Telecom Fraud Detection Based On Graph Embedding

At present,driven by huge economic benefits,telecommunications fraud has become a highly specialized black industry chain with a clear division of labor.Fraudsters carry out fraud from multiple sources and multiple channels.The model is complex and changeable,the behavior is vague and concealed,and the technology continues to upgrade and intensity Keep increasing.According to statistics from the Ministry of Public Security,since 2010,the number of telecom fraud reports has exploded at an annual rate of 30%,and a large number of unreported small fraud cases have occurred frequently.In particular,all kinds of new types of telecommunications fraud,with high concealment and high success rate,pose a serious threat to residents' personal property and the entire society.Therefore,improving the detection capabilities of new types of fraud is a problem faced by current anti-fraud technologies.After repeated research on scam samples,this thesis proposes a telecommunication fraud detection method based on graph embedding,which is used to mine the potential features of call correlation graphs between fraudsters and users.First,this method is based on attribute graph embedding,which not only saves the network structure information,but also incorporates the node statistical features extracted for the call behavior,the purpose is to represent the node as a low-dimensional vector.Secondly,this method belongs to semi-supervised learning.In the supervised part,the corresponding label is predicted,and in the unsupervised part,the node context vector is predicted.Both share parameters and optimize the model at the same time.Finally,for the weak association problem of telecommunication networks,this thesis proposes a weighted random walk context sampling method,and improves the objective function,which significantly improves the detection ability of new types of telecommunications fraud.Based on the entire process of the algorithm,this thesis also designs a corresponding telecommunications fraud detection system based on graph embedding,including the basic data processing platform,the core algorithm platform and the corresponding application platform.In order to verify the effect of the model,experiments were conducted on the real data set on the live network and two simulation data sets,including the node classification task and the link prediction task.The accuracy has been significantly improved,proving that the graph embedding algorithm proposed in this thesis Outstanding anti-telecom fraud effect.