Risk analysis method and system for account data

By collecting and dynamically analyzing account file behavior data in real time, a multi-level anomaly identification system and sensitive information model are constructed, which solves the challenge of traditional methods in distinguishing between normal and abnormal situations and achieves more accurate risk management.

CN119048102AInactive Publication Date: 2024-11-29BEIJING TRUSFORT TECH CO LTD

Patent Information

Application Number
CN202411524195.X
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-10-30
Publication Date
2024-11-29
Estimated Expiration
Not applicable · inactive patent

AI Technical Summary

Technical Problem

Traditional static rules and fixed threshold methods are difficult to effectively distinguish between normal and abnormal account file behavior, resulting in high false positive rates and high risk of false negatives.

Method used

Real-time collection of account file behavior data; identification of initial anomalies based on preset abnormal behavior standards; dynamic selection of features and construction of a multi-level anomaly identification system; and risk level judgment by combining sensitive information dynamic selection model and natural language processing technology.

Benefits of technology

It improves the accuracy and real-time nature of risk identification, reduces the risk of false alarms and missed alarms, and provides flexible and systematic risk management support.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119048102A_ABST
    Figure CN119048102A_ABST
Patent Text Reader

Abstract

The invention provides an account data risk analysis method and system, and relates to the technical field of account data analysis. According to the method, multi-dimensional data is collected in real time, key features are dynamically selected and extracted, a multi-stage anomaly recognition system and a sensitive information dynamic selection model are constructed, and deep analysis is performed in combination with a natural language processing technology, so that the accuracy and real-time performance of risk recognition are improved, the flexibility and systematicness of an analysis process are ensured, and the risk recognition efficiency is improved. According to the method, the risk of misinformation and missed information is effectively balanced, normal and abnormal behaviors can be defined more accurately even in complex and changeable working scenes, and more reliable and intelligent data analysis support is provided for risk management of enterprises.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of account data analysis, and in particular to a method and system for risk analysis of account data. Background Art

[0002] In today's digital age, enterprises are facing increasingly complex cybersecurity and risk management challenges. With the expansion of business scale and the surge in data volume, behavioral data monitoring of account files has become a key link in ensuring enterprise information security and operational efficiency.

[0003] The behavioral data of account files covers multiple dimensions such as login time, location, operation type, amount, and result, and these behaviors may present diverse patterns in different business scenarios. In daily work, some seemingly abnormal account operations may actually be normal work needs. For example, large-scale fund transfers may involve legitimate business transactions rather than fraudulent behavior. This ambiguity poses a huge challenge to risk monitoring. Traditional static rules and fixed threshold methods often find it difficult to effectively distinguish between normal and abnormal, resulting in a high false alarm rate and a high risk of missed reports.

[0004] Therefore, it is necessary to provide a risk analysis method and system for account data to solve the above technical problems. Summary of the invention

[0005] In order to solve the above technical problems, the present invention provides a risk analysis method and system for account data for solving the problem that traditional methods are difficult to accurately define normal and abnormal risk behaviors in account file behaviors.

[0006] The present invention provides a method for risk analysis of account data, the method comprising the following steps: Collect behavioral data of account files in real time and identify initial abnormal behavior data of accounts based on preset abnormal behavior standards; Set up a dynamic selection mechanism for feature extraction, dynamically select features based on the initial abnormal behavior data of the account and extract them as first-level abnormal features; Based on the primary abnormal features, the secondary abnormal features associated with the primary abnormal features are identified and extracted through the pre-trained association model; Based on the relationship between historical account abnormal behavior data and sensitive information, a dynamic selection model for sensitive information is established; The first-level abnormal features and the second-level abnormal features are used as inputs, and the sensitive information dynamic selection model dynamically outputs the corresponding sensitive information, which are used as the primary monitoring point and the secondary monitoring point, respectively, to form the focus of risk monitoring; Based on the focus of risk monitoring, natural language processing technology is used in combination with sensitive information definition standards to monitor the behavior of account files and make risk level judgments.

[0007] Preferably, the specific steps of collecting the behavior data of the account file in real time and identifying the initial abnormal behavior data of the account based on the preset abnormal behavior standard are: Capture the behavioral data of account files in real time through monitoring systems and logging functions, including but not limited to login time, login location, operation type, operation amount and operation results; Set abnormal behavior standards according to the business rules of the enterprise, including but not limited to the number of failed logins within a set period, abnormal large-value transactions, and operations during unusual time periods; Compare and analyze the behavior data of the account file with the set abnormal behavior standards, identify the behavior data that meets the abnormal behavior standards, and mark it as the initial abnormal behavior data.

[0008] Preferably, the specific steps of setting a dynamic selection mechanism for feature extraction and dynamically selecting features based on the initial abnormal behavior data of the account and extracting them as primary abnormal features are: Use statistical methods to evaluate the correlation between features and behaviors in the behavioral data of historical account files and preliminarily screen features; Evaluate the correlation values ​​of the initially screened features, and preset multiple threshold segments corresponding to the features; The features that do not belong to the threshold segment are regarded as the first-level abnormal features to build a dynamic selection mechanism; According to the dynamic selection mechanism, specific features are extracted from the initial abnormal behavior data as the first-level abnormal features.

[0009] Preferably, the specific steps of identifying and extracting the secondary abnormal features associated with the primary abnormal features through the pre-trained association model based on the primary abnormal features are: According to the business logic, a statistical method is used to identify the correlation between historical abnormal features, and a correlation model is constructed based on the correlation between historical abnormal features, and a correlation value range is set; Based on the extracted primary abnormal features, other abnormal behavior data associated with the primary abnormal features are determined according to the constructed association model and the set association value range and identified as secondary abnormal features; Extract the identified secondary abnormal features.

[0010] Preferably, the specific steps of establishing a dynamic selection model for sensitive information based on the relationship between historical account abnormal behavior data and sensitive information are: Collect historical account abnormal behavior data, and identify the correlation between account abnormal behavior data and account sensitive information through data analysis methods; The correlation between the identified account abnormal behavior data and the account sensitive information is used as the training set, and a sensitive information dynamic selection model is constructed by training the machine learning model.

[0011] Preferably, the specific steps of taking the primary abnormal features and the secondary abnormal features as inputs respectively, and dynamically outputting the corresponding sensitive information using the sensitive information dynamic selection model as the primary monitoring point and the secondary monitoring point respectively, to form the focus of risk monitoring are: Input the first-level abnormal features into the dynamic selection model of sensitive information, output the sensitive information corresponding to the first-level abnormal features and use it as the main monitoring point; Input the secondary abnormal features into the dynamic selection model of sensitive information, output the sensitive information corresponding to the secondary abnormal features and use it as the monitoring point; Integrate the selected primary monitoring points and secondary monitoring points to form a focal area for risk monitoring.

[0012] Preferably, the specific steps of forming a focus for risk monitoring, using natural language processing technology and combining sensitive information definition standards to monitor the behavior of account files and make risk level judgments are: Use natural language processing technology to parse the operation behavior description of the account file and extract key information, including but not limited to sensitive operation type, amount involved, timestamp, operation object and error prompt information; Screen and judge the extracted key information according to the preset sensitive information definition standards, and determine the sensitive information that needs to be paid attention to in the key information; Based on the sensitive information that requires attention and the focus areas of risk monitoring, the risk level of the operation behavior of the account files is judged.

[0013] A risk analysis system for account data, the analysis system comprising: An abnormal data identification module is used to collect the behavior data of account files in real time and identify the initial abnormal behavior data of the account based on the preset abnormal behavior standards; The dynamic extraction module is used to set up a dynamic selection mechanism for feature extraction, dynamically select features based on the initial abnormal behavior data of the account and extract them as first-level abnormal features; An identification and extraction module, used for identifying and extracting secondary abnormal features associated with the primary abnormal features through a pre-trained association model based on the primary abnormal features; Model building module, used to build a dynamic selection model for sensitive information based on the relationship between historical account abnormal behavior data and sensitive information; The screening module is used to take the primary abnormal features and the secondary abnormal features as inputs respectively, and dynamically output the corresponding sensitive information using the sensitive information dynamic selection model, which serve as the primary monitoring point and the secondary monitoring point respectively, forming the focus of risk monitoring; The judgment module is used to monitor the behavior of account files and make risk level judgments based on the focus of risk monitoring, using natural language processing technology and combining sensitive information definition standards.

[0014] Compared with the related art, the risk analysis method and system for account data provided by the present invention have the following beneficial effects: The present invention collects multi-dimensional data in real time, dynamically selects and extracts key features, constructs a multi-level anomaly recognition system and a dynamic selection model for sensitive information, and combines natural language processing technology for in-depth analysis. This not only improves the accuracy and real-time nature of risk identification, but also ensures the flexibility and systematic nature of the analysis process, effectively balancing the risks of false positives and missed negatives, so that even in complex and changeable work scenarios, normal and abnormal behaviors can be more accurately defined, providing more reliable and intelligent data analysis support for corporate risk management. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 A flow chart of a method for risk analysis of account data provided by the present invention; Figure 2 A system block diagram of an account data risk analysis system provided by the present invention. DETAILED DESCRIPTION

[0016] The present invention will be further described below in conjunction with the accompanying drawings and implementation modes.

[0017] Embodiment 1 like Figure 1 As shown, a risk analysis method for account data includes the following steps: S100, collecting behavior data of account files in real time, and identifying initial abnormal behavior data of the account based on preset abnormal behavior standards; In the specific implementation process, the specific implementation steps of step S100 are: S101. Capture the behavior data of the account file in real time through the monitoring system and log recording function, including but not limited to login time, login location, operation type, operation amount and operation result; Specifically, a monitoring system is deployed at key nodes of the enterprise network, including but not limited to servers and switches, to ensure that all activities of account files are fully covered, and logs are collected including but not limited to login time, login location through IP address resolution, transfer or query operation type, operation amount and operation result.

[0018] For example, an employee of a company logs into the financial system through the company's intranet IP address at 9 a.m. and performs a transfer operation of 100,000 yuan. The operation is successful. This behavior will be captured by the monitoring system and a log record will be generated. Then the corresponding data can be collected: "Time: 2023-04-01 09:00:00, Location: Company Intranet IP (specific IP address), Operation Type: Transfer, Operation Amount: 100,000 yuan, Operation Result: Success.

[0019] S102. Set abnormal behavior standards according to the business rules of the enterprise, including but not limited to the number of failed logins within a set period, abnormal large-amount transaction standards, and non-routine time period operation standards; Specifically, it should be noted that the abnormal behavior standards are set according to the company's business rules, and specific thresholds are set for each abnormal behavior standard. In this example, the threshold for failed logins is set at 5 times / hour, the large transaction standard is 100,000 yuan / transaction, and the abnormal large transaction standard is that a single transaction amount exceeding 100,000 yuan is considered a large transaction and requires special attention. The standard for operations during non-routine time periods is after get off work.

[0020] S103, comparing and analyzing the behavior data of the account file with the set abnormal behavior standard, identifying the behavior data that meets the abnormal behavior standard, and marking it as initial abnormal behavior data; Specifically, Python tools are used to clean the behavioral data of account files captured in real time, including data cleaning and format conversion, to ensure data consistency and analyzability, and to compare the preprocessed behavioral data with the set abnormal behavior standards one by one. Specifically, for each abnormal behavior standard, check whether there is behavioral data that meets its conditions, and for behavioral data that meets the abnormal behavior standards, identify it as initial abnormal behavior data through semantic labels.

[0021] For example, if a certain account is captured to have made multiple transfer operations in one day, and the amount of one transfer is 200,000 yuan, which exceeds the large transaction standard set by the enterprise (100,000 yuan), then this transfer operation will be compared and analyzed with the set abnormal behavior standard, and it will be identified that it meets the abnormal behavior standard of "large transaction". Therefore, it will be marked as the initial abnormal behavior data and recorded for further analysis.

[0022] S200, setting a dynamic selection mechanism for feature extraction, dynamically selecting features based on the initial abnormal behavior data of the account and extracting them as first-level abnormal features; In the specific implementation process, the specific implementation steps of step S200 are: S201, using statistical methods to evaluate the correlation between features and behaviors in the behavior data of historical account files, and preliminarily screening features; Specifically, account file behavior data is collected from historical databases, log files, transaction records, etc., and cleaned to remove duplicate values ​​and invalid information, and to identify which fields or attributes can be regarded as features. For example, in this embodiment, the login time can be divided into features such as "whether it is a non-working time login" and "distribution of login time points"; the operation type can be refined into specific behavior features such as "transfer", "withdrawal", and "password modification". Then, the Pearson correlation coefficient in the statistical method is used to analyze the correlation between each feature and the abnormal behavior, and the correlation analysis results are obtained. Based on the results of the correlation analysis, the features that are significantly correlated with the abnormal behavior are preliminarily screened out.

[0023] For example, if correlation analysis finds that the two features of "login outside working hours" and "multiple failed logins in a short period of time" are highly correlated with known abnormal behaviors (such as account theft), then these two features can be preliminarily screened out as the basis for subsequent analysis.

[0024] S202, evaluating the correlation values ​​of the initially screened features, and presetting a plurality of threshold segments corresponding to the features; Specifically, according to business needs and the statistical characteristics of historical data, multiple threshold segments are set for different features, where the threshold segments can be set based on the distribution characteristics of the features, such as the mean and standard deviation. For each initially screened feature, its correlation value (such as the correlation coefficient) is calculated and compared with the set threshold segment for judgment.

[0025] For example, the feature of “non-working hours login” sets two threshold segments: high (>0.8), medium (0.5-0.8), and low (<0.5). Through calculation, it is found that the correlation value of this feature is 0.75, which belongs to the medium threshold segment.

[0026] S203, determining that the feature does not belong to the threshold segment is regarded as a first-level abnormal feature to construct a dynamic selection mechanism; Specifically, the features are assigned to different threshold segments according to their correlation values ​​as a selection strategy, and the dynamic selection mechanism is set with the selection strategy.

[0027] S204, extracting specific features from the initial abnormal behavior data as primary abnormal features according to a dynamic selection mechanism; Specifically, according to the dynamic selection mechanism, the features are assigned to different threshold segments according to their correlation values, and the data containing these features are screened out from the initial abnormal behavior data and extracted as the first-level abnormal feature values.

[0028] S300, based on the primary abnormal feature, identifying and extracting the secondary abnormal feature associated with the primary abnormal feature through a pre-trained association model; In the specific implementation process, the specific implementation steps of step S300 are: S301. According to the business logic, a statistical method is used to identify the correlation between historical abnormal features, and a correlation model is constructed according to the correlation between the historical abnormal features, and a correlation value range is set; Specifically, abnormal behavior records in historical account data are collected, including but not limited to login failures, abnormal transaction amounts, operations during non-working hours, etc. The correlation between historical abnormal features is analyzed through correlation analysis methods according to business logic. For example, when a behavior pattern of making a large transfer immediately after a login failure is found, it may be highly correlated, indicating the existence of potential risks. Then, the correlation between historical abnormal features is used as a training set, and a decision tree model is trained to build an association model to quantify the strength of the association between features. Finally, one or more correlation value ranges are set for the association model to determine whether the association between two or more features has reached a level that requires attention.

[0029] Exemplarily, a correlation value range is set, such as "after more than 3 failed login attempts, the subsequent large transfer (the amount exceeds 5 times the average daily transaction amount of the account) is considered highly correlated."

[0030] S302: Based on the extracted primary abnormal features, other abnormal behavior data associated with the primary abnormal features are determined according to the constructed association model and the set association value range and are marked as secondary abnormal features; Specifically, the first-level abnormal feature extracted in step S204 is used as input and input into the association model. According to the association model and the set association value range, other abnormal behavior data associated with the first-level abnormal feature is determined, and the abnormal behavior data determined to be associated with the first-level abnormal feature is identified as a second-level abnormal feature.

[0031] S303: Extract the marked secondary abnormal features.

[0032] S400. Based on the relationship between historical account abnormal behavior data and sensitive information, a dynamic selection model for sensitive information is established; In the specific implementation process, the specific implementation steps of step S400 are: S401. Collect historical account abnormal behavior data, and identify the correlation between the account abnormal behavior data and account sensitive information through data analysis methods; S402: Using the correlation correspondence between the identified account abnormal behavior data and the account sensitive information as a training set, a dynamic selection model for sensitive information is constructed by training a neural network model in a machine learning model.

[0033] S500, taking the primary abnormal feature and the secondary abnormal feature as input respectively, and using the sensitive information dynamic selection model to dynamically output the corresponding sensitive information, which are respectively used as the primary monitoring point and the secondary monitoring point, to form the focus of risk monitoring; In the specific implementation process, the specific implementation steps of step S500 are: S501, input the first-level abnormal features into the sensitive information dynamic selection model, output the sensitive information corresponding to the first-level abnormal features and use it as the main monitoring point; Specifically, the first-level abnormal features extracted in step S204 are used as input data and sent to the sensitive information dynamic selection model previously trained by machine learning. The sensitive information dynamic selection model performs calculations and analysis based on the characteristics of the first-level abnormal features and the association rules between these features and sensitive information in historical data. Finally, the sensitive information closely related to the first-level abnormal features is output and used as the main monitoring point.

[0034] For example, the first-level abnormal feature is "multiple large-value transfer operations in a short period of time". After this feature is input into the model, the sensitive information dynamic selection model outputs the sensitive information associated with this feature as "account balance", "transfer target account information" and "transfer operation time", and these sensitive information are identified as the main monitoring points for current risk monitoring.

[0035] S502, input the secondary abnormal features into the sensitive information dynamic selection model, output the sensitive information corresponding to the secondary abnormal features and use it as a monitoring point; Specifically, the secondary abnormal features extracted in step S303 are used as input data and sent to the sensitive information dynamic selection model previously trained by machine learning. The sensitive information dynamic selection model performs calculation and analysis based on the characteristics of the secondary abnormal features and the association rules between these features and sensitive information in historical data, and finally outputs sensitive information closely related to the secondary abnormal features as the monitoring point.

[0036] S503, integrating the selected primary monitoring point and secondary monitoring point to form a focal area for risk monitoring; Specifically, the main monitoring points and secondary monitoring points output by steps S501 and S502 are integrated to form a comprehensive risk monitoring focus area, which includes the most direct risk signals (main monitoring points) and also covers auxiliary information that may indirectly point to risks (secondary monitoring points), providing rich data support for subsequent risk level judgments.

[0037] Exemplarily, the focus areas of risk monitoring may include "account balance", "transfer target account information", "transfer operation time" (primary monitoring points) and "login IP address", "login time", "historical login location records" (secondary monitoring points).

[0038] S600, based on the focus of risk monitoring, using natural language processing technology and combining sensitive information definition standards, monitor the behavior of account files and make risk level judgments; In the specific implementation process, the specific implementation steps of step S600 are: S601. Use natural language processing technology to parse the operation behavior description of the account file and extract key information, including but not limited to sensitive operation type, amount involved, timestamp, operation object, and error prompt information; For example, in this embodiment, an account operation behavior is described as: "User A transferred 10,000 yuan to account B through online banking at 2023-04-01 14:30, and the operation was successful." After text preprocessing, the word segmentation result is obtained: "User A transferred 10,000 yuan to account B through online banking at 2023-04-01 14:30, and the operation was successful." NER identifies sensitive entities: timestamp "2023-04-01 14:30", amount "10,000 yuan", operation object "account B", and dependency syntactic analysis reveals that the key operation is "transfer" and associates the time, amount and operation object.

[0039] S602: Screen and judge the extracted key information according to the preset sensitive information definition standard to determine the sensitive information that needs to be paid attention to in the key information; Specifically, the extracted key information is compared with the preset sensitive information definition standards to check whether there are any matches, and multiple logical rules are set to evaluate the risk of the information. For example, if the transfer amount exceeds the preset large transaction threshold, the amount is considered high-risk information.

[0040] S603. Based on the sensitive information that needs attention and the focus areas of risk monitoring, determine the risk level of the operation behavior of the account file; Specifically, sensitive information (such as large transactions, transactions at abnormal times, etc.) and features of the monitoring focus areas (such as abnormal login patterns, unconventional operation paths, etc.) are summarized to form a complete set of risk factors. The risk assessment model pre-trained and constructed by the enterprise outputs the risk level score. The risk assessment model is trained through a machine learning algorithm model and a data set of historical account file behavior data.

[0041] Embodiment 2 like Figure 2As shown, a risk analysis system for account data applied to a risk analysis method for account data specifically includes: An abnormal data identification module is used to collect the behavior data of account files in real time and identify the initial abnormal behavior data of the account based on the preset abnormal behavior standards; The dynamic extraction module is used to set up a dynamic selection mechanism for feature extraction, dynamically select features based on the initial abnormal behavior data of the account and extract them as first-level abnormal features; An identification and extraction module, used for identifying and extracting secondary abnormal features associated with the primary abnormal features through a pre-trained association model based on the primary abnormal features; Model building module, used to build a dynamic selection model for sensitive information based on the relationship between historical account abnormal behavior data and sensitive information; The screening module is used to take the primary abnormal features and the secondary abnormal features as inputs respectively, and dynamically output the corresponding sensitive information using the sensitive information dynamic selection model, which serve as the primary monitoring point and the secondary monitoring point respectively, forming the focus of risk monitoring; The judgment module is used to monitor the behavior of account files and make risk level judgments based on the focus of risk monitoring, using natural language processing technology and combining sensitive information definition standards.

[0042] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0043] Those skilled in the art can understand that all or part of the steps in the various methods of the above embodiments can be completed by instructing related hardware through a program, and the program can be stored in a computer-readable storage medium, and the storage medium includes a read-only memory (ROM), a random access memory (RAM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), a one-time programmable read-only memory (OTPROM), an electronically erasable rewritable read-only memory (EEPROM), a compact disc (CD-ROM) or other optical disc storage, magnetic disk storage, magnetic tape storage, or any other computer-readable medium that can be used to carry or store data.

[0044] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

Claims

1. A risk analysis method for account data, characterized in that: The analytical method comprises the following steps: Collect behavioral data of account files in real time and identify initial abnormal behavior data of accounts based on preset abnormal behavior standards; Set up a dynamic selection mechanism for feature extraction, dynamically select features based on the initial abnormal behavior data of the account and extract them as first-level abnormal features; Based on the primary abnormal features, the secondary abnormal features associated with the primary abnormal features are identified and extracted through the pre-trained association model; Based on the relationship between historical account abnormal behavior data and sensitive information, a dynamic selection model for sensitive information is established; The first-level abnormal features and the second-level abnormal features are used as inputs, and the sensitive information dynamic selection model dynamically outputs the corresponding sensitive information, which are used as the primary monitoring point and the secondary monitoring point, respectively, to form the focus of risk monitoring; Based on the focus of risk monitoring, natural language processing technology is used in combination with sensitive information definition standards to monitor the behavior of account files and make risk level judgments.

2. The method for risk analysis of account data according to claim 1, characterized in that: The specific steps of collecting the behavior data of the account file in real time and identifying the initial abnormal behavior data of the account based on the preset abnormal behavior standard are as follows: Capture the behavioral data of account files in real time through monitoring systems and logging functions, including but not limited to login time, login location, operation type, operation amount and operation results; Set abnormal behavior standards according to the business rules of the enterprise, including but not limited to the number of failed logins within a set period, abnormal large-value transactions, and operations during unusual time periods; Compare and analyze the behavior data of the account file with the set abnormal behavior standards, identify the behavior data that meets the abnormal behavior standards, and mark it as the initial abnormal behavior data.

3. The method for risk analysis of account data according to claim 1, characterized in that: The specific steps of setting a dynamic selection mechanism for feature extraction, dynamically selecting features based on the initial abnormal behavior data of the account and extracting them as first-level abnormal features are as follows: Use statistical methods to evaluate the correlation between features and behaviors in the behavioral data of historical account files and preliminarily screen features; Evaluate the correlation values ​​of the initially screened features, and preset multiple threshold segments corresponding to the features; The features that do not belong to the threshold segment are regarded as the first-level abnormal features to build a dynamic selection mechanism; According to the dynamic selection mechanism, specific features are extracted from the initial abnormal behavior data as the first-level abnormal features.

4. The method for risk analysis of account data according to claim 1, characterized in that: The specific steps of identifying and extracting the secondary abnormal features associated with the primary abnormal features through the pre-trained association model based on the primary abnormal features are: According to the business logic, a statistical method is used to identify the correlation between historical abnormal features, and a correlation model is constructed based on the correlation between historical abnormal features, and a correlation value range is set; Based on the extracted primary abnormal features, other abnormal behavior data associated with the primary abnormal features are determined according to the constructed association model and the set association value range and identified as secondary abnormal features; Extract the identified secondary abnormal features.

5. The method for risk analysis of account data according to claim 1, characterized in that: The specific steps of establishing a dynamic selection model for sensitive information based on the relationship between historical account abnormal behavior data and sensitive information are as follows: Collect historical account abnormal behavior data, and identify the correlation between account abnormal behavior data and account sensitive information through data analysis methods; The correlation between the identified account abnormal behavior data and the account sensitive information is used as the training set, and a sensitive information dynamic selection model is constructed by training the machine learning model.

6. The method for risk analysis of account data according to claim 1, characterized in that: The specific steps of taking the primary abnormal features and the secondary abnormal features as inputs, and dynamically outputting the corresponding sensitive information using the sensitive information dynamic selection model as the primary monitoring point and the secondary monitoring point, respectively, to form the focus of risk monitoring are as follows: Input the first-level abnormal features into the dynamic selection model of sensitive information, output the sensitive information corresponding to the first-level abnormal features and use it as the main monitoring point; Input the secondary abnormal features into the dynamic selection model of sensitive information, output the sensitive information corresponding to the secondary abnormal features and use it as the monitoring point; Integrate the selected primary monitoring points and secondary monitoring points to form a focal area for risk monitoring.

7. The method for risk analysis of account data according to claim 1, characterized in that: The specific steps of forming a focus on risk monitoring, using natural language processing technology and combining sensitive information definition standards to monitor the behavior of account files and make risk level judgments are as follows: Use natural language processing technology to parse the operation behavior description of the account file and extract key information, including but not limited to sensitive operation type, amount involved, timestamp, operation object and error prompt information; Screen and judge the extracted key information according to the preset sensitive information definition standards, and determine the sensitive information that needs to be paid attention to in the key information; Based on the sensitive information that requires attention and the focus areas of risk monitoring, the risk level of the operation behavior of the account files is judged.

8. A risk analysis system for account data, using a risk analysis method for account data according to any one of claims 1 to 7, characterized in that: The analysis system comprises: An abnormal data identification module is used to collect the behavior data of account files in real time and identify the initial abnormal behavior data of the account based on the preset abnormal behavior standards; The dynamic extraction module is used to set up a dynamic selection mechanism for feature extraction, dynamically select features based on the initial abnormal behavior data of the account and extract them as first-level abnormal features; An identification and extraction module, used for identifying and extracting secondary abnormal features associated with the primary abnormal features through a pre-trained association model based on the primary abnormal features; Model building module, used to build a dynamic selection model for sensitive information based on the relationship between historical account abnormal behavior data and sensitive information; The screening module is used to take the primary abnormal features and the secondary abnormal features as inputs respectively, and dynamically output the corresponding sensitive information using the sensitive information dynamic selection model, which serve as the primary monitoring point and the secondary monitoring point respectively, forming the focus of risk monitoring; The judgment module is used to monitor the behavior of account files and make risk level judgments based on the focus of risk monitoring, using natural language processing technology and combining sensitive information definition standards.

Citation Information

Patent Citations

  • Account risk assessment system based on electronic commerce

    CN116503130A

  • Method and device for detecting fraudulent account

    CN117993914A

  • Abnormal behavior detection method and device, electronic equipment and storage medium

    CN118194205A

Cited By

  • Radar data intelligent identification method and system based on deep learning

    CN120611130A

  • A radar data intelligent recognition method and system based on deep learning

    CN120611130B