CN113989018A - Risk management method, risk management device, electronic equipment and medium - Google Patents

Abstract

The disclosure provides a risk management method, a risk management device, an electronic device and a medium. The risk management method and the risk management device can be used in the technical field of big data. The risk management method comprises the following steps: acquiring enterprise data, wherein the enterprise data comprises a party and an event relation between the party, the event relation comprises m category data, each category data comprises n event information, m is an integer greater than or equal to 1, and n is an integer greater than or equal to 1; constructing a knowledge graph according to enterprise data, wherein the party and event information are nodes of the knowledge graph, the event relationship is the side of the knowledge graph, and the event information is used for explaining the corresponding side; calculating reduction weight of n event information edges between two parties based on the knowledge graph constructed by each category data

Calculating an aggregation weight ω of m class-data edges between two parties based on the reduction weights_vu(ii) a And performing risk management on the enterprise according to the aggregation weight.

Description

Risk management methods, devices, electronic equipment and media

technical field

The present disclosure relates to the technical field of big data, and more particularly, to a risk management method, apparatus, electronic device and medium.

Background technique

At this stage, my country's small and micro financing market has huge potential. According to data from relevant departments, there are more than 100 million business entities in my country's market, including more than 70 million individual industrial and commercial households, which has a broad operating space. As an important pillar of economic development and social stability, small and micro enterprises play an indispensable role in promoting the orderly flow of talents, maintaining market vitality, and promoting technological innovation.

Compared with large and medium-sized enterprises, small and micro enterprises are still in a weak position in market competition, and financing problems make it difficult for them to ensure the stability and continuity of operations. Bank loans are an important means of corporate financing. In order to relieve the operating pressure of small and micro enterprises and meet the strong financing needs, the small and micro loan business of banks has gradually expanded, and various credit products have emerged. The accompanying problem is that the low loan repayment ability of small and micro enterprises may lead to a high loan overdue repayment ratio.

The traditional financial risk control system mainly relies on expert rules and various indicators in the Basel Agreement, while small and micro enterprises have a natural weakness in providing their own information, and their "opaque" and "internalized" asymmetric data information makes banks It is difficult to control the substantial credit risks of small and micro enterprise customers, and the management of credit products for small and micro enterprises is much more difficult than that of large enterprises. If an undifferentiated risk control model is adopted, it is impossible to reasonably predict the risks of most small and micro enterprises, and a new data analysis and processing method needs to be introduced to solve the problems of small and micro credit business.

SUMMARY OF THE INVENTION

In view of this, the present disclosure provides a knowledge graph-based risk management method, apparatus, electronic device and computer-readable storage medium with good risk prediction effect and convenient use.

根据所述归约权重，计算两个所述当事方之间的m个类别数据边的聚合权重ω_vu；以及根据所述聚合权重对企业进行风险管理。One aspect of the present disclosure provides a knowledge graph-based risk management method, including: acquiring enterprise data, wherein the enterprise data includes a party and an event relationship between the parties, and the event relationship includes m categories of data , each of the category data includes n pieces of event information, where m is an integer greater than or equal to 1, and n is an integer greater than or equal to 1; construct a knowledge graph according to the enterprise data, wherein the parties and the The event information is the node of the knowledge graph, the event relationship is the edge of the knowledge graph, and the event information is used to explain the corresponding edge; based on the knowledge graph constructed by each of the category data, two the reduction weights of the n event information edges between the parties

According to the reduction weights, an aggregated weight ω _vu of m class data edges between two of the parties is calculated; and the enterprise is risk managed according to the aggregated weights.

According to the risk management method based on the knowledge graph according to the embodiment of the present disclosure, by establishing the knowledge graph and calculating the aggregate weight of the edges in the knowledge graph, the risk management of the enterprise can be performed. The risk of loan duration will affect its upstream and downstream related customers, thereby improving the ability to provide inclusive financial services for small and micro enterprises. In the present disclosure, event information is used as a node of the knowledge graph to describe the corresponding edge, so that the edge does not need additional attributes, thereby reducing the redundancy of the edge, thereby making the knowledge graph respond more quickly when used.

In some embodiments, the calculation method of the reduction weight is as follows:

其中，v表示两个所述当事方中的头节点，u表示两个所述当事方中的尾节点，E(v，u)表示头节点为v、尾节点为u的n个所述事件信息边的集合，E(v)表示头节点为v的有向边的集合，

表示边l′在对应的事件发生时刻的初始权重，t表示当前时刻，

表示边l′的时间权重，

表示边l在对应的事件发生时刻的初始权重，

表示边l的时间权重；以及

Among them, v represents the head node of the two parties, u represents the tail node of the two parties, and E(v, u) represents n all the nodes whose head node is v and tail node is u. E(v) represents the set of directed edges whose head node is v,

represents the initial weight of edge l' at the time of the corresponding event, t represents the current moment,

represents the time weight of edge l′,

represents the initial weight of edge l at the time of the corresponding event,

represents the time weight of edge l; and

The calculation method of the aggregation weight is as follows:

其中，R(v，u)表示头节点为v、尾节点为u的m个所述类别数据边的集合，C_r表示与不同的所述类别数据对应的常数系数。

Wherein, R(v, u) represents a set of m edges of the class data with the head node v and the tail node u, and C _r represents a constant coefficient corresponding to different class data.

In some embodiments, the m categories of data include at least one of capital flow data, investment data, guarantee data, and person-enterprise association data.

In some embodiments, the n pieces of event information include event information under the same category of data for n time periods.

In some embodiments, before the construction of the knowledge graph according to the enterprise data, the method further includes cleaning the enterprise data, wherein the cleaning of the enterprise data includes: data deduplication, feature complementing in the data, and One of the exception feature handling.

In some embodiments, the constructing the knowledge graph according to the enterprise data comprises: constructing a schema layer according to the parties and m pieces of the category data, wherein the schema layer comprises according to the parties and m Nodes established by m categories in the category data, and links between nodes established according to events in each category data; and importing the data in the category data into the corresponding schema layer.

In some embodiments, the category data is structured data, and the importing data in the category data into a corresponding schema layer includes: converting the category data into resource description framework data; and converting the resource description The frame data is imported into the corresponding schema layer.

In some embodiments, the performing risk management on the enterprise according to the aggregated weight includes: performing label propagation on the nodes of the knowledge graph according to the aggregated weight; dividing the community according to the result of the label propagation to obtain a community group size; according to the community size, calculate the degree centrality of the nodes of the knowledge graph; and use the degree centrality as an input of a risk prediction model to perform risk prediction.

In some embodiments, the performing risk management on the enterprise according to the aggregated weight includes: performing label propagation on the nodes of the knowledge graph according to the aggregated weight; and performing risk prediction according to a result of the label propagation.

In some embodiments, the risk management method based on the knowledge graph further includes visualizing the knowledge graph, wherein the visualizing the knowledge graph includes: performing node retrieval based on the knowledge graph, performing node retrieval based on the knowledge graph at least one of subgraph walking, path exploration based on the knowledge graph, and self-loop exploration based on the knowledge graph.

In some embodiments, the node retrieval includes: in response to an input node name, displaying a node related to the node name and associated information of the node; the sub-graph wandering includes: in response to a manual click on the event The operation of the relationship, displaying the node corresponding to the clicked event relationship and all the edges sent by the node; the path exploration includes: acquiring and displaying the path relationship between the two nodes; and the self-loop exploration includes: acquiring the path Nodes whose relationships are closed loops are displayed.

第二计算模块，所述第二计算模块用于根据所述归约权重，计算两个所述当事方之间的m个类别数据边的聚合权重ω_vu；以及管理模块，所述管理模块用于根据所述聚合权重对企业进行风险管理。Another aspect of the present disclosure provides a knowledge graph-based risk management device, comprising: an acquisition module, the acquisition module is configured to acquire enterprise data, wherein the enterprise data includes parties and event relationships between parties , the event relationship includes m pieces of category data, each of which includes n pieces of event information, where m is an integer greater than or equal to 1, and n is an integer greater than or equal to 1; a building module, the building module is used for A knowledge graph is constructed according to the enterprise data, wherein the parties and the event information are nodes of the knowledge graph, the event relationship is an edge of the knowledge graph, and the event information is used to explain the corresponding the edge; the first calculation module, the first calculation module is used to calculate the normalization of n event information edges between two parties based on the knowledge graph constructed by each of the category data. about weight

a second calculation module, which is configured to calculate, according to the reduction weights, the aggregated weights ω _vu of m class data edges between two of the parties; and a management module, the management module For risk management of the enterprise according to the aggregated weight.

Another aspect of the present disclosure provides an electronic device including one or more processors and one or more memories, wherein the memories are used to store executable instructions, the executable instructions being executed by the processor When executed, the method described above is implemented.

Another aspect of the present disclosure provides a computer-readable storage medium storing computer-executable instructions that, when executed, are used to implement the method as described above

Description of drawings

The above and other objects, features and advantages of the present disclosure will become more apparent from the following description of embodiments of the present disclosure with reference to the accompanying drawings, in which:

FIG. 1 schematically shows an exemplary system architecture to which methods and apparatuses according to embodiments of the present disclosure can be applied;

FIG. 2 schematically shows a flowchart of a risk management method based on knowledge graph according to an embodiment of the present disclosure;

FIG. 3 schematically shows a schematic diagram of a knowledge graph according to an embodiment of the present disclosure;

FIG. 4 schematically shows a flowchart of constructing a knowledge graph according to enterprise data according to an embodiment of the present disclosure;

FIG. 5 schematically shows a flowchart of importing data in category data into a corresponding schema layer according to an embodiment of the present disclosure;

FIG. 6 schematically shows a flowchart of risk management for an enterprise according to an aggregated weight according to an embodiment of the present disclosure;

FIG. 7 schematically shows a flowchart of risk management for an enterprise according to an aggregated weight according to an embodiment of the present disclosure;

FIG. 8 schematically shows a flowchart of a visualized knowledge graph according to an embodiment of the present disclosure;

FIG. 9 schematically shows a flowchart of a visualized knowledge graph according to an embodiment of the present disclosure;

FIG. 10 schematically shows a flowchart of a visualized knowledge graph according to an embodiment of the present disclosure;

FIG. 11 schematically shows a flowchart of a visualized knowledge graph according to an embodiment of the present disclosure;

FIG. 12 schematically shows a block diagram of a risk management apparatus based on a knowledge graph according to an embodiment of the present disclosure;

13 schematically shows a block diagram of an electronic device according to an embodiment of the present disclosure.

Detailed ways

Hereinafter, embodiments of the present disclosure will be described with reference to the accompanying drawings. It should be understood, however, that these descriptions are exemplary only, and are not intended to limit the scope of the present disclosure. In the following detailed description, for convenience of explanation, numerous specific details are set forth in order to provide a thorough understanding of the embodiments of the present disclosure. It will be apparent, however, that one or more embodiments may be practiced without these specific details. Also, in the following description, descriptions of well-known structures and techniques are omitted to avoid unnecessarily obscuring the concepts of the present disclosure. In the technical solution of the present disclosure, the acquisition, storage and application of the user's personal information involved all comply with the relevant laws and regulations, take necessary confidentiality measures, and do not violate public order and good customs.

The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the present disclosure. The terms "comprising", "comprising" and the like as used herein indicate the presence of stated features, steps, operations and/or components, but do not preclude the presence or addition of one or more other features, steps, operations or components.

Where expressions like "at least one of A, B, or C, etc.," are used, they should generally be interpreted in accordance with the meaning of the expression as commonly understood by those skilled in the art (eg, "has A, B, or C, etc." At least one of the "systems" shall include, but not be limited to, systems with A alone, B alone, C alone, A and B, A and C, B and C, and/or A, B, C, etc. ). The terms "first" and "second" are only used for descriptive purposes, and should not be construed as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Thus, features defined as "first", "second" may expressly or implicitly include one or more of said features.

At this stage, my country's small and micro financing market has huge potential. According to data from relevant departments, there are more than 100 million business entities in my country's market, including more than 70 million individual industrial and commercial households, which has a broad operating space. As an important pillar of economic development and social stability, small and micro enterprises play an indispensable role in promoting the orderly flow of talents, maintaining market vitality, and promoting technological innovation.

Compared with large and medium-sized enterprises, small and micro enterprises are still in a weak position in market competition, and financing problems make it difficult for them to ensure the stability and continuity of operations. Bank loans are an important means of corporate financing. In order to relieve the operating pressure of small and micro enterprises and meet the strong financing needs, the small and micro loan business of banks has gradually expanded, and various credit products have emerged. The accompanying problem is that the low loan repayment ability of small and micro enterprises may lead to a high loan overdue repayment ratio.

The traditional financial risk control system mainly relies on expert rules and various indicators in the Basel Agreement, while small and micro enterprises have a natural weakness in providing their own information, and their "opaque" and "internalized" asymmetric data information makes banks It is difficult to control the substantial credit risks of small and micro enterprise customers, and the management of credit products for small and micro enterprises is much more difficult than that of large enterprises. If an undifferentiated risk control model is adopted, it is impossible to reasonably predict the risks of most small and micro enterprises, and a new data analysis and processing method needs to be introduced to solve the problems of small and micro credit business.

Most of the current duration management models still have certain limitations: they are only modeled by the data characteristics of small and micro enterprise customers themselves, while the data in the financial industry is mostly related data involving multiple customers, which makes the use of data and information ineffective. At the same time, it is impossible to consider the impact of the risk of a customer's loan duration on its upstream and downstream related customers, ignoring an important aspect of loan duration management, and affecting the ability to provide inclusive financial services for small and micro enterprises.

根据归约权重，计算两个当事方之间的m个类别数据边的聚合权重ω_vu；以及根据聚合权重对企业进行风险管理。Embodiments of the present disclosure provide a knowledge graph-based risk management method, a risk management apparatus, an electronic device, a computer-readable storage medium, and a computer program. The risk management method based on knowledge graph includes: acquiring enterprise data, wherein the enterprise data includes the parties and the event relationship between the parties, the event relationship includes m category data, each category data includes n event information, wherein , m is an integer greater than or equal to 1, and n is an integer greater than or equal to 1; construct a knowledge graph based on enterprise data, in which the parties and event information are nodes of the knowledge graph, the event relationship is the edge of the knowledge graph, and the event information is used for Interpret the corresponding edges; calculate the reduction weights of n event information edges between two parties based on the knowledge graph constructed by each category of data

According to the reduction weights, the aggregated weights ω _vu of the m class data edges between two parties are calculated; and the enterprise is risk managed according to the aggregated weights.

It should be noted that the knowledge graph-based risk management method, risk management device, electronic device, computer-readable storage medium and computer program of the present disclosure can be used in the field of big data, and can also be used in any field other than the field of big data, For example, in the financial field, the field of the present disclosure is not limited here.

FIG. 1 schematically shows an exemplary system architecture 100 to which a knowledge graph-based risk management method, a risk management apparatus, an electronic device, a computer-readable storage medium and a computer program can be applied according to an embodiment of the present disclosure. It should be noted that FIG. 1 is only an example of a system architecture to which the embodiments of the present disclosure can be applied, so as to help those skilled in the art to understand the technical content of the present disclosure, but it does not mean that the embodiments of the present disclosure cannot be used for other A device, system, environment or scene.

As shown in FIG. 1 , the system architecture 100 according to this embodiment may include terminal devices 101 , 102 , and 103 , a network 104 and a server 105 . The network 104 is a medium used to provide a communication link between the terminal devices 101 , 102 , 103 and the server 105 . The network 104 may include various connection types, such as wired, wireless communication links, or fiber optic cables, among others.

The user can use the terminal devices 101, 102, 103 to interact with the server 105 through the network 104 to receive or send messages and the like. Various communication client applications may be installed on the terminal devices 101 , 102 and 103 , such as shopping applications, web browser applications, search applications, instant messaging tools, email clients, social platform software, etc. (only examples).

The terminal devices 101, 102, 103 may be various electronic devices having a display screen and supporting web browsing, including but not limited to smart phones, tablet computers, laptop computers, desktop computers, and the like.

The server 105 may be a server that provides various services, such as a background management server (just an example) that provides support for websites browsed by users using the terminal devices 101 , 102 , and 103 . The background management server can analyze and process the received user requests and other data, and feed back the processing results (such as web pages, information, or data obtained or generated according to user requests) to the terminal device.

It should be noted that, the risk management method based on the knowledge graph provided by the embodiments of the present disclosure may generally be executed by the server 105 . Correspondingly, the risk management apparatus provided by the embodiments of the present disclosure may generally be provided in the server 105 . The knowledge graph-based risk management method provided by the embodiments of the present disclosure may also be executed by a server or server cluster that is different from the server 105 and can communicate with the terminal devices 101 , 102 , 103 and/or the server 105 . Correspondingly, the risk management apparatus provided by the embodiments of the present disclosure may also be provided in a server or a server cluster that is different from the server 105 and can communicate with the terminal devices 101 , 102 , 103 and/or the server 105 .

It should be understood that the numbers of terminal devices, networks and servers in FIG. 1 are merely illustrative. There can be any number of terminal devices, networks and servers according to implementation needs.

Based on the scenario described in FIG. 1 , the following will describe the risk management method based on the knowledge graph according to the embodiment of the present disclosure in detail through FIGS. 2 to 11 .

FIG. 2 schematically shows a flowchart of a risk management method based on a knowledge graph according to an embodiment of the present disclosure.

As shown in FIG. 2 , the risk management method based on the knowledge graph of this embodiment includes operations S210 to S250 .

In operation S210, enterprise data is acquired, wherein the enterprise data includes a party and an event relationship between the parties, the event relationship includes m types of data, and each type of data includes n pieces of event information, where m is greater than or equal to An integer of 1, and n is an integer greater than or equal to 1.

It should be noted that the parties can be both enterprises, the parties can also be individuals, and the parties can also be enterprises and individuals. Here, an event relationship can exist between enterprises and enterprises, an event relationship can exist between individuals and individuals, and an event relationship can exist between companies and individuals.

The event relationship includes m categories of data, and as a possible implementation, the m categories of data include at least one of capital flow data, investment data, guarantee data, and person-enterprise association data.

It can be understood that the capital flow data can be the capital flow transaction data between parties, that is, the capital flow transaction data between enterprises and enterprises, the capital flow transaction data between individuals, or the capital flow transaction data between individuals. Transaction data for the flow of funds between businesses and individuals.

Investment data can be understood as the investment relationship data between the parties, that is, the investment relationship between enterprises and enterprises, the investment relationship between individuals and individuals, and the investment relationship between enterprises and individuals.

Guarantee data can be understood as the guarantee relationship data between the parties, that is, the guarantee relationship between enterprises and enterprises, the guarantee relationship between individuals and individuals, and the guarantee relationship between enterprises and individuals.

Person-enterprise association data can be understood as the association relationship data between enterprises and individuals, for example, the legal representative, second responsible person, insurance legal beneficiary, financial director, shareholder, general manager, unit contact person of an enterprise and the enterprise , Chairman and other responsible persons.

Wherein, each category of data includes n pieces of event information. As a possible implementation manner, the n pieces of event information include event information under the same category of data in n time periods. For example, under the category data of capital flow data, events that occur in each time period are classified as one event information. For example, if the time period is set as a quarter, and there can be 4 time periods in a year, the parties and the parties There are 4 event information under the capital flow data between parties. Each event information may include, but is not limited to, funds outflow parties, transaction funds amount, transaction time, and funds inflow parties.

For another example, under the category data of investment data, events that occur in each time period are classified as one event information. For example, if the time period is set as a quarter, and there can be 4 time periods in a year, the parties and There are 4 event information under investment data between parties. Each event information may include, but is not limited to, the investing party, the investment amount, the time the investment occurred, and the funded party.

For another example, under the category data of guarantee data, events that occur in each time period are classified as one event information. There are 4 event information under the guarantee data between the parties. Each event information may include, but is not limited to, the insured party, the insured amount, the insured time, and the insured party.

For another example, under the category data of person-enterprise association data, events that occur in each time period are classified as one event information, and the legal representative, second responsible person, and insurance legal beneficiary of an enterprise and the enterprise in each time period. , financial director, shareholder, general manager, unit contact person, chairman of the board and other persons in charge, etc., are each associated with one event information. Further, for example, if the time period is set as a quarter, and there can be 4 time periods in a year, the enterprise shall contact the legal representative, the second person in charge, the legal beneficiary of insurance, the financial director, the shareholder, the general manager, and the unit of the enterprise. There are 4 event information respectively for the person, chairman and other responsible persons. Each event information may include, but is not limited to, the business party, the individual's job title, the start date and the end date.

Of course, the division of time periods is not limited to division by quarters, and the division of time periods may be based on years, months, days, or any time interval, and there are no restrictions on time periods here.

In operation S220, a knowledge graph is constructed according to the enterprise data, wherein the parties and event information are nodes of the knowledge graph, the event relationship is an edge of the knowledge graph, and the event information is used to explain the corresponding edge.

For example, referring to Figure 3, there is a capital flow transaction and investment relationship between enterprise A and enterprise B, wherein, enterprise A is the capital outflow party, the transaction capital amount is 50,000, the transaction time is t ₁ , and enterprise B is the capital inflow party; Enterprise A is the investor, the investment amount is 1 million, the transaction time is t ₂ , and the enterprise B is the fundee.

There is a guarantee relationship between enterprise A and enterprise C. Among them, enterprise A is the guarantor, the guarantee amount is 500,000 yuan, the guarantee time is t ₃ , and enterprise C is the insured party; there is a person-enterprise relationship between enterprise A and individual a , individual a is a shareholder of enterprise A, the starting time is t ₄ , and the ending time is t ₅ ; there is a guarantee relationship between enterprise D and enterprise B, where enterprise B is the guarantor and the guarantee amount is 400,000 yuan, The guarantee time is t ₆ , and enterprise D is the insured party.

There is a capital flow transaction between enterprise D and individual a. Among them, enterprise D is the capital outflow party, the transaction capital amount is 70,000, the transaction time is t ₇ , and individual a is the capital inflow party; there is a person between enterprise C and individual b. Enterprise association relationship, individual b is the financial director of enterprise C, where the starting time is t ₈ and the ending time is t ₉ ; there is a person-enterprise association relationship between enterprise C and individual c, and individual c is the chairman of enterprise C. , where the starting time is t ₁₀ , and the ending time is t ₁₁ .

Based on the above enterprise data, enterprise A, enterprise B, enterprise C, enterprise D, individual a, individual b and individual c can be used as nodes of the knowledge graph, and each node can be connected by edges according to the event relationship; is the capital outflow party, the transaction capital amount is 50,000, the transaction time is t ₁ , and the company B is the capital inflow party”, “Company A is the investor, the investment amount is 1 million, the transaction time is t ₂ , and the company B is the capital recipient. ", "Enterprise A is the guarantor, the guarantee amount is 500,000 yuan, the guarantee period is t ₃ , and enterprise C is the insured party", "Individual a is the shareholder of enterprise A, and the starting time is t ₄ , and the ending time is t 4 . is t ₅ ”, “Enterprise B is the guarantor, the guarantee amount is 400,000, the guarantee time is t ₆ , and the enterprise D is the insured party”, “Enterprise D is the fund outflow party, the transaction capital amount is 70,000, and the transaction time is t ₇ , individual a is the fund inflow party”, “individual b is the financial director of enterprise C, where the starting time is t ₈ and the ending time is t ₉ ” and “individual c is the chairman of the company C, where, The starting service time is t ₁₀ , and the end service time is t ₁₁ ” as the nodes of the knowledge graph, which are used to explain the corresponding edges.

As a possible implementation manner, as shown in FIG. 4 , operation S220 to construct a knowledge graph according to enterprise data includes operations S221 to S222.

In operation S221, a schema layer is constructed according to the parties and m category data, wherein the schema layer includes nodes established according to the parties and m categories in the m category data, and established according to events in each category data edges between nodes.

In operation S222, the data in the category data is imported into the corresponding schema layer.

Among them, the schema layer can be understood as the structure layer of the knowledge graph. After the structure of the knowledge graph is built by operating S211, the specific data in the category data can be imported into the corresponding schema layer. Therefore, by operating S221 and operating S222, it is convenient to Build a knowledge graph.

Further, as shown in FIG. 5 , the category data is structured data, and operation S222 to import the data in the category data into the corresponding schema layer includes operations S2221 to S2222 .

In operation S2221, the category data is converted into resource description framework data.

In operation S2222, the resource description framework data is imported into the corresponding schema layer.

Through operations S2221 to S2222, the data in the category data can be easily imported into the corresponding schema layer, thereby facilitating the construction of a knowledge graph.

作为一种可实施的方式，归约权重的计算方法如下：In operation S230, based on the knowledge graph constructed by each category of data, the reduction weights of n event information edges between two parties are calculated

As an implementable way, the calculation method of the reduction weight is as follows:

其中，v表示两个当事方中的头节点，u表示两个当事方中的尾节点，E(v，u)表示头节点为v、尾节点为u的n个事件信息边的集合，E(v)表示头节点为v的有向边的集合，

表示边l′在对应的事件发生时刻的初始权重，t表示当前时刻，

表示边l′的时间权重，

表示边l在对应的事件发生时刻的初始权重，

表示边l的时间权重。

Among them, v represents the head node in the two parties, u represents the tail node in the two parties, E(v, u) represents the set of n event information edges with the head node v and the tail node u , E(v) represents the set of directed edges whose head node is v,

represents the initial weight of edge l' at the time of the corresponding event, t represents the current moment,

represents the time weight of edge l′,

represents the initial weight of edge l at the time of the corresponding event,

represents the time weight of edge l.

In operation S240, according to the reduction weights, the aggregated weights ω _vu of the m class data edges between the two parties are calculated. As an implementable manner, the calculation method of the aggregate weight is as follows:

其中，R(v，u)表示头节点为v、尾节点为u的m个类别数据边的集合，C_r表示与不同的类别数据对应的常数系数。

Among them, R(v, u) represents the set of m class data edges whose head node is v and tail node is u, and C _r represents constant coefficients corresponding to different class data.

In operation S250, risk management is performed on the enterprise according to the aggregated weight.

As a possible implementation manner, as shown in FIG. 6 , operation S250 to perform risk management on the enterprise according to the aggregated weight includes operations S251 to S254 .

In operation S251, label propagation is performed on the nodes of the knowledge graph according to the aggregated weights. For example, a target node can be set, with the target node as the center, there are multiple neighbor nodes connected to the target node, the connection between the target node and each neighbor node has an aggregation weight, and according to the aggregation weight, it can be determined which neighbor node the target node propagates. Tag of.

For further example, for example, the target node is A, the target node A has neighbor node B, neighbor node C and neighbor node D, the label of target node A is label 1, the label of neighbor node B and neighbor node C is label 2, and the neighbor node C has label 2. The label of node D is label 3. According to the aggregation weight, it can be determined whether the new label of target node A propagates label 2 or label 3. If target node A propagates label 2, the new label of target node A is label 2; if target node A propagates label 3, then target node A's new label is label 2. The new tab is Tab 3.

In operation S252, the community is divided according to the result of label propagation, and the community size is obtained. It can be understood that after label propagation of the knowledge graph, new labels of each node can be obtained, and each node can be divided into communities according to the new labels. For example, the nodes with the same label 1 are divided into The nodes of label 2 are divided into community 2, and the nodes of the same label 3 are divided into community 3. The number of nodes and edges of each community can be understood as the scale of the community.

In operation S253, according to the community size, the degree centrality of the nodes of the knowledge graph is calculated.

In operation S254, risk prediction is performed using the degree centrality as an input of the risk prediction model. Therefore, by entering the degree centrality into the risk prediction model, the risk prediction model can predict the risk of each node, for example, which node is an overdue customer, which node is a non-overdue customer, and which node is an overdue customer.

As another possible implementation manner, as shown in FIG. 7 , operation S250 to perform risk management on the enterprise according to the aggregated weight includes operations S255 to S256 .

In operation S255, label propagation is performed on the nodes of the knowledge graph according to the aggregated weights. For example, a target node can be set, with the target node as the center, there are multiple neighbor nodes connected to the target node, the connection between the target node and each neighbor node has an aggregation weight, and according to the aggregation weight, it can be determined which neighbor node the target node propagates. Tag of.

For further illustration, for example, the target node is A, the target node A has neighbor node B, neighbor node C and neighbor node D, the label of target node A is a non-risk customer, the labels of neighbor node B and neighbor node C are non-risk customers, neighbors Node D's label is a risky customer. According to the aggregation weight, it can be determined whether the new label of target node A propagates risk customers or non-risk customers. If target node A propagates risk customers, the new label of target node A is risk customers; if target node A propagates non-risk customers, then target node A propagates non-risk customers. A's new label is a non-risk customer.

In operation S256, risk prediction is performed according to the result of label propagation. It is understandable that after label propagation is performed on the knowledge graph, a new label of each node can be obtained, and a prediction result of whether the node is a risk customer or a non-risk customer can be obtained according to the new label.

According to the risk management method based on the knowledge graph according to the embodiment of the present disclosure, by establishing the knowledge graph and calculating the aggregate weight of the edges in the knowledge graph, the risk management of the enterprise can be performed. The risk of loan duration will affect its upstream and downstream related customers, thereby improving the ability to provide inclusive financial services for small and micro enterprises. In the present disclosure, event information is used as a node of the knowledge graph to describe the corresponding edge, so that the edge does not need additional attributes, thereby reducing the redundancy of the edge, thereby making the knowledge graph respond more quickly when used.

In some embodiments of the present disclosure, with reference to FIG. 2 , before operation S220 constructs a knowledge graph according to enterprise data, operation S001 is further included.

In operation S001, the enterprise data is cleaned, wherein the cleaning of the enterprise data includes one of: data deduplication, feature complementation in the data, and abnormal feature processing in the data. In some specific examples, data deduplication can be understood as removing duplicate data in the data. For example, enterprise A included in the enterprise data is the guarantor with a guarantee amount of 2 million, enterprise B is the insured and enterprise B is the insured person, the guarantee amount of 2 million guaranteed by the receiving enterprise A for it is the duplicate data, and one of them can be deleted in the data deduplication operation.

In some specific examples, the completion of features in the data can be understood as missing features in the data, and the missing features can be filled in. For example, the company A included in the company data is the guarantor, the guarantee amount is empty, and the company B If the insured is the insured, the guaranteed amount is the missing feature, which can be filled.

In some specific examples, the processing of abnormal features in the data can be understood as abnormal features in the data, and the abnormal features can be replaced or deleted. For example, enterprise A included in the enterprise data is the guarantor, the guarantee amount is negative, and the enterprise B is the insured, and the guaranteed amount is an abnormal feature, which can be replaced or deleted.

By cleaning enterprise data, relatively standard enterprise data can be obtained as the basic data for building a knowledge graph, so that the established knowledge graph is more accurate.

In some embodiments of the present disclosure, as shown in FIG. 2 , the knowledge graph-based risk management method further includes operation S260.

In operation S260, the knowledge graph is visualized, wherein, as shown in FIG. 8-FIG. 11, the visualization of the knowledge graph in operation S260 includes at least one of operations S261 to S264.

In operation S261, node retrieval is performed based on the knowledge graph. In some examples, the node retrieval may include: in response to the input node name, displaying the node related to the node name and the associated information of the node, for example, taking the target node A as the input point, that is, displaying the connection with the target node A All the neighbor nodes of , the neighbor nodes of the neighbor node, the edge between the target node A and each neighbor node, the edge between the neighbor node and the neighbor node, and the event information node used to describe the corresponding edge. From this, we can understand the position and importance of the target node A in the entire knowledge graph.

In operation S262, a subgraph walk is performed based on the knowledge graph. In some examples, the subgraph walk may include, in response to an operation of manually clicking on an event relationship, exposing the node corresponding to the clicked event relationship and all edges emanating from the node. It can be understood that when a user wants to view the nodes and edges related to an event relationship, he can click on the event relationship based on the knowledge graph, and then the node corresponding to the clicked event relationship and all edges sent by the node can be displayed.

In operation S263, path exploration is performed based on the knowledge graph. In some examples, the path exploration includes: obtaining and displaying the path relationship between two nodes, wherein, by clicking on the two nodes based on the knowledge graph, the path relationship between the two nodes can be obtained and displayed.

In operation S264, at least one of self-loop exploration is performed based on the knowledge graph. In some examples, the self-loop exploration includes: acquiring and displaying nodes whose path relationships are closed loops. For example, the knowledge graph includes node A investing in node B, node B guaranteeing node C, node D is the director of node C, and node D's funds flow into node A. Therefore, there is a closed loop from node A to node A in the above knowledge graph. By visualizing the knowledge graph, the path relationship from node A to node A and all nodes can be obtained and displayed.

Diversifying the use of the knowledge graph through operations S261 to S264 can increase the usability and practicability of the knowledge graph.

The risk management method based on the knowledge graph according to the embodiment of the present disclosure is described in detail below. It is to be understood that the following description is intended to be illustrative only and not specific to the present disclosure.

According to the embodiments of the present disclosure, it is mainly divided into the construction of knowledge graph and the application of knowledge graph, and the construction of knowledge graph is mainly divided into knowledge acquisition, knowledge modeling, knowledge extraction, knowledge storage and knowledge visualization.

Specifically, knowledge acquisition is to screen out the structured data that supports the construction of knowledge graphs from in-line data; knowledge modeling refers to building an ontology model from general domain knowledge and domain expert experience, and using it as the pattern layer of the knowledge graph; knowledge extraction According to the structure of the ontology model, the structured data in the knowledge acquisition process is converted into RDF format data (resource description framework data), so as to complete the construction of the knowledge graph data layer; knowledge storage is to store the constructed knowledge graph in the RDF format. In the graph database of the data; knowledge visualization accesses the graph database through the http service to realize the visualization of the knowledge graph and present it to the credit business personnel.

Among them, in the knowledge acquisition, the target customers of this disclosure are small and micro enterprises that have successfully used the quick loan products, and the time range is only to take the samples with the loan term (expiration date) of the IOU from May 2018 to April 2019. The length of the sliding time period is set to 12 months, that is, the observation period of each sample is 12 months, so the time window 12 months before the expiration date of each small and micro enterprise is taken as the time window.

Considering that small and micro enterprises are less connected to each other and the relationship is sparse, it is impossible to comprehensively calculate the risk propagation path between small and micro enterprises; moreover, in actual business, it is also necessary to put small and micro enterprises into the entire financial business network for modeling , instead of only focusing on the association information between small and micro enterprises, therefore, focusing on the target small and micro enterprise customers, capital flow relationship, investment relationship, guarantee relationship and human-enterprise relationship as the path, the related non-small and micro enterprise loans are expanded. Customers, the specific extension rules are as follows.

(1) Fund flow relationship: Take the target customer as the center and take all the fund flow transaction data within the corresponding time window. Due to the huge number of transactions, the transaction data of each small and micro enterprise is aggregated by month, that is, the transaction data of the same counterparty The transaction amounts are summed and a new transaction record is generated.

(2) Investment relationship: Take the target customer as the center and take all the investment data existing in the corresponding time window.

(3) Guarantee relationship: Take the target customer as the center, take all the data of the guarantee circle in the corresponding time window, and split the guarantee circle, that is, the target customer and any other customer in the guarantee circle will generate a new guarantee circle. guarantee data records.

(4) Person-enterprise relationship: take the target customer as the center, and take all the person-enterprise related data existing in the corresponding time window, including the target customer’s second person in charge, legal representative, insurance legal beneficiary, financial director, shareholders , general manager, unit contact person, chairman of the board and other responsible persons. For the associated individual customers, all the transaction data of the capital flow within the corresponding time window are also taken, and the processing method is the same as the above-mentioned capital flow relationship data.

After completing the customer expansion, data cleaning work is also required, mainly to adjust the different characteristics of different data, including record deduplication, feature filling and exception handling.

(1) Deduplication of records: In the original records of capital flow transaction data, if both parties to the transaction are intra-bank customers, the same transaction will be recorded twice. The difference between the two records is that the loan direction is different, so all records need to be unified. Loan direction, remove duplicate records.

(2) Feature filling: First, we must analyze the reasons for missing features, and then decide how to deal with missing features according to different reasons. There are basically three processing methods: deleting samples, deleting feature variables, and filling missing values. Specifically, non-bank customers do not have an ID primary key in the data. Due to the large number of samples involved, they should not be deleted. The MD5 encrypted customer account number is used instead of the ID primary key to fill in the missing values.

(3) Exception handling: To confirm the abnormal value of the data, firstly, it should be determined in combination with the results of business and exploratory data analysis. Secondly, some methods commonly used in statistics can be used to determine it, and then the abnormal value should be processed accordingly. Eliminate the impact of outliers on model training. For example, the transaction amount in the cash flow transaction data should be a positive number, delete the cash flow transaction data with negative transaction amount.

Among them, in knowledge modeling, knowledge graph is mainly divided into data layer and schema layer in logical structure. The data layer contains a large amount of factual information, namely (entity, relationship, entity) or (entity, attribute, attribute value) and other three Tuple representation, storing these data in a graph database will form a large-scale entity-relationship network, and then form a knowledge graph. The schema layer is the core of the knowledge graph, built on the data layer, and stores the refined knowledge. The ontology model is usually used to manage the schema layer, that is, the support ability of the ontology model for axioms, rules and constraints is used to standardize the relationship between entities, relationships, and objects such as entity types and attributes.

The present disclosure adopts a top-down approach to construct the financial knowledge graph of small and micro enterprises. First, define the ontology, or data schema, for the knowledge graph, and then generate the data layer of the knowledge graph. This construction method is generally suitable for domain knowledge graphs. Construct. In the process of defining the ontology, first start with the top-level concept, and then gradually refine it to form a well-structured hierarchical structure; after defining the ontology, add entity data into the concept of the ontology one by one, and the underlying entity data is based on The relationship between ontology concepts automatically generates a knowledge graph.

According to the above method, the relevant knowledge in the financial field is deeply excavated, the data set in the knowledge acquisition stage is investigated as a whole, and the financial ontology model of small and micro enterprises is constructed by analyzing the semantic association between concepts and attributes in the field. The ontology model includes five types of entity nodes: enterprise and/or individual customers, human-enterprise association events, capital transaction events, guarantee events, and investment events, including four types of entity nodes: enterprise-individual, capital inflow-outflow, guarantee-guaranteed, and investment-invested four. The class association relationship, the attributes of each type of entity node are as follows.

(1) Corporate/individual customers: corporate ID primary key, corporate name and corporate category.

(2) Person-enterprise related events: personal position, start time and end time.

(3) Fund transaction event: transaction event ID primary key, transaction event amount and transaction event.

(4) Guarantee event: Guarantee circle ID primary key and guarantee circle establishment time.

(5) Investment event: investment event ID primary key, investment event amount, investment event start time and investment event end time.

The present disclosure instantiates the association relationship between different customers as an entity node, rather than a simple relationship link, for example, instantiate a fund transaction event node to replace the fund transaction link. This is because, in the financial knowledge graph of small and micro enterprises, the relationship links also contain a lot of information, such as transaction amount and transaction time. If an edge represents this relationship, a lot of key information will be lost. Therefore, four types of relationship events are instantiated in the financial knowledge graph of small and micro enterprises to display the attributes of the relationship in a more comprehensive and flattened manner.

At present, the knowledge graph data layer generally adopts the RDF (Resource Description Framework) model to represent data. RDF is a standard data model for representing and exchanging machine-understandable information developed by the W3C World Wide Web Consortium for the Semantic Web. It uses Web Identifiers (URIs) to identify resources, and uses attributes and attribute values to describe resources. In an RDF graph, each resource has a (HTTP URI) as its unique address. An RDF graph is defined as a finite set of triples (s, p, o); each triple represents a fact statement, where s is the subject, p is the predicate, and o is the object ); (s, p, o) means that there is a connection p between s and o, or that s has an attribute p and its value is o. A triple in RDF is sometimes called a statement, and we also call it a piece of knowledge in the knowledge graph.

RDF-oriented database is a way to extract knowledge from relational database. Database table names are directly mapped to classes in RDF, fields are mapped to attributes of classes, and relationships between different classes can be derived from tables representing relationships .

Among them, in knowledge storage, one of the core issues of knowledge graph data management in RDF format is how to efficiently store RDF datasets and quickly answer SPARQL queries. In general, there are two completely different sets of ideas. One is that we can use existing mature database management systems (such as relational database systems) to store knowledge graph data, and convert SPARQL queries oriented to RDF knowledge graphs into queries oriented to such mature database management systems, such as relational database oriented The second is to directly develop a native knowledge graph data storage and query system (Native RDF, that is, a graph database system) for RDF knowledge graph data, considering that The characteristics of RDF knowledge graph management are optimized from the bottom layer of the database system.

The present disclosure chooses a native knowledge graph storage management solution for RDF graphs, because when a large number of relationships need to be described, the traditional relational database is already overwhelmed, and it can bear the situation that there are many entities but the relationship between entities is slightly simple. For small and micro enterprises, the relationship between entities is very complex, and it is often necessary to record data in the relationship, and most of the operations on the data are related to the relationship, the RDF graph database is a more reasonable choice. It can not only bring us the improvement of operating performance, but also greatly improve the efficiency of system development and reduce maintenance costs.

Among them, in the knowledge visualization, a knowledge graph platform for small and micro enterprises can be built to provide a human-computer interaction interface for loan personnel, and assist loan personnel in dynamic management of small and micro enterprise customers. The platform will realize the visualization of the knowledge graph of small and micro enterprises, as well as four main functions of graph-based node retrieval, subgraph walking, path exploration and self-loop exploration, which are described as follows.

(1) Node retrieval: Node retrieval is the most basic and core function of the knowledge graph of small and micro enterprises. For a given target node name, the platform can respond quickly and display the target node and related information in the visual interface. The visual interface will display the attribute information of the target node and the neighbor nodes within a certain number of hops on the graph, that is, the associated entities of the target entity. At the same time, since the knowledge graph can realize the deep correlation query operation, the credit personnel can obtain the structure of the entire relationship network where the target node is located, such as the complete upstream and downstream enterprise chain where the target node is located, the complete guarantee circle where the target node is located, etc., so as to understand The position and importance of the target node in the entire relationship network.

(2) Subgraph walking: In the knowledge graph of small and micro enterprises, each node has different numbers and types of edges, and accordingly, the number and types of its neighbor nodes are also quite different. Therefore, There is no fixed pattern to comprehensively describe the feature profile of nodes. Credit officers can obtain node association information in a targeted manner by heuristically walking on the node neighborhood subgraph. The subgraph walk method comes from the random walk algorithm based on the graph. Random walk is a method of extracting the structural features of the graph. In short, the random walk algorithm builds several random walkers, each of which is a random walker. The random walker is initialized from a node, and then randomly visits a neighbor node of the current node in each random walk step.

The biggest difference between the subgraph walk and random walk on the knowledge graph of small and micro enterprises is that the nodes to be visited in the next step can be selected manually instead of random methods. For example, the guarantee relationship between enterprises is an important but very important feature of small and micro enterprises. For sparse data, for the target small and micro enterprise node, if it has multiple edges of the guarantee relationship on the knowledge graph, the credit officer can choose to walk along the edge of the guarantee relationship; if the node does not have a guarantee relationship, then You can choose other relationships for sub-graph walks to obtain comprehensive feature portraits of different enterprises in a targeted manner.

(3) Path exploration: For two given target node names, the small and micro enterprise knowledge graph platform can explore whether there is a path on the graph between the two nodes, and the loan officer can check whether the small and micro enterprise customer is related to the industry's leading enterprises or listed companies. Whether there is a path for the enterprise and whether there is a path with the loan defaulting enterprise in the industry provides a reference for risk assessment. In addition, there may be no direct relationship between the two enterprises, that is, there is no edge between the two corresponding nodes in the graph, but there may be various relationships through intermediate nodes, such as competition relationship, holding relationship, investment and financing relationship, etc. The loan officer can obtain the indirect relationship between the two enterprises through the analysis of the channel between the two enterprises.

(4) Self-loop exploration: For a given target node name, the knowledge graph platform for small and micro enterprises can retrieve whether there are one or more self-loop paths in the graph. Presented to loan officers. The self-loop on the financial knowledge map of small and micro enterprises means that small and micro customers may have the risk of arbitrage, that is, the loan issued to this customer flows through other multi-party customers and then is transferred back to itself, which requires the focus of credit personnel and separate analysis. .

The application of knowledge graph is mainly divided into edge weight modeling, community division, graph feature calculation and risk propagation prediction.

Among them, in the edge weight modeling, the financial knowledge graph of small and micro enterprises constructed above can be regarded as a directed multi-graph after ignoring the attributes on the nodes, that is, on the basis of the directed simple graph, There are more than one edge between two nodes in the graph. For the financial knowledge graph of small and micro enterprises, the capital flow relationship and the investment relationship are both directed edges. The direction of the capital flow relationship represents the direction of capital outflow and inflow, and the direction of the investment relationship represents the investment and the investee. There may be multiple relationships between the two clients, and the two clients may have capital flow transactions in different months, and there will be multiple capital flow transaction connections. Therefore, the knowledge graph is a multiple directed graph. In order to model the graph features of nodes, the multiple graphs need to be reduced to a simple graph first, and then the network feature indicators can be calculated intuitively.

Based on the constructed knowledge graph, assuming that the set of association relationships contained in the graph is {R ₁ , R ₂ , R ₃ ... R _n }, G is divided into multiple relation graphs according to different relationships, such as G _r1 , G _r2 , G _r3 ...G _rn , each relational graph G _ri only contains the edges with the relation ri in the knowledge graph, and the types of edges between any two connected nodes in G _ri are the same, but the time of connecting the edges is the same. It may be different from other attributes. The weights are calculated for the edges according to the difference of attribute values. Finally, the weights of all the edges between the two nodes are summed and reduced to one edge.

归约后的新连边权重为

其中，

为图G_ri中的节点v的出边邻居节点集合，v表示两个当事方中的头节点，u表示两个当事方中的尾节点，E(v，u)表示头节点为v、尾节点为u的n个事件信息边的集合，E(v)表示头节点为v的有向边的集合，

表示边l′在对应的事件发生时刻的初始权重，t表示当前时刻，

表示边l′的时间权重，

表示边l在对应的事件发生时刻的初始权重，

表示边l的时间权重。资金流关系中初始权重为交易事件的交易金额属性值，投资关系中初始权重为投资事件的投资金额属性值，其余关系中初始权重均为1；边的创建时刻t_l距离当前时刻t跨度越大，时间权重越小，这条边的重要性就越低。Specifically, for a node v in a relational graph G _ri , if the node

The weight of the new connection after reduction is

in,

is the set of outgoing neighbor nodes of the node v in the graph G _ri , v represents the head node of the two parties, u represents the tail node of the two parties, and E(v, u) represents that the head node is v , the set of n event information edges whose tail node is u, E(v) represents the set of directed edges whose head node is v,

represents the initial weight of edge l' at the time of the corresponding event, t represents the current moment,

represents the time weight of edge l′,

represents the initial weight of edge l at the time of the corresponding event,

represents the time weight of edge l. The initial weight in the capital flow relationship is the transaction amount attribute value of the transaction event, the initial weight in the investment relationship is the investment amount attribute value of the investment event, and the initial weight in the other relationships is 1; the creation time t _l of the edge is more than the current time t span. The smaller the time weight, the less important this edge is.

其中，N_out(v)为新的关联关系图谱中节点v的出边邻居节点集合，R(v，u)表示头节点为v、尾节点为u的m个类别数据边的集合，C_r表示与不同的类别数据对应的常数系数。After the aggregation of multiple edges in the relational graph G _ri is completed, all relational graphs are merged to obtain a new relational relation graph. For the node v in the graph, if the node u∈N _out (v), then the new edge weight

Among them, N _out (v) is the set of outgoing neighbor nodes of node v in the new relationship graph, R(v, u) is the set of m types of data edges whose head node is v and tail node is u, C _r Represents constant coefficients corresponding to different categories of data.

Finally, a directed simple graph with only one edge in one direction between any two adjacent nodes is obtained, and the greater the weight of the edge, the more closely the two nodes are related.

Among them, in the community division, due to the large number of nodes in the graph, the differences in the results of the global calculation of node graph features are small, and a large number of repeated values appear, and the calculation results cannot be used as features. Therefore, the label propagation algorithm is applied to divide the community of the new graph based on the edge weight, and the graph features of the nodes are calculated in the divided community subgraph.

The basic process of community division by label propagation is to first assign a different community label to each node in the graph, and then these labels start to propagate in the knowledge graph. In each step of propagation, each node will be based on the neighbor nodes. In the case of the label, update its own label. Specifically, each node will choose to join the label category with the largest sum of edge weights among neighbor nodes. As the label propagates, the final set of closely connected nodes will reach a consensus, and the labels on them will no longer change. The community division algorithm based on label propagation will select the label propagation path according to the weight of the edges between nodes, that is, the degree of connection tightness, and finally divide the closely related enterprise entities into the same community. Class division makes it easier to determine potential information such as the group in which the enterprise is located, and risk communication.

Calculating the graph features of nodes in different community subgraphs will ignore the edges between different communities, resulting in information waste. In order to make full use of data information, the present disclosure defines the node features calculated in the community as local feature information ;Abstract each community as a node, and the weighted summation of all the edges between the two communities is used as the edge between the community nodes, and a community association network is obtained, and the characteristics of the community nodes in the network are calculated as Global feature information of original client nodes in the community. Considering global features and local features comprehensively, limited data can be fully utilized.

Among them, in the graph feature calculation, the present disclosure selects six indicators of degree centrality, feature vector centrality, betweenness centrality, close centrality, HITS value and PAGERANK value to calculate the features of local and global dimensions, and use these graph features Combined with traditional variable features, a logistic regression model is established to predict the loan overdue risk of small and micro customers. In the end, a total of 12 variables are entered into the model, including the degree centrality feature of the global dimension, the edge closeness centrality feature of the local dimension, and the hub of the local dimension. The three graph features of the value feature indicate that this feature has the highest degree of discrimination for overdue risks of small and micro enterprise customers, and can be used as a model variable to make up for the feature dimension of traditional prediction models.

Among them, in the risk propagation prediction, in addition to calculating the graph features, the present disclosure also applies the label propagation algorithm to classify and predict the default nodes. The label propagation algorithm believes that the label of each node should be the same as the label of most of its neighbors, that is, when When a small and micro enterprise customer has no overdue risk, but multiple neighboring customers closely related to it have overdue risk, the risk is very likely to spread to the non-overdue customer. Warning to avoid.

First, add a label to each node to represent whether the customer's loan is overdue, and then iteratively propagate on the graph to update the node's label. After a certain number of iterations, whether the labels of the nodes in the statistical graph have changed, and it is found that some nodes with unknown labels are propagated as risk nodes. The unknown labels represent out-of-bank customers who do not hold loans in our bank, and are propagated as risk nodes for subsequent loans. Access provides reference; there are a small number of customer labels that have never been overdue and spread to overdue customers, indicating that these customers are easily passed on by risks, and they are the focus of duration management.

Based on the above risk management method based on knowledge graph, the present disclosure also provides a risk management device 10 based on knowledge graph. The risk management device 10 will be described in detail below with reference to FIG. 12 .

FIG. 12 schematically shows a structural block diagram of the risk management apparatus 10 based on the knowledge graph according to an embodiment of the present disclosure.

The knowledge graph-based risk management device 10 includes an acquisition module 1 , a construction module 2 , a first calculation module 3 , a second calculation module 4 and a management module 5 .

Obtaining module 1, the obtaining module 1 is configured to perform operation S210: obtaining enterprise data, wherein the enterprise data includes a party and an event relationship between the parties, the event relationship includes m types of data, and each type of data includes n data Event information, where m is an integer greater than or equal to 1, and n is an integer greater than or equal to 1.

Building module 2, the building module 2 is used to perform operation S220: build a knowledge graph according to enterprise data, wherein the parties and event information are nodes of the knowledge graph, the event relationship is an edge of the knowledge graph, and the event information is used to explain the corresponding side.

A first calculation module 3, the first calculation module 3 is configured to perform operation S230: based on the knowledge graph constructed by each category of data, calculate the reduction weights of n event information edges between two parties

The second calculation module 4 is configured to perform operation S240: according to the reduction weight, calculate the aggregation weight ω _vu of the m category data edges between the two parties.

Management module 5. The management module 5 is configured to perform operation S250: perform risk management on the enterprise according to the aggregated weight.

Since the above-mentioned risk management apparatus 10 is set up based on the risk management method, the beneficial effects of the above-mentioned risk management apparatus 10 are the same as those of the risk management method, which will not be repeated here.

In addition, according to the embodiment of the present disclosure, any number of modules in the acquisition module 1, the construction module 2, the first calculation module 3, the second calculation module 4, and the management module 5 may be combined in one module for implementation, or any of them may be implemented in one module. A module can be split into multiple modules. Alternatively, at least part of the functionality of one or more of these modules may be combined with at least part of the functionality of other modules and implemented in one module.

According to an embodiment of the present disclosure, at least one of the acquisition module 1 , the building module 2 , the first computing module 3 , the second computing module 4 , and the management module 5 may be at least partially implemented as a hardware circuit, such as a field programmable gate array (FPGA), Programmable Logic Array (PLA), System-on-Chip, System-on-Substrate, System-on-Package, Application-Specific Integrated Circuit (ASIC), or any other reasonable means by which circuits can be integrated or packaged such as hardware or Firmware, or any one of software, hardware and firmware, or any appropriate combination of any of them.

Alternatively, at least one of the acquisition module 1, the building module 2, the first computing module 3, the second computing module 4, and the management module 5 may be implemented, at least in part, as a computer program module that, when executed, may perform the corresponding function.

13 schematically shows a block diagram of an electronic device suitable for implementing a knowledge graph-based risk management method according to an embodiment of the present disclosure.

As shown in FIG. 13 , an electronic device 900 according to an embodiment of the present disclosure includes a processor 901 that can be loaded into a random access memory (RAM) 903 according to a program stored in a read only memory (ROM) 902 or from a storage portion 908 program to perform various appropriate actions and processes. The processor 901 may include, for example, a general-purpose microprocessor (eg, a CPU), an instruction set processor and/or a related chipset, and/or a special-purpose microprocessor (eg, an application-specific integrated circuit (ASIC)), and the like. The processor 901 may also include on-board memory for caching purposes. The processor 901 may include a single processing unit or multiple processing units for performing different actions of the method flow according to the embodiments of the present disclosure.

In the RAM 903, various programs and data necessary for the operation of the electronic device 900 are stored. The processor 901 , the ROM 902 and the RAM 903 are connected to each other through a bus 904 . The processor 901 performs various operations of the method flow according to the embodiment of the present disclosure by executing the programs in the ROM 902 and/or the RAM 903 . Note that the program may also be stored in one or more memories other than the ROM 902 and the RAM 903 . The processor 901 may also perform various operations of the method flow according to the embodiments of the present disclosure by executing programs stored in the one or more memories.

According to an embodiment of the present disclosure, the electronic device 900 may also include an input/output (I/O) interface 905 which is also connected to the bus 904 . Electronic device 900 may also include one or more of the following components connected to I/O interface 905: input portion 906 including keyboard, mouse, etc.; including components such as cathode ray tube (CRT), liquid crystal display (LCD), etc., and An output section 907 of speakers and the like; a storage section 908 including a hard disk and the like; and a communication section 909 including a network interface card such as a LAN card, a modem, and the like. The communication section 909 performs communication processing via a network such as the Internet. Drive 910 is also connected to input/output (I/O) interface 905 as needed. A removable medium 911, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc., is mounted on the drive 910 as needed so that a computer program read therefrom is installed into the storage section 908 as needed.

The present disclosure also provides a computer-readable storage medium. The computer-readable storage medium may be included in the device/apparatus/system described in the above embodiments; it may also exist alone without being assembled into the device/system. device/system. The above-mentioned computer-readable storage medium carries one or more programs, and when the above-mentioned one or more programs are executed, implement the method according to the embodiment of the present disclosure.

According to an embodiment of the present disclosure, the computer-readable storage medium may be a non-volatile computer-readable storage medium, such as, but not limited to, portable computer disks, hard disks, random access memory (RAM), read only memory (ROM) , erasable programmable read only memory (EPROM or flash memory), portable compact disk read only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing. In this disclosure, a computer-readable storage medium may be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. For example, according to embodiments of the present disclosure, a computer-readable storage medium may include one or more memories other than ROM 902 and/or RAM 903 and/or ROM 902 and RAM 903 described above.

Embodiments of the present disclosure also include a computer program product comprising a computer program containing program code for performing the method illustrated in the flowchart. When the computer program product is run in a computer system, the program code is used to cause the computer system to implement the methods of the embodiments of the present disclosure.

When the computer program is executed by the processor 901, the above-described functions defined in the system/apparatus of the embodiment of the present disclosure are performed. According to embodiments of the present disclosure, the systems, apparatuses, modules, units, etc. described above may be implemented by computer program modules.

In one embodiment, the computer program may rely on a tangible storage medium such as an optical storage device, a magnetic storage device, or the like. In another embodiment, the computer program may also be transmitted, distributed in the form of a signal over a network medium, and downloaded and installed through the communication section 909, and/or installed from a removable medium 911. The program code embodied by the computer program may be transmitted using any suitable network medium, including but not limited to: wireless, wired, etc., or any suitable combination of the foregoing.

In such an embodiment, the computer program may be downloaded and installed from the network via the communication portion 909, and/or installed from the removable medium 911. When the computer program is executed by the processor 901, the above-described functions defined in the system of the embodiment of the present disclosure are performed. According to embodiments of the present disclosure, the above-described systems, apparatuses, apparatuses, modules, units, etc. can be implemented by computer program modules.

According to the embodiments of the present disclosure, the program code for executing the computer program provided by the embodiments of the present disclosure may be written in any combination of one or more programming languages, and specifically, high-level procedures and/or object-oriented programming may be used. programming language, and/or assembly/machine language to implement these computational programs. Programming languages include, but are not limited to, languages such as Java, C++, python, "C" or similar programming languages. The program code may execute entirely on the user computing device, partly on the user device, partly on a remote computing device, or entirely on the remote computing device or server. In the case of a remote computing device, the remote computing device may be connected to the user computing device through any kind of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computing device (eg, using an Internet service provider business via an Internet connection).

The flowchart and block diagrams in the Figures illustrate the architecture, functionality, and operation of possible implementations of systems, methods and computer program products according to various embodiments of the present disclosure. In this regard, each block in the flowchart or block diagrams may represent a module, segment, or portion of code that contains one or more logical functions for implementing the specified functions executable instructions. It should also be noted that, in some alternative implementations, the functions noted in the blocks may occur out of the order noted in the figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently, or the blocks may sometimes be executed in the reverse order, depending upon the functionality involved. It is also noted that each block of the block diagrams or flowchart illustrations, and combinations of blocks in the block diagrams or flowchart illustrations, can be implemented in special purpose hardware-based systems that perform the specified functions or operations, or can be implemented using A combination of dedicated hardware and computer instructions is implemented.

Those skilled in the art will appreciate that various combinations and/or combinations of features recited in various embodiments and/or claims of the present disclosure are possible, even if such combinations or combinations are not expressly recited in the present disclosure. In particular, various combinations and/or combinations of the features recited in the various embodiments of the present disclosure and/or in the claims may be made without departing from the spirit and teachings of the present disclosure. All such combinations and/or combinations fall within the scope of this disclosure.

Embodiments of the present disclosure have been described above. However, these examples are for illustrative purposes only, and are not intended to limit the scope of the present disclosure. Although the various embodiments are described above separately, this does not mean that the measures in the various embodiments cannot be used in combination to advantage. The scope of the present disclosure is defined by the appended claims and their equivalents. Without departing from the scope of the present disclosure, those skilled in the art can make various substitutions and modifications, and these substitutions and modifications should all fall within the scope of the present disclosure.

Claims (14)

1. A risk management method based on knowledge graph is characterized by comprising the following steps:

acquiring enterprise data, wherein the enterprise data comprises a party and an event relation between the party, the event relation comprises m category data, each category data comprises n event information, m is an integer greater than or equal to 1, and n is an integer greater than or equal to 1;

constructing a knowledge graph according to the enterprise data, wherein the parties and the event information are nodes of the knowledge graph, the event relation is an edge of the knowledge graph, and the event information is used for explaining the corresponding edge;

calculating reduction weights of n event information edges between the two parties based on the knowledge graph constructed for each of the category data

Calculating an aggregation weight ω of m class-data edges between the two parties according to the reduction weight_vu(ii) a And

and carrying out risk management on the enterprises according to the aggregation weight.

2. The method of knowledge-graph-based risk management according to claim 1, wherein the reduction weight is calculated as follows:

wherein v represents a head node in two of the parties, u represents a tail node in two of the parties, E (v, u) represents a set of n event information edges with a head node of v and a tail node of u, E (v) represents a set of directed edges with a head node of v,

indicating the initial weight of the edge l' at the corresponding event occurrence time, t indicating the current time,

the temporal weight of the edge l' is represented,

indicating the initial weight of the edge/at the time of the corresponding event occurrence,

represents the temporal weight of the edge l; and

the calculation method of the aggregation weight is as follows:

wherein R (v, u) represents a set of m said class data edges with a head node v and a tail node u, C_rRepresenting constant coefficients corresponding to different ones of the category data.

3. The method of knowledge-graph-based risk management according to claim 1, wherein the m category data includes at least one of capital stream data, investment data, warranty data, and corporate-related data.

4. The method of knowledge-graph-based risk management according to claim 1, wherein the n event information comprises event information under the same category data for n time periods.

5. The method of knowledge-graph based risk management according to claim 1, further comprising, prior to said building a knowledge-graph from the enterprise data:

cleansing the enterprise data, wherein cleansing the enterprise data comprises: one of data deduplication, feature in data completion, and anomalous feature processing in data.

6. The method of knowledge-graph based risk management according to claim 1, wherein said building a knowledge-graph from the enterprise data comprises:

constructing a schema layer from the party and the m category data, wherein the schema layer comprises nodes established from the party and m categories of the m category data, and edges between the nodes established from events in each of the category data; and

and importing the data in the category data into a corresponding mode layer.

7. The method of knowledge-graph-based risk management according to claim 6, wherein the category data is structured data, and importing the data in the category data into a corresponding schema layer comprises:

converting the category data into resource description framework data; and

and importing the resource description framework data into a corresponding mode layer.

8. The method of knowledge-graph-based risk management according to claim 1, wherein said risk managing an enterprise according to the aggregation weight comprises:

performing label propagation on the nodes of the knowledge graph according to the aggregation weight;

carrying out community division according to the label propagation result to obtain the community scale;

calculating the degree centrality of the nodes of the knowledge graph according to the community scale; and

and performing risk prediction by taking the centrality as an input of a risk prediction model.

9. The method of knowledge-graph-based risk management according to claim 1, wherein said risk managing an enterprise according to the aggregation weight comprises:

performing label propagation on the nodes of the knowledge graph according to the aggregation weight; and

and predicting the risk according to the label propagation result.

10. The method of knowledge-graph-based risk management according to any one of claims 1-9, further comprising visualizing the knowledge-graph, wherein the visualizing the knowledge-graph comprises: at least one of node retrieval based on the knowledge graph, sub-graph walking based on the knowledge graph, path exploration based on the knowledge graph, and self-loop exploration based on the knowledge graph.

11. The knowledge-graph based risk management method of claim 10,

the node retrieval includes: responding to the input node name, and displaying a node related to the node name and associated information of the node;

the subgraph wandering comprises: responding to the operation of manually clicking the event relation, and displaying a node corresponding to the clicked event relation and all edges sent by the node;

the path exploration comprises the following steps: acquiring and displaying a path relation between two nodes; and

the self-loop exploration comprises the following steps: and acquiring and displaying the nodes with the closed-loop path relation.

12. A knowledge-graph-based risk management device, comprising:

the system comprises an acquisition module, a processing module and a processing module, wherein the acquisition module is used for acquiring enterprise data, the enterprise data comprises a party and an event relation between the party, the event relation comprises m category data, each category data comprises n event information, m is an integer which is greater than or equal to 1, and n is an integer which is greater than or equal to 1;

a construction module, configured to construct a knowledge graph according to the enterprise data, where the party and the event information are nodes of the knowledge graph, the event relationship is an edge of the knowledge graph, and the event information is used to explain the corresponding edge;

a first calculation module for calculating reduction weights of n event information edges between the two parties based on the knowledge graph constructed for each of the category data

A second calculation module for calculating an aggregation weight ω of m class-data edges between the two parties according to the reduction weight_vu(ii) a And

and the management module is used for carrying out risk management on the enterprises according to the aggregation weight.

13. An electronic device, comprising:

one or more processors;

one or more memories for storing executable instructions that, when executed by the processor, implement the method of any of claims 1-11.

14. A computer-readable storage medium having stored thereon executable instructions that when executed by a processor implement a method according to any one of claims 1 to 11.

Images