CN116777606B - Intelligent anti-fraud system and method for bank credit based on relation graph - Google Patents
Intelligent anti-fraud system and method for bank credit based on relation graph Download PDFInfo
- Publication number
- CN116777606B CN116777606B CN202310818794.1A CN202310818794A CN116777606B CN 116777606 B CN116777606 B CN 116777606B CN 202310818794 A CN202310818794 A CN 202310818794A CN 116777606 B CN116777606 B CN 116777606B
- Authority
- CN
- China
- Prior art keywords
- fraud
- rule
- state
- model
- data
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 48
- 238000012545 processing Methods 0.000 claims abstract description 11
- 238000004364 calculation method Methods 0.000 claims abstract description 9
- 238000004140 cleaning Methods 0.000 claims abstract description 9
- 238000012360 testing method Methods 0.000 claims description 28
- 238000012549 training Methods 0.000 claims description 22
- 239000011159 matrix material Substances 0.000 claims description 21
- 238000013210 evaluation model Methods 0.000 claims description 19
- 238000013135 deep learning Methods 0.000 claims description 17
- 230000008569 process Effects 0.000 claims description 16
- 230000006870 function Effects 0.000 claims description 11
- 238000013461 design Methods 0.000 claims description 10
- 230000008859 change Effects 0.000 claims description 7
- 238000011835 investigation Methods 0.000 claims description 7
- 238000006243 chemical reaction Methods 0.000 claims description 4
- 238000010276 construction Methods 0.000 claims description 4
- 230000014509 gene expression Effects 0.000 claims description 4
- 230000010354 integration Effects 0.000 claims description 4
- 238000011144 upstream manufacturing Methods 0.000 claims description 4
- 230000000694 effects Effects 0.000 claims description 3
- 230000001788 irregular Effects 0.000 claims description 3
- 238000004891 communication Methods 0.000 claims description 2
- 238000011156 evaluation Methods 0.000 claims description 2
- 238000000605 extraction Methods 0.000 claims description 2
- 230000004927 fusion Effects 0.000 claims description 2
- 230000003993 interaction Effects 0.000 claims description 2
- 230000002159 abnormal effect Effects 0.000 description 9
- 238000010586 diagram Methods 0.000 description 5
- 230000001276 controlling effect Effects 0.000 description 4
- 230000005856 abnormality Effects 0.000 description 3
- 230000009471 action Effects 0.000 description 3
- 230000006399 behavior Effects 0.000 description 3
- 238000002372 labelling Methods 0.000 description 3
- 238000012986 modification Methods 0.000 description 3
- 230000004048 modification Effects 0.000 description 3
- 241000700189 Hystrix <Rodentia> Species 0.000 description 2
- 230000008878 coupling Effects 0.000 description 2
- 238000010168 coupling process Methods 0.000 description 2
- 238000005859 coupling reaction Methods 0.000 description 2
- 238000013136 deep learning model Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 238000007726 management method Methods 0.000 description 2
- 230000004044 response Effects 0.000 description 2
- 238000009825 accumulation Methods 0.000 description 1
- 125000002015 acyclic group Chemical group 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000015556 catabolic process Effects 0.000 description 1
- 238000012937 correction Methods 0.000 description 1
- 238000007405 data analysis Methods 0.000 description 1
- 238000013523 data management Methods 0.000 description 1
- 238000007418 data mining Methods 0.000 description 1
- 238000013499 data model Methods 0.000 description 1
- 230000007123 defense Effects 0.000 description 1
- 238000006731 degradation reaction Methods 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000007689 inspection Methods 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 230000006855 networking Effects 0.000 description 1
- 238000012797 qualification Methods 0.000 description 1
- 230000001105 regulatory effect Effects 0.000 description 1
- 238000012552 review Methods 0.000 description 1
- 238000010079 rubber tapping Methods 0.000 description 1
- 238000012216 screening Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q40/00—Finance; Insurance; Tax strategies; Processing of corporate or income taxes
- G06Q40/03—Credit; Loans; Processing thereof
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F9/00—Arrangements for program control, e.g. control units
- G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
- G06F9/44—Arrangements for executing specific programs
- G06F9/448—Execution paradigms, e.g. implementations of programming paradigms
- G06F9/4482—Procedural
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N5/00—Computing arrangements using knowledge-based models
- G06N5/02—Knowledge representation; Symbolic representation
- G06N5/022—Knowledge engineering; Knowledge acquisition
- G06N5/025—Extracting rules from data
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N7/00—Computing arrangements based on specific mathematical models
- G06N7/01—Probabilistic graphical models, e.g. probabilistic networks
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Software Systems (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Business, Economics & Management (AREA)
- Finance (AREA)
- Mathematical Physics (AREA)
- Computing Systems (AREA)
- Evolutionary Computation (AREA)
- Data Mining & Analysis (AREA)
- Accounting & Taxation (AREA)
- Artificial Intelligence (AREA)
- Algebra (AREA)
- Probability & Statistics with Applications (AREA)
- Computational Mathematics (AREA)
- General Business, Economics & Management (AREA)
- Technology Law (AREA)
- Mathematical Analysis (AREA)
- Mathematical Optimization (AREA)
- Pure & Applied Mathematics (AREA)
- Strategic Management (AREA)
- Marketing (AREA)
- Economics (AREA)
- Development Economics (AREA)
- Computational Linguistics (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a system and a method for intelligent anti-fraud of bank credit based on a relationship graph, and belongs to the technical field of financial credit. The invention comprises the following steps: s10: processing and cleaning the access data by using a cleaning platform, storing the processed and cleaned data into a fraud database, and constructing a new fraud database; s20: formulating an anti-fraud strategy through fraud data in a fraud database constructed in the S10, deriving fraud rules and fraud models according to the formulated anti-fraud strategy, and configuring the derived fraud rules and fraud models; s30: the method comprises the steps of carrying out combined operation on the fraud model, the fraud rule and the expert model formulated according to the issuing file in the S20, splicing the fraud model into a fraud model with complex calculation logic, and executing anti-fraud work by utilizing the spliced fraud model.
Description
Technical Field
The invention relates to the technical field of financial credit, in particular to a system and a method for intelligent anti-fraud of bank credit based on a relationship graph.
Background
With the continuous development of the networking age, the online fraud mode is stored endlessly, the traditional anti-fraud means comprise the initial manual detection, a black-and-white list, a rule engine, a supervised learning algorithm and the current non-supervised learning, so far, fraud and anti-fraud modes can be changed variously, and the method is the same as the method. Under the temptation of huge interests, fraudsters continuously expand teams, upgrade technologies and change attack modes, thereby bringing huge threats to individuals and enterprises.
Because the fraud means are very different day by day, and fraud personnel and strategy personnel are in the role of attack and defense, if the strategy personnel cannot react at the first time, a great deal of data analysis and mining are required to be carried out afterwards, and new characteristics and rules can be extracted. In addition, policies need to be updated on an irregular basis and kept strictly secret, and existing credit anti-fraud systems cannot identify customers with higher risk, and thus cannot minimize the potential fraud risk that exists.
Disclosure of Invention
The invention aims to provide a system and a method for intelligent anti-fraud of bank credit based on a relation graph, which are used for solving the problems in the background technology.
In order to solve the technical problems, the invention provides the following technical scheme: a method for intelligent anti-fraud of bank credit based on a relationship graph, the method comprising:
s10: accessing the credit investigation system and the equipment embedded point information of the client into an intelligent anti-fraud system through an API (application program interface), accessing the three-party data information of the client into the intelligent anti-fraud system through an API of a plain road, accessing a risk file issued by a central office into the intelligent anti-fraud system in a file interaction mode, identifying a fraud strategy in the risk file issued by the central office, interfacing the intelligent anti-fraud system with a cleaning platform through the API, processing and cleaning the client credit investigation report information, the client equipment embedded point information, the three-party data information and the identified fraud strategy information in the credit investigation system by using the cleaning platform, storing the processed and cleaned data into a fraud database, and constructing a new fraud database, wherein the API generally refers to an application programming interface, and the processing and cleaning refers to correction processing of credit investigation report data, in-row white list data, in-row blacklist data, plain road data and three-party model tapping ports;
s20: formulating an anti-fraud strategy through fraud data in a fraud database constructed in the S10, deriving fraud rules and fraud models according to the formulated anti-fraud strategy, and configuring the derived fraud rules and fraud rule sets;
s30: combining the fraud model, fraud rules and expert model formulated according to the issuing file in the S20 to splice the fraud model into a complex calculation logic fraud model, and executing anti-fraud work by using the spliced fraud model;
s40: carrying out service combination on the data accessed in the S10, the fraud rules and fraud models derived in the S20 and the expert model formulated according to the downlink issuing file to derive a matrix anti-fraud model;
s50: completing node configuration of a decision canvas according to the name of the decision stream and the description of the decision stream, determining the connection condition among the configuration nodes according to possible results output by the configuration nodes, and configuring an anti-fraud decision stream through a DAG (DAG-directed acyclic graph) based on the determination results;
s60: controlling the derived fraud rules in S20 and the state circulation of the anti-fraud decision flow configured in S50, and completing compliance approval of the rules and the decision flow;
s70: according to the configured call name, call number, call decision flow (the decision flow is the anti-fraud decision flow configured in S50) and call description, the required parameters of a feed (the feed refers to the system of submitting a personal basic data to a bank or the system of a financial institution after the personal basic data is prepared) are checked in a page dictionary, whether the required parameters are necessary input items or not is defined, an output result is automatically checked according to an anti-display result, the whole call link depends on state flow in S60, and the states of rules, rule sets, formula editors and decision matrixes used in the decision flow are required to pass approval;
the system is connected with a plurality of external interfaces as judging indexes of fraud rules, the return values of the external interfaces are a group of data, the page dictionary of the system is the return data of the associated interfaces, when in use, the corresponding external interfaces are found through the dictionary, and the corresponding return data are obtained through the found external interfaces;
s80: inquiring the case list and case detail information of the person to check whether the checking result is reasonable;
fraud results of the anti-fraud system are classified into two types, one type is approved by manual intervention, the other type is approved by the system, and cases approved by the system and manual intervention can be checked through inquiring a list of the cases, detailed information of the cases is checked, and whether the inspection results are reasonable or not is checked based on the checking information.
Further, the step S20 includes:
s201: formulating an anti-fraud strategy according to fraud data in the fraud database constructed in the step S10, adding fraud rules according to the formulated anti-fraud strategy, and defining the names, types and descriptions of the added fraud rules;
s202: configuring the running condition of the newly-added fraud rule in S201, wherein the data required by the running condition of the newly-added fraud rule is derived from the access data in S10, and obtaining a final execution rule according to the configured running condition of the newly-added fraud rule, wherein the configured running condition supports a newly-added single condition or a newly-added condition group;
s203: performing back display by using indexes (indexes refer to data in a fraud database used by the final execution rule when the final execution rule is executed) used by the final execution rule obtained in the step S202, matching an execution return result of the final execution rule with the back display result, and testing whether the rule execution is accurate or not according to the matching result, wherein the indexes are manually input;
s204: configuring names and descriptions of the newly-added fraud rule sets according to the fraud data in the fraud database constructed in the S10 to configure the content as the definition of the newly-added fraud rule sets;
s205: selecting the newly-added fraud rules defined in the step S201, configuring the execution sequence of the selected newly-added fraud rules, wherein the execution sequence comprises sequential execution (a plurality of groups of newly-added fraud rules are sequentially executed according to the configuration sequence, if the execution return result of a group of newly-added fraud rules is false, the subsequent newly-added fraud rules are not executed any more) and traversal execution (all the configured newly-added fraud rules are executed, combining rule sets configured in the step S204 according to the execution result, and returning the combined rule sets);
s206: and (3) performing a related logic test by using an index used by the newly-added fraud rule set (the index refers to data in a fraud database used by the newly-added fraud rule set when the newly-added fraud rule set is executed), and testing whether the newly-added fraud rule set is executed accurately according to a logic test result, wherein the related logic test refers to judging whether an output result configured by the test rule set meets expectations or not.
Further, a personal anti-fraud evaluation model is built through the fraud model derived in the step S20, personal fraud scores are trained through the built personal anti-fraud evaluation model, and the specific training process is as follows:
the personal information is acquired, the acquired information is divided into structured data, semi-structured data and unstructured data according to data types, mutual relations among entities, attributes and entities in the unstructured data and the semi-structured data are extracted, the structured data and access data in S10 are subjected to data integration, the integrated data and an extraction result are fused, and a personal anti-fraud evaluation model is constructed according to the fusion result;
structured data refers to data that can be represented and stored using a relational database; unstructured data refers to data without a fixed structure; semi-structured data refers to data model structures associated in a form that does not conform to a relational database or other data table, but contains associated labels that separate semantic elements and hierarchy records and fields;
constructing a deep learning personal anti-fraud evaluation model, training the deep learning personal anti-fraud evaluation model through an open source packet of tensorflow (tensorflow is an end-to-end open source machine learning platform) and keras (keras is an open source artificial neural network library written by Python), developing training language by taking Python as the model, and selecting training samples and test samples, wherein the ratio of the training samples to the test samples is 8:3;
the importance of the selected 30 model entering features is evaluated through the importance of the input disturbance features, after the evaluation, the 30 model entering features are arranged according to the order of the importance of the disturbance features from high to low, different thresholds are sequentially selected to screen the model entering features, the screened model entering features are trained for multiple times by using a determined deep learning personal anti-fraud evaluation model, the model entering features are determined through the training effect of the deep learning personal anti-fraud evaluation model, an optimal deep learning personal anti-fraud evaluation model is output according to the determined model entering features, and the output model is stored;
in the training process, a model learning curve is drawn to show the change conditions of a loss function, the accuracy of a training sample and the accuracy of a test sample in the model training process along with the model iteration process, and the convergence condition of the deep learning model is judged according to the change conditions;
after the optimal deep learning personal anti-fraud scoring model is established, predicting enterprise fraud probability through the optimal deep learning personal anti-fraud evaluation model, carrying out standard scoring card conversion processing on the predicted enterprise fraud probability to obtain enterprise fraud scores, then checking the distribution condition of the overall fraud scores of the training sample through a normal checking method, if the distribution of the fraud scores does not accord with the score result of normal distribution, adjusting the distribution condition of the fraud scores through a score adjustment and score conversion method, and determining the final enterprise fraud score according to the adjustment result;
the disturbance characteristics refer to characteristics of changing results when the characteristics are used and predicted after characteristic disturbance is carried out on the model; by testing and adjusting disturbance characteristics, the interpretability and the understandability of the model are improved, so that the model meets the requirements of practical application;
the in-mold feature refers to a variable with better performance selected by a related method of feature engineering; the model fitting effect is ensured to a large extent by the model entering feature;
the deep learning algorithm is suitable for parallel realization of large-scale training samples, so that the running efficiency of anti-fraud recognition is higher, the realization path and the realization scene for carrying out enterprise fraud risk recognition based on high-dimensional features and massive training samples are expanded, and the application scene of the enterprise risk recognition assisted personal credit service in the credit field is expanded by applying the deep learning algorithm. Meanwhile, the timing update of the learning model can be realized through the iterative tuning of the deep learning model so as to adapt to the continuous update of fraud means, and the interception strength of high-risk fraud enterprises is ensured.
Further, the step S30 includes:
s301: defining a formula editor name, a formula editor type and a formula editor description;
s302: defining logic of the formula editor defined in S301 when the formula editor is actually executed, integrating the logic of the fraud model derived in S20, the fraud rule and the expert model formulated according to the file issued in the central row according to the access data in S10 and the fraud rule derived in S20, configuring the fraud model of the complex calculation logic through the formula editor based on the integration result, and outputting a required result by utilizing the fraud model of the complex design logic, wherein the formula editor introduces a Fel expression calculation engine;
s303: defining the output result of the fraud model of the complex design logic in S302, matching the defined output result with the output result obtained in S302, and checking the fraud model of the configured complex design logic according to the matching result;
for example: log (Log) p (personal score model) +ln (personal credit model) +revenue score model = new derivative model.
Further, the step S40 includes:
s401: defining a matrix name, a matrix description and a matrix type;
s402: performing service combination according to the access data in S10, the fraud rules and fraud rule sets derived in S20, forming an irregular matrix according to the combination result, and outputting the combination result;
s403: and configuring a plurality of groups of results according to the selectable content of the matrix, testing the configuration results, and deriving a matrix anti-fraud model according to the testing results.
Further, the S50 includes:
s501: defining a decision flow name and a decision flow description;
s502: completing node configuration of a decision canvas, wherein the decision canvas can select nodes as access data in S10, fraud rules and fraud rule sets derived in S20, a formula editor in S30 and a decision matrix in S40, carrying out arbitrary combination connection on the configured nodes, and outputting a final decision result;
s503: and testing the flow configured by the decision, and checking the accuracy of the configured anti-fraud decision flow according to the test result.
Further, the step S60 includes:
s601: controlling regular state circulation;
the method comprises the steps that a rule list is newly built, the rule in the newly built rule list is initialized, the rule initial state is a to-be-checked state, a rule creator is obtained through a query rule interface provided by the rear end, if a login person obtained by the front end is a creator, an edit button is displayed, otherwise, a check button is displayed, the approval state in the rule list is updated to be the to-be-checked state, an administrator approves the rule in the to-be-checked state, if the approval state of the rule is updated to be the approval passing state, the approval passing state is indicated, if the approval state of the rule is updated to be the to-be-checked state, the approval rejecting is indicated, the rule set can be used by a decision flow when the rule set is in the to-be-approved and the approval passing state is indicated, the rule set can be disabled when the rule set is not referenced, the update rule set is in the approval rejecting state is the rule set is the approval passing state, and the rule set is all updated to be the approval passing state is the approval passing state;
s602: controlling rule set state circulation;
creating a rule set, initializing the rule in the new rule set to enable the rule initial state to be a to-be-checked state, acquiring a rule creator through a query rule interface provided by the rear end, displaying an edit button if a login acquired by the front end is a creator, otherwise displaying a check button, updating the approval state in a rule set table to be the to-be-checked state, enabling an administrator to examine the rule in the to-be-checked state, enabling the rule set to be approved if the approval state of the rule set is updated to be approved, enabling the rule set to be approved and approved if the approval state of the rule set is updated to be the to-be-checked state, enabling the rule set to be used by a decision flow when the rule set is not referred to, enabling the rule set to be disabled when the rule set is not referred to, enabling the rule set to be updated to be approved when the rule set is updated, and enabling the rule set to be updated and the rule in the decision flow to be approved if the rule set is updated to be approved;
s603: controlling decision stream state circulation;
the method comprises the steps of creating a rule set, initializing rules in the new rule set, enabling the initial state of the rules to be a state to be checked, obtaining a rule creator through a query rule interface provided by the rear end, displaying an editing button if a login person obtained by the front end is a creator, otherwise displaying a check button, updating the approval state in a rule set table to be the state to be checked, enabling an administrator to examine the rules in the state to be checked, enabling the approval state of the rule set to be approved, enabling the rules to be approved, enabling the approval state of the rule set to be approved, enabling the rule set to be used by a decision flow when the rule set is in the state to be approved and approved, enabling the rule set to be disabled when the rule set is not referred to, enabling the update rule set to be in the state to be approved when the rule set is updated, enabling the update rule set to be in the state to be approved when the rule set is approved, and enabling all update states of the rules in the rule set to be approved.
Further, the specific process of formulating the anti-fraud policy in S20 by using the fraud data in the fraud database constructed in S10 is as follows:
(1) formulating an admission strategy through the equipment embedded point information, wherein the equipment embedded point information comprises: device GPS, device IP and device number; for example: the account number two times of login is smaller than amin, the equipment for logging in the same account number two times is different, and the IP attribution province inconsistency displayed by the account number two times of login/account movement distance abnormality (movement distance > maximum movement distance) within login interval time, the number of the accounts associated with the same IP is too large or the number of the accounts associated with the same equipment is too large within the period of near b hours, a is more than 0, and b is more than 0;
(2) issuing files through a central row to formulate a risk strategy; for example: the account number is logged in a sensitive time period (point c-point d in early morning) and a region with high risk in different places, such as Yunnan province, the login times of the same customer in v minutes are more than or equal to X, the same account number is logged in T devices in Y days, the current device is strange device, meanwhile, the current IP attribution is a region with high risk (such as Yunnan province), and c, d, v, X, Y, T is larger than zero;
(3) formulating a policy by three-party data information accessed in the form of an API, the three-party data information comprising: multi-head information, property information, blacklist information, vehicle information, and revenue information; for example: the number of reiterations of the associated mobile phone number within 7 days of the same residence address;
(4) formulating a strong rule strategy through a credit report; the credit report includes: reporting heads, identity information, marital information, resident information, professional information, scoring information, credit transaction information summaries, non-credit transaction information summaries, public information summaries, query record summaries, debit account information, credit agreement information, related repayment liability information, post-paid business information, arrears information, civil decision records, enforcement records, administrative penalty records, housing public accumulation fund payment records, low-security rescue records, enforcement qualification records, administrative rewards records, other labeling and statement information (other labeling and statement information is mainly used for describing objection labeling and statement information related to personal basic information or related to whole credit reports) and query record information (query records record records that describe the case that personal credit reports were queried in the past two years); for example: judging whether the client has overdue behaviors or not;
(5) formulating a strategy according to the association condition between the information in the newly built fraud database and the client information; for example: the emergency contact of the client has blacklist behavior, the emergency contact of the client has fraud behavior, and whether the client is a risk client is judged.
A bank credit intelligent anti-fraud system based on a relationship graph, the credit intelligent anti-fraud system comprises an upstream system part, a front-end service, a back-end application, a data service and a registration center;
the front-end service comprises a definition parameter-in service, a definition parameter-out service, a definition call event, a definition rule set, a definition formula editor and a definition decision matrix service;
the back-end application comprises decision flow construction, a rule executor, a rule set executor and data access;
the data service comprises MySQL, REDIS and binlog, wherein MySQL represents a relational database management system, REDIS represents a remote dictionary service, and binlog represents a binary log for recording all database table structure changes and table data modifications;
the upstream system inlet is communicated with the front-end service, the front-end service is communicated with the back-end application, and the back-end application is used for respectively transmitting communication information to the data service and the registration center;
the credit intelligent anti-fraud system adopts Spring Cloud (Spring Cloud is an ordered set of a series of frameworks) as a basic framework, uses a rubber (the rubber is an open source item issued by Netflix, and mainly functions to provide a software load balancing algorithm and service call of a client) to make load balancing, openFeign (OpenFeign is a declarative and templated HTTP client) to make service call, hystrix (Hystrix is an open source library for processing delay and fault tolerance of a distributed system) to make service degradation, eureka is a service discovery framework developed by Netflix, is a REST-based service and is mainly used for positioning middle layer services running in an AWS domain) as a registration center, and uses Apollo (Apollo is based on an ActiveMQ prototype and is a faster, more reliable and easier message proxy tool) as a configuration center, so that the stability of an application framework is improved.
Compared with the prior art, the invention has the following beneficial effects:
1. the invention introduces DAG (Directed Acyclic Graph) diagram, namely directed acyclic diagram. The method is applied to decision flow construction and execution. The high coupling of decision flows with rules, rule sets, and scoring card models is solved. The low coupling and high cohesion of functions are ensured, and the reusability and portability of program modules are greatly enhanced. The diversity of fraud model configuration is ensured, multiple modes are provided for free combination of fraud models, the combination of black and white lists is included, and execution of the next node can be freely combined according to the type of the return value of the node.
2. The invention makes decision flow make its construction work by introducing executor concept, each back end executor processes its own internal logic to complete independent function calculation, and has independent function, plug and play, and can expand infinitely, break the natural mode of decision engine in market, introduce execution logic of self-defined fraud model, and only need to develop execution logic alone, decision flow does not need to make any change, so that anti-fraud model can be exerted to the maximum extent.
3. By introducing a Fel lightweight, efficient expression calculation engine in the formula editor, the Fel engine can typically execute tens of millions of expressions per second, and both our expert model and the trained fraud model can get a fast response. The execution of Fel is mainly realized by functions, operators (+, -and the like are Fel functions), all the functions can be replaced, and the expansion function is very simple, so that the response speed of a fraud model is ensured, and real-time feedback of fraud results is truly realized.
Drawings
The accompanying drawings are included to provide a further understanding of the invention and are incorporated in and constitute a part of this specification, illustrate the invention and together with the embodiments of the invention, serve to explain the invention. In the drawings:
FIG. 1 is a schematic workflow diagram of a system and method for intelligent anti-fraud of bank credit based on a relationship graph of the present invention;
FIG. 2 is a logic architecture flow diagram of a credit anti-fraud system of the intelligent anti-fraud system and method for bank credit based on a relationship graph of the present invention.
Detailed Description
The following description of the embodiments of the present invention will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present invention, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
Referring to fig. 1 and 2, the present invention provides the following technical solutions:
embodiment 1: an anti-fraud indicator (anti-fraud indicator refers to data in a fraud database) is accessed.
The anti-fraud index display page comprises index newly-added, index modified, index enabled, index disabled and index calling details;
clicking the new options, inputting new index data by the client, and storing the input data by the background;
clicking the editing option, leading the data information in the list into the front end, modifying and storing the data in the list by the client, and modifying the change information by the background according to the client storage information;
clicking the query option, and screening data according to the input parameters by the background and paging back;
clicking an enabling option, and modifying the data state from disabled to enabled in the background;
clicking a disable option, and modifying the data state from enable to disable in the background;
clicking the call detail option and calling the detail list in the background.
Embodiment 2: and (5) decision flow configuration.
Step 1: the configuration of the decision flow process is completed through a dragging form, and the name of the process is defined;
step 2: designing a decision flow:
step 2.1: outputting pass/reject/pass review by inputting rules, rule sets, formula editors and decision matrices;
step 2.2: the configuration of the decision flow is completed by adding nodes, drawing connecting lines, configuring nodes and the like on canvas in the newly added page of the decision flow;
step 2.3: to start as a starting node for the decision flow;
step 2.4: in the configuration window of the rule, the "selection rule" can pull down all configured rules in the selection decision engine, and the "rule result" can be the connection line of the data exception by selecting the connection line or rejecting the selection connection line, and when the data acquisition is abnormal (such as network exception, data missing, etc.), the check box of the "data acquisition exception handling" is selected, and the trend of the decision flow is the connection line of the data exception at this time;
step 2.5: in the configuration window of the rule set, the rule set can be selected by pulling down all the configured rule sets in the decision engine, and the rule set result can be the connection line through selection or connection line rejection, when the data acquisition is abnormal (such as network abnormality, data missing and the like), the check box of the data acquisition abnormality processing is selected, and the decision flow is in the trend of the data abnormal connection line;
step 2.6: the decision node indicates that a decision flow has a decision return value when executing the decision flow, the decision node can have a plurality of decision return values, and in a configuration window of the decision, the option of pulling down the decision return value includes passing, rejecting, suggesting passing, suggesting rejecting, customizing, last node value and transferring to check, and the last node value is selected to be returned, so that the execution result of the last node of the decision node needs to be returned when the decision flow is executed to the decision node; if the 'decision return value' selects the customization, the key and the value in the customization content need to be configured, the customization return value is provided with a constant/variable drop-down box, and if the customization return value does not select, the customization return value defaults to select the constant; when a constant is selected, the function of an input box is not regulated, when a variable is selected, a field name drop-down selection box appears, a drop-down list binds field contents in data management, the field contents are displayed in two stages, the first stage uses a field dimension, and the second stage uses a field name;
step 2.7: in the executing process of each node (except for starting, ending and deciding), if the data acquisition is abnormal, skipping the selection continuously executed by the node, checking the abnormal processing of the data acquisition, pulling down the path executed when the selected data acquisition is abnormal, after the selection, recording the result of the node as abnormal, if the node is selected to be continuously executed, and the abnormal processing of the data acquisition is not checked, ending the executing step 2.8) after the data acquisition is abnormal, ending the node which is the ending node of each branch in the decision flow, wherein the decision flow can be configured with a plurality of ending nodes, and the nodes have no specific configuration content;
step 2.8: after the filling value of the field required by the decision flow is input in the page, the test result can be output after clicking the test in the page.
Embodiment 3: event design is invoked.
Step 1: the call event page includes call event details, call event name, call event number, call event description, call decision flow, request parameters (request parameters refer to decision inflow parameters selected above), output parameters (output parameters refer to outflow parameters of decision flow selected above), and editing (in detail page editing).
Step 2: the configuration of the input and output parameters of the decision flow is called, page checking and filling are carried out, the input and output parameter templates are fixed, and the execution is carried out according to the templates;
step 3: designing a class diagram;
step 3.1: designing and inquiring, wherein the front end requests a background paging inquiry interface management list;
step 3.2: designing new addition and storage, and requesting a background to store interface data and configuration data information by the front end;
step 3.3: design editing, wherein the front end brings in data, clicks and saves the data, and the front end requests the background to update interface data and configuration table data information;
step 3.4: design starting, and a front end requests a background to update a data state;
step 3.5: design disabling, and front end requesting background updating data state;
step 3.6: designing test, wherein the front end requests the background to return a test result;
step 3.7: the method comprises the steps that an acquisition history version is designed and used for version switching in an editing page, and a front end requests a background to acquire history version data information;
step 4: and (5) designing a database.
Embodiment 4: deriving a matrix anti-fraud model.
For example:
| model A | Model B | Rule A | Rule B | Formula editor A |
| 10 | 50 | True | true | a |
| 20 | 60 | false | false | b |
| 30 | 70 | c | ||
| 40 | 80 | d |
The results of the matrix whose output results are 4*5 are freely combined. A scalable matrix of 4*5 results is output.
It is noted that relational terms such as first and second, and the like are used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Moreover, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus.
Finally, it should be noted that: the foregoing description is only a preferred embodiment of the present invention, and the present invention is not limited thereto, but it is to be understood that modifications and equivalents of some of the technical features described in the foregoing embodiments may be made by those skilled in the art, although the present invention has been described in detail with reference to the foregoing embodiments. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.
Claims (5)
1. A bank credit intelligent anti-fraud method based on a relation graph is characterized in that: the method comprises the following steps:
s10: accessing the credit investigation system and the equipment embedded point information of the client into an intelligent anti-fraud system through an API (application program interface), accessing the three-party data information of the client into the intelligent anti-fraud system through an API of a plain road, accessing a risk file issued by a central office into the intelligent anti-fraud system in a file interaction mode, identifying a fraud strategy in the risk file issued by the central office, interfacing the intelligent anti-fraud system with a cleaning platform through the API, processing and cleaning the client credit investigation report information, the client equipment embedded point information, the three-party data information and the identified fraud strategy information in the credit investigation system by using the cleaning platform, storing the processed and cleaned data into a fraud database, and constructing a new fraud database;
s20: formulating an anti-fraud strategy through fraud data in a fraud database constructed in the S10, deriving fraud rules and fraud models according to the formulated anti-fraud strategy, and configuring the derived fraud rules and fraud rule sets;
the S20 includes:
s201: formulating an anti-fraud strategy according to fraud data in the fraud database constructed in the step S10, adding fraud rules according to the formulated anti-fraud strategy, and defining the names, types and descriptions of the added fraud rules;
s202: configuring the running condition of the newly-added fraud rule in S201, wherein the data required by the running condition of the newly-added fraud rule is derived from the access data in S10, and obtaining a final execution rule according to the configured running condition of the newly-added fraud rule, wherein the configured running condition supports a newly-added single condition or a newly-added condition group;
s203: performing inverse display by using the index used by the final execution rule obtained in the step S202, matching the execution return result of the final execution rule with the inverse display result, and testing whether the rule execution is accurate according to the matching result, wherein the index value is manually input;
s204: configuring names and descriptions of the newly-added fraud rule sets according to the fraud data in the fraud database constructed in the S10 to configure the content as the definition of the newly-added fraud rule sets;
s205: selecting the newly-added fraud rule defined in S201, and configuring the execution sequence of the selected newly-added fraud rule, wherein the execution sequence comprises sequential execution and traversal execution;
s206: using the index used by the newly added fraud rule set as a relevant logic test, and testing whether the execution of the newly added fraud rule set is accurate or not according to the logic test result;
s30: combining the fraud model, fraud rules and expert model formulated according to the issuing file in the S20 to splice the fraud model into a complex calculation logic fraud model, and executing anti-fraud work by using the spliced fraud model;
the S30 includes:
s301: defining a formula editor name, a formula editor type and a formula editor description;
s302: defining logic of the formula editor defined in S301 when the formula editor is actually executed, integrating the logic of the fraud model derived in S20, the fraud rule and the expert model formulated according to the file issued in the central row according to the access data in S10 and the fraud rule derived in S20, configuring the fraud model of the complex calculation logic through the formula editor based on the integration result, and outputting a required result by utilizing the fraud model of the complex design logic, wherein the formula editor introduces a Fel expression calculation engine;
s303: defining the output result of the fraud model of the complex design logic in S302, matching the defined output result with the output result obtained in S302, and checking the fraud model of the configured complex design logic according to the matching result;
s40: carrying out service combination on the data accessed in the S10, the fraud rules and fraud models derived in the S20 and the expert model formulated according to the downlink issuing file to derive a matrix anti-fraud model;
the S40 includes:
s401: defining a matrix name, a matrix description and a matrix type;
s402: performing service combination according to the access data in S10, the fraud rules and fraud rule sets derived in S20, forming an irregular matrix according to the combination result, and outputting the combination result;
s403: configuring a plurality of groups of results according to the selectable content of the matrix, testing the configuration results, and deriving a matrix anti-fraud model according to the test results;
s50: completing node configuration of a decision canvas according to the name of the decision stream and the description of the decision stream, determining the connection condition among the configuration nodes according to the output result of each configuration node, and configuring an anti-fraud decision stream through a DAG based on the determination result;
the S50 includes:
s501: defining a decision flow name and a decision flow description;
s502: completing node configuration of a decision canvas, wherein the decision canvas can select nodes as access data in S10, fraud rules and fraud rule sets derived in S20, a formula editor in S30 and a decision matrix in S40, carrying out arbitrary combination connection on the configured nodes, and outputting a final decision result;
s503: testing the flow configured by the decision, and checking the accuracy of the configured anti-fraud decision flow according to the test result;
s60: controlling the derived fraud rules in S20 and the state circulation of the anti-fraud decision flow configured in S50, and completing compliance approval of the rules and the decision flow;
s70: according to the configured call name, call number, call decision flow and call description, checking the parameters required by the member in the page dictionary, defining whether the required parameters are necessary items or not, automatically checking the output result according to the back display result, and the whole call link depends on the state flow in S60, wherein the states of rules, rule sets, formula editors and decision matrixes used in the decision flow must be approved;
s80: and checking whether the auditing result is reasonable or not by inquiring the case list and the case detail information of the person.
2. The intelligent anti-fraud method for bank credit based on relation map according to claim 1, characterized in that: constructing a personal anti-fraud evaluation model through the fraud model derived in the step S20, and training personal fraud scores through the constructed personal anti-fraud evaluation model, wherein the specific training process is as follows:
the personal information is acquired, the acquired information is divided into structured data, semi-structured data and unstructured data according to data types, mutual relations among entities, attributes and entities in the unstructured data and the semi-structured data are extracted, the structured data and access data in S10 are subjected to data integration, the integrated data and an extraction result are fused, and a personal anti-fraud evaluation model is constructed according to the fusion result;
constructing a deep learning personal anti-fraud evaluation model, training the deep learning personal anti-fraud evaluation model through an open source packet of tensorf low and keras, developing a training language by taking python as a model, and selecting a training sample and a test sample with the proportion of 8:3;
the importance of the selected 30 model entering features is evaluated through the importance of the input disturbance features, after the evaluation, the 30 model entering features are arranged according to the order of the importance of the disturbance features from high to low, different thresholds are sequentially selected to screen the model entering features, the screened model entering features are trained for multiple times by using a determined deep learning personal anti-fraud evaluation model, the model entering features are determined through the training effect of the deep learning personal anti-fraud evaluation model, an optimal deep learning personal anti-fraud evaluation model is output according to the determined model entering features, and the output model is stored;
in the training process, a model learning curve is drawn to show the change conditions of a loss function, the accuracy of a training sample and the accuracy of a test sample in the model training process along with the model iteration process, and the convergence condition of a deep learning personal anti-fraud evaluation model is judged according to the change conditions;
after the optimal deep learning personal anti-fraud scoring model is established, the enterprise fraud probability is predicted through the optimal deep learning personal anti-fraud evaluation model, standard scoring card conversion processing is carried out on the predicted enterprise fraud probability to obtain enterprise fraud scores, then the distribution situation of the overall fraud scores of the training samples is checked through a normal checking method, if the distribution of the fraud scores does not accord with the score result of normal distribution, the distribution situation of the fraud scores is adjusted through a score adjustment and score conversion method, and the final enterprise fraud scores are determined according to the adjustment result.
3. The intelligent anti-fraud method for bank credit based on the relation map according to claim 2, wherein: the S60 includes:
s601: controlling regular state circulation;
the method comprises the steps that a rule list is newly built, rules in the new rule list are initialized, the initial state of the rules is a to-be-checked state, a rule creator is obtained through a query rule interface provided by the rear end, if a login person obtained by the front end is a creator, an edit button is displayed, otherwise, a check button is displayed, the approval state in the rule list is updated to be the to-be-checked state, an administrator approves the rules in the to-be-checked state, if the approval state of the rules is updated to be the approval passing state, the approval passing state is indicated, if the approval state of the rules is updated to be the to-be-checked state, the approval rejecting is indicated, the rule set is used by a decision flow when the state is to be approved and the approval passing state is indicated, the rule set is forbidden when the rule set is not referenced, if the approval rejecting is updated, the rule set is updated to be the approval passing state, and the approval state of the rules in the rule set is all updated to be the approval passing state;
s602: controlling rule set state circulation;
the method comprises the steps that a rule set is newly established, rules in the newly established rule set are initialized, the initial state of the rules is a state to be checked, a rule creator is obtained through an inquiry rule interface provided by the rear end, if a login person obtained by the front end is a creator, an edit button is displayed, otherwise, a check button is displayed, the approval state in an update rule set table is the state to be checked, an administrator approves the rules in the state to be checked, if the approval state of the rule set is updated to be approved, the approval is passed, if the approval state of the rule set is updated to be the state to be checked, approval rejection is indicated, the rule set is used by a decision flow when the state of the rule set is to be checked and approved, the rule set is forbidden when the rule set is not referenced, the state of the update rule set is approved rejection, and the states of the rules in the rule set and the decision flow are approved when the approval is passed;
s603: controlling decision stream state circulation;
the method comprises the steps of creating a rule set, initializing rules in the new rule set, enabling the initial state of the rules to be a state to be checked, obtaining a rule creator through a query rule interface provided by the rear end, displaying an editing button if a login person obtained by the front end is the creator, otherwise displaying a check button, updating the approval state in a rule set table to be the state to be checked, enabling an administrator to examine the rules in the state to be checked, enabling the approval state of the rule set to be the approval state, enabling the rule set to be used by a decision flow when the rule set is in the state to be checked and the approval state to be the approval state, disabling the rule set to be the approval state when the rule set is updated when the rule set is not referred, and enabling the update state of the rule set to be the approval state when the rule set is approved to be the approval state is the approval state.
4. A method for intelligent anti-fraud of bank credit based on a relationship graph according to claim 3, characterized in that: the specific process of making the anti-fraud policy by using the fraud data in the fraud database constructed in S10 in S20 is as follows:
(1) formulating an admission strategy through the embedded point information of the equipment;
(2) issuing files through a central row to formulate a risk strategy;
(3) formulating a strategy through three-party data information accessed in an API form;
(4) formulating a strong rule strategy through a credit report;
(5) and formulating a strategy according to the association condition between the information in the newly built fraud database and the client information.
5. A relational graph based intelligent anti-fraud system for bank credit applied to the relational graph based intelligent anti-fraud method for bank credit of any of claims 1 to 4, characterized in that: the credit intelligent anti-fraud system comprises an upstream system feed, a front-end service, a back-end application, a data service and a registration center;
the front-end service comprises a definition parameter-in service, a definition parameter-out service, a definition call event, a definition rule set, a definition formula editor and a definition decision matrix service;
the back-end application comprises decision flow construction, a rule executor, a rule set executor and data access;
the data services include MySQL, REDIS, and binlog;
the upstream system inlet and the front-end service are communicated with each other, the front-end service and the back-end application are communicated with each other, and the back-end application is used for transmitting communication information to the data service and the registration center respectively.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310818794.1A CN116777606B (en) | 2023-07-05 | 2023-07-05 | Intelligent anti-fraud system and method for bank credit based on relation graph |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310818794.1A CN116777606B (en) | 2023-07-05 | 2023-07-05 | Intelligent anti-fraud system and method for bank credit based on relation graph |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN116777606A CN116777606A (en) | 2023-09-19 |
| CN116777606B true CN116777606B (en) | 2024-01-23 |
Family
ID=87987819
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310818794.1A Active CN116777606B (en) | 2023-07-05 | 2023-07-05 | Intelligent anti-fraud system and method for bank credit based on relation graph |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116777606B (en) |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US7686214B1 (en) * | 2003-05-12 | 2010-03-30 | Id Analytics, Inc. | System and method for identity-based fraud detection using a plurality of historical identity records |
| CN110727922A (en) * | 2019-10-11 | 2020-01-24 | 集奥聚合(北京)人工智能科技有限公司 | Anti-fraud decision model construction method based on multi-dimensional data flow |
| CN112767136A (en) * | 2021-01-26 | 2021-05-07 | 天元大数据信用管理有限公司 | Credit anti-fraud identification method, credit anti-fraud identification device, credit anti-fraud identification equipment and credit anti-fraud identification medium based on big data |
| CN113850665A (en) * | 2021-09-14 | 2021-12-28 | 江苏中交车旺科技有限公司 | Logistics financial knowledge graph-based fraud prevention and control method and system |
| US11277437B1 (en) * | 2020-10-14 | 2022-03-15 | Paypal, Inc. | Automated device data retrieval and analysis platform |
| CN116225417A (en) * | 2023-05-08 | 2023-06-06 | 无锡锡商银行股份有限公司 | Financial platform decision engine management system and method based on big data |
| CN116228391A (en) * | 2022-12-28 | 2023-06-06 | 支付宝(杭州)信息技术有限公司 | Risk identification method and device, storage medium and electronic equipment |
Family Cites Families (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20150039512A1 (en) * | 2014-08-08 | 2015-02-05 | Brighterion, Inc. | Real-time cross-channel fraud protection |
| WO2019118868A1 (en) * | 2017-12-14 | 2019-06-20 | Resonate Networks, Inc. | Fraud detection in data sets using bayesian networks |
| CN111523676B (en) * | 2020-04-17 | 2024-04-12 | 第四范式（北京）技术有限公司 | Method and device for assisting machine learning model to be online |
-
2023
- 2023-07-05 CN CN202310818794.1A patent/CN116777606B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US7686214B1 (en) * | 2003-05-12 | 2010-03-30 | Id Analytics, Inc. | System and method for identity-based fraud detection using a plurality of historical identity records |
| CN110727922A (en) * | 2019-10-11 | 2020-01-24 | 集奥聚合(北京)人工智能科技有限公司 | Anti-fraud decision model construction method based on multi-dimensional data flow |
| US11277437B1 (en) * | 2020-10-14 | 2022-03-15 | Paypal, Inc. | Automated device data retrieval and analysis platform |
| CN112767136A (en) * | 2021-01-26 | 2021-05-07 | 天元大数据信用管理有限公司 | Credit anti-fraud identification method, credit anti-fraud identification device, credit anti-fraud identification equipment and credit anti-fraud identification medium based on big data |
| CN113850665A (en) * | 2021-09-14 | 2021-12-28 | 江苏中交车旺科技有限公司 | Logistics financial knowledge graph-based fraud prevention and control method and system |
| CN116228391A (en) * | 2022-12-28 | 2023-06-06 | 支付宝(杭州)信息技术有限公司 | Risk identification method and device, storage medium and electronic equipment |
| CN116225417A (en) * | 2023-05-08 | 2023-06-06 | 无锡锡商银行股份有限公司 | Financial platform decision engine management system and method based on big data |
Non-Patent Citations (3)
| Title |
|---|
| Internet Financial Fraud Detection Based on a Distributed Big Data Approach With Node2vec;Hangjun Zhou;《IEEE Access》;第9卷;43378-43386 * |
| 基于贝叶斯网络的商业银行操作风险度量研究;任达;重庆理工大学学报(自然科学);第32卷(第07期);207-214 * |
| 改进的贝叶斯网络模型在保险欺诈挖掘中的应用;陈亮;河南城建学院学报;第21卷(第01期);50-53 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN116777606A (en) | 2023-09-19 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US10452523B1 (en) | System and method for state based intelligent test generation | |
| EP4010817A1 (en) | Graph outcome determination in domain-specific execution environment | |
| US9558230B2 (en) | Data quality assessment | |
| EP4236197A2 (en) | Micro-loan system | |
| EP2228726A2 (en) | A method and system for task modeling of mobile phone applications | |
| CN105556552A (en) | Fraud detection and analysis | |
| An | Data analysis | |
| Luo et al. | Overview of intelligent online banking system based on HERCULES architecture | |
| Khan et al. | Predictors without borders: behavioral modeling of product adoption in three developing countries | |
| US20160092811A1 (en) | Business rules framework | |
| El Ayeb et al. | Community detection for mobile money fraud detection | |
| CN117291722A (en) | Object management method, related device and computer readable medium | |
| CN111177653B (en) | Credit evaluation method and device | |
| CN116777606B (en) | Intelligent anti-fraud system and method for bank credit based on relation graph | |
| CN113918534A (en) | Policy processing system and method | |
| CN116228402A (en) | Financial credit investigation feature warehouse technical support system | |
| CN113407734B (en) | Method for constructing knowledge graph system based on real-time big data | |
| Moudani et al. | Fraud detection in mobile telecommunication | |
| US20210357942A1 (en) | Method and apparatus for identifying risky vertices | |
| US20220215142A1 (en) | Extensible Agents in Agent-Based Generative Models | |
| WO2022150343A1 (en) | Generation and evaluation of secure synthetic data | |
| CN114756224A (en) | Activity execution method and device | |
| CN114723548A (en) | Data processing method, apparatus, device, medium, and program product | |
| Di Pilla et al. | Listening to what the system tells us: Innovative auditing for distributed systems | |
| CN112907254A (en) | Fraud transaction identification and model training method, device, equipment and storage medium |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |