Enterprise risk updating method, device, equipment, medium and program product

Abstract

The invention provides an enterprise risk updating method which can be applied to the technical field of big data. The method comprises the following steps: in response to monitored dynamic data of a target enterprise, retrieving a target entity, obtaining tense attribute data of the target entity, and performing comparative analysis based on the tense attribute data and the dynamic data to obtain a change index; based on a preset hop count, retrieving an association graph taking the target entity as a central entity, performing information propagation based on the association graph through a preset information propagation model, and transmitting risk information in the change index from the central entity to other enterprise entities in the association graph to obtain propagation information of the other enterprise entities; carrying out risk prediction based on the change index or propagation information of the to-be-processed enterprise through a multi-dimensional risk assessment model to obtain a risk increment; and calculating the sum of the risk increment and the current risk probability to obtain an updated risk probability. The invention further provides an enterprise risk updating device and equipment, a storage medium and a program product.

Description

Technical Field

[0001] This application relates to the field of big data technology, specifically to a method, apparatus, equipment, medium, and program product for updating enterprise risks. Background Technology

[0002] In the process of enterprises applying for financial services, efficient risk identification by financial institutions is a key barrier to ensuring asset quality and preventing financial risks. Some financial services are processed by having the enterprise client submit materials, which the financial institution then conducts an initial risk assessment. After passing the initial risk assessment, the financial institution still needs to go through multiple approval stages, or the approval process may be delayed due to time constraints, before finally issuing a decision to grant or deny the application. Only after granting the application will the relevant funds be disbursed to the enterprise client, thus ensuring the safety of their financial assets.

[0003] Currently, in cases involving multiple approval stages or delayed approval processes, financial institutions still rely on approval personnel to make empirical judgments based on the initial static materials submitted by the client. This decision-making approach makes it difficult to dynamically track changes in the client's risk. For example, some risk events occur after the company submits its materials, and existing risk identification methods cannot capture these risks in real time. Furthermore, since multiple companies applying for this financial business may be related, a dynamic risk event in one company may affect other companies. Existing risk detection methods cannot effectively identify real-time risks caused by updates to dynamic risk events between companies. This results in problems such as delayed capture of key risk signals, inability to quickly identify abnormal changes, and biased risk assessment criteria. Consequently, financial institutions cannot provide timely warnings of potential risks before disbursing funds, and cannot effectively protect the safety of their assets. Summary of the Invention

[0004] In view of the above problems, this application provides a method, apparatus, equipment, medium and program product for updating enterprise risks.

[0005] According to the first aspect of this application, a method for updating enterprise risk is provided, comprising: responding to the monitoring of dynamic data of a target enterprise, retrieving the target entity corresponding to the target enterprise from a pre-constructed temporal knowledge graph, and obtaining the temporal attribute data of the target entity; performing comparative analysis based on the temporal attribute data and dynamic data to obtain change indicators; wherein, the dynamic data indicates changes in business conditions or public opinion, and the temporal attribute data includes attributes corresponding to multiple different times; retrieving a relationship graph with the target entity as the central entity from the temporal knowledge graph based on a preset number of hops, and performing information propagation based on the relationship graph through a preset information propagation model to transmit risk information in the change indicators from the central entity to other enterprise entities in the relationship graph, thereby obtaining propagation information of other enterprise entities; the relationship graph includes multiple enterprise entities associated with the central entity; when an enterprise entity in the relationship graph matches any enterprise to be processed, performing risk prediction based on the change indicators or propagation information of the enterprise to be processed through a pre-trained multi-dimensional risk assessment model to obtain a risk increment; obtaining the current risk probability of the enterprise to be processed, and calculating the sum of the risk increment and the current risk probability to obtain an updated risk probability, thereby determining the processing result of the business to be processed based on the updated risk probability.

[0006] According to an embodiment of this application, a comparative analysis is performed based on temporal attribute data and dynamic data to obtain the following change indicators: when the dynamic data indicates the operating status, the operating indicators in the dynamic data are extracted, the historical average of the corresponding operating indicators in the temporal attribute data is calculated, and the difference between the operating indicators and the historical average is calculated to obtain the absolute fluctuation range; the ratio of the absolute fluctuation range to the historical average is calculated to obtain the relative fluctuation range, and the relative fluctuation range is used as the change indicator. The operating indicators include at least one of the following: profit margin, operating revenue growth rate, net profit growth rate, and asset-liability ratio.

[0007] According to an embodiment of this application, the method of obtaining change indicators by comparing and analyzing temporal attribute data and dynamic data further includes: extracting operating indicators from dynamic data when dynamic data indicates operating conditions, extracting historical indicators corresponding to operating indicators from temporal attribute data, and sorting historical indicators and operating indicators by time to obtain a time-series indicator sequence; performing trend prediction based on the time-series indicator sequence through a temporal convolutional network to obtain indicator trends, and using indicator trends as change indicators.

[0008] According to the embodiments of this application, the change indicators obtained by comparing and analyzing temporal attribute data and dynamic data include: when dynamic data indicates public opinion information, the dynamic data is used to identify the sentiment tendency through a preset public opinion analysis model to obtain a sentiment tendency score; based on preset weights, the number of clicks and the number of reposts in the dynamic data are weighted and summed to obtain the dissemination volume, and the dissemination volume is standardized to obtain the dissemination popularity; the sentiment tendency score and the dissemination popularity are used as change indicators.

[0009] According to an embodiment of this application, pre-constructing a temporal knowledge graph includes: using a pre-trained entity extraction model to extract entities from collected multidimensional data to obtain an entity set, wherein the multidimensional data includes at least one of enterprise business registration information, enterprise financial statements, enterprise public opinion information, and enterprise business application materials; for any entity in the entity set, obtaining an entity dataset related to the entity in the multidimensional data, and dividing the entity dataset into multiple temporal subsets according to time information, and extracting features from the temporal subsets to obtain entity features; storing the entity features and corresponding time information in association as temporal attribute data of the entity; using a pre-trained relation recognition model to identify the relationship types between entities to obtain an entity relation triplet set, and constructing a knowledge graph based on the triplet set to obtain a temporal knowledge graph, wherein the relation types include at least one of supply chain association, guarantee association, or equity association.

[0010] According to an embodiment of this application, information is propagated based on an association graph using a preset information propagation model to transmit risk information in changing indicators from a central entity to other enterprise entities in the association graph, thereby obtaining propagation information of other enterprise entities. This includes: determining the first edge weight corresponding to the first relationship type in the preset information propagation model based on the first relationship type between the central entity and the enterprise entity associated with the first hop in the association graph; propagating information about changing indicators based on the first edge weight to obtain propagation information of the first-hop enterprise entity; at any hop after the first hop, obtaining the second relationship type between the current enterprise entity and the enterprise entity associated with the next hop, determining the second edge weight corresponding to the second relationship type in the information propagation model; and propagating the propagation information of the current enterprise entity based on the second edge weight to obtain propagation information of the next-hop enterprise entity.

[0011] According to an embodiment of this application, propagating the propagation information of the current enterprise entity based on the second edge weight to obtain the propagation information of the next-hop enterprise entity includes: performing a linear transformation on the propagation information of the current enterprise entity based on the edge weight to obtain transformed information; determining an attenuation factor based on the distance between the next-hop enterprise entity and the central entity; and obtaining the propagation information of the next-hop enterprise entity based on the product of the attenuation factor and the transformed information.

[0012] According to an embodiment of this application, obtaining the current risk probability of the enterprise to be processed includes: in response to receiving a risk question about the enterprise to be processed, searching a preset temporal knowledge graph based on the risk question to obtain search results, the search results including nodes and / or edges related to the risk question in the temporal knowledge graph and the association information of the nodes and / or edges; extracting features based on the search results to obtain multi-dimensional risk features, and performing risk prediction based on the multi-dimensional risk features through a multi-dimensional risk assessment model to obtain the initial risk probability of the enterprise to be processed, and using the initial risk probability as the current risk probability.

[0013] According to an embodiment of this application, pre-training a multi-dimensional risk assessment model includes: repeatedly training a preset risk prediction network based on pre-acquired multi-dimensional risk features of samples and corresponding sample risk probabilities until the fusion loss value of the risk prediction network reaches a preset loss threshold, then using the optimal risk prediction network as the multi-dimensional risk assessment model; the method for obtaining the fusion loss value includes: performing risk prediction based on the multi-dimensional risk features of samples through the risk prediction network to obtain a first predicted value; randomly grouping the multi-dimensional risk features of samples to obtain a first sub-feature and a second sub-feature, and performing risk prediction based on the first sub-feature through the risk prediction model to obtain a second predicted value; performing risk prediction based on the second sub-feature through the risk prediction model to obtain a third predicted value, and calculating the sum of the second and third predicted values ​​to obtain a fourth predicted value; calculating a loss value to characterize the first predicted value and the sample risk probability to obtain a full prediction loss value; calculating a loss value to characterize the fourth predicted value and the sample risk probability to obtain a component prediction loss value; and performing a weighted summation of the full prediction loss value and the component prediction loss value to obtain a fusion loss value.

[0014] According to an embodiment of this application, risk increment is obtained by using a pre-trained multi-dimensional risk assessment model to predict risks based on the changing indicators or dissemination information of the enterprise to be processed. This includes: aggregating multiple dissemination information or changing indicators of the enterprise to be processed based on a preset time interval to obtain aggregated information; and using the multi-dimensional risk assessment model to predict risks based on the aggregated information to obtain risk increment.

[0015] The second aspect of this application provides an enterprise risk update device, comprising: a dynamic comparison and analysis module, used to respond to the monitoring of dynamic data of a target enterprise, retrieve the target entity corresponding to the target enterprise from a pre-constructed temporal knowledge graph, obtain the temporal attribute data of the target entity, and perform comparative analysis based on the temporal attribute data and dynamic data to obtain change indicators; wherein, the dynamic data indicates changes in business conditions or public opinion, and the temporal attribute data includes attributes corresponding to multiple different times; and an information dissemination module, used to retrieve a relationship graph centered on the target entity from the temporal knowledge graph based on a preset number of hops, and perform information dissemination based on the relationship graph using a preset information dissemination model. The system comprises several modules: a risk propagation module, a risk increment assessment module, and a risk update module. The risk increment module is used to predict the risk based on the changing indicators or propagation information of the enterprise to be processed, using a pre-trained multi-dimensional risk assessment model when an enterprise entity in the association graph matches any enterprise to be processed. The risk update module is used to obtain the current risk probability of the enterprise to be processed and calculate the sum of the risk increment and the current risk probability to obtain the updated risk probability, which is then used to determine the processing result of the business to be processed.

[0016] A third aspect of this application provides an electronic device comprising: one or more processors; and a memory for storing one or more computer programs, wherein the one or more processors execute the one or more computer programs to implement the steps of the method described above.

[0017] A fourth aspect of this application also provides a computer-readable storage medium having a computer program or instructions stored thereon, which, when executed by a processor, implement the steps of the above-described method.

[0018] The fifth aspect of this application also provides a computer program product, including a computer program or instructions that, when executed by a processor, implement the steps of the above-described method. Attached Figure Description

[0019] The above-mentioned contents, other objects, features and advantages of this application will become clearer from the following description of embodiments with reference to the accompanying drawings, in which:

[0020] Figure 1 The illustrations depict application scenarios of the enterprise risk update method, apparatus, device, medium, and program products according to embodiments of this application.

[0021] Figure 2 A flowchart illustrating an enterprise risk update method according to an embodiment of this application is shown schematically;

[0022] Figure 3 A schematic diagram illustrating the association diagram according to an embodiment of this application is shown.

[0023] Figure 4 This illustration illustrates the process of identifying indicator changes by comparing and analyzing current dynamic data and historical data according to an embodiment of this application.

[0024] Figure 5 This illustration schematically depicts a process for iteratively updating the enterprise risk probability according to an embodiment of this application.

[0025] Figure 6 The diagram illustrates the training process of a multi-dimensional risk assessment model according to an embodiment of this application.

[0026] Figure 7 A schematic diagram illustrating the structure of an enterprise risk update apparatus according to an embodiment of this application is shown; and

[0027] Figure 8 A block diagram schematically illustrates an electronic device suitable for implementing a risk update method according to an embodiment of this application. Detailed Implementation

[0028] The embodiments of this application will now be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of this application. In the following detailed description, numerous specific details are set forth to provide a thorough understanding of the embodiments of this application for ease of explanation. However, it will be apparent that one or more embodiments may be implemented without these specific details. Furthermore, descriptions of well-known structures and technologies are omitted in the following description to avoid unnecessarily obscuring the concepts of this application.

[0029] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of this application. The terms “comprising,” “including,” etc., as used herein indicate the presence of the stated features, steps, operations, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, or components.

[0030] All terms used herein (including technical and scientific terms) have the meanings commonly understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein are to be interpreted in a manner consistent with the context of this specification, and not in an idealized or overly rigid way.

[0031] When using expressions such as "at least one of A, B and C", they should generally be interpreted in accordance with the meaning that is commonly understood by those skilled in the art (e.g., "a system having at least one of A, B and C" should include, but is not limited to, a system having A alone, a system having B alone, a system having C alone, a system having A and B, a system having A and C, a system having B and C, and / or a system having A, B and C, etc.).

[0032] In the technical solution of this application, the user information (including but not limited to user personal information, user image information, user device information, such as location information) and data (including but not limited to data used for analysis, stored data, and displayed data) involved are all information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, storage, use, processing, transmission, provision, disclosure, and application of the relevant data all comply with relevant laws, regulations, and standards, and necessary measures have been taken to ensure that they do not violate public order and good morals. Corresponding operation entry points are provided for users to choose to authorize or refuse.

[0033] In scenarios involving automated decision-making using personal information, the methods, devices, and systems provided in this application all offer users corresponding entry points for choosing to agree to or reject the automated decision-making results; if the user chooses to reject, the process proceeds to the expert decision-making stage. Here, "automated decision-making" refers to the activity of automatically analyzing and assessing risks in a decision-making process through computer programs, and then making a decision. Here, "expert decision-making" refers to the activity of making decisions by personnel who specialize in a particular field, possess specialized experience, knowledge, and skills, and have reached a certain level of professional expertise.

[0034] Figure 1 The illustration shows an application scenario diagram of the enterprise risk update method according to an embodiment of this application.

[0035] like Figure 1 As shown, application scenario 100 according to this embodiment may include a first terminal device 101, a second terminal device 102, a third terminal device 103, a network 104, and a server 105. The network 104 serves as a medium for providing communication links between the first terminal device 101, the second terminal device 102, the third terminal device 103, and the server 105. The network 104 may include various connection types, such as wired or wireless communication links, or fiber optic cables, etc.

[0036] Users can use the first terminal device 101, the second terminal device 102, and the third terminal device 103 to interact with the server 105 via the network 104 to receive or send messages, etc. Various communication client applications can be installed on the first terminal device 101, the second terminal device 102, and the third terminal device 103, such as shopping applications, web browser applications, search applications, instant messaging tools, email clients, social media platform software, etc. (for example only).

[0037] The first terminal device 101, the second terminal device 102, and the third terminal device 103 can be various electronic devices with displays and support web browsing, including but not limited to smartphones, tablets, laptops, and desktop computers.

[0038] Server 105 can be a server that provides various services, such as a backend management server that supports websites browsed by users using the first terminal device 101, the second terminal device 102, and the third terminal device 103 (this is just an example). The backend management server can analyze and process data such as received user requests, and feed back the processing results (such as web pages, information, or data obtained or generated according to user requests) to the terminal devices.

[0039] It should be noted that the enterprise risk update method provided in this application embodiment can generally be executed by server 105. Correspondingly, the enterprise risk update device provided in this application embodiment can generally be located in server 105. The enterprise risk update method provided in this application embodiment can also be executed by a server or server cluster that is different from server 105 and capable of communicating with the first terminal device 101, the second terminal device 102, the third terminal device 103, and / or server 105. Correspondingly, the enterprise risk update device provided in this application embodiment can also be located in a server or server cluster that is different from server 105 and capable of communicating with the first terminal device 101, the second terminal device 102, the third terminal device 103, and / or server 105.

[0040] It should be understood that Figure 1 The number of terminal devices, networks, and servers shown is merely illustrative. Depending on implementation needs, any number of terminal devices, networks, and servers can be included.

[0041] The following will be based on Figure 1 The described scene, through Figures 2-6 The enterprise risk update method according to the embodiments of this application will be described in detail.

[0042] Figure 2 A flowchart illustrating an enterprise risk update method according to an embodiment of this application is shown.

[0043] like Figure 2 As shown, the enterprise risk update method in this embodiment includes operations S210 to S240.

[0044] In S210, in response to the detection of dynamic data of the target enterprise, the target entity corresponding to the target enterprise is retrieved from the pre-constructed temporal knowledge graph, and the temporal attribute data of the target entity is obtained. Based on the comparison and analysis of the temporal attribute data and dynamic data, change indicators are obtained. Among them, the dynamic data indicates the change information of business status or public opinion, and the temporal attribute data includes multiple attributes corresponding to different times.

[0045] In S220, based on a preset number of hops, an association graph centered on the target entity is retrieved from the temporal knowledge graph. Information is then propagated based on the association graph using a preset information propagation model to transfer risk information from the central entity to other enterprise entities in the association graph, thereby obtaining the propagation information of other enterprise entities. The association graph includes multiple enterprise entities associated with the central entity.

[0046] In S230, when a corporate entity in the association graph matches any enterprise to be processed, a risk increment is obtained by predicting the risk based on the change indicators or propagation information of the enterprise to be processed through a pre-trained multi-dimensional risk assessment model.

[0047] In S240, the current risk probability of the enterprise to be processed is obtained, and the sum of the risk increment and the current risk probability is calculated to obtain the updated risk probability, so as to determine the processing result of the business to be processed based on the updated risk probability.

[0048] In S210, during the process of a company applying for financial services, the corporate client has submitted the relevant application materials. The financial institution conducts a preliminary risk assessment based on the materials. After passing the preliminary risk assessment, the financial institution still goes through multiple process nodes for approval before finally giving a result of granting or denying the application. Only after giving a result of granting the application will the relevant funds be disbursed to the corporate client. These corporate clients who have submitted application materials but whose relevant funds have not yet been disbursed are the target companies.

[0049] In S210, the dynamic data of the target enterprise being monitored can be at least one of the following: the target enterprise's current overview information, current credit status, current asset pledge information, current operating status, and current negative public opinion.

[0050] In S210, temporal attribute data includes attributes for each time period. For example, at time t1, the attributes are: "Revenue: 10 million; Abnormal public opinion: 5 times; Dissemination volume: 200;", at time t2, the attributes are: "Profit margin: 15%; Financial information: Stock price increase", and at time t3, the attribute is: "Equity change: Shareholder A reduced holdings by 5%". In other words, the temporal attribute data records the attributes of the enterprise entity at each time point.

[0051] In S210, the comparative analysis based on temporal attribute data and dynamic data can be performed by selecting temporal attribute data and dynamic data within a set interval, or by comparing and analyzing the attribute information and dynamic data from the closest previous time.

[0052] In S220, the temporal knowledge graph connects various enterprise entities through entity relationships. For example, enterprise A is associated with enterprise B through relationship 1, enterprise B is associated with enterprise C through relationship 2, and enterprise C is associated with enterprise D through relationship 3. Then, enterprise A is considered to be associated with enterprise D through relationship 1, relationship 2, and relationship 3. Each of these relationships is considered to be one hop. For example, if relationships 1, 2, and 3 are each one hop, then enterprise A and enterprise D are associated through a three-hop relationship.

[0053] For example, with a preset hop count of 2, enterprise entities that are associated with the target enterprise within 2 hops are retrieved from the temporal knowledge graph to obtain an association graph. The association graph includes enterprise entities and entity relationships. If the association graph is represented by a graph structure, then the enterprise entities in the association graph are the nodes of the graph structure, and the entity relationships are the edges of the graph structure.

[0054] Figure 3 A schematic diagram of an association diagram according to an embodiment of this application is shown.

[0055] like Figure 3 As shown, in the two-hop association graph with enterprise O1 as the central entity, the first hop connects enterprise A1 through relation 1, enterprise B1 through relation 2, and enterprise C1 through relation 3. In the second hop, enterprise A1 connects to enterprise A11 through relation 2, enterprise A12 through relation 2, and enterprise A13 through relation 3. Enterprise B1 connects to enterprise B11 through relation 1, enterprise B12 through relation 2, and enterprise B13 through relation 2. Enterprise C1 has no enterprise associated with it in the second hop.

[0056] In S220, the preset information propagation model can be a graph convolution model. The input to the graph convolution model is a graph; therefore, the relationship graph can be input into the graph convolution model in the form of a graph structure. Information propagation occurs through the message passing mechanism of the graph convolution model, thereby obtaining information about the impact of changing indicators on other business entities. This impact information will affect the risk probability of other business entities. For example, if company A is a supplier of company B, and company B is a supplier of company C, when dynamic data indicates that company C's cash flow is disrupted, company B may experience an increased risk probability due to its inability to remit payments. This impact further affects company A, leading to an increased risk probability for company A.

[0057] In S220, the preset information propagation model can also be a neural network model, which takes the change data of the central entity and the entity relationship as input and outputs the propagation information obtained by the enterprise corresponding to the entity relationship.

[0058] In S220, after information is disseminated, the dissemination information of other enterprise entities and the time of monitoring dynamic data can be added to the attributes of the corresponding enterprise entities in the temporal knowledge graph.

[0059] In S230, after the information in the association diagram is propagated, the information of enterprises A, B, C, and D is updated. The central entity updates the change indicators, and the other related enterprise entities update the propagated information. Currently, the enterprises that have submitted application materials but have not yet received the relevant business payments include A, C, F, G, and H. Since the information of enterprises A and C has been updated, the risk increment is predicted based on the updated information, and the risk increment of enterprises A and C caused by the dynamic data is obtained.

[0060] In S240, the risk probabilities of companies A and C can be the initial risk probabilities assessed by A and C when submitting their business application materials, or they can be risk probabilities that have been updated at a previous time. For example, if company A's initial risk probability is 0.5, and dynamic data of company D is detected at time t1, company A passively receives the information disseminated at time t1, with a corresponding risk increment probability of 0.05, then company A's risk probability is updated to 0.55 at time t1. If dynamic data of company C is detected at time t2, company A passively receives the information disseminated at time t1, with a corresponding risk increment probability of 0.08, then company A's risk probability is updated to 0.63 at time t1.

[0061] In S240, risk interpretation information can also be generated based on changing indicators or dissemination information to explain risk increments. For example, if the risk increment is 0.06, the corresponding risk interpretation information is: the predicted future profit margin is 5%, which is a 5% decrease compared to the present.

[0062] In S240, determining the processing result of the pending business based on the updated risk probability may include: for enterprises whose updated risk probability is greater than the first risk threshold (e.g., 0.8), a decision to not process the application and not issue funds is given; for enterprises whose updated risk probability is in the medium-high risk range (e.g., 0.5~0.8), the approver makes a decision by combining the approval materials submitted by the enterprise and the risk interpretation information of previous risk probability updates.

[0063] For example, for specific types of financial transactions, differentiated approval strategies and control measures can be implemented based on updated risk probabilities. For instance, for identified medium- to high-risk customers / enterprises, stricter guarantee and credit enhancement requirements can be automatically triggered, more prudent credit limits / terms can be set (e.g., increasing or decreasing limits), the frequency of inspections after the disbursement of financial funds can be increased (e.g., frequency limits or increases), or clear admission / removal trigger conditions can be set (e.g., customer admission / removal standards). For low-risk customers, policies for different financial products can be automatically approved.

[0064] According to the embodiments of this application, for financial transactions involving multiple approval nodes or those with delayed approval, dynamic information can be captured in real time from the submission of application materials to the disbursement of relevant funds. Based on the dynamic information, changes in relevant information of the target enterprise and its related enterprises can be identified, effectively capturing changes in risk and updating risk probabilities. This allows financial institutions to make decisions based on the latest risk probabilities when making final decisions, thus ensuring the safety of the financial institution's assets.

[0065] According to an embodiment of this application, a comparative analysis is performed based on temporal attribute data and dynamic data to obtain the following change indicators: when the dynamic data indicates the operating status, the operating indicators in the dynamic data are extracted, the historical average of the corresponding operating indicators in the temporal attribute data is calculated, and the difference between the operating indicators and the historical average is calculated to obtain the absolute fluctuation range; the ratio of the absolute fluctuation range to the historical average is calculated to obtain the relative fluctuation range, and the relative fluctuation range is used as the change indicator. The operating indicators include at least one of the following: profit margin, operating revenue growth rate, net profit growth rate, and asset-liability ratio.

[0066] For example, based on a preset time period, such as the past year, the profit margin of the past year is extracted from the temporal attributes, and the average value is calculated, such as the monthly average, quarterly average, semi-annual average, or annual average. The current profit margin is extracted from the newly acquired dynamic data, and the relative fluctuation range between the current profit margin and the profit margin of the past year is calculated to obtain the fluctuation range of the profit margin.

[0067] For example, the absolute fluctuation range can be expressed as:

[0068]

[0069] Where Abs represents the absolute fluctuation range of the operating indicator. This refers to the operational indicators extracted from dynamic data. This represents the historical average of operating indicators.

[0070] For example, operating indicators can be at least one of the following: operating revenue, net profit, gross profit margin, net profit margin, return on equity, debt-to-equity ratio, current ratio, quick ratio, inventory turnover, and accounts receivable turnover. The debt-to-equity ratio (total liabilities / total assets × 100%) reflects the company's financial leverage and long-term solvency risk. The current ratio (current assets / current liabilities) measures short-term solvency; the quick ratio (current assets - inventory) / current liabilities measures short-term solvency excluding inventory; inventory turnover (cost of goods sold / average inventory) measures inventory turnover speed, reflecting supply chain efficiency; accounts receivable turnover (operating revenue / average accounts receivable) measures collection speed, reflecting customer credit management level; and total asset turnover (operating revenue / average total assets) measures asset utilization efficiency.

[0071] According to the embodiments of this application, by monitoring changes in the operating conditions of a target enterprise, change indicators reflecting changes in operating conditions are calculated. These change indicators represent the enterprise's solvency, liquidity level, and development sustainability. By dynamically capturing changes in the enterprise's solvency and other aspects, the risk changes of the enterprise can be effectively identified, which is beneficial for obtaining risk changes affected by dynamic data changes in real time.

[0072] According to an embodiment of this application, the method of obtaining change indicators by comparing and analyzing temporal attribute data and dynamic data further includes: extracting operating indicators from dynamic data when dynamic data indicates operating conditions, extracting historical indicators corresponding to operating indicators from temporal attribute data, and sorting historical indicators and operating indicators by time to obtain a time-series indicator sequence; performing trend prediction based on the time-series indicator sequence through a temporal convolutional network to obtain indicator trends, and using indicator trends as change indicators.

[0073] For example, the indicator trend can be a forecasted operating indicator, such as a forecasted profit margin or a forecasted turnover.

[0074] For example, the trend of the indicator can be the predicted direction and magnitude of change in the operating indicator, such as the percentage decrease in profit margin.

[0075] For example, since the temporal knowledge graph stores time-series data of operating indicators, such as revenue of R1 and profit margin of D1 at time t1, revenue of R2 and profit margin of D2 at time t2, and revenue of R3 and profit margin of D3 at time t3, the time-series attribute data corresponding to a certain operating indicator are sorted according to time order to obtain the time-series indicator sequence of the operating indicator. The time-series indicator sequence is then input into a temporal convolutional network for trend prediction to obtain the predicted future profit margin.

[0076] For example, historical indicators include multiple dimensions, so that when making predictions, the temporal convolutional network can combine multiple indicators such as operating revenue, net profit, gross profit margin, net profit margin, return on net assets, debt-to-equity ratio, current ratio, quick ratio, inventory turnover, and accounts receivable turnover to obtain accurate indicator trends.

[0077] For example, since temporal convolutional networks are trained on a large amount of sample data, they can make trend predictions more accurately. Furthermore, large models can also be used for trend prediction.

[0078] According to the embodiments of this application, for relevant financial businesses, the expectation of financial institutions issuing funds is that the enterprise will be able to repay the funds on schedule in the future. By predicting future trends based on time series indicators, a metric that can reflect whether the enterprise will be able to make the corresponding repayment in the future can be obtained. The predicted indicator trend can better reflect the repayment risk corresponding to the financial business compared with the current operating indicators, which is more conducive to financial institutions identifying the enterprise's risks and thus assisting financial institutions in making accurate decisions.

[0079] According to the embodiments of this application, the change indicators obtained by comparing and analyzing temporal attribute data and dynamic data include: when dynamic data indicates public opinion information, the dynamic data is used to identify the sentiment tendency through a preset public opinion analysis model to obtain a sentiment tendency score; based on preset weights, the number of clicks and the number of reposts in the dynamic data are weighted and summed to obtain the dissemination volume, and the dissemination volume is standardized to obtain the dissemination popularity; the sentiment tendency score and the dissemination popularity are used as change indicators.

[0080] For example, public opinion information about a company may also cause changes in the company's risk factors. For instance, if a company encounters public opinion risks, its market share may decline, leading to a decrease in future solvency and an increase in risk. In response, by monitoring public opinion information related to the target company on the Internet, such as news, a pre-set public opinion analysis model can be used to identify the sentiment tendency of dynamic data and obtain a sentiment tendency score. This sentiment tendency score reflects the market's preference for the company, which may potentially affect the company's operating performance in the future. Furthermore, considering the dissemination volume of public opinion information, the number of clicks and reposts of the public opinion information is obtained to calculate the dissemination heat. The sentiment tendency score and dissemination heat are used as indicators of change.

[0081] For example, the product of sentiment tendency score and dissemination popularity is further calculated to obtain the sentiment influence factor, which is used as a change indicator.

[0082] Figure 4 This illustration illustrates the process of identifying indicator changes by combining current dynamic data and historical data according to an embodiment of this application.

[0083] like Figure 4 As shown, when conducting comparative analysis, the first step is to obtain current dynamic data, which can be the business status of customers or enterprises, or public opinion about customers. The current dynamic data is then compared and analyzed with historical data to obtain change indicators. These change indicators can indicate situations such as a sharp drop in revenue, a sudden increase in debt, or a surge in public opinion.

[0084] According to the embodiments of this application, by monitoring and capturing public opinion risk information that affects the timely repayment of funds by enterprises, obtaining change indicators based on public opinion risk information, and disseminating information among related enterprises, the changes in the future repayment ability of enterprises caused by external information such as public opinion information are incorporated when identifying risk increments, which is conducive to more accurately identifying and updating the risk probability of enterprises.

[0085] According to an embodiment of this application, pre-constructing a temporal knowledge graph includes: using a pre-trained entity extraction model to extract entities from collected multidimensional data to obtain an entity set, wherein the multidimensional data includes at least one of enterprise business registration information, enterprise financial statements, enterprise public opinion information, and enterprise business application materials; for any entity in the entity set, obtaining an entity dataset related to the entity in the multidimensional data, and dividing the entity dataset into multiple temporal subsets according to time information, and extracting features from the temporal subsets to obtain entity features; storing the entity features and corresponding time information in association as temporal attribute data of the entity; using a pre-trained relation recognition model to identify the relationship types between entities to obtain an entity relation triplet set, and constructing a knowledge graph based on the triplet set to obtain a temporal knowledge graph, wherein the relation types include at least one of supply chain association, guarantee association, or equity association.

[0086] For example, collecting multidimensional data includes: preparing source data tables, indicator description documents, and report description documents. The source data tables primarily provide data on abnormal fluctuations, such as basic information on historical customer balances, credit status, asset collateral information, operating conditions, and negative public opinion information. This information includes corresponding time information. The indicator description documents clearly define fields and, based on actual business needs, provide information including indicator names, indicator processing logic (e.g., explanations of profit margin calculation methods), indicator definitions, and a thesaurus of related synonyms and professional terminology, including term names and explanations. For example, revenue difference could be the difference between the second half and the first half of the year; negative customer information includes news reports and related public opinion data. The report description documents are primarily knowledge documents related to the risk assessment report expected by the user, including report titles (e.g., please generate a credit risk report for customer xx), and report content descriptions including indicators related to business changes and investment risks.

[0087] For example, a temporal knowledge graph is constructed. A pre-trained entity extraction model is used to extract entities from the collected multidimensional data, resulting in enterprise entities. For the temporal attribute data corresponding to these enterprise entities, an indicator knowledge base and a configured vector index are first built. The indicator base mainly includes indicator settings, synonym configurations, table metadata definitions, and professional terminology settings. Based on the documents provided by the business, indicator definitions and question-answering domains are constructed. When professional terms cannot be recognized, synonym aliases are configured to improve performance. The constructed indicators are used as attribute data for the enterprise entities. These attribute data correspond to time information.

[0088] For example, the relationship types between entities are identified to obtain a set of entity relationship triples, and a temporal knowledge graph is constructed based on the set of triples.

[0089] According to embodiments of this application, the temporal knowledge graph stores the temporal attribute data of enterprises, thereby facilitating the comparison of the enterprise's temporal attribute data with dynamic data to obtain change indicators. The temporal knowledge graph stores the relationships between enterprises, thereby enabling information dissemination based on these relationships. When the information of a certain enterprise changes, its corresponding risk probability is updated, and the change information is disseminated to related enterprises to prompt them to update their risk as well. Thus, not only is the risk probability of the target enterprise updated, but also the risk probabilities of related enterprises are updated simultaneously. This helps financial institutions accurately identify the risk probabilities of each enterprise and prevent systemic financial risks.

[0090] According to an embodiment of this application, information is propagated based on an association graph using a preset information propagation model to transmit risk information in changing indicators from a central entity to other enterprise entities in the association graph, thereby obtaining propagation information of other enterprise entities. This includes: determining the first edge weight corresponding to the first relationship type in the preset information propagation model based on the first relationship type between the central entity and the enterprise entity associated with the first hop in the association graph; propagating information about changing indicators based on the first edge weight to obtain propagation information of the first-hop enterprise entity; at any hop after the first hop, obtaining the second relationship type between the current enterprise entity and the enterprise entity associated with the next hop, determining the second edge weight corresponding to the second relationship type in the information propagation model; and propagating the propagation information of the current enterprise entity based on the second edge weight to obtain propagation information of the next-hop enterprise entity.

[0091] For example, the preset information dissemination model can be a neural network model, which takes the change data of the central entity and the entity relationship as input, and outputs the dissemination information obtained by the enterprise corresponding to the entity relationship.

[0092] like Figure 3 As shown, the information of the changing indicators in enterprise O1 is propagated to other enterprise nodes. In the first hop, the weight of the first side corresponding to relation 1 in the information propagation model is determined. Based on the weight of the first side, the information of the changing indicators is propagated to enterprise A1, and the propagation information of A1 is determined. The weight of the second side corresponding to relation 2 is determined. Based on the weight of the second side, the information of the changing indicators is propagated to enterprise B1, and the propagation information of B1 is determined. Similarly, the weight of the third side corresponding to relation 3 is determined, and the information of the changing indicators is propagated to enterprise C1.

[0093] For example, in the second hop, such as the second hop corresponding to enterprise A1, the propagation information of enterprise A1 is propagated based on the second side weight to obtain the propagation information of enterprise A11 in the next hop.

[0094] According to the embodiments of this application, by propagating information in the changing indicators from the central enterprise entity to other enterprise entities in the association graph based on the edge weight pairs corresponding to the relationship type, risk identification is no longer conducted in isolation based solely on the information of the target enterprise itself, but rather by combining the relevant information of the related enterprises. Since the enterprise's risk is affected by its related enterprises, risk identification based on the influence relationship between related enterprises helps financial institutions effectively capture the potential risks of enterprises.

[0095] According to an embodiment of this application, propagating the propagation information of the current enterprise entity based on the second edge weight to obtain the propagation information of the next-hop enterprise entity includes: performing a linear transformation on the propagation information of the current enterprise entity based on the edge weight to obtain transformed information; determining an attenuation factor based on the distance between the next-hop enterprise entity and the central entity; and obtaining the propagation information of the next-hop enterprise entity based on the product of the attenuation factor and the transformed information.

[0096] For example, considering the attenuation characteristics of information propagation, that is, the risk impact of the monitored dynamic data on the target enterprise is direct, while the risk impact on related enterprises is indirect, and the farther away from the central enterprise entity, the smaller the impact, the attenuation factor is determined based on the distance between the next-hop enterprise entity and the central entity. The propagation information of the next-hop enterprise entity is obtained by multiplying the attenuation factor and the transformation information. For example, if the distance between the first-hop related entity and the central entity is 1 hop, the attenuation factor can be 0.8; if the distance between the second-hop related entity and the central entity is 2 hops, the attenuation factor can be 0.6.

[0097] According to the embodiments of this application, when disseminating information, the attenuation of information is taken into consideration. For enterprises that are farther away from the target enterprise, a larger attenuation factor is set, thereby obtaining dissemination information that is more in line with the characteristics of information dissemination, and the risk increment identification based on the dissemination information is more accurate.

[0098] According to an embodiment of this application, obtaining the current risk probability of the enterprise to be processed includes: in response to receiving a risk question about the enterprise to be processed, searching a preset temporal knowledge graph based on the risk question to obtain search results, the search results including nodes and / or edges related to the risk question in the temporal knowledge graph and the association information of the nodes and / or edges; extracting features based on the search results to obtain multi-dimensional risk features, and performing risk prediction based on the multi-dimensional risk features through a multi-dimensional risk assessment model to obtain the initial risk probability of the enterprise to be processed, and using the initial risk probability as the current risk probability.

[0099] For example, when a company submits its initial application materials, an initial risk assessment is conducted. The reviewers at the initial process node raise questions related to the company's risks. These questions are then segmented and quantified to obtain vectorized questions. A temporal knowledge graph is retrieved based on these vectorized questions to obtain search results. Feature extraction is then performed on the search results to obtain multidimensional risk features. Based on these multidimensional risk features, a multidimensional risk assessment model is used to predict the risk and obtain the initial risk probability of the company to be processed. This initial risk probability is then used as the current risk probability.

[0100] For example, the multidimensional risk assessment model can be a large model. After integrating the search results, the large model provides an initial risk probability. Furthermore, a risk assessment report corresponding to the initial risk probability is generated according to a preset assessment report template. The risk assessment report contains a structured list of various attribute data from the search results.

[0101] For example, a large-scale model is generated using query statements to analyze the questions raised by approval personnel, and a target risk assessment report is generated using an intelligent data insight model. The query statement generation model primarily generates query statements, translating user questions into language recognizable by the model. To improve the learning quality of the analysis model, existing user risk Q&A templates are provided for fine-tuning and training, requiring over 1000 sample data points for model fine-tuning. Continuous optimization is performed to enhance semantic understanding capabilities specific to business scenarios, resulting in the intelligent data insight model. The large-scale model automatically recommends data charts based on the query results, vectorizing the query results. Then, the intelligent data insight model performs multi-dimensional analysis to output a credit approval risk assessment. The intelligent data insight model undergoes multiple fine-tuning training sessions using query results and credit domain knowledge. To improve the learning quality of the analysis model, at least 10 high-quality credit approval risk reports from users are required, with the number potentially increased depending on the complexity of the scenario. Through multi-dimensional analysis and continuous iteration, the target risk assessment report is output.

[0102] For example, the initial risk probability is stored in a risk probability database so that it can be quickly retrieved when the risk probability is obtained.

[0103] For example, after an enterprise updates the risk probability by identifying risk increments based on changing indicators or disseminated information, it stores the latest risk probability in the risk probability database in association with the update time, so that it can be quickly retrieved when the risk probability is obtained.

[0104] For example, when updating the risk probability, the risk interpretation information can also be output as a risk assessment report.

[0105] Figure 5The illustration shows a schematic diagram of the process of iteratively updating the enterprise risk probability according to an embodiment of this application.

[0106] like Figure 5 As shown, based on the static materials of corporate clients, including client overview, credit profile, asset collateral and pledge information, etc., and dynamic data, including corporate operating data (such as debt repayment, operation, cash flow), performance records, industry analysis, etc., a historical-current comparative analysis is conducted to obtain change indicators. After dissemination, a multi-dimensional risk assessment model is used to obtain the risk increment based on the change indicators. This increment is summed with the current risk probability and updated using a large model based on risk interpretation information. A risk assessment report is then generated. The risk probability and risk assessment report are then saved and fed back so that they can be read and updated further when new dynamic data is received.

[0107] According to an embodiment of this application, risk prediction is performed based on the retrieval results of the temporal knowledge graph to obtain an initial risk probability. Since the retrieval results include multi-dimensional data collected when the enterprise customer submits the application materials, as well as relevant data of related enterprises, the accurate risk probability before the update can be obtained by multi-dimensional analysis when performing risk prediction based on this information. Thus, the accurate updated risk probability can be obtained by summing and calculating.

[0108] According to an embodiment of this application, pre-training a multi-dimensional risk assessment model includes: repeatedly training a preset risk prediction network based on pre-acquired multi-dimensional risk features of samples and corresponding sample risk probabilities until the fusion loss value of the risk prediction network reaches a preset loss threshold, then using the optimal risk prediction network as the multi-dimensional risk assessment model; the method for obtaining the fusion loss value includes: performing risk prediction based on the multi-dimensional risk features of samples through the risk prediction network to obtain a first predicted value; randomly grouping the multi-dimensional risk features of samples to obtain a first sub-feature and a second sub-feature, and performing risk prediction based on the first sub-feature through the risk prediction model to obtain a second predicted value; performing risk prediction based on the second sub-feature through the risk prediction model to obtain a third predicted value, and calculating the sum of the second and third predicted values ​​to obtain a fourth predicted value; calculating a loss value to characterize the first predicted value and the sample risk probability to obtain a full prediction loss value; calculating a loss value to characterize the fourth predicted value and the sample risk probability to obtain a component prediction loss value; and performing a weighted summation of the full prediction loss value and the component prediction loss value to obtain a fusion loss value.

[0109] Figure 6 The training method for a multidimensional risk assessment model is illustrated.

[0110] like Figure 6As shown, firstly, all sample risk features are input into the risk prediction network to obtain the corresponding first predicted value. Then, all sample risk features are divided into two parts: the first sub-feature and the second sub-feature, which are then input into the risk prediction network to obtain the corresponding second and third predicted values. The sum of the two is calculated to obtain the fourth predicted value. The loss value between the first predicted value and the labeled sample risk probability is calculated to obtain the total loss. The loss value between the first predicted value and the labeled sample risk probability is calculated to obtain the component loss. The total loss and the component loss are weighted and summed to obtain the fusion loss value. The fusion loss value is used for iterative training until the fusion loss value reaches the preset loss threshold. The optimal risk prediction network is then used as the multi-dimensional risk assessment model.

[0111] For example, the weight of the full prediction loss value is 0.7, and the weight of the component prediction loss value is 0.3.

[0112] For example, dividing the risk characteristics of the entire sample into two parts could be as follows: "Profit margin: 3%; turnover: 1 million" could be divided into "Profit margin 3%" and "Turnover: 1 million"; or it could be divided according to the data ratio into "Profit margin 2%, turnover 660,000" and "Profit margin 1%, turnover 330,000". When splitting according to the data ratio, a large model can be used to scientifically split the data. For example, a large model can be used to split the data according to the correspondence between turnover and profit margin.

[0113] According to embodiments of this application, by iteratively training the model, the risk prediction model can more accurately capture the relationship between changes in risk information and risk increments, thereby avoiding prediction distortion caused by missing data during incremental prediction and improving the accuracy of incremental prediction.

[0114] According to an embodiment of this application, risk increment is obtained by using a pre-trained multi-dimensional risk assessment model to predict risks based on the changing indicators or dissemination information of the enterprise to be processed. This includes: aggregating multiple dissemination information or changing indicators of the enterprise to be processed based on a preset time interval to obtain aggregated information; and using the multi-dimensional risk assessment model to predict risks based on the aggregated information to obtain risk increment.

[0115] Since the amount of dynamic data obtained from each monitoring is relatively small, information can be aggregated based on a preset time interval, such as every 8 hours, for multiple communication information or change indicators of the enterprise to be processed. After obtaining the aggregated information, the incremental information volume is used to predict the risk of the aggregated information through a multi-dimensional risk assessment model to obtain the risk increment.

[0116] According to the embodiments of this application, considering that the amount of change indicators or data obtained after a single detection is small and frequent updates waste computing resources, a time interval is set according to the needs of the scenario. The change indicators or propagation information within the time interval are periodically aggregated before risk prediction is performed. This can combine propagation information or change indicators from multiple moments to perform incremental prediction. On the one hand, it saves computing resources, and on the other hand, the amount of aggregated information is greater, making incremental prediction more accurate.

[0117] Based on the above-mentioned enterprise risk updating method, this application also provides an enterprise risk updating device. The following will be combined with... Figure 7 The device is described in detail.

[0118] Figure 7 A schematic block diagram of an enterprise risk update apparatus according to an embodiment of this application is shown.

[0119] like Figure 7 As shown, the enterprise risk update device 700 of this embodiment includes a dynamic comparison analysis module 710, an information dissemination module 720, a risk increment assessment module 730, and a risk update module 740.

[0120] The dynamic comparison analysis module 710 is used to respond to the monitoring of dynamic data of the target enterprise, retrieve the target entity corresponding to the target enterprise from a pre-constructed temporal knowledge graph, obtain the temporal attribute data of the target entity, and perform comparative analysis based on the temporal attribute data and dynamic data to obtain change indicators; wherein, the dynamic data indicates the change information of business status or public opinion, and the temporal attribute data includes attributes corresponding to multiple different times. In one embodiment, the dynamic comparison analysis module 710 can be used to perform the operation S210 described above, which will not be repeated here.

[0121] The information dissemination module 720 is used to retrieve a relationship graph centered on the target entity from the temporal knowledge graph based on a preset number of hops. Information is then disseminated based on this relationship graph using a preset information dissemination model to transfer risk information from the central entity to other enterprise entities in the relationship graph, thereby obtaining the dissemination information from these other enterprise entities. The relationship graph includes multiple enterprise entities associated with the central entity. In one embodiment, the information dissemination module 720 can be used to execute the operation S220 described above, which will not be repeated here.

[0122] The risk increment assessment module 730 is used to predict the risk increment based on the changing indicators or propagation information of the enterprise to be processed when an enterprise entity in the association graph matches any enterprise to be processed. This prediction is achieved using a pre-trained multi-dimensional risk assessment model. In one embodiment, the risk increment assessment module 730 can be used to perform the operation S230 described above, which will not be repeated here.

[0123] The risk update module 740 is used to obtain the current risk probability of the enterprise to be processed, and calculate the sum of the risk increment and the current risk probability to obtain the updated risk probability, so as to determine the processing result of the business to be processed based on the updated risk probability. In one embodiment, the risk update module 740 can be used to perform the operation S240 described above, which will not be repeated here.

[0124] According to an embodiment of this application, the dynamic comparison analysis module 710 is further configured to: extract operating indicators from the dynamic data when the dynamic data indicates the operating status, calculate the historical average of the corresponding operating indicators in the temporal attribute data, and calculate the difference between the operating indicators and the historical average to obtain the absolute fluctuation range; calculate the ratio of the absolute fluctuation range to the historical average to obtain the relative fluctuation range, and use the relative fluctuation range as a change indicator. The operating indicators include at least one of the following: profit margin, operating revenue growth rate, net profit growth rate, and asset-liability ratio.

[0125] According to an embodiment of this application, the dynamic comparison analysis module 710 is further configured to: extract operating indicators from the dynamic data when the dynamic data indicates the operating status, extract historical indicators corresponding to the operating indicators from the temporal attribute data, sort the historical indicators and operating indicators according to time to obtain a time-series indicator sequence; perform trend prediction based on the time-series indicator sequence through a temporal convolutional network to obtain the indicator trend, and use the indicator trend as a change indicator.

[0126] According to an embodiment of this application, the dynamic comparison analysis module 710 is further configured to: when dynamic data indicates public opinion information, identify the sentiment tendency of the dynamic data through a preset public opinion analysis model to obtain a sentiment tendency score; based on preset weights, perform a weighted summation of the number of clicks and the number of reposts in the dynamic data to obtain the dissemination volume, and standardize the dissemination volume to obtain the dissemination popularity; and use the sentiment tendency score and dissemination popularity as change indicators.

[0127] According to an embodiment of this application, the device 700 further includes a knowledge graph construction module 701, configured to: extract entities from the collected multidimensional data using a pre-trained entity extraction model to obtain an entity set, wherein the multidimensional data includes at least one of enterprise business registration information, enterprise financial statements, enterprise public opinion information, and enterprise business application materials; for any entity in the entity set, obtain the entity dataset related to the entity in the multidimensional data, and divide the entity dataset into multiple temporal subsets according to time information, extract features from the temporal subsets respectively to obtain entity features; store the entity features and corresponding time information in association as temporal attribute data of the entity; identify the relationship type between entities using a pre-trained relationship recognition model to obtain an entity relationship triplet set, and construct a knowledge graph based on the triplet set to obtain a temporal knowledge graph, wherein the relationship type includes at least one of supply chain association, guarantee association, or equity association.

[0128] According to an embodiment of this application, the information propagation module 720 is further configured to: determine the first edge weight corresponding to the first relationship type in the preset information propagation model based on the first relationship type between the central entity in the association graph and the enterprise entity associated with the first hop; propagate information about the changing indicators based on the first edge weight to obtain the propagation information of the first hop enterprise entity; in any hop after the first hop, obtain the second relationship type between the current enterprise entity and the enterprise entity associated with the next hop, determine the second edge weight corresponding to the second relationship type in the information propagation model; and propagate information about the current enterprise entity based on the second edge weight to obtain the propagation information of the next hop enterprise entity.

[0129] According to an embodiment of this application, the information propagation module 720 is further configured to: perform a linear transformation on the propagation information of the current enterprise entity based on the edge weight to obtain transformed information; determine an attenuation factor based on the distance between the next-hop enterprise entity and the central entity; and obtain the propagation information of the next-hop enterprise entity based on the product of the attenuation factor and the transformed information.

[0130] According to an embodiment of this application, the device 700 further includes a current risk acquisition module 702, configured to: in response to receiving a risk question from a company to be processed, perform a search in a preset temporal knowledge graph based on the risk question to obtain search results, the search results including nodes and / or edges related to the risk question in the temporal knowledge graph and the association information of the nodes and / or edges; perform feature extraction based on the search results to obtain multi-dimensional risk features, perform risk prediction based on the multi-dimensional risk features through a multi-dimensional risk assessment model to obtain the initial risk probability of the company to be processed, and use the initial risk probability as the current risk probability.

[0131] According to an embodiment of this application, the device 700 further includes a multi-dimensional risk assessment model training module 703, used for: repeatedly training a preset risk prediction network based on pre-acquired multi-dimensional risk features of samples and corresponding sample risk probabilities until the fusion loss value of the risk prediction network reaches a preset loss threshold, then using the optimal risk prediction network as the multi-dimensional risk assessment model; the method for obtaining the fusion loss value includes: performing risk prediction based on the multi-dimensional risk features of samples through the risk prediction network to obtain a first predicted value; randomly grouping the multi-dimensional risk features of samples to obtain a first sub-feature and a second sub-feature, and performing risk prediction based on the first sub-feature through the risk prediction model to obtain a second predicted value; performing risk prediction based on the second sub-feature through the risk prediction model to obtain a third predicted value, and calculating the sum of the second and third predicted values ​​to obtain a fourth predicted value; calculating a loss value to characterize the first predicted value and the sample risk probability to obtain a full prediction loss value; calculating a loss value to characterize the fourth predicted value and the sample risk probability to obtain a component prediction loss value; and performing a weighted summation of the full prediction loss value and the component prediction loss value to obtain a fusion loss value.

[0132] According to an embodiment of this application, the risk increment assessment module 730 is further configured to: aggregate multiple dissemination information or change indicators of the enterprise to be processed based on a preset time interval to obtain aggregated information; and predict the risk of the aggregated information through a multi-dimensional risk assessment model to obtain the risk increment.

[0133] According to embodiments of this application, any multiple modules among the dynamic comparison analysis module 710, information dissemination module 720, risk increment assessment module 730, and risk update module 740 can be merged into one module, or any one of these modules can be split into multiple modules. Alternatively, at least some of the functions of one or more of these modules can be combined with at least some of the functions of other modules and implemented in one module. According to embodiments of this application, at least one of the dynamic comparison analysis module 710, information dissemination module 720, risk increment assessment module 730, and risk update module 740 can be at least partially implemented as hardware circuitry, such as a field-programmable gate array (FPGA), a programmable logic array (PLA), a system-on-a-chip, a system-on-a-substrate, a system-on-package, an application-specific integrated circuit (ASIC), or any other reasonable means of integrating or packaging circuitry, or implemented in software, hardware, or firmware, or in any suitable combination of any of these three implementation methods. Alternatively, at least one of the dynamic comparison analysis module 710, information dissemination module 720, risk increment assessment module 730, and risk update module 740 can be at least partially implemented as a computer program module, which can perform corresponding functions when the computer program module is run.

[0134] Figure 8 A block diagram schematically illustrates an electronic device suitable for implementing an enterprise risk update method according to an embodiment of this application.

[0135] like Figure 8 As shown, an electronic device 800 according to an embodiment of this application includes a processor 801, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 802 or a program loaded from a storage portion 808 into a random access memory (RAM) 803. The processor 801 may include, for example, a general-purpose microprocessor (e.g., a CPU), an instruction set processor and / or an associated chipset and / or a special-purpose microprocessor (e.g., an application-specific integrated circuit (ASIC)), etc. The processor 801 may also include onboard memory for caching purposes. The processor 801 may include a single processing unit or multiple processing units for performing different actions of the method flow according to an embodiment of this application.

[0136] RAM 803 stores various programs and data required for the operation of electronic device 800. Processor 801, ROM 802, and RAM 803 are interconnected via bus 804. Processor 801 executes various operations of the method flow according to embodiments of this application by executing programs in ROM 802 and / or RAM 803. It should be noted that the programs may also be stored in one or more memories other than ROM 802 and RAM 803. Processor 801 may also execute various operations of the method flow according to embodiments of this application by executing programs stored in said one or more memories.

[0137] According to embodiments of this application, the electronic device 800 may further include an input / output (I / O) interface 805, which is also connected to a bus 804. The electronic device 800 may also include one or more of the following components connected to the input / output (I / O) interface 805: an input section 806 including a keyboard, mouse, etc.; an output section 807 including a cathode ray tube (CRT), liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 808 including a hard disk, etc.; and a communication section 809 including a network interface card such as a LAN card, modem, etc. The communication section 809 performs communication processing via a network such as the Internet. A drive 810 is also connected to the input / output (I / O) interface 805 as needed. A removable medium 811, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed on the drive 810 as needed so that computer programs read from it can be installed into the storage section 808 as needed.

[0138] This application also provides a computer-readable storage medium, which may be included in the device / apparatus / system described in the above embodiments; or it may exist independently and not assembled into the device / apparatus / system. The computer-readable storage medium carries one or more programs, which, when executed, implement the method according to the embodiments of this application.

[0139] According to embodiments of this application, the computer-readable storage medium can be a non-volatile computer-readable storage medium, such as including 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 thereof. In this application, the computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. For example, according to embodiments of this application, the computer-readable storage medium may include ROM 802 and / or RAM 803 and / or one or more memories other than ROM 802 and RAM 803 described above.

[0140] Embodiments of this application also include a computer program product comprising a computer program containing program code for performing the methods shown in the flowchart. When the computer program product is run on a computer system, the program code is used to enable the computer system to implement the enterprise risk update method provided in the embodiments of this application.

[0141] When the computer program is executed by the processor 801, it performs the functions defined in the system / apparatus of this application embodiment. According to the embodiments of this application, the systems, apparatuses, modules, units, etc., described above can be implemented by computer program modules.

[0142] In one embodiment, the computer program may rely on a tangible storage medium such as an optical storage device or a magnetic storage device. In another embodiment, the computer program may also be transmitted and distributed in the form of signals over a network medium, and may be downloaded and installed via the communication section 809, and / or installed from a removable medium 811. The program code contained in the computer program can be transmitted using any suitable network medium, including but not limited to: wireless, wired, etc., or any suitable combination thereof.

[0143] In such an embodiment, the computer program can be downloaded and installed from a network via the communication section 809, and / or installed from the removable medium 811. When the computer program is executed by the processor 801, it performs the functions defined in the system of this application embodiment. According to the embodiments of this application, the systems, devices, apparatuses, modules, units, etc., described above can be implemented by computer program modules.

[0144] According to embodiments of this application, program code for executing the computer programs provided in the embodiments of this application can be written in any combination of one or more programming languages. Specifically, these computational programs can be implemented using high-level procedural and / or object-oriented programming languages, and / or assembly / machine languages. Programming languages ​​include, but are not limited to, languages ​​such as Java, C++, Python, "C", or similar programming languages. The program code can be executed entirely on the user's computing device, partially on the user's device, partially on a remote computing device, or entirely on a remote computing device or server. In cases involving remote computing devices, the remote computing device can be connected to the user's computing device via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computing device (e.g., via the Internet using an Internet service provider).

[0145] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram or flowchart, and combinations of blocks in a block diagram or flowchart, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0146] Those skilled in the art will understand that the features described in the various embodiments of this application can be combined and / or combined in various ways, even if such combinations or combinations are not explicitly described in this application. In particular, the features described in the various embodiments of this application can be combined and / or combined in various ways without departing from the spirit and teachings of this application. All such combinations and / or combinations fall within the scope of this application.

Claims

1. A method for updating enterprise risk, characterized in that, The method includes: In response to the detection of dynamic data of the target enterprise, the target entity corresponding to the target enterprise is retrieved from the pre-constructed temporal knowledge graph, and the temporal attribute data of the target entity is obtained. Based on the temporal attribute data and the dynamic data, a comparative analysis is performed to obtain change indicators. The dynamic data indicates the change information of business status or public opinion, and the temporal attribute data includes multiple attributes corresponding to different times. Based on a preset number of hops, an association graph centered on the target entity is retrieved from the temporal knowledge graph. Information is then propagated based on the association graph using a preset information propagation model to transfer the risk information in the change index from the central entity to other enterprise entities in the association graph, thereby obtaining the propagation information of other enterprise entities. The association graph includes multiple enterprise entities associated with the central entity. When a business entity in the association graph matches any business to be processed, a risk increment is obtained by predicting the risk based on the change indicators or propagation information of the business to be processed using a pre-trained multi-dimensional risk assessment model. Obtain the current risk probability of the enterprise to be processed, and calculate the sum of the risk increment and the current risk probability to obtain the updated risk probability, so as to determine the processing result of the business to be processed based on the updated risk probability.

2. The method according to claim 1, characterized in that, The comparative analysis based on the temporal attribute data and the dynamic data yields the following change indicators: When the dynamic data indicates the operating status, the operating indicators in the dynamic data are extracted, the historical average of the corresponding operating indicators in the temporal attribute data is calculated, and the difference between the operating indicators and the historical average is calculated to obtain the absolute fluctuation range. The ratio of the absolute fluctuation range to the historical average is calculated to obtain the relative fluctuation range. The relative fluctuation range is used as a change indicator. The operating indicator includes at least one of the following: profit margin, operating revenue growth rate, net profit growth rate, and debt-to-equity ratio.

3. The method according to claim 1, characterized in that, The comparative analysis based on the temporal attribute data and the dynamic data to obtain the change index also includes: When the dynamic data indicates the operating status, the operating indicators are extracted from the dynamic data, and the historical indicators corresponding to the operating indicators are extracted from the temporal attribute data. The historical indicators and the operating indicators are sorted by time to obtain a time-series indicator sequence. The trend of the indicator is obtained by performing trend prediction based on the time series indicator sequence through a temporal convolutional network, and the indicator trend is used as the indicator of change.

4. The method according to claim 1, characterized in that, The comparative analysis based on the temporal attribute data and the dynamic data yields the following change indicators: When the dynamic data indicates public opinion information, the dynamic data is used to identify sentiment tendency through a preset public opinion analysis model to obtain a sentiment tendency score; based on preset weights, the number of clicks and reposts in the dynamic data are weighted and summed to obtain the dissemination volume, and the dissemination volume is standardized to obtain the dissemination popularity. The sentiment score and the popularity of dissemination are used as indicators of change.

5. The method according to claim 1, characterized in that, The temporal knowledge graph is pre-constructed, including: Using a pre-trained entity extraction model, entity extraction is performed on the collected multidimensional data to obtain an entity set. The multidimensional data includes at least one of the following: enterprise business registration information, enterprise financial statements, enterprise public opinion information, and enterprise business application materials. For any entity in the entity set, obtain the entity dataset related to the entity in the multidimensional data, and divide the entity dataset into multiple temporal subsets according to time information. Perform feature extraction on each temporal subset to obtain entity features. Store the entity features and corresponding time information in association as the temporal attribute data of the entity. Using a pre-trained relation recognition model, the types of relationships between entities are identified, resulting in a set of entity relation triples. A knowledge graph is then constructed based on the set of triples to obtain a temporal knowledge graph. The relation types include at least one of the following: supply chain association, guarantee association, or equity association.

6. The method according to claim 1, characterized in that, The step of disseminating information based on the association graph using a preset information dissemination model to transfer risk information in the changing indicators from the central entity to other enterprise entities in the association graph, thereby obtaining dissemination information from other enterprise entities, includes: Based on the first relationship type between the central entity in the association graph and the enterprise entity associated with the first hop, the weight of the first edge corresponding to the first relationship type in the preset information propagation model is determined. Based on the first edge weight, information is propagated to the changing index to obtain the propagation information of the first-hop enterprise entity; In any hop after the first hop, obtain the second relationship type between the current enterprise entity and the enterprise entity associated with the next hop, determine the second edge weight corresponding to the second relationship type in the information propagation model, and propagate the propagation information of the current enterprise entity based on the second edge weight to obtain the propagation information of the next hop enterprise entity.

7. The method according to claim 6, characterized in that, Based on the second edge weight, the propagation information of the current enterprise entity is propagated to obtain the propagation information of the next-hop enterprise entity, including: Based on the edge weights, the propagation information of the current enterprise entity is linearly transformed to obtain the transformed information; The attenuation factor is determined based on the distance between the next-hop enterprise entity and the central entity, and the propagation information of the next-hop enterprise entity is obtained by multiplying the attenuation factor with the transformation information.

8. The method according to claim 1, characterized in that, The process of obtaining the current risk probability of the enterprise to be processed includes: In response to receiving a risk question from a company to be processed, a search is performed in a preset temporal knowledge graph based on the risk question to obtain search results. The search results include nodes and / or edges in the temporal knowledge graph related to the risk question and the association information of the nodes and / or edges. Feature extraction is performed on the search results to obtain multidimensional risk features. Based on the multidimensional risk features, the multidimensional risk assessment model is used to predict the risk and obtain the initial risk probability of the enterprise to be processed. The initial risk probability is used as the current risk probability.

9. The method according to claim 8, characterized in that, Pre-training the multi-dimensional risk assessment model includes: The preset risk prediction network is repeatedly trained based on the pre-acquired multidimensional risk features of the samples and the corresponding sample risk probabilities until the fusion loss value of the risk prediction network reaches the preset loss threshold. Then the optimal risk prediction network is used as the multidimensional risk assessment model. The method for obtaining the fusion loss value includes: The risk prediction network performs risk prediction based on the multidimensional risk characteristics of the sample to obtain a first predicted value. The sample's multidimensional risk features are randomly grouped to obtain a first sub-feature and a second sub-feature. The risk prediction model is then used to predict the risk based on the first sub-feature to obtain a second predicted value. The risk prediction model is then used to predict the risk based on the second sub-feature to obtain a third predicted value. Finally, the sum of the second predicted value and the third predicted value is calculated to obtain a fourth predicted value. Calculate the loss value used to characterize the first predicted value and the sample risk probability to obtain the full prediction loss value; calculate the loss value used to characterize the fourth predicted value and the sample risk probability to obtain the component prediction loss value; The fusion loss value is obtained by weighted summing of the full prediction loss value and the component prediction loss value.

10. The method according to claim 1, characterized in that, The risk increment obtained by the pre-trained multi-dimensional risk assessment model based on the changing indicators or dissemination information of the enterprise to be processed includes: Based on a preset time interval, multiple pieces of communication information or changing indicators of the enterprise to be processed are aggregated to obtain aggregated information. The risk increment is obtained by using the multi-dimensional risk assessment model to predict the risk of the aggregated information.

11. A business risk update device, characterized in that, The device includes: The dynamic comparison and analysis module is used to respond to the monitoring of dynamic data of the target enterprise, retrieve the target entity corresponding to the target enterprise from the pre-constructed temporal knowledge graph, obtain the temporal attribute data of the target entity, and perform comparative analysis based on the temporal attribute data and the dynamic data to obtain change indicators; wherein, the dynamic data indicates the change information of business status or public opinion, and the temporal attribute data includes multiple attributes corresponding to different times; The information dissemination module is used to retrieve the association graph with the target entity as the central entity in the temporal knowledge graph based on a preset number of hops, and to disseminate information based on the association graph through a preset information dissemination model, so as to transmit the risk information in the change index from the central entity to other enterprise entities in the association graph, and obtain the dissemination information of other enterprise entities. The association graph includes multiple enterprise entities associated with the central entity. The risk increment assessment module is used to predict the risk increment based on the changing indicators or propagation information of the enterprise to be processed when the enterprise entity in the association graph matches any enterprise to be processed; and to obtain the risk increment. The risk update module is used to obtain the current risk probability of the enterprise to be processed, and calculate the sum of the risk increment and the current risk probability to obtain the updated risk probability, so as to determine the processing result of the business to be processed based on the updated risk probability.

12. An electronic device, comprising: One or more processors; Memory, used to store one or more computer programs. The characteristic feature is that the one or more processors execute the one or more computer programs to implement the steps of the method according to any one of claims 1 to 10.

13. A computer-readable storage medium having a computer program or instructions stored thereon, characterized in that, When the computer program or instructions are executed by a processor, they implement the steps of the method according to any one of claims 1 to 10.

14. A computer program product, comprising a computer program or instructions, characterized in that, When the computer program or instructions are executed by a processor, they implement the steps of the method according to any one of claims 1 to 10.

Images