CN-121980435-A - AI-based anti-fraud full-link intelligent regulation and control method and device

CN121980435ACN 121980435 ACN121980435 ACN 121980435ACN-121980435-A

Abstract

The invention provides an anti-fraud full-link intelligent regulation method and device based on AI, the method comprises the steps of obtaining multi-source heterogeneous data, generating standardized tags according to data source types, obtaining configuration of configurators on a visual arrangement interface on an anti-fraud model, packaging all the configuration into an executable model, respectively generating executable sentences suitable for a flow batch scene according to the data source type of each standardized tag in the executable model, a computing source engine and logic relations of each standardized tag, starting operation by adopting a differential scheduling mechanism, generating and storing an anti-fraud identification result, obtaining a three-dimensional evaluation result through a three-dimensional evaluation model fused by flow batch data, and automatically triggering iterative optimization of the executable model based on the three-dimensional evaluation result. The invention realizes the rapid construction, real-time operation and continuous iterative optimization of the anti-fraud model, solves the problems of poor timeliness, low flexibility and poor accuracy of the prior art, and can rapidly and accurately realize the anti-fraud identification.

Inventors

ZHENG JIDONG
LAI CHUNMEI
HUANG LING
WANG JIACHENG
HUANG MENGHUI

Assignees

福建福诺移动通信技术有限公司

Dates

Publication Date: 20260505
Application Date: 20251205

Claims (10)

1. The anti-fraud full-link intelligent regulation method based on the AI is characterized by comprising the following steps: Step S1, multi-source heterogeneous data are obtained, and standardized labels comprising stream labels and batch labels are respectively generated according to data source types; S2, acquiring the configuration of the anti-fraud model by a configurator on a visual arrangement interface, and packaging all the configuration into an executable model, wherein the executable model comprises a streaming model, a mixed model and an offline model; Step S3, according to the data source type of each standardized label in the executable model, a calculation source engine and the logic relation of each standardized label, executable sentences suitable for a flow scene are respectively generated, and a differential scheduling mechanism is adopted to start and run the corresponding executable sentences, so that an anti-fraud identification result is generated and stored; S4, constructing a three-dimensional evaluation model for flow batch data fusion, and automatically quantifying the executable model from three dimensions of accuracy, timeliness and balance to obtain a three-dimensional evaluation result; And step S5, based on the three-dimensional evaluation result, parameter intelligent iteration of the model and optimization of the data link are automatically triggered, and an optimized executable model is obtained.
2. The AI-based anti-fraud full-link intelligent regulation method of claim 1, wherein the step S3 specifically includes: step S31, obtaining the logic relationship among the data source type, the calculation source engine and each standardized label of each standardized label in the executable model; Step S32, generating a streaming executable statement, a mixed executable statement and a batch executable statement according to the data source type and the logic relation, wherein the streaming executable statement is bound with a real-time computing source engine, a streaming label is acquired from a real-time data source to generate a main label, the batch label is acquired from an offline data source to serve as an auxiliary label for judging, the mixed executable statement is mainly taken as the real-time computing source engine, the offline computing source engine serves as the auxiliary, the streaming label is acquired from the real-time data source, the batch label is acquired from the offline data source to serve as the auxiliary label for equally associating and judging, and the batch executable statement is bound with the offline computing source engine, and the main label is acquired from the offline data source and the streaming label is acquired from the real-time data source to serve as the auxiliary label for judging; And step S33, starting and running the streaming executable sentences and the mixed executable sentences through a real-time scheduling mechanism, starting and running the batch executable sentences through a timing task scheduling mechanism, and generating and storing an anti-fraud identification result.
3. The AI-based anti-fraud full-link intelligent regulation method of claim 2, wherein the step S33 further comprises: Step S331, monitoring data throughput of a real-time data source, task CPU load of a real-time computing source engine and label computing complexity in real time, dynamically distributing or recycling computing examples by combining a cluster resource pool state of a real-time scheduling mechanism through a resource demand prediction model based on time sequence and load characteristics, and carrying out resource isolation and priority scheduling on a real-time streaming task and an offline batch task.
4. The AI-based anti-fraud full-link intelligent regulation method of claim 2, wherein the step S33 further comprises: And step S332, when the failure of the real-time streaming task is detected, automatically backtracking streaming data and batch data in the latest preset period for reprocessing, and when the data accumulation of the real-time data source is serious, automatically switching to a streaming batch mixed processing mode, temporarily storing the accumulated streaming data in an offline database, and starting offline calculation.
5. The AI-based anti-fraud full-link intelligent regulation method of claim 4, wherein step S332 further comprises: And monitoring the data quality and the computing performance of the standardized tag in real time, and automatically performing level adjustment and state switching on the execution state of the standardized tag in the corresponding model according to the data quality and the computing performance of the standardized tag.
6. The AI-based anti-fraud full-link intelligent regulation method according to any one of claims 1 to 5, wherein the accuracy in step S4 is obtained by comparing real-time case-related stream data with batch-type risk clearance, the timeliness is calculated based on a time difference between stream tag generation and treatment execution, and the balance-related batch-type complaint data and case-related changes are comprehensively studied and judged.
7. The AI-based anti-fraud full-link intelligent regulation method according to any one of claims 1 to 5, wherein step S5 further includes: step S51, when analysis finds that the contribution degree of the first flow label in novel fraud is improved to a preset contribution threshold value and the accuracy rate is reduced due to the fact that the first flow label associated with the first flow label is updated with data, the weight of the first flow label is automatically adjusted according to a preset weight adjustment algorithm, and the weight of the first flow label is correspondingly reduced; Step S52, when the anti-fraud recognition result of the executable model is monitored to have the complaint rate increased and the case-related quantity reaches the preset case quantity, the execution frequency or the treatment intensity of the executable model is reduced, the risk number hit by the executable model is screened for the second time, and the number hit by a people library is automatically rejected or degraded; Step S53, when the batch tag data loss rate is continuously overhigh or the flow tag abnormal value duty ratio exceeds a threshold value through backtracking analysis, data optimization and data quality management are automatically carried out; And S54, for all the optimization points discovered through backtracking, performing simulation operation on the historical data aiming at the optimized executable model before formal deployment, and only when the evaluation index of the pre-run reaches a preset standard, the optimized executable model is formally released and is executed.
8. The AI-based anti-fraud full-link intelligent regulation method of any of claims 1 to 5, wherein real-time data sources of the streaming labels are bound to Kafka and a database and a Flink computing node is used as a computing source engine, and offline data sources of computing source engines of the batch labels are bound to dis and a database and Hive computing tasks are used as computing source engines.
9. The AI-based anti-fraud full-link intelligent regulation method according to any one of claims 1 to 5, wherein the configuration of step S2 includes, but is not limited to, a mix of flow label hooks, setting rule values, weight values and logical relationships of standardized labels, and setting risk determination rules.
10. AI-based anti-fraud full-link intelligent regulation device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, characterized in that the processor implements the AI-based anti-fraud full-link intelligent regulation method of any of claims 1 to 9 when executing the computer program.

Description

AI-based anti-fraud full-link intelligent regulation and control method and device Technical Field The invention relates to the technical field of communication information processing, in particular to an anti-fraud full-link intelligent regulation and control method and device based on AI. Background In the communications industry, there is a need to discover telecommunication fraud and process it in a timely manner. The patent publication No. CN120186262A is based on the telecommunication fraud treatment method and system of the model prejudgement and AI outbound evidence, the scheme takes communication data as input, and carries out risk classification on users by constructing a risk scoring model with fixed rules, thereby triggering corresponding shutdown treatment operation. The regional fraud risk assessment method, device, equipment, medium and program product with the patent publication number of CN118780602A divides a management and control range into high, medium and low risk regions according to the case density of different regions, and sets differentiated shutdown proportion and disposal priority for different regions. The scheme integrates user attributes and historical communication behavior data, a person tag library is built, secondary screening is conducted on users marked with high risks by a model, and clear normal users are removed to reduce false shut-down rate. However, the above prior art solution has the following problems: 1. The amount of false shut-down and customer complaints remain large. 2. The fraud behavior and the normal communication behavior are approaching, and the existing anti-fraud strategy is difficult to quickly respond to the new case-related situation. Disclosure of Invention In order to solve the problems in the prior art, the invention provides an anti-fraud full-link intelligent regulation and control method and device based on AI, which can rapidly and accurately realize anti-fraud identification. In order to achieve the above purpose, the invention adopts the following technical scheme: In a first aspect, the present invention provides an AI-based anti-fraud full-link intelligent regulation method, including: Step S1, multi-source heterogeneous data are obtained, and standardized labels comprising stream labels and batch labels are respectively generated according to data source types; S2, acquiring the configuration of the anti-fraud model by a configurator on a visual arrangement interface, and packaging all the configuration into an executable model, wherein the executable model comprises a streaming model, a mixed model and an offline model; Step S3, according to the data source type of each standardized label in the executable model, a calculation source engine and the logic relation of each standardized label, executable sentences suitable for a flow scene are respectively generated, and a differential scheduling mechanism is adopted to start and run the corresponding executable sentences, so that an anti-fraud identification result is generated and stored; S4, constructing a three-dimensional evaluation model for flow batch data fusion, and automatically quantifying the executable model from three dimensions of accuracy, timeliness and balance to obtain a three-dimensional evaluation result; And step S5, based on the three-dimensional evaluation result, parameter intelligent iteration of the model and optimization of the data link are automatically triggered, and an optimized executable model is obtained. The method has the beneficial effects that the method takes the standardized labels integrated with the flow batch as the basis, autonomously completes the construction, configuration and operation and maintenance of the anti-fraud model through visual arrangement, automatically converts the visual arrangement model configuration into executable codes and dispatches and operates, realizes the real-time response and batch concurrent processing of the model, finally carries out backtracking verification, realizes the rapid construction, real-time operation and continuous iterative optimization of the anti-fraud model, and solves the pain points with poor timeliness, low flexibility and poor accuracy in the prior art, thereby being capable of rapidly and accurately realizing the anti-fraud identification. Optionally, the step S3 specifically includes: step S31, obtaining the logic relationship among the data source type, the calculation source engine and each standardized label of each standardized label in the executable model; Step S32, generating a streaming executable statement, a mixed executable statement and a batch executable statement according to the data source type and the logic relation, wherein the streaming executable statement is bound with a real-time computing source engine, a streaming label is acquired from a real-time data source to generate a main label, the batch label is acquired from an offline data source to serve as an auxiliar