision-making.Furthermore,in order to provide a support platform for the above algorithm,and to adapt to the ever-changing fraud methods,this paper proposes an intelligent anti-fraud decision-making platform for telecommunication data.Through this platform,users can upload the corresponding telecommunications data,use the multi-source heterogeneous feature engineering provided by the platform to uniformly encapsulate and model the data,and use the fraud phone recognition algorithm,fraud pattern detection algorithm,and fraud trend analysis algorithm by the platform.According to different scenarios and goals,make rapid anti-fraud decision-making and analysis to help and guide the work of relevant personnel.The key algorithm of this article lies in the comprehensive anti-fraud decision-making,fraud pattern discovery and trend analysis.Based on the telecom anti-fraud scenario,this topic effectively solves the identification problem of fraudulent calls through comprehensive anti-fraud decision-making,and proposes "fraud recognition algorithm based on code width learning network" and "model parallel training method based on width learning".The former combines the noise reduction self-encoding and the width learning system,and innovatively proposes an algorithm,which greatly reduces the training time on the basis of further improving the accuracy;the latter proposes a corresponding method based on the former algorithm.The parallelized algorithm further reduces the training prediction time,solves the problem of memory explosion,and better meets the timeliness requirements of anti-fraud.At the same time,through fraud pattern discovery and trend analysis,the "semi-supervised network-based fraud pattern discovery algorithm" and "time-sequence-based fraud trend discovery and analysis method" were respectively proposed.The former uses the semi-supervised graph network and borrows the graph structure to further explore fraud and find more fraudulent calls;the latter based on time series results,comprehensively using clustering and mathematical statistics to analyze fraud trends.Firstly,this article expounds the background and practical significance of the whole research,and analyzes the current situation of domestic and foreign research on this subject.And analyze the needs of the entire platform,analyze the research significance of anti-fraud integrated decision-making algorithms and fraud pattern discovery and trend analysis algorithms.According to different purposes,there are three main functions,they are multi-source heterogeneous telecommunication data feature engineering,anti-fraud intelligent integrated decision-making,and fraud Mode and trend mining,respectively.Next,research the key issues which the system needs.Then,the architecture and modules of the platform are implemented,and introduced the classes and interfaces inside the modules,and expound how the platform processes requests and finally gets a response through the interaction and cooperation between the modules.Finally,described the deployment and testing of the platform,and the work is briefly summarized and prospected.