Risk monitoring method, device, equipment and computer readable storage medium

By generating and parsing the risk data of operational data, combining pre-trained models and judgment functions, the problem of inaccurate risk analysis in the financial industry is solved, comprehensive risk detection of business operations is achieved, and analysis accuracy is improved.

CN110991871BActive Publication Date: 2024-11-12WEBANK (CHINA)
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN201911205139.9
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2019-11-29
Publication Date
2024-11-12
Estimated Expiration
2039-11-29

AI Technical Summary

Technical Problem

The existing technology is not accurate enough in the financial industry to collect and analyze the operational behavior data of the business system in real time, resulting in some risk points being missed.

Method used

By generating the risk data corresponding to the operation data, determining the operation type, and parsing it into target data, using pre-trained risk monitoring models or judgment functions to analyze whether there is a risk, and adjusting neural network parameters in combination with real-time recording and historical data to improve the analysis accuracy.

Benefits of technology

A comprehensive risk analysis of business operations is achieved, the accuracy of risk analysis is improved, data collection is prevented, and the ability to detect potential risks is enhanced.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN110991871B_ABST
    Figure CN110991871B_ABST
Patent Text Reader

Abstract

The present invention discloses a risk monitoring method, including: if it is monitored that the writing of operation data based on a target account is completed, then risk data corresponding to the operation data is generated; determining the operation type to which the operation data belongs, and parsing the risk data into target data corresponding to the operation type; based on the target data, determining whether the business operation corresponding to the target account has risks. The present invention also discloses a risk monitoring device, equipment and computer-readable storage medium. The present invention records the operation data in real time to generate the risk data of the target account, and parses the risk data into target data, so that the data sources based on the risk analysis are comprehensive, preventing the situation where data collection is easily missed, and finally determining whether the business operation corresponding to the target account has risks through the target data, thereby improving the accuracy of risk analysis.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of financial technology (Fintech) technology, and in particular to risk monitoring methods, devices, equipment and computer-readable storage media. Background Art

[0002] With the development of computer technology, more and more technologies are applied in the financial field. Traditional finance is gradually transforming to Fintech, and data analysis technology is no exception. However, due to the security and real-time requirements of the financial industry, higher requirements are also placed on data analysis technology. In the daily service process of banks and other financial institutions, due to some objective factors, such as system failures, and subjective factors, such as artificial changes in limit amounts, the business system has risks in the process of external services, causing a certain degree of economic losses to banks and other financial institutions, such as the inability to conduct related businesses and increased risks. Therefore, risk assessment of the operation behavior of the business system is an important task that banks and other financial institutions must do.

[0003] In the prior art, the operational behavior data of the business system is recorded in a database. Then, when analyzing whether there are risky operations in the business system, data is exported from the database of the business system, that is, the operational behavior data in the database is collected, and risk analysis is performed based on the collected data.

[0004] However, the data collected by existing technologies are all instantaneous. If they are not collected at the time, they will be overwritten afterwards. In other words, there is no in-depth analysis of the business system and the collection of more detailed real-time operation behavior data, resulting in the collected data being incomplete. Some risk points may be missed. For example, in order to evade an audit, the system (which may be caused by human operation or system defects, problematic design, etc.) turns on a switch and turns off the switch afterwards, causing major risks that cannot be detected afterwards.

[0005] Therefore, existing data risk analysis is not accurate enough. Summary of the invention

[0006] The main purpose of the present invention is to propose a risk monitoring method, device, equipment and computer-readable storage medium, aiming to improve the accuracy of risk analysis.

[0007] To achieve the above object, the present invention provides a risk monitoring method, which comprises the following steps:

[0008] If it is monitored that the writing of the operation data based on the target account is completed, risk data corresponding to the operation data is generated;

[0009] Determine the operation type to which the operation data belongs, and parse the risk data into target data corresponding to the operation type;

[0010] Based on the target data, determine whether a business operation corresponding to the target account has risks.

[0011] Preferably, the step of determining the operation type to which the operation data belongs and parsing the risk data into target data corresponding to the operation type includes:

[0012] Determine the operation type to which the operation data belongs, and based on the operation type, determine the field value of each field in a preset general format, wherein the preset general format includes at least one field;

[0013] Based on a preset common format and the field value, target data corresponding to the risk data is generated.

[0014] Preferably, the step of determining whether the business operation corresponding to the target account has risks based on the target data includes:

[0015] If it is determined that the operation data does not belong to the preset scenario data, the target data is input into a pre-trained risk monitoring model to obtain a corresponding risk level value;

[0016] Based on the risk level value, determine whether the business operation corresponding to the target account has risks.

[0017] Preferably, the step of inputting the target data into a pre-trained risk monitoring model to obtain a corresponding risk level value comprises:

[0018] If it is determined that the target data includes at least two operation sequence data, the operation sequence data are input into a pre-trained risk monitoring model to obtain each output value output by the risk monitoring model;

[0019] The final output value is taken as the risk level value corresponding to the target data.

[0020] Preferably, before the step of inputting the target data into a pre-trained risk monitoring model to obtain a corresponding risk level value, the risk monitoring method further comprises:

[0021] Collecting historical target accounts, obtaining historical business operations corresponding to the historical target accounts, and historical target data corresponding to the historical business operations, and setting the historical target data as input of the initial model;

[0022] Setting the historical risk level value corresponding to the historical business operation as the output of the initial model;

[0023] Initializing the neural network parameters of the initial model and inputting the historical target data into the initial model to obtain the current cycle result;

[0024] Determine the current error between the current cycle result and the historical risk level value, and adjust the initialized neural network parameters according to the current error to obtain the neural network parameters of the initial model in the next cycle, and cyclically execute the step of inputting the historical target data into the initial model to obtain the current cycle result;

[0025] If it is detected that the current error is the smallest, the risk monitoring model is constructed based on the initial model and the neural network parameters corresponding to the current error.

[0026] Preferably, the step of determining whether the business operation corresponding to the target account has risks based on the target data includes:

[0027] If it is determined that the operation data belongs to the preset scene data, determining a judgment function corresponding to the operation type;

[0028] Based on the judgment function and the target data, it is determined whether the business operation corresponding to the target account has risks.

[0029] Preferably, the step of determining whether the business operation corresponding to the target account has risks based on the judgment function and the target data includes:

[0030] Determine whether there is a field value corresponding to the judgment function in the target data;

[0031] If so, it is determined that the business operation corresponding to the target account is at risk.

[0032] In addition, to achieve the above-mentioned purpose, the present invention further provides a risk monitoring device, the risk monitoring device comprising:

[0033] A monitoring module, configured to generate risk data corresponding to the operation data if it is detected that the operation data based on the target account has been written;

[0034] A parsing module, used to determine the operation type to which the operation data belongs, and parse the risk data into target data corresponding to the operation type;

[0035] A determination module is used to determine whether a business operation corresponding to the target account has risks based on the target data.

[0036] Preferably, the parsing module is also used for:

[0037] Determine the operation type to which the operation data belongs, and based on the operation type, determine the field value of each field in a preset general format, wherein the preset general format includes at least one field;

[0038] Based on a preset common format and the field value, target data corresponding to the risk data is generated.

[0039] Preferably, the determining module is further used for:

[0040] If it is determined that the operation data does not belong to the preset scenario data, the target data is input into a pre-trained risk monitoring model to obtain a corresponding risk level value;

[0041] Based on the risk level value, determine whether the business operation corresponding to the target account has risks.

[0042] Preferably, the determining module is further used for:

[0043] If it is determined that the target data includes at least two operation sequence data, the operation sequence data are input into a pre-trained risk monitoring model to obtain each output value output by the risk monitoring model;

[0044] The final output value is taken as the risk level value corresponding to the target data.

[0045] Preferably, the risk monitoring model further includes a training module, and the training module is further used for:

[0046] Collecting historical target accounts, obtaining historical business operations corresponding to the historical target accounts, and historical target data corresponding to the historical business operations, and setting the historical target data as input of the initial model;

[0047] Setting the historical risk level value corresponding to the historical business operation as the output of the initial model;

[0048] Initializing the neural network parameters of the initial model and inputting the historical target data into the initial model to obtain the current cycle result;

[0049] Determine the current error between the current cycle result and the historical risk level value, and adjust the initialized neural network parameters according to the current error to obtain the neural network parameters of the initial model in the next cycle, and cyclically execute the step of inputting the historical target data into the initial model to obtain the current cycle result;

[0050] If it is detected that the current error is the smallest, the risk monitoring model is constructed based on the initial model and the neural network parameters corresponding to the current error.

[0051] Preferably, the determining module is further used for:

[0052] If it is determined that the operation data belongs to the preset scene data, determining a judgment function corresponding to the operation type;

[0053] Based on the judgment function and the target data, it is determined whether the business operation corresponding to the target account has risks.

[0054] Preferably, the determining module is further used for:

[0055] Determine whether there is a field value corresponding to the judgment function in the target data;

[0056] If so, it is determined that the business operation corresponding to the target account is at risk.

[0057] In addition, to achieve the above-mentioned purpose, the present invention also provides a risk monitoring device, which includes: a memory, a processor, and a risk monitoring program stored in the memory and executable on the processor, and the risk monitoring program implements the steps of the risk monitoring method described above when executed by the processor.

[0058] In addition, to achieve the above-mentioned purpose, the present invention also provides a computer-readable storage medium, on which a risk monitoring program is stored, and when the risk monitoring program is executed by a processor, the steps of the risk monitoring method as described above are implemented.

[0059] The risk monitoring method proposed in the present invention generates risk data corresponding to the operation data if it is detected that the operation data based on the target account has been written; determines the operation type to which the operation data belongs, and parses the risk data into target data corresponding to the operation type; and determines whether the business operation corresponding to the target account is risky based on the target data. The present invention records the operation data in real time to generate the risk data of the target account, and parses the risk data into target data, so that the data sources based on the risk analysis are comprehensive, preventing the situation where data collection is easily missed, and finally determines whether the business operation corresponding to the target account is risky through the target data, thereby improving the accuracy of risk analysis. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 It is a schematic diagram of the device structure of the hardware operating environment involved in the embodiment of the present invention;

[0061] Figure 2 This is a flow chart of the first embodiment of the risk monitoring method of the present invention;

[0062] Figure 3 This is a schematic diagram of the architecture of a risk monitoring device in the first embodiment of the risk monitoring method of the present invention;

[0063] Figure 4 Schematic diagram of the second embodiment of the risk monitoring method of the present invention.

[0064] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0065] It should be understood that the specific embodiments described herein are only used to explain the present invention, and are not used to limit the present invention.

[0066] like Figure 1 As shown, Figure 1 It is a schematic diagram of the device structure of the hardware operating environment involved in the embodiment of the present invention.

[0067] The device in the embodiment of the present invention may be a PC or a server device.

[0068] like Figure 1 As shown, the device may include: a processor 1001, such as a CPU, a network interface 1004, a user interface 1003, a memory 1005, and a communication bus 1002. Among them, the communication bus 1002 is used to realize the connection and communication between these components. The user interface 1003 may include a display screen (Display), an input unit such as a keyboard (Keyboard), and the optional user interface 1003 may also include a standard wired interface and a wireless interface. The network interface 1004 may optionally include a standard wired interface and a wireless interface (such as a WI-FI interface). The memory 1005 may be a high-speed RAM memory, or it may be a stable memory (non-volatile memory), such as a disk memory. The memory 1005 may also be a storage device independent of the aforementioned processor 1001.

[0069] Those skilled in the art will understand that Figure 1 The device structure shown in the figure does not constitute a limitation of the device, and may include more or less components than shown in the figure, or combine certain components, or arrange the components differently.

[0070] like Figure 1 As shown, the memory 1005 as a computer storage medium may include an operating system, a network communication module, a user interface module, and a risk monitoring program.

[0071] Among them, the operating system is a program for managing and controlling risk monitoring equipment and software resources, supporting the operation of network communication modules, user interface modules, risk monitoring programs and other programs or software; the network communication module is used to manage and control the network interface 1002; the user interface module is used to manage and control the user interface 1003.

[0072] exist Figure 1 In the risk monitoring device shown, the risk monitoring device calls the risk monitoring program stored in the memory 1005 through the processor 1001, and executes the operations in each embodiment of the risk monitoring method described below.

[0073] Based on the above hardware structure, an embodiment of the risk monitoring method of the present invention is proposed.

[0074] Reference Figure 2 , Figure 2 This is a flow chart of a first embodiment of a risk monitoring method of the present invention, the method comprising:

[0075] Step S10: if it is detected that the writing of the operation data based on the target account is completed, then risk data corresponding to the operation data is generated;

[0076] Step S20, determining the operation type to which the operation data belongs, and parsing the risk data into target data corresponding to the operation type;

[0077] Step S30: Based on the target data, determine whether the business operation corresponding to the target account has risks.

[0078] The risk monitoring method of this embodiment is applied to the risk monitoring equipment of financial institutions such as financial management institutions or banks. The risk monitoring equipment can be a terminal, a robot or a PC device. For the convenience of description, the risk monitoring equipment is described as a monitoring equipment. The monitoring equipment includes a business system, a first database and a second database, wherein the first database is used to record the operation data of the business system, and the second database is used to record the risk data corresponding to the operation data. In the actual operation process, the operation behavior of the customer or the salesperson in the business system will be recorded, that is, the business system of the monitoring device will write the operation data corresponding to the operation behavior based on the current account into the first database, and then generate the risk data corresponding to the operation data in the first database in the second database. The framework of the monitoring device of this embodiment is based on the existing business system and the first database, and then adds a second database, such as Figure 3 In the specific implementation, the data of the second database is recorded in binlog (binary log, binary log, log data reflecting the real data changes of the database). Of course, other methods can also be used to reflect the data changes of the current operation behavior, such as providing records before the operation and results after the operation.

[0079] Binlog is a binary structure. After the binlog parsing layer, the monitoring device parses the risk data into data in a more general processing format, that is, the target data, and then sends the target data to a distributed message queue, such as Kafka (a distributed message queue that can quickly process large-scale messages). The target data in the message queue is consumed by a distributed stream processor, such as spark streaming (a distributed stream processing framework) or flink (a distributed stream processing framework). The choice of stream processor is not limited. As long as it can support horizontal expansion when the data volume is large and use more machines to ensure processing timeliness, it can be selected.

[0080] The process of stream processor consumption is also the process of risk analysis. The stream processor specifically includes a rule engine module and a recurrent neural network model. The rule engine module is used to determine whether the current operation behavior hits the risk rule. In the specific implementation, the risk rule is specifically a judgment function, that is, to determine whether the current operation line satisfies the judgment function. The rule engine module can be implemented using open source Drools (an open source rule system) and the like. Of course, other existing rule systems can also be used as long as they are easy to write and understand, have high enough operating efficiency, and do not affect the timeliness of data processing. The recurrent neural network model is also used to determine whether the current operation behavior is risky. The principle is to train an intelligent model based on prior knowledge, such as historical operation data and corresponding historical operation risks, so as to determine whether the current operation behavior is risky. The recurrent neural network model is not limited to LSTM (Long Short-Term Memory, a time recursive neural network), GRU (Gated Recurrent Unit, a variant of LSTM), etc.

[0081] This embodiment only needs to obtain relatively comprehensive target data. During the risk analysis process, the target data is input into the corresponding rule engine module or recurrent neural network model to determine whether the current business operation has risks.

[0082] The following is a detailed description of each step:

[0083] Step S10: If it is monitored that the writing of the operation data based on the target account is completed, risk data corresponding to the operation data is generated.

[0084] In this embodiment, if a customer or salesperson performs an operation on the current target account, such as transferring money, the business system writes the operation data of the current operation, such as adding a new transfer record, into the first database. When the monitoring device detects that the operation data based on the target account is written, it generates risk data corresponding to the operation data in the second database. During the specific implementation, a binlog data is specifically generated.

[0085] Furthermore, when it is monitored that the writing of the operation data based on the target account is completed, the operation type of the current operation data is determined, and based on the operation type, risk data corresponding to the current operation data is generated in the second database, wherein the operation type includes addition, deletion and modification, etc., that is, different risk data is generated in the second database according to the different operation types of the current operation data.

[0086] Assume that the current business system has a table A, which contains fields A1, A2, ...An. When the operation types corresponding to the current operation behaviors are add, delete and modify, the operation data recorded in the first database and the binlog data recorded in the second database binlog, i.e., risk data, are shown in the following table:

[0087]

[0088]

[0089] For the addition operation, the records in the first database and the binlog records in the second database can both check the added data; after deletion, the data recorded in the first database disappears, but the binlog record of the second database will have a deletion data; the difference between the two is more obvious for the modification operation. The first database only has the modified data, while the binlog record of the second database records the data before and after the modification.

[0090] It can be seen that binlog records are very important for risk analysis of operation lines. If a business system (which may be caused by human operation or business system defects, problematic design, etc.) turns on a switch to avoid auditing, and then turns it off afterwards, causing major risks, but the first database shows that it is normal afterwards, the binlog records in the second database can truly restore the course of the event.

[0091] Since the data of the binlog records in the second database is several times that of the records in the first database, in specific implementation, a distributed architecture is used to parse the binlog data in the binlog records.

[0092] Step S20 determines the operation type to which the operation data belongs, and parses the risk data into target data corresponding to the operation type.

[0093] In this embodiment, the monitoring device first determines the operation type to which the current operation data belongs, where the operation type includes adding, deleting, and modifying, etc. Then, according to the operation type, the risk data is parsed into target data, that is, the binlog data is parsed into target data.

[0094] During specific implementation, a general format is defined in advance, and the risk data is parsed into various fields according to the general format, and each field is filled into the position corresponding to the general format, so as to obtain the target data.

[0095] Specifically, step S20 includes:

[0096] Determine the operation type to which the operation data belongs, and based on the operation type, determine the field value of each field in a preset general format, wherein the preset general format includes at least one field;

[0097] In this step, the monitoring device determines the operation type to which the current operation data belongs, and determines the field value of each field in the preset general format based on the operation type, wherein the preset general format includes at least one field, and in a specific implementation, the preset general format includes four fields.

[0098] For ease of understanding, suppose table A of the first database contains fields A1, A2, ...An, which define common formats for various database operations:

[0099] {

[0100] operate_type:insert|delete|upate,

[0101] operate_dt:2019-01-01 09:10:40,

[0102] A1_has_old_value:true|false,

[0103] A2_has_old_value:true|false, ...

[0105] An_has_old_value:true|false,

[0106] A1_old_value:

[0107] A2_old_value: ...

[0109] An_old_value:

[0110] A1_has_new_value:true|false,

[0111] A2_has_new_value:true|false, ...

[0113] An_has_new_value:true|false,

[0114] A1_new_value:

[0115] A2_new_value: ..

[0117] An_new_value:

[0118] }

[0119] Among them, operate_type indicates the operation type, which mainly includes three types: insert, delete, and update. Of course, in actual application, other operation types can be divided according to needs, such as general operation and special operation types; operate_dt indicates the operation time corresponding to the operation data; the data of each field is expanded into 4 parts, namely [xxx]_has_old_value, [xxx]_old_value, [xxx]_has_new_value, [xxx]_new_value, [xxx] is used to represent the risk field in the risk data. If the risk data includes the risk field A1, the preset general format is A1_has_old_value, A1_old_value, lue,A1_has_new_value,A1_new_value, and then, according to the operation type, determine the field value of each field. For example, if the current operation type is add, the field value of A1_has_old_value is false, the field value of A1_old_value is null, the field value of A1_has_new_value is true, and the field value of A1_new_value is the value after insertion, that is, A1; when the current operation type is delete, the field value of A1_has_old_value is true, the field value of A1_old_value is the value before deletion, that is, A1, the field value of A1_has_new_value is false, and the field value of A1_new_value is null, etc.

[0120] The preset general format is a pre-set format, the purpose of which is to expand the current risk data into four copies, which is convenient for data unification and for subsequent risk analysis to use the same set of analysis standards to analyze and judge all risk data. Of course, it can also be set to other numbers according to actual needs.

[0121] Based on a preset common format and the field value, target data corresponding to the risk data is generated.

[0122] The final fields and field values ​​are the target data, as shown in the following table:

[0123]

[0124]

[0125] As shown in the table above, the monitoring device includes the process and results of the add, delete, and modify operations in one record. For example, for a delete operation, the monitoring device records the value before the deletion; for a modify operation, the monitoring device records the value before and after the modification, respectively.

[0126] If the current risk data is A1 and the operation type is add, then the final target data is A1_has_old_value: false, A1_old_value: null value, A1_has_new_value: true, A1_new_value: A1.

[0127] Step S30 determines whether the business operation corresponding to the target account has risks based on the target data.

[0128] In this embodiment, the monitoring device uses two risk methods, namely manual rule determination and machine learning determination, to perform risk analysis on the target data by selecting appropriate risk analysis methods, thereby determining whether the business operations corresponding to the target account are risky.

[0129] Specifically, step S30 includes:

[0130] If it is determined that the operation data belongs to the preset scene data, determining a judgment function corresponding to the operation type;

[0131] In this step, the monitoring device determines whether the current operation data belongs to the preset scenario data, where the preset scenario data is a pre-set data type. For example, if the transaction amount is greater than 5,000, a risk alarm will be generated. Then, the operation data of new transfers exceeding 5,000 can be set as the preset scenario data. That is, some scenarios are relatively simple, the alarm reasons are easy to understand, and the operation data with strong explanatory power are set as preset scenario data.

[0132] If it is determined that the current operation data belongs to the preset scenario data, then the judgment function corresponding to the operation type corresponding to the current operation data is determined, that is, each type of operation type corresponds to a judgment function. When it is determined that the operation data of the target account belongs to the preset scenario data, it is determined that the current risk analysis adopts manual rule judgment. Therefore, it is necessary to determine which judgment function to use to perform risk analysis on it.

[0133] For ease of understanding, in the specific implementation, the following judgment function is defined to reflect the process and results of the operation:

[0134]

[0135]

[0136] From the above table, we can see that different judgment functions are used to analyze the risk of the current operation behavior according to different operation types. The specific judgment function can be set according to the actual situation. For example, the judgment function corresponding to the general operation in the above table is operation_type=update and xx_new_value>xx_old_value, etc. Of course, in the actual operation process, various judgment functions can also be combined to determine whether an operation is risky, such as:

[0137] The judgment function is: predict:insertDataLike(A1|A2,V1|V2) to judge that there is risk

[0138] predict means assertion, such as insertDataLike(A1|A2,V1|V2), which means that a record with A1=V1,A2=V2 is inserted, which is considered risky.

[0139] Based on the judgment function and the target data, it is determined whether the business operation corresponding to the target account has risks.

[0140] In this step, the monitoring device determines whether there is a risk in the business operation corresponding to the target account based on the judgment function and the target data, that is, determines whether the target data satisfies the judgment function so that the judgment function is established.

[0141] Specifically, determining whether a field value corresponding to the judgment function exists in the target data;

[0142] In this step, determine whether the target data contains a field value corresponding to the judgment function. For example, if the judgment function is predict:insertDataLike(A1|A2,V1|V2), determine whether the target data contains has_new_value:true, new_value:A1=V1,A2=V2.

[0143] If so, it is determined that the business operation corresponding to the target account has risks;

[0144] If it is determined to exist, it means that the conditions for manual rule determination are met, and it is determined that the current business operation has risks. If it does not exist, it is determined that the current business operation does not have risks.

[0145] Further, obtaining the operation time of the operation data, and determining a corresponding judgment function based on the operation time and the operation type;

[0146] Based on the judgment function and the target data, it is determined whether the business operation corresponding to the target account has risks.

[0147] It is understandable that in the actual operation process, there will be a series of business operations performed by customers or salesmen on the same account, that is, combined operations, such as adding a new data at the beginning and then modifying the data. At this time, it is necessary to conduct risk analysis on this series of business operations. Specifically, the combination rule, that is, the combined judgment function, is used to perform risk analysis. The operation time of the operation data, that is, the operation time of the business operation, is obtained, and then the sequence of a series of business operations for the target account is determined according to the operation time, and the corresponding judgment function is determined according to the corresponding operation type. The judgment function at this time is generally a combined function.

[0148] For example, if the overdraft attribute of an account is changed to "allowed", and the account is overdrafted, it is judged as risk-free:

[0149] predict: exists updateDataTo (account number|whether overdraft is allowed, acct1|no, acct1|yes) and updateDataTo (account number|account overdraft status, acct1|no overdraft, acct1|overdraft) determines that there is no risk, where Exists indicates whether there is a specified behavior before querying the account.

[0150] If the present embodiment detects that the writing of the operation data based on the target account is completed, the risk data corresponding to the operation data is generated; the operation type to which the operation data belongs is determined, and the risk data is parsed into the target data corresponding to the operation type; based on the target data, it is determined whether the business operation corresponding to the target account is at risk. The present invention records the operation data in real time to generate the risk data of the target account, and parses the risk data into the target data, so that the data sources based on the risk analysis are comprehensive, preventing the situation where data collection is easily missed, and finally determining whether the business operation corresponding to the target account is at risk through the target data, thereby improving the accuracy of the risk analysis.

[0151] Furthermore, based on the first embodiment of the risk monitoring method of the present invention, a second embodiment of the risk monitoring method of the present invention is proposed.

[0152] The difference between the second embodiment of the risk monitoring method and the first embodiment of the risk monitoring method is that, referring to Figure 4 , step S30 comprises:

[0153] Step S31, if it is determined that the operation data does not belong to the preset scenario data, the target data is input into the pre-trained risk monitoring model to obtain a corresponding risk level value;

[0154] Step S32: Based on the risk level value, determine whether the business operation corresponding to the target account has risks.

[0155] This embodiment uses a machine learning risk analysis method for some complex scenarios, that is, operation data that does not belong to the preset scenario data, so that the monitoring equipment can be applied to various application scenarios.

[0156] The following is a detailed description of each step:

[0157] Step S31, if it is determined that the operation data does not belong to the preset scenario data, the target data is input into the pre-trained risk monitoring model to obtain the corresponding risk level value.

[0158] In this embodiment, if the monitoring device determines that the operation data of the target account does not belong to the preset scenario data, it means that the situation is relatively complicated. At this time, the risk analysis method of machine learning can be used to input the target data into the pre-trained risk monitoring model to obtain the corresponding risk level value. In specific implementation, the risk monitoring model can be obtained using recurrent neural network training.

[0159] Specifically, before step S31, the risk monitoring method further includes:

[0160] Collecting historical target accounts, obtaining historical business operations corresponding to the historical target accounts, and historical target data corresponding to the historical business operations, and setting the historical target data as input of the initial model;

[0161] In this step, prior knowledge is collected in advance, such as historical target accounts, and historical business operations corresponding to historical target accounts and historical target data corresponding to historical business operations are obtained, and the target data is used as a training set as the input of the initial model. In order to ensure the high availability of the risk monitoring model obtained by subsequent training, the training set must be sufficient, that is, the prior knowledge collected must be sufficient, generally based on a data volume of millions.

[0162] Setting the historical risk level value corresponding to the historical business operation as the output of the initial model;

[0163] In this step, the historical risk level value corresponding to the historical business operation is set as the output of the initial model, wherein the historical risk level value is set manually. It can be understood that the historical data is known, so whether the historical business operation is risky is known. Based on the known results, the historical risk level value of the current historical business operation is determined. In the actual processing process, the actual risk loss can be evaluated to determine the risk level value of the current historical business operation. After the historical risk level values ​​of all historical business operations are evaluated, the historical risk level value is used as the output of the initial model.

[0164] Based on the historical target data and the historical risk level values, the initial model is trained into the risk monitoring model.

[0165] After obtaining the input and output of the initial model, the model is trained according to the recurrent neural network to train the initial model into a risk monitoring model.

[0166] Specific:

[0167] Initializing the neural network parameters of the initial model and inputting the historical target data into the initial model to obtain the current cycle result;

[0168] Determine the current error between the current cycle result and the historical risk level value, and adjust the initialized neural network parameters according to the current error to obtain the neural network parameters of the initial model in the next cycle, and cyclically execute the step of inputting the historical target data into the initial model to obtain the current cycle result;

[0169] If it is detected that the current error is the smallest, the risk monitoring model is constructed based on the initial model and the neural network parameters corresponding to the current error.

[0170] In this embodiment, a recurrent neural network is used to train the risk monitoring model. The principle is to initialize the neural network parameters, that is, to randomly select values, and use historical target data as the initial input to obtain a cycle result, and then reversely adjust the neural network parameters based on the cycle result. Specifically, the cycle result is compared with the historical risk level value, and the originally set neural network parameters are reversely adjusted according to the error. Through the cyclic adjustment, until the error between the current cycle result and the historical risk level value is minimized, the neural network parameters are optimal at this time, and the current neural network can be used to construct the risk monitoring model. In the specific implementation, forward propagation and reverse propagation are used to achieve it, as shown in the following example:

[0171] Assume that there is a batch of historical target accounts S, and their corresponding target data are Seq1, Seq2, Seq3, Seqn; where Seq = {X1, X2, ..Xm} represents the operation behavior of a certain account within a period of time, that is, {X1, X2, ..Xm} represents a series of business operations, in which each individual business operation is closely related to other business operations and has a sequence, so it is called operation sequence data, that is, the target data contains several operation sequence data. The risk monitoring model will learn such a mapping relationship:

[0172] Seq{t1,t2,...tm}->Y;

[0173] Among them, tm represents the risk level value of each operation sequence data, 0≤Y≤1, 0 represents no risk at all, and 1 represents the highest risk.

[0174] Specifically, the operation sequence data is input into the recurrent neural network, the output vector of the last stage of the recurrent neural network is taken, and the vector is mapped to a risk level value Y.

[0175] The recurrent neural network cyclically calculates the current business operation Xt and the result of the previous business operation in the sequence:

[0176] o t =σ(w it (x t ,h t-1 )+b ti );

[0177] Among them, σ is the sigmoid function, w it and b it is the neural network parameter, which is obtained through back propagation learning. it is the offset parameter, which is used to characterize the intersection point on the y-axis when the function is mapped to the coordinate axis. it is the weight value, initialize w during training it and b it , that is, random values, such as random values ​​between 0 and 1, are obtained by forward propagation. t Since we know whether a business operation is risky during training, we can compare the forward propagation result with the training target (that is, the expected value) and reversely correct w it and b it , until the result of the forward propagation has the smallest error with the training target, w it and b it i means that each operation in the operation sequence must be input into the neural network for training, h t-1 It follows the network output of the previous step, that is, the output of each step will be input to the next calculation, that is, the cyclic network, the first item of the operation sequence ht-1 It is also the initial value, and finally the o calculated last time in the sequence t is the Y value.

[0178] The neural network model has the ability to represent learning and can automatically learn the before-and-after correlation of operation sequence data, combining a series of operation behaviors of an account to determine whether there is risk. After the initial model is trained into a risk monitoring model, similar risky operation behaviors in history can be found in the future and a close score can be given to determine whether there is risk.

[0179] Further, step S31 includes:

[0180] If it is determined that the target data includes at least two operation sequence data, the operation sequence data are input into a pre-trained risk monitoring model to obtain each output value output by the risk monitoring model;

[0181] In this step, if it is determined that the target data contains several operation sequence data, that is, the operation behaviors for the current target account are a series, then each operation sequence data is input into the pre-trained risk monitoring model to obtain each output value of the risk monitoring model.

[0182] The final output value is taken as the risk level value corresponding to the target data.

[0183] Then, the final output value is taken as the risk level value of the target data. This is because a series of operations on the target account may result in the previous operations being invalid. For example, an account does not allow overdrafts at the beginning, but an excess amount is transferred. Later, it is changed to allow overdrafts. At this time, there is no risk of overdrafts in the account. Therefore, the last operation can be taken as the final operation.

[0184] In the actual processing process, the last sorted operation sequence data in the operation sequence data can be determined as the input of the risk monitoring model, and other operation sequence data do not need to be input, and the same result can be obtained.

[0185] This embodiment selects different risk analysis methods to perform risk analysis according to different application scenarios. For some simple scenarios, that is, the operation data belonging to the preset scenario data, manual rules are used to determine the risk. For some complex scenarios, that is, the operation data that does not belong to the preset scenario data, a machine learning risk analysis method is used. The two methods are combined to make the monitoring equipment applicable to various application scenarios and have strong interpretability.

[0186] The present invention also provides a risk monitoring device. The risk monitoring device of the present invention comprises:

[0187] A monitoring module, configured to generate risk data corresponding to the operation data if it is detected that the operation data based on the target account has been written;

[0188] A parsing module, used to determine the operation type to which the operation data belongs, and parse the risk data into target data corresponding to the operation type;

[0189] A determination module is used to determine whether a business operation corresponding to the target account has risks based on the target data.

[0190] Furthermore, the parsing module is also used for:

[0191] Determine the operation type to which the operation data belongs, and based on the operation type, determine the field value of each field in a preset general format, wherein the preset general format includes at least one field;

[0192] Based on a preset common format and the field value, target data corresponding to the risk data is generated.

[0193] Furthermore, the determining module is also used for:

[0194] If it is determined that the operation data does not belong to the preset scenario data, the target data is input into a pre-trained risk monitoring model to obtain a corresponding risk level value;

[0195] Based on the risk level value, determine whether the business operation corresponding to the target account has risks.

[0196] Furthermore, the determining module is also used for:

[0197] If it is determined that the target data includes at least two operation sequence data, the operation sequence data are input into a pre-trained risk monitoring model to obtain each output value output by the risk monitoring model;

[0198] The final output value is taken as the risk level value corresponding to the target data.

[0199] Furthermore, the risk monitoring model also includes a training module, and the training module is also used to:

[0200] Collecting historical target accounts, obtaining historical business operations corresponding to the historical target accounts, and historical target data corresponding to the historical business operations, and setting the historical target data as input of the initial model;

[0201] Setting the historical risk level value corresponding to the historical business operation as the output of the initial model;

[0202] Initializing the neural network parameters of the initial model and inputting the historical target data into the initial model to obtain the current cycle result;

[0203] Determine the current error between the current cycle result and the historical risk level value, and adjust the initialized neural network parameters according to the current error to obtain the neural network parameters of the initial model in the next cycle, and cyclically execute the step of inputting the historical target data into the initial model to obtain the current cycle result;

[0204] If it is detected that the current error is the smallest, the risk monitoring model is constructed based on the initial model and the neural network parameters corresponding to the current error.

[0205] Furthermore, the determining module is also used for:

[0206] If it is determined that the operation data belongs to the preset scene data, determining a judgment function corresponding to the operation type;

[0207] Based on the judgment function and the target data, it is determined whether the business operation corresponding to the target account has risks.

[0208] Furthermore, the determining module is also used for:

[0209] Determine whether there is a field value corresponding to the judgment function in the target data;

[0210] If so, it is determined that the business operation corresponding to the target account is at risk.

[0211] The present invention also provides a computer-readable storage medium.

[0212] A risk monitoring program is stored on the computer-readable storage medium of the present invention, and the risk monitoring program implements the steps of the risk monitoring method described above when executed by a processor.

[0213] Among them, the method implemented when the risk monitoring program running on the processor is executed can refer to the various embodiments of the risk monitoring method of the present invention, and will not be repeated here.

[0214] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or system 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, article or system. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the existence of other identical elements in the process, method, article or system including the element.

[0215] The serial numbers of the above embodiments of the present invention are only for description and do not represent the advantages or disadvantages of the embodiments.

[0216] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) as described above, and includes a number of instructions for a terminal device (which can be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in each embodiment of the present invention.

[0217] The above are only preferred embodiments of the present invention, and are not intended to limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A risk monitoring method, characterized in that: Applied to the server, the risk monitoring method comprises the following steps: If it is monitored that the operation data based on the target account is written into the first database, risk data corresponding to the operation data is generated in the second database, wherein the data in the second database is recorded in binlog; Determine the operation type to which the operation data belongs, and based on the operation type, determine the field value of each field in a preset common format, wherein the preset common format includes at least one field; Based on a preset common format and the field value, generating target data corresponding to the risk data; If it is determined that the operation data does not belong to the preset scenario data, the target data is input into the pre-trained risk monitoring model to obtain a corresponding risk level value; based on the risk level value, it is determined whether the business operation corresponding to the target account has a risk; If it is determined that the operation data belongs to the preset scenario data, determine the judgment function corresponding to the operation type; determine whether there is a field value corresponding to the judgment function in the target data; if so, determine that there is a risk in the business operation corresponding to the target account.

2. The risk monitoring method according to claim 1, characterized in that: The step of inputting the target data into the pre-trained risk monitoring model to obtain the corresponding risk level value comprises: If it is determined that the target data includes at least two operation sequence data, each of the operation sequence data is input into a pre-trained risk monitoring model to obtain each output value output by the risk monitoring model; The final output value is taken as the risk level value corresponding to the target data.

3. The risk monitoring method according to claim 2, characterized in that: Before the step of inputting the target data into the pre-trained risk monitoring model to obtain the corresponding risk level value, the risk monitoring method further includes: Collecting historical target accounts, obtaining historical business operations corresponding to the historical target accounts, and historical target data corresponding to the historical business operations, and setting the historical target data as input of the initial model; Setting the historical risk level value corresponding to the historical business operation as the output of the initial model; Initializing the neural network parameters of the initial model and inputting the historical target data into the initial model to obtain the current cycle result; Determine the current error between the current cycle result and the historical risk level value, and adjust the initialized neural network parameters according to the current error to obtain the neural network parameters of the initial model in the next cycle, and cyclically execute the step of inputting the historical target data into the initial model to obtain the current cycle result; If it is detected that the current error is the smallest, the risk monitoring model is constructed based on the initial model and the neural network parameters corresponding to the current error.

4. A risk monitoring device, characterized in that: The risk monitoring device comprises: A monitoring module, configured to generate risk data corresponding to the operation data in a second database if it is detected that the operation data based on the target account in the first database is written completely, wherein the data in the second database is recorded in binlog; a determination module, configured to determine the type of operation to which the operation data belongs, and based on the type of operation, determine the field value of each field in a preset general format, wherein the preset general format includes at least one field; A parsing module, used for generating target data corresponding to the risk data based on a preset general format and the field value; The determination module is also used to input the target data into a pre-trained risk monitoring model to obtain a corresponding risk level value if it is determined that the operation data does not belong to the preset scenario data; determine whether the business operation corresponding to the target account is risky based on the risk level value; and to determine the judgment function corresponding to the operation type if it is determined that the operation data belongs to the preset scenario data; determine whether there is a field value corresponding to the judgment function in the target data; if so, determine that the business operation corresponding to the target account is risky.

5. A risk monitoring device, characterized in that: The risk monitoring device comprises: a memory, a processor, and a risk monitoring program stored in the memory and executable on the processor. When the risk monitoring program is executed by the processor, the steps of the risk monitoring method according to any one of claims 1 to 3 are implemented.

6. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a risk monitoring program, which, when executed by a processor, implements the steps of the risk monitoring method according to any one of claims 1 to 3.