Fraud number identification method based on space-time diagram

Abstract

The invention discloses a fraud number identification method based on a space-time diagram, and the method comprises the steps: generating a telecommunication network through a full call record, constructing the space-time diagram for identifying a fraud number, and continuously intercepting a plurality of time-space subgraphs from the diagram according to a time interval T through employing a Snapshot method; calculating a call feature index vector of each user in the time period corresponding to each time-space sub-graph based on the call record of the user in the time period corresponding to each time-space sub-graph, and extracting a neighbor user having the same called party as each user from the time-space sub-graph for each user; constructing and training a space-time diagram neuralnetwork for identifying fraud numbers; and inputting the call feature index vectors of the target user to be identified and all neighbor users thereof in all time-space sub-graphs into a space-time graph neural network, and judging whether the number of the target user to be identified is a suspected fraud number according to the output. The invention belongs to the technical field of information, and can dynamically capture the change of a number calling behavior so as to realize accurate identification of a fraud number.

Description

Technical field [0001] The invention relates to a method for identifying fraud numbers based on time-space diagrams, and belongs to the field of information technology. Background technique [0002] At present, the identification of fraudulent numbers mainly includes the following technical solutions: [0003] 1) Based on the number marking method, the defects of this type of scheme are lagging, inaccurate marking, and inability to deal with gang fraud; [0004] 2) Methods based on speech and semantics, such as keyword matching. The problem with this type of scheme is that the detection process is complex, and because fraud methods emerge in an endless stream, fraudsters will change their statements, resulting in low recognition accuracy; [0005] 3) Based on the feature extraction method, due to the anti-detection awareness of the fraudsters, they will pretend to be a normal number dialing behavior, dial the phone at a normal frequency, or use the number change platform to constantly...

AI Technical Summary

Problems solved by technology

[0003] 1) Based on the method of number marking, the disadvantages of this type of scheme are that there is hysteresis, inaccurate marking, and inability to deal with gang fraud;

[0004] 2) Methods based on speech and semantics, such as keyword matching. The problem with this type of scheme is that the detection process is complicated, and because fraudulent methods emerge in endlessly, fraudsters will change their words, resulting in low recognition accuracy;

[0005] 3) Based on the method of feature extraction, because fraudsters have anti-investigation awareness, they will pretend to be dialing normal numbers, make calls with normal frequency, or use the number changing platform to change numbers continuously, and the fraud mode is becoming more and more serious. Covert, so the recognition accuracy is also low

This technical solution is based on the "multi-point-point" communication relationship structure model of fraudulent calls, and will construct a static map according to the user's call signaling within a period of time, but for the periodic call pattern of number change fraud, that is Fraudsters will use multiple calling numbers to make fraudulent calls in sequence. Since the call frequency of each number is very low, it is difficult for this technical solution to dynamically capture the changes in the calling behavior of fraudulent numbers during this period.

Images