CN113535809A - Distributed multi-kernel decision making system and method for credit granting industry - Google Patents

Distributed multi-kernel decision making system and method for credit granting industry Download PDF

Info

Publication number
CN113535809A
CN113535809A CN202110703252.0A CN202110703252A CN113535809A CN 113535809 A CN113535809 A CN 113535809A CN 202110703252 A CN202110703252 A CN 202110703252A CN 113535809 A CN113535809 A CN 113535809A
Authority
CN
China
Prior art keywords
decision
module
data
rule
result
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.)
Pending
Application number
CN202110703252.0A
Other languages
Chinese (zh)
Inventor
黄滔
王新宇
鲁萍
王新根
陈伟
胡云飞
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Shenzhen Bangsheng Real Time Intelligent Technology Co ltd
Original Assignee
Shenzhen Bangsheng Real Time Intelligent Technology Co ltd
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Shenzhen Bangsheng Real Time Intelligent Technology Co ltd filed Critical Shenzhen Bangsheng Real Time Intelligent Technology Co ltd
Priority to CN202110703252.0A priority Critical patent/CN113535809A/en
Publication of CN113535809A publication Critical patent/CN113535809A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
    • G06F16/24—Querying
    • G06F16/245—Query processing
    • G06F16/2458—Special types of queries, e.g. statistical queries, fuzzy queries or distributed queries
    • G06F16/2471—Distributed queries
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
    • G06F16/24—Querying
    • G06F16/245—Query processing
    • G06F16/2455—Query execution
    • G06F16/24564—Applying rules; Deductive queries
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
    • G06F16/27—Replication, distribution or synchronisation of data between databases or within a distributed database system; Distributed database system architectures therefor
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
    • G06F16/28—Databases characterised by their database models, e.g. relational or object models
    • G06F16/284—Relational databases
    • G06F16/288—Entity relationship models

Abstract

The invention discloses a distributed multi-kernel decision making system and method for the credit industry, which comprises the following processes: the first step is to build an integral decision network, then to build different decision nodes, and to store the decision nodes on a server, wherein the decision nodes can be stored in different servers; the second step is that the system calls different decision nodes on the server by using RPC, and summarizes the results in a result output module; and the third step is to obtain the final decision result. The system provided by the invention enables a plurality of decision nodes to be executed in parallel, and speeds up decision making. In addition, the invention also defines a rule updating language, and business personnel can update the rules of the existing decision system by using the language, thereby greatly facilitating the daily use of the system.

Description

Distributed multi-kernel decision making system and method for credit granting industry
Technical Field
The invention relates to the field of application development in the credit granting industry, in particular to a distributed multi-kernel decision making system and method for the credit granting industry in the field of needing a large number of transactions.
Background
Decision engines are one of the cores of financial technology and automatically determine whether a current operation should be performed by using a series of rules. For example, in a credit granting decision, a bank needs to determine whether or not to credit the current user and how much credit is appropriate according to information such as the credit status and work income of the current user. The decision engine has the following values that firstly, a business department can quickly define, update and maintain business rules, a scoring model and a decision process; secondly, business analysis personnel do not need IT support and can quickly deploy and implement business rules and decisions; thirdly, the results of the decision engine can be shared among multiple systems. The decision engine is typically comprised of a client side and a server side. The client is used for business personnel to configure and update rules used in the engine, and the server comprises different models used for decision making.
The decision model used in the current server generally includes the following:
1) hard coding rules: hard-coded rules refer to a user writing a rule into a model in some form that every time the model attempts to match all rules contained in the library. For each input, a valid output will only be produced if it can be completely matched with the existing rules.
2) Rete algorithm: the Rete algorithm is a forward chain reasoning algorithm, and the core idea of the Rete algorithm is to adopt an increment matching concept and dynamically construct a matching tree by using separated matching items according to contents so as to achieve the effect of obviously reducing the calculation amount. The Rete algorithm utilizes two characteristics of time redundancy and structural similarity to improve the matching efficiency of the system mode, so that the matching speed is independent of the number of rules. The Rete algorithm contains two processes: matching tree construction and fact matching. The matching tree structure is to construct a directed acyclic graph composed of root nodes, Alpha networks and Beta networks, wherein the Alpha networks represent conditions of rules, each condition corresponds to an Alpha node, the rule patterns form the Beta networks, each Beta node represents a part of the pattern, and the whole directed acyclic graph is called Rete network. Fact matching is the process of letting a fact move along the Rete network until the rule conclusion is reached.
3) Deep learning based decision model: with the rapid development of deep learning technology, the accuracy of a decision model built by using a neural network becomes higher and higher, so that the financial field can train the model by using huge data volume of the model to obtain a better model with good effect.
There are problems with existing decision making systems. From the aspect of clients, most of the clients of the existing decision making systems can only be used for viewing the predicted result, and the existing rules in the model cannot be effectively edited. From the server side, there are also several problems: one is that existing decision systems may still employ a method of hard coding rules in the code, which is inconvenient for future expansion and management. Secondly, some systems using Rete network as decision engine may cause the Alpha network to be constructed to be very complex and occupy a large amount of memory space due to complex scenes. Thirdly, the rules in the decision engine are generally executed sequentially, so that when one rule is very complex, the time consumed for decision making can be very large. In summary, decision-making systems in the current credit industry cannot bear increasingly high business requirements.
Disclosure of Invention
The invention aims to provide a distributed multi-kernel decision making system and a distributed multi-kernel decision making method aiming at the defects of the existing decision making system in the credit industry. The rules and the models contained in the current decision engine can be managed more conveniently by a user, and partial nodes in the decision engine are distributed and deployed on different servers, so that the decision speed of the system is increased.
The purpose of the invention is realized by the following technical scheme: a distributed multi-kernel decision making system for credit granting industry comprises an information obtaining and inquiring module, a data structure abstract module, a rule and model management module, a decision making module and a result output module. The information acquisition and query module, the data structure abstraction module and the rule and model management module are deployed at a client, and the decision module and the result output module are deployed at a server.
The information acquisition and query module is used for receiving data which is sent from a client and needs to be decided and querying received historical data, and sending the data which needs to be decided currently to the data structure abstraction module, wherein the data which needs to be decided comprises: time of application, applicant name, applicant ID, applicant company, and approval status.
The data structure abstraction module is used for abstracting the received data which needs to be subjected to decision making into a data structure which can be processed by a model and a rule, and sending the abstracted data structure to the decision making module in a mode of calling RPC. The basic data in the data structure obtained by the last abstraction comprises the applicant ID, the applicant credit status, the application amount and the repayment period, and is used for inputting into the decision module.
And the rule and model management module is used for increasing, deleting, modifying and checking the rules or models in the current system. Because the decision-making system is an optimized Rete network, the input of the Alpha network in the Rete network uses a uniform format, the matching work is put into the Beta network for carrying out, and a part of nodes in the Beta network are distributed on other servers, the nodes can work asynchronously in actual operation, and the decision-making waiting time is reduced; therefore, the rules are stored in the nodes of the Rete network, and the rule and model management module converts the input script language into the corresponding rules and then performs corresponding matching operation on the nodes in the Rete network.
The decision module is used for managing the whole Rete network, sending the data structure obtained by the data structure abstraction module in an abstract mode to a server where the Rete network child node is located, and sorting the returned result and sending the returned result to the result output module.
And the result output module returns the final result to the client, and persistently stores the application information and the result in the system together, so that the system is convenient to consult in the future.
Further, the user inputs the data to be decided by manual input or batch import.
The invention also provides a decision method based on the distributed multi-kernel decision system, which specifically comprises the following steps:
(1) the data needing to be decided is obtained from the client through the information obtaining and inquiring module, and is converted into data of a data structure which can be processed by a model and a rule in the decision system through the data structure abstract module, so that the data can be input into the decision module.
(2) Rules and models existing in the decision-making system are managed through a rule and model management module. And updating the rules to a server where a Rete node corresponding to the decision system is located. For the newly added rule, the rule and model management module abstracts the new rule into a new Rete node and sends the node to a certain server for use.
(3) The server sends the data structure abstracted by the data structure abstraction module to the server where the Rete network child node is located through the decision module according to the existing rules and the input data, and sorts the returned result and sends the result to the result output module.
(4) And the server returns the result to the client through the result output module, and the system stores the result, so that the result is convenient to look up in the future.
Further, if the newly added rule is similar to the original rule, the corresponding modification can be directly performed on the node of the original rule.
The invention has the beneficial effects that:
1) the distributed multi-kernel decision making system is used in the credit industry, the decision making system at the server side is exposed in a service form, and the client side can use the whole decision making system only through simple configuration and interface calling without developing a complex module.
2) The distributed multi-kernel decision making system for the credit industry has the advantages that business personnel can manage rules in the system by using a unified script language, so that the system is greatly convenient for non-technical personnel to use, and the flexibility of the system is improved.
3) In the distributed multi-kernel decision making system for credit service, nodes in a Rete network are stored on a server in the form of individual services, so that more complex rules can be stored, different inputs can be free from waiting each other, the nodes can work independently, and then input service data is determined according to information contained in input and returned results, so that the use experience of a user is improved.
Drawings
Fig. 1 is a schematic structural diagram of the system.
Fig. 2 is a service flow diagram of the present system.
Detailed Description
The following describes embodiments of the present invention in further detail with reference to the accompanying drawings.
The invention relates to a distributed multi-kernel decision making system for the credit industry, which generally consists of an optimized version of Rete network, namely, the nodes of the network are not only simple regular nodes, but also other decision models as one node, and the nodes can be distributed on different servers. For example, if the user requests a loan application, and enters the details of the loan into the system, the system uses the optimized Rete network to comprehensively determine whether the loan should be approved. The technology provides a method for constructing a distributed multi-kernel decision making system and updating an existing decision making engine, so that the decision making engine can be continuously updated according to an actual service scene while a decision making result can be quickly obtained, and the timeliness of the system is guaranteed. The decision system comprises an information acquisition and query module, a rule and model management module, a data structure abstraction module, a decision module and a result output module. The relationship between the modules in the system is shown in fig. 1. Assume that the input application data is in the format of
Figure BDA0003131052860000041
Where value is the detailed information contained in the application form. The information acquisition and query module, the data structure abstraction module and the rule and model management module are deployed at a client, and the decision module and the result output module are deployed at a server.
The distributed multi-kernel decision making system for the credit industry is an optimized Rete network, the input of an Alpha network in the Rete network uses a uniform format, matching work is put into a Beta network for carrying out, a part of nodes in the Beta network are distributed on other servers, the nodes can work asynchronously during actual operation, and the decision waiting time is reduced; the rules are thus deposited in the nodes of the Rete network.
The information acquisition and query module acquires data needing to be decided and historical data received by query from the client through manual input or batch import by a user, sends the data needing to be decided currently to the data structure abstraction module, and can query historical input information and corresponding decision results through unique data such as application form ID and the like. After the input data is acquired, the system stores the input data in a persistent mode. The data that needs to be decided includes: time of application, applicant name, applicant ID, applicant company, and approval status.
And the rule and model management module is used for defining a data structure which can be processed by the rules and models in the system and matching rules in the Beta network, and adding, deleting and checking the rules or models contained in the current system. The module achieves the purpose of managing the rules by changing the Rete nodes in the Rete network, and when nodes are newly added, the newly added nodes need to be sent to a server, and the calling mode of the newly added nodes is specified. And the rule and model management module converts the input script language into a corresponding rule and then performs corresponding matching operation on the nodes in the Rete network.
The data structure abstraction module is used for abstracting the received data needing decision making into a model and a ruleA data structure for processing, specifically an input application dataiConverting into an input data format defined by the decision model, and recording the converted data as di＝(ki,vi). And sending the abstracted data structure to a decision module in a mode of calling RPC. The basic data in the final abstracted data structure includes the applicant ID, the applicant credit status, the application amount, and the repayment period.
And the decision module is used for managing the whole Rete network and making decisions on the input data. After data sent by the last module is obtained, the data is sent to all Rete nodes, and data d is assumediIs sent to a certain Rete NodekAfter the calculation is finished, the result is expressed as Resk＝(keyi,resk) Is sent to the result output module, wherein keyiIs key, res of the input datakIs the result of the match of the data obtained by the settlement of the node, generally indicated as true or false.
And the result output module is used for integrating the results returned by the nodes, then sending the results back to the client, splicing the results with the input data and then storing the results. The module maintains a table, the number of Rete nodes contained in the Rete network when each input key and data corresponding to the current key are input is recorded in the table, and after the result of one node is sent to the result output module, whether the current key is contained in the table or not is searched according to the key in the result. If the key is not contained, the input is calculated, and the returned result is discarded; if the current key is contained, the result obtained by the input corresponding to the key in the previous node is true. Then checking the result of the node, if the returned result is true, subtracting one from the value corresponding to the key in the table, if the value corresponding to the key is 0, returning the input result to the client, and if the value corresponding to the key is not 0, updating the value corresponding to the key in the table to the value; if the returned result is false, then this key is removed from the table and the input result is immediately returned to the client. The overall business process in this system is shown in fig. 2.
The invention also provides a decision method based on the distributed multi-kernel decision system, which specifically comprises the following steps:
(1) the data needing to be decided is obtained from the client through the information obtaining and inquiring module, and is converted into data of a data structure which can be processed by a model and a rule in the decision system through the data structure abstract module, so that the data can be input into the decision module.
(2) Rules and models existing in the decision-making system are managed through a rule and model management module. And updating the rules to a server where a Rete node corresponding to the decision system is located. For the newly added rule, the rule and model management module abstracts the new rule into a new Rete node and sends the node to a certain server for use. If the newly added rule is similar to the original rule, the corresponding modification can be directly carried out on the node of the original rule.
(3) The server sends the data structure abstracted by the data structure abstraction module to the server where the Rete network child node is located through the decision module according to the existing rules and the input data, and sorts the returned result and sends the result to the result output module.
(4) And the server returns the result to the client through the result output module, and the system stores the result, so that the result is convenient to look up in the future.
This section will further describe, by way of example, a distributed multi-kernel decision making system for the trust industry described above.
Assume that some loan application input data for the system is shown in table 1:
TABLE 1
Name of applicant Bob
Applicant ID 123456X
Application for note number 312378123
Applicant credit status Good effect
Amount of application 100000
Application for reimbursement duration 3months
Home address XXXXX
…… ……
The system acquires the input information and then sends the information to the data structure abstraction module, the abstraction module combines the information stored in the database with the input information, and the data is processed to obtain the data shown in table 2 by taking the data as an example:
TABLE 2
Applicant ID 123456X
Application for note number 312378123
Applicant credit status Good effect
Applicant family member credit status Good effect
Amount of application 100000
Working income situation of applicant 300000
…… ……
And after the input data is processed, the data structure abstract module sends the data to the decision module. The decision module firstly extracts the ID of the current applicant and the ID of the application form number from the input data, and combines the two IDs to be used as the key of the current input data. And then the decision module sends the data to each decision node in a key-value form, and then sends a currently input key and the contained node number size to the result output module to inform the result output module to wait for the arrival of the result of each node.
In the result output module, the module maintains a table as shown in Table 3:
TABLE 3
…… ……
ID1+ID2 size
…… ……
The size in the table indicates how many nodes of the result are left without output for the current input. When an incoming result is true, the corresponding size is reduced by one, and when the size is 0, the record is removed from the table and the result of the consent application is returned to the client; and if the record is false, removing the record from the table, and immediately returning the result of rejecting the application to the client.
And respectively carrying out rule matching on the input data by using a system constructed based on a Rete algorithm of a general type and a current system. 1000000 data inputs were used in the test procedure and the performance ratio for both systems is shown in table 4:
TABLE 4
Figure BDA0003131052860000061
Figure BDA0003131052860000071
The above-described embodiments are intended to illustrate rather than to limit the invention, and any modifications and variations of the present invention are within the spirit of the invention and the scope of the appended claims.

Claims (4)

1. A distributed multi-kernel decision making system for credit granting industry is characterized by comprising an information acquisition and query module, a data structure abstraction module, a rule and model management module, a decision making module and a result output module. The information acquisition and query module, the data structure abstraction module and the rule and model management module are deployed at a client, and the decision module and the result output module are deployed at a server.
The information acquisition and query module is used for receiving data which is sent from a client and needs to be decided and querying received historical data, and sending the data which needs to be decided currently to the data structure abstraction module, wherein the data which needs to be decided comprises: time of application, applicant name, applicant ID, applicant company, and approval status.
The data structure abstraction module is used for abstracting the received data which needs to be subjected to decision making into a data structure which can be processed by a model and a rule, and sending the abstracted data structure to the decision making module in a mode of calling RPC. The basic data in the data structure obtained by the last abstraction comprises the applicant ID, the applicant credit status, the application amount and the repayment period, and is used for inputting into the decision module.
And the rule and model management module is used for increasing, deleting, modifying and checking the rules or models in the current system. Because the decision-making system is an optimized Rete network, the input of the Alpha network in the Rete network uses a uniform format, the matching work is put into the Beta network for carrying out, and a part of nodes in the Beta network are distributed on other servers, the nodes can work asynchronously in actual operation, and the decision-making waiting time is reduced; therefore, the rules are stored in the nodes of the Rete network, and the rule and model management module converts the input script language into the corresponding rules and then performs corresponding matching operation on the nodes in the Rete network.
The decision module is used for managing the whole Rete network, sending the data structure obtained by the data structure abstraction module in an abstract mode to a server where the Rete network child node is located, and sorting the returned result and sending the returned result to the result output module.
And the result output module returns the final result to the client, and persistently stores the application information and the result in the system together, so that the system is convenient to consult in the future.
2. The distributed multi-kernel decision making system for the trust industry of claim 1, wherein the user imports the data needed to make the decision by hand or by batch.
3. A decision method based on the distributed multi-kernel decision system of claim 1, specifically comprising the steps of:
(1) the data needing to be decided is obtained from the client through the information obtaining and inquiring module, and is converted into data of a data structure which can be processed by a model and a rule in the decision system through the data structure abstract module, so that the data can be input into the decision module.
(2) Rules and models existing in the decision-making system are managed through a rule and model management module. And updating the rules to a server where a Rete node corresponding to the decision system is located. For the newly added rule, the rule and model management module abstracts the new rule into a new Rete node and sends the node to a certain server for use.
(3) The server sends the data structure abstracted by the data structure abstraction module to the server where the Rete network child node is located through the decision module according to the existing rules and the input data, and sorts the returned result and sends the result to the result output module.
(4) And the server returns the result to the client through the result output module, and the system stores the result, so that the result is convenient to look up in the future.
4. A decision method as claimed in claim 3, characterized in that if the newly added rule is similar to the original rule, the corresponding modification can be made directly on the node of the original rule.
CN202110703252.0A 2021-06-24 2021-06-24 Distributed multi-kernel decision making system and method for credit granting industry Pending CN113535809A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110703252.0A CN113535809A (en) 2021-06-24 2021-06-24 Distributed multi-kernel decision making system and method for credit granting industry

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110703252.0A CN113535809A (en) 2021-06-24 2021-06-24 Distributed multi-kernel decision making system and method for credit granting industry

Publications (1)

Publication Number Publication Date
CN113535809A true CN113535809A (en) 2021-10-22

Family

ID=78125794

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110703252.0A Pending CN113535809A (en) 2021-06-24 2021-06-24 Distributed multi-kernel decision making system and method for credit granting industry

Country Status (1)

Country Link
CN (1) CN113535809A (en)

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104679790A (en) * 2013-12-03 2015-06-03 富士通株式会社 Distributed rule engine system, distributed rule engine construction method and rule processing method
CN106777029A (en) * 2016-12-08 2017-05-31 中国科学技术大学 A kind of distributed rule automotive engine system and its construction method
CN112668944A (en) * 2021-01-26 2021-04-16 天元大数据信用管理有限公司 Enterprise wind control method, device, equipment and medium based on big data credit investigation
CN112700329A (en) * 2021-01-27 2021-04-23 永辉云金科技有限公司 Response method of wind control rule engine and wind control rule engine

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104679790A (en) * 2013-12-03 2015-06-03 富士通株式会社 Distributed rule engine system, distributed rule engine construction method and rule processing method
CN106777029A (en) * 2016-12-08 2017-05-31 中国科学技术大学 A kind of distributed rule automotive engine system and its construction method
CN112668944A (en) * 2021-01-26 2021-04-16 天元大数据信用管理有限公司 Enterprise wind control method, device, equipment and medium based on big data credit investigation
CN112700329A (en) * 2021-01-27 2021-04-23 永辉云金科技有限公司 Response method of wind control rule engine and wind control rule engine

Similar Documents

Publication Publication Date Title
CN103336790B (en) Hadoop-based fast neighborhood rough set attribute reduction method
US9613164B2 (en) System and method for data management in large data networks
CN107766511A (en) Intelligent answer method, terminal and storage medium
US7308436B2 (en) Distributed data mining and compression method and system
CN103336791B (en) Hadoop-based fast rough set attribute reduction method
CN111506621B (en) Data statistical method and device
CN107247799A (en) Data processing method, system and its modeling method of compatible a variety of big data storages
CN109857833B (en) Rule engine implementation method and device and electronic equipment
CN112667860A (en) Sub-graph matching method, device, equipment and storage medium
US11461317B2 (en) Method, apparatus, system, device, and storage medium for answering knowledge questions
US6389410B1 (en) Method for minimizing the number of sorts required for a query block containing window functions
CN111292201A (en) Method for pushing field operation and maintenance information of power communication network based on Apriori and RETE
CN116541166A (en) Super-computing power scheduling server and resource management method
CN113535809A (en) Distributed multi-kernel decision making system and method for credit granting industry
CN115329078A (en) Text data processing method, device, equipment and storage medium
CN115510116A (en) Data directory construction method, device, medium and equipment
CN108256086A (en) Data characteristics statistical analysis technique
CN114022295A (en) Group fraud identification method and system
EP1202191A2 (en) Data management method and storage medium storing data management program
CN113946568A (en) Data management system and method
Andreichikov et al. Genetic Algoritms for the Synthesis of Innovative Engineering Solutions
Xia et al. A novel data schema integration framework for the human-centric services in smart city
CN113641654B (en) Marketing treatment rule engine method based on real-time event
CN116541414B (en) Unified ontology connection query method, medium and system based on multi-source heterogeneous data
CN113726342B (en) Segmented difference compression and inert decompression method for large-scale graph iterative computation

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