Risk measurement method and device based on multi-modal data and causal hypergraph, and equipment
By constructing a causal hypergraph and training a risk prediction model, the problem of low efficiency in identifying risk sources in target financial systems in existing technologies is solved, and rapid and accurate risk source identification is achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- 湖南工商大学
- Filing Date
- 2026-05-26
- Publication Date
- 2026-07-28
AI Technical Summary
Existing risk measurement methods struggle to identify risk sources arising from multi-institutional interactions and feedback within a target financial system, resulting in low identification efficiency and susceptibility to human intervention.
A risk measurement method based on multimodal data and causal hypergraph is adopted. By acquiring numerical and textual data of the financial system, a causal hypergraph is constructed, message passing operations are performed, risk transmission characteristics are generated, and a risk prediction model is trained to identify risk sources.
It improves the efficiency and accuracy of identifying risk sources in the target financial system, reduces identification time, enables rapid identification of risk sources, and reduces reliance on manual identification.
Smart Images
Figure CN122264937B_ABST
Abstract
Description
Technical Field
[0001] This application relates to the fields of risk identification technology and artificial intelligence technology, and in particular to risk measurement methods, apparatus and equipment based on multimodal data and causal hypergraphs. Background Technology
[0002] As a core infrastructure of the modern economic system, the target financial system primarily undertakes key functions such as capital allocation, risk management, and payment settlement. With the continuous operation of the target financial system, the expansion of its business scale, and changes in the external environment, potential risks will gradually accumulate, eventually forming risk sources.
[0003] However, existing risk measurement methods struggle to identify risk sources within a target financial system, hindering the efficiency of risk source identification. This is because existing methods use binary graphs to model risk relationships between financial institutions. In reality, risk contagion in financial systems often involves multiple institutions interacting and feeding back to each other simultaneously. For example, several banks might be impacted simultaneously due to jointly holding the same type of risky asset, or multiple institutions might experience risk resonance through nested derivative contracts. These many-to-many group interactions exceed the expressive capacity of binary graphs, thus inherently limiting their representation. Analysts must meticulously examine numerous point-to-point relationships to infer the potential scope of group influence. This process significantly increases the time required to identify risk sources within the target financial system and is susceptible to human intervention, thus hindering the efficiency of risk source identification. Summary of the Invention
[0004] This application provides a risk measurement method, apparatus, and device based on multimodal data and causal hypergraphs to solve the technical problem that the existing risk measurement methods are difficult to identify risk sources in the target financial system and are not conducive to improving the identification efficiency of risk sources in the target financial system.
[0005] In a first aspect, embodiments of this application provide a risk measurement method based on multimodal data and causal hypergraphs, applied to electronic devices, the risk measurement method comprising: Acquire multimodal data from a preset financial system, which includes both numerical and textual data. Based on the numerical and textual data of the preset financial system, the causal hypergraph of the preset financial system is determined. Message passing operations are performed on the causal hypergraph of the preset financial system to generate the risk transmission characteristics of the preset financial system. The risk transmission characteristics of the preset financial system, the true risk level of the preset financial system, and the true risk contribution information of the preset financial system are combined into a sample. Different samples are combined into a training set, and the risk prediction model is trained using the training set. The total loss of the risk prediction model is obtained based on a predefined method. When the total loss is less than a preset value, the trained risk prediction model is saved. The trained risk prediction model generates the risk level and risk contribution information of the target financial system. Based on the risk level and risk contribution information of the target financial system, the risk sources of the target financial system are identified.
[0006] In one possible implementation of the first aspect, determining the causal hypergraph of the preset financial system based on numerical data and text data of the preset financial system includes: A feature extraction model is used to extract features from the numerical data and text data of the preset financial system, respectively, to obtain the feature vectors of the numerical data and the text data of the preset financial system. The feature vectors of the numerical data and the text data of the preset financial system are concatenated to obtain the global feature vector of the preset financial system. The global feature vector of the preset financial system is then input into the causal inference model. The causal strength of each preset financial institution is generated through the causal inference model. The causal hypergraph of the preset financial system is then constructed based on the causal strength of each preset financial institution.
[0007] In one possible implementation of the first aspect, the step of forming a training set from different samples, training the risk prediction model using the training set, obtaining the total loss of the risk prediction model based on a predefined method, and saving the trained risk prediction model when the total loss is less than a preset value includes: Different samples are combined into a training set, and the risk prediction model is trained using the training set. During the training process of the risk prediction model, the average prediction loss of the risk prediction model is generated through the average prediction loss model, the independence constraint loss of the risk prediction model is generated through the constraint loss model, the sparsity constraint loss of the risk prediction model is generated through the sparse model, and the regularization term loss of the risk prediction model is generated through the regularization function. Based on the average prediction loss, independence constraint loss, sparsity constraint loss, regularization loss, and total loss model of the risk prediction model, the total loss of the risk prediction model is generated. When the total loss is less than a preset value, the risk prediction model is trained and the trained risk prediction model is saved.
[0008] In one possible implementation of the first aspect, generating the risk level and risk contribution information of the target financial system through a trained risk prediction model, and identifying risk sources of the target financial system based on the risk level and risk contribution information, includes: Acquire multimodal data of the target financial system, which includes both numerical and textual data of the target financial system. A feature extraction model is used to extract features from the numerical data and text data of the target financial system, respectively, to obtain the feature vectors of the numerical data and text data of the target financial system. The feature vectors of the numerical data and text data of the target financial system are then concatenated to obtain the global feature vector of the target financial system. The global feature vector of the target financial system is input into the causal inference model. The causal strength of each target financial institution is generated through the causal inference model. The causal hypergraph of the target financial system is constructed through the causal strength of each target financial institution. Message passing operation is performed on the causal hypergraph of the target financial system to generate the risk transmission characteristics of the target financial system. The risk transmission characteristics of the target financial system are input into the trained risk prediction model. The trained risk prediction model generates the current risk level and current risk contribution information of the target financial system. The current risk contribution information of the target financial system includes the risk contribution value corresponding to each target financial institution. When the current risk level of the target financial system is greater than the preset level, the risk contribution value of each target financial institution is sorted, and the target financial institution with the largest risk contribution value is selected as the risk source of the target financial system.
[0009] In one possible implementation of the first aspect, the preset real risk contribution information of the financial system includes the risk contribution value corresponding to each preset financial institution; The numerical data of the preset financial system includes the interbank lending rate and the debt ratio of each preset financial institution. The text data of the preset financial system includes the news, announcements and research report summaries of each preset financial institution. The preset financial institutions are the financial institutions in the preset financial system. The numerical data of the target financial system includes the interbank lending rate and the debt ratio of each target financial institution. The text data of the target financial system includes the news, announcements, and research report summaries of each target financial institution. The target financial institutions are the financial institutions in the target financial system.
[0010] In one possible implementation of the first aspect, the average prediction loss model is defined as follows: ; This represents the average prediction loss of the risk prediction model; This represents the total number of quantiles sampled by the risk prediction model in each training iteration; Indicates the first Each sampled quantile level value; Indicates the quantile level value; This represents the actual variable at time point t; This represents the predicted value output by the risk prediction model at time point t.
[0011] In one possible implementation of the first aspect, the constraint loss model is defined as follows: , ; This represents the independence constraint loss of the risk prediction model, used to decouple the structural contagion characterization from macro-environmental characteristics; Represents a statistical function; This represents the invariant feature extracted at time point t; Represents a linear projection function; This represents the conditional feature at time point t. This represents the interference characteristics at time point t. The sparse model is defined as follows: ; This represents the sparsity constraint loss of the risk prediction model, used to control for sparsity. To reduce complexity and prevent overfitting; This represents the interaction component at time point t, at the quantile level. Indicates the quantile level value; This represents the L1 norm.
[0012] In one possible implementation of the first aspect, the total loss model is defined as follows: ; This represents the total loss of the risk prediction model; This represents the independence constraint loss of the risk prediction model; This represents the sparsity constraint loss of the risk prediction model; This represents the regularization loss of the risk prediction model; , , These are the first weight parameter, the second weight parameter, and the third weight parameter, which are used to control the relative importance of the independence constraint loss, the sparsity constraint loss, and the regularization loss of the risk prediction model in the total loss model, respectively.
[0013] Secondly, embodiments of this application provide a risk measurement device based on multimodal data and causal hypergraphs, applied to electronic devices, including: The acquisition module is used to acquire multimodal data of a preset financial system, which includes numerical data and text data of the preset financial system. The determination module is used to determine the causal hypergraph of the preset financial system based on the numerical data and text data of the preset financial system. The generation module is used to perform message passing operations on the causal hypergraph of the preset financial system, generate the risk transmission characteristics of the preset financial system, and combine the risk transmission characteristics of the preset financial system, the real risk level of the preset financial system, and the real risk contribution information of the preset financial system into a sample. The training module is used to form a training set from different samples, use the training set to train the risk prediction model, obtain the total loss of the risk prediction model based on a predefined method, and save the trained risk prediction model when the total loss is less than a preset value. The identification module is used to generate the risk level and risk contribution information of the target financial system through the trained risk prediction model, and to identify the risk sources of the target financial system based on the risk level and risk contribution information of the target financial system.
[0014] Thirdly, embodiments of this application provide an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the risk measurement method described in the first aspect above.
[0015] Fourthly, embodiments of this application provide a computer-readable storage medium storing a computer program that, when executed by a processor, implements the risk measurement method described in the first aspect above.
[0016] Fifthly, embodiments of this application provide a computer program product that, when run on an electronic device, causes the electronic device to execute the risk measurement method described in the first aspect above.
[0017] The beneficial effects of the embodiments of this application are as follows: Firstly, the risk level and risk contribution information of the target financial system are generated through the trained risk prediction model. Based on the risk level and risk contribution information of the target financial system, the risk sources of the target financial system are identified. Since there is no need for manual identification of the risk sources of the target financial system, the identification time of the risk sources of the target financial system is reduced, which is conducive to improving the identification efficiency of the risk sources of the target financial system. Secondly, causal hypergraphs directly express the collective linkages between three or more institutions through hyperedges. The hyperedge containing the risk source can simultaneously represent all groups of institutions impacted by the same causal factor. Therefore, when identifying risk sources based on causal hypergraphs, it is unnecessary to traverse redundant paired connections. By simply locating the hyperedge carrying the strongest causal relationship and its central institution, the risk source can be quickly identified. This shortens the time window from risk occurrence to source location, improving the timeliness and accuracy of risk identification in the target financial system. Attached Figure Description
[0018] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0019] Figure 1 This is an application scenario diagram of the risk measurement method provided in the embodiments of this application; Figure 2 This is a flowchart illustrating the risk measurement method provided in the embodiments of this application; Figure 3 A flowchart illustrating the implementation of S202 provided in this application embodiment; Figure 4 A schematic block diagram of a risk measurement device provided in the embodiments of this application; Figure 5 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation
[0020] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application. All other embodiments obtained by those skilled in the art based on the embodiments in this application without inventive effort are within the scope of protection of this application.
[0021] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.
[0022] It should be understood that in the description of this application and the appended claims, the terms "first," "second," "third," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance. The terms "comprising," "including," "having," and their variations all mean "including but not limited to," unless otherwise specifically emphasized.
[0023] Furthermore, the technical solutions of the various embodiments can be combined with each other, but only if they are based on the ability of those skilled in the art to implement them. When the combination of technical solutions is contradictory or cannot be implemented, it should be considered that such combination of technical solutions does not exist and is not within the scope of protection claimed in this application.
[0024] The flowchart shown in the attached diagram is for illustrative purposes only and does not necessarily include all content and operations / steps, nor does it necessarily have to be performed in the order described. For example, some operations / steps can be broken down, combined, or partially merged, so the actual execution order may change depending on the actual situation.
[0025] The risk measurement method provided in this application can be applied to electronic devices, including but not limited to servers, mobile phones, tablets, wearable devices, vehicle-mounted devices, and laptops. This application does not impose any restrictions on the specific type of electronic device.
[0026] Please see Figure 1 , Figure 1 The application scenario diagram of the risk measurement method provided in the embodiments of this application is described in detail below: Electronic devices access a data storage system to obtain multimodal data from a preset financial system. The multimodal data from the preset financial system includes numerical data and text data.
[0027] In this embodiment, the electronic device accesses the data storage system and obtains multimodal data from a preset financial system through the data storage system. It can retrieve the required data from the data storage system in real time according to the demand, which greatly reduces the local storage pressure on the electronic device.
[0028] Please see Figure 2 , Figure 2 This is a flowchart illustrating the risk measurement method provided in this application embodiment, which can be applied to electronic devices.
[0029] like Figure 2 As shown, the risk measurement method provided in this application includes the following steps, detailed below: S201, Obtain multimodal data of the preset financial system. The multimodal data of the preset financial system includes numerical data and text data of the preset financial system. The numerical data of the preset financial system includes the interbank lending rate and the debt ratio of each preset financial institution. The text data of the preset financial system includes the news, announcements, and research report summaries of each preset financial institution. The preset financial institutions are the financial institutions in the preset financial system.
[0030] The news items corresponding to the pre-defined financial institutions include their operational updates and major events.
[0031] Among them, the announcements corresponding to the pre-set financial institutions refer to the legally disclosed documents such as financial reports, major events, personnel changes, and business adjustments issued by the pre-set financial institutions in accordance with regulatory requirements and information disclosure standards. The research report summary corresponding to the pre-selected financial institution refers to the core viewpoints and conclusions of a professional research institution that analyzes and evaluates the pre-selected financial institution's operating conditions, asset quality, development prospects, and risk status.
[0032] S202, Based on the numerical data and text data of the preset financial system, determine the causal hypergraph of the preset financial system; S203, perform message passing operation on the causal hypergraph of the preset financial system to generate the risk transmission characteristics of the preset financial system, and combine the risk transmission characteristics of the preset financial system, the real risk level of the preset financial system and the real risk contribution information of the preset financial system into a sample. For example, a message passing operation is performed on the causal hypergraph of a pre-defined financial system to generate risk transmission characteristics of the pre-defined financial system. A sample is then formed by combining these risk transmission characteristics, the true risk level of the pre-defined financial system, and the true risk contribution information of the pre-defined financial system, including: Message passing operations are performed on the causal hypergraph of the preset financial system. When the number of message passing operations on the causal hypergraph of the preset financial system reaches a preset number, the final features of all nodes on the causal hypergraph of the preset financial system are pooled to generate the risk transmission features of the preset financial system. The risk transmission features of the preset financial system, the true risk level of the preset financial system, and the true risk contribution information of the preset financial system are combined into a sample.
[0033] The preset financial system's real risk contribution information includes the risk contribution value corresponding to each preset financial institution.
[0034] S204: Different samples are combined into a training set. The risk prediction model is trained using the training set. The total loss of the risk prediction model is obtained based on a predefined method. When the total loss is less than a preset value, the trained risk prediction model is saved. The step of forming a training set from different samples, training the risk prediction model using the training set, obtaining the total loss of the risk prediction model based on a predefined method, and saving the trained risk prediction model when the total loss is less than a preset value includes: Different samples are combined into a training set, and the risk prediction model is trained using the training set. During the training process of the risk prediction model, the average prediction loss of the risk prediction model is generated through the average prediction loss model, the independence constraint loss of the risk prediction model is generated through the constraint loss model, the sparsity constraint loss of the risk prediction model is generated through the sparse model, and the regularization term loss of the risk prediction model is generated through the regularization function. Regularization functions include, but are not limited to, L1 regularization, L2 regularization, and Dropout.
[0035] The Dropout function is a function that randomly discards items.
[0036] Based on the average prediction loss, independence constraint loss, sparsity constraint loss, regularization loss, and total loss model of the risk prediction model, the total loss of the risk prediction model is generated. When the total loss is less than a preset value, the risk prediction model is trained and the trained risk prediction model is saved.
[0037] The average prediction loss model is defined as follows: ; This represents the average prediction loss of the risk prediction model; This represents the total number of quantiles sampled by the risk prediction model in each training iteration; Indicates the first Each sampled quantile level value; Indicates the quantile level value; This represents the actual variable at time point t; This represents the predicted value output by the risk prediction model at time point t.
[0038] In the scenario of financial system risk prediction, sample data such as historical risk indicator data, market transaction data, capital flow data, and credit risk control data of the financial industry are arranged in order of numerical value. The overall data value range is divided into hierarchical equal parts according to a preset ratio. The quantitative value corresponding to each dividing boundary node is the quantile level value.
[0039] The constrained loss model is defined as follows: , ; This represents the independence constraint loss of the risk prediction model, used to decouple the structural contagion characterization from macro-environmental characteristics; Represents a statistical function; This represents the invariant feature extracted at time point t; Represents a linear projection function; This represents the conditional feature at time point t. This represents the interference characteristics at time point t. The sparse model is defined as follows: ; This represents the sparsity constraint loss of the risk prediction model, used to control for sparsity. To reduce complexity and prevent overfitting; This represents the interaction component at time point t, at the quantile level. Indicates the quantile level value; This represents the L1 norm.
[0040] Among them, the invariant feature refers to the inherent financial risk control parameter that is unaffected by short-term time fluctuations, instantaneous transaction disturbances, and temporary fund movements during the dynamic evolution of financial time-series data, market fluctuations, and real-time changes in business transaction behavior, and has long-term stability and inherent attribute invariance.
[0041] Among them, the invariable characteristics include the entity's identity attributes, industry business category, basic account qualifications, and inherent credit rating.
[0042] The total loss model is defined as follows: ; This represents the total loss of the risk prediction model; This represents the independence constraint loss of the risk prediction model; This represents the sparsity constraint loss of the risk prediction model; This represents the regularization loss of the risk prediction model; , , These are the first weight parameter, the second weight parameter, and the third weight parameter, which are used to control the relative importance of the independence constraint loss, the sparsity constraint loss, and the regularization loss of the risk prediction model in the total loss model, respectively.
[0043] S205 generates the risk level and risk contribution information of the target financial system through the trained risk prediction model, and identifies the risk sources of the target financial system based on the risk level and risk contribution information of the target financial system.
[0044] The numerical data of the target financial system includes the interbank lending rate and debt ratio of each target financial institution. The text data of the target financial system includes news, announcements, and research report summaries of each target financial institution. The target financial institution is the financial institution of the target financial system.
[0045] The news related to the target financial institution includes its operational dynamics and major events.
[0046] Among them, the announcements corresponding to the target financial institutions refer to the legally disclosed documents such as financial reports, major events, personnel changes, and business adjustments issued by the target financial institutions in accordance with regulatory requirements and information disclosure standards. The research report summary corresponding to the target financial institution refers to the core viewpoints and conclusions of a professional research institution that analyzes and evaluates the target financial institution's operating conditions, asset quality, development prospects, and risk status.
[0047] The process of generating the risk level and risk contribution information of the target financial system through the trained risk prediction model, and identifying the risk sources of the target financial system based on the risk level and risk contribution information, includes: Acquire multimodal data of the target financial system, which includes both numerical and textual data of the target financial system. A feature extraction model is used to extract features from the numerical data and text data of the target financial system, respectively, to obtain the feature vectors of the numerical data and text data of the target financial system. The feature vectors of the numerical data and text data of the target financial system are then concatenated to obtain the global feature vector of the target financial system. The global feature vector of the target financial system is input into the causal inference model. The causal strength of each target financial institution is generated through the causal inference model. The causal hypergraph of the target financial system is constructed through the causal strength of each target financial institution. Message passing operation is performed on the causal hypergraph of the target financial system to generate the risk transmission characteristics of the target financial system. The risk transmission characteristics of the target financial system are input into the trained risk prediction model. The trained risk prediction model generates the current risk level and current risk contribution information of the target financial system. The current risk contribution information of the target financial system includes the risk contribution value corresponding to each target financial institution. When the current risk level of the target financial system is greater than the preset level, the risk contribution value of each target financial institution is sorted, and the target financial institution with the largest risk contribution value is selected as the risk source of the target financial system.
[0048] For example, a message passing operation is performed on the causal hypergraph of the target financial system to generate risk transmission characteristics of the target financial system, including: Message passing operations are performed on the causal hypergraph of the target financial system. When the number of message passing operations on the causal hypergraph of the target financial system reaches the target number, a pooling operation is performed on the final features of all nodes on the causal hypergraph of the target financial system to generate the risk transmission features of the target financial system.
[0049] Among them, message passing is the process of exchanging information between nodes and hyperedges.
[0050] In this process, a causal strength threshold is set, and target financial institutions with causal strength higher than the threshold are selected as target nodes. Target financial institutions with causal strength lower than the threshold are temporarily excluded from the scope of hyperedge construction, thereby reducing the complexity of the entire topology network. The joint interaction relationship between multiple target financial institutions is identified. When multiple target financial institutions work together to have a synergistic effect on the system result, the nodes corresponding to these target financial institutions are connected by a hyperedge. All identified hyperedges are combined with each target node to form a causal hypergraph of the target financial system.
[0051] In this context, the hyperedges of a causal hypergraph can connect multiple nodes simultaneously, naturally expressing many-to-many relationships between three or more institutions. By using causal hypergraphs to depict the time-varying path of risk propagation along the chain of relationships within a target financial institution at different times, it is shown that risk contagion is not merely a point-to-point relationship, but rather manifests more as many-to-many group interactions. Therefore, it overcomes the technical limitations of binary graphs in representing group interactions, providing a more realistic structural basis for accurately depicting time-varying risk contagion.
[0052] The causal strength can be calculated using causal inference methods such as multivariate Granger causality tests and transit entropy. Based on the above calculation methods, this application defines the causal strength model as follows: ; in, and They represent two different target financial institutions; At time t, the The central institution for the first Normalized causal strength of each neighboring institution; This indicates that at time t, the first... The central institution for the first The original causal strength of each neighboring institution; It is an exponential function; Indicates except the first All institutions other than the central institution; The central institution is a financial institution that is currently being analyzed and is being used as a super-edge core to build a group.
[0053] Neighboring institutions are the target financial institutions influenced by the central institution.
[0054] Based on the hyperedge construction model, threshold, and maximum hyperedge size, the target financial institution Generate candidate superedges: The hyperedge construction model is defined as follows: ; This indicates that at time t, the value of the first time is... Each central institution is a hyperedge formed by the central node, which contains the central node and all neighboring nodes that meet the conditions.
[0055] {i} contains only the target financial institutions A collection of single elements; ∪ is the union symbol, used to merge the central node with other institutions that meet certain conditions into a single set; At time t, the target financial institution For target financial institutions Normalized causality; The threshold value represents the causal strength, and is a preset threshold value. This represents the maximum size of the superedge, which is a preset upper limit on the number of neighboring institutions. It indicates the maximum number of institutions that can be included in the superedge, excluding the central institution. Neighboring organizations.
[0056] By integrating the model, all generated hyperedges are collected into a hyperedge set; The integrated model is defined as follows: ; Denotes the set of superedges at time point t; This indicates the total number of central institutions.
[0057] Using a weighted model, the hyperedge weight is defined as the aggregation of causal strengths within the group: The weighted model is defined as follows: ; e represents a superedge, which is a set of groups consisting of a central institution and its neighboring institutions that satisfy the causal strength condition; This represents the weight of the hyperedge, used to quantify the average strength of causal influence within the group corresponding to that hyperedge; This represents all causal direction pairs belonging to the superedge e; This represents the cardinality of the superedge, that is, the total number of institutions contained in the superedge; To suppress structural jitter, temporal smoothing and sparsity constraints can be added. Temporal smoothing means penalizing weight changes between adjacent time points, preventing sudden and large jumps in the interconnected structure and instead allowing for gradual changes. Sparsity constraints limit the size of hyperedges, preventing a single hyperedge from encompassing too many mechanisms and avoiding the model treating unimportant weak connections as valid links. The effect of this is that even with periodic switching, the entire interconnected structure maintains an interpretable, gradual trend, preventing drastic changes due to short-term noise or sampling fluctuations.
[0058] Among these, accurately identifying the sources of risk in the financial system can promptly pinpoint the root causes of risks, effectively block the spread of risks, enhance the stability and security of the financial system, and provide a scientific basis for regulatory decisions, market control, and business operations.
[0059] For ease of explanation, the following example is provided: For example, a province may have 100 target financial institutions. After identifying financial institution A as a risk source from these 100 institutions using risk measurement methods, subsequent analysis only needs to focus on monitoring financial institution A. There is no need to traverse and analyze all entities and massive amounts of data, which significantly reduces the amount of data that needs to be processed and significantly improves the efficiency of risk analysis.
[0060] The beneficial effects of the embodiments of this application are as follows: Firstly, the risk level and risk contribution information of the target financial system are generated through the trained risk prediction model. Based on the risk level and risk contribution information of the target financial system, the risk sources of the target financial system are identified. Since there is no need for manual identification of the risk sources of the target financial system, the identification time of the risk sources of the target financial system is reduced, which is conducive to improving the identification efficiency of the risk sources of the target financial system. Secondly, causal hypergraphs directly express the collective linkages between three or more institutions through hyperedges. The hyperedge containing the risk source can simultaneously represent all groups of institutions impacted by the same causal factor. Therefore, when identifying risk sources based on causal hypergraphs, it is unnecessary to traverse redundant paired connections. By simply locating the hyperedge carrying the strongest causal relationship and its central institution, the risk source can be quickly identified. This shortens the time window from risk occurrence to source location, improving the timeliness and accuracy of risk identification in the target financial system.
[0061] Please see Figure 3 , Figure 3 The implementation flowchart of S202 provided in the embodiments of this application is described in detail below: S301, using a feature extraction model, features are extracted from the numerical data and text data of the preset financial system respectively, to obtain the feature vectors of the numerical data and text data of the preset financial system respectively. S302, the feature vectors of the numerical data and the text data of the preset financial system are concatenated to obtain the global feature vector of the preset financial system. The global feature vector of the preset financial system is input into the causal inference model. The causal strength of each preset financial institution is generated through the causal inference model. The causal hypergraph of the preset financial system is constructed through the causal strength of each preset financial institution.
[0062] A causal strength threshold is set, and preset financial institutions with causal strength higher than the threshold are selected as preset nodes. Preset financial institutions with causal strength lower than the threshold are temporarily excluded from the scope of hyperedge construction, thereby reducing the complexity of the entire topology network. The joint interaction relationship between multiple preset financial institutions is identified. When multiple preset financial institutions work together to have a synergistic effect on the system result, the nodes corresponding to these preset financial institutions are connected by a hyperedge. All identified hyperedges are combined with each preset node to form a causal hypergraph of the preset financial system.
[0063] In this embodiment, a causal hypergraph of the preset financial system is constructed by the causal strength of each preset financial institution, which can clearly present the synergistic effect of multiple preset financial institutions, avoid the risk prediction model from omitting key risk correlation information, and improve the comprehensiveness of risk identification.
[0064] For the risk measurement method described in the above embodiments, please refer to [link / reference]. Figure 4 , Figure 4 This is a schematic block diagram of the risk measurement device provided in the embodiments of this application. Figure 4 The risk measurement device 400 shown can be applied to, for example... Figure 1 The application scenario diagram shows electronic devices. The following section uses electronic devices as an example to illustrate this. Figure 4 The risk measurement device 400 shown will be described in detail. The risk measurement device 400 may include an acquisition module 401, a determination module 402, a generation module 403, a training module 404, and an identification module 405.
[0065] The acquisition module 401 is used to acquire multimodal data of a preset financial system, which includes numerical data and text data of the preset financial system. Module 402 is used to determine the causal hypergraph of the preset financial system based on the numerical data and text data of the preset financial system. The generation module 403 is used to perform message passing operations on the causal hypergraph of the preset financial system, generate the risk transmission characteristics of the preset financial system, and combine the risk transmission characteristics of the preset financial system, the real risk level of the preset financial system, and the real risk contribution information of the preset financial system into a sample. The training module 404 is used to form a training set from different samples, train the risk prediction model using the training set, obtain the total loss of the risk prediction model based on a predefined method, and save the trained risk prediction model when the total loss is less than a preset value. The identification module 405 is used to generate the risk level and risk contribution information of the target financial system through the trained risk prediction model, and to identify the risk sources of the target financial system based on the risk level and risk contribution information of the target financial system.
[0066] It should be noted that the various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other.
[0067] The beneficial effects of the embodiments of this application are as follows: Firstly, the risk level and risk contribution information of the target financial system are generated through the trained risk prediction model. Based on the risk level and risk contribution information of the target financial system, the risk sources of the target financial system are identified. Since there is no need for manual identification of the risk sources of the target financial system, the identification time of the risk sources of the target financial system is reduced, which is conducive to improving the identification efficiency of the risk sources of the target financial system. Secondly, causal hypergraphs directly express the collective linkages between three or more institutions through hyperedges. The hyperedge containing the risk source can simultaneously represent all groups of institutions impacted by the same causal factor. Therefore, when identifying risk sources based on causal hypergraphs, it is unnecessary to traverse redundant paired connections. By simply locating the hyperedge carrying the strongest causal relationship and its central institution, the risk source can be quickly identified. This shortens the time window from risk occurrence to source location, improving the timeliness and accuracy of risk identification in the target financial system.
[0068] Please see Figure 5 , Figure 5 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application.
[0069] like Figure 5 As shown, Figure 5The electronic device includes: at least one processor 20, a memory 21, and a computer program 22 stored in the memory 21 and executable on the at least one processor 20, wherein the processor 20 executes the computer program 22 to implement the steps in any of the above method embodiments.
[0070] The electronic device may include, but is not limited to, processor 20 and memory 21. Those skilled in the art will understand that... Figure 5 This is merely an example of an electronic device and does not constitute a limitation on electronic devices. It may include more or fewer components than shown in the illustration, or combinations of certain components, or different components. For example, it may also include input / output devices, network access devices, etc.
[0071] The processor 20 is used to run a computer program 22 stored in the memory 21, and performs the following steps when executing the computer program 22: Acquire multimodal data from a preset financial system, which includes both numerical and textual data. Based on the numerical and textual data of the preset financial system, the causal hypergraph of the preset financial system is determined. Message passing operations are performed on the causal hypergraph of the preset financial system to generate the risk transmission characteristics of the preset financial system. The risk transmission characteristics of the preset financial system, the true risk level of the preset financial system, and the true risk contribution information of the preset financial system are combined into a sample. Different samples are combined into a training set, and the risk prediction model is trained using the training set. The total loss of the risk prediction model is obtained based on a predefined method. When the total loss is less than a preset value, the trained risk prediction model is saved. The trained risk prediction model generates the risk level and risk contribution information of the target financial system. Based on the risk level and risk contribution information of the target financial system, the risk sources of the target financial system are identified.
[0072] The processor 20 may be a central processing unit (CPU), or it may be other general-purpose processors, digital signal processors, field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor.
[0073] In some embodiments, the memory 21 may be an internal storage unit of the electronic device, such as a hard disk or memory. In other embodiments, the memory 21 may be an external storage device of the electronic device, such as a plug-in hard disk, SmartMediaCard (SMC), SecureDigital (SD) card, or FlashCard.
[0074] Furthermore, the memory 21 may include both internal storage units and external storage devices of the electronic device. The memory 21 is used to store the operating system, applications, boot loader, data, and other programs, such as the program code of the computer program. The memory 21 can also be used to temporarily store data that has been output or will be output.
[0075] It should be noted that the information interaction and execution process between the above-mentioned devices / units are based on the same concept as the method embodiments of this application. For details on their specific functions and technical effects, please refer to the method embodiments section, and they will not be repeated here.
[0076] This application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps described in the various method embodiments above.
[0077] The computer-readable storage medium may also be an external storage device of the risk measurement device or electronic device, such as a plug-in hard drive, smart media card (SMC), secure digital (SD) card, flash card, or non-transitory computer-readable storage medium equipped on the risk measurement device or electronic device.
[0078] Since the computer program stored in the computer-readable storage medium can execute any of the risk measurement methods based on multimodal data and causal hypergraphs provided in the embodiments of this application, the computer-readable storage medium can achieve the beneficial effects that any of the risk measurement methods based on multimodal data and causal hypergraphs provided in the embodiments of this application can achieve, as detailed in the preceding embodiments, and will not be repeated here.
[0079] This application provides a computer program product that, when run on an electronic device, causes the electronic device to perform the aforementioned risk measurement method.
[0080] When a computer program is loaded into an electronic device, it can perform the following steps: Acquire multimodal data from a preset financial system, which includes both numerical and textual data. Based on the numerical and textual data of the preset financial system, the causal hypergraph of the preset financial system is determined. Message passing operations are performed on the causal hypergraph of the preset financial system to generate the risk transmission characteristics of the preset financial system. The risk transmission characteristics of the preset financial system, the true risk level of the preset financial system, and the true risk contribution information of the preset financial system are combined into a sample. Different samples are combined into a training set, and the risk prediction model is trained using the training set. The total loss of the risk prediction model is obtained based on a predefined method. When the total loss is less than a preset value, the trained risk prediction model is saved. The trained risk prediction model generates the risk level and risk contribution information of the target financial system. Based on the risk level and risk contribution information of the target financial system, the risk sources of the target financial system are identified.
[0081] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium.
[0082] Based on this understanding, all or part of the processes in the methods of the above embodiments of this application can be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium includes: an entity or device for carrying computer program code to an electronic device, a recording medium, a computer memory, a read-only memory (ROM), a random access memory (RAM), an electrical carrier signal, a telecommunication signal, and a software distribution medium.
[0083] In the above embodiments, the descriptions of each embodiment have different focuses. For parts that are not described in detail or recorded in a certain embodiment, please refer to the relevant descriptions of other embodiments.
[0084] The above are merely preferred embodiments of this application and do not limit the patent scope of this application. Any equivalent structural or procedural transformations made using the content of this application's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of this application.
Claims
1. A method of risk measure based on multi-modal data and causal hypergraphs, characterized in that, The risk measurement method, applied to electronic devices, includes: Acquire multimodal data from a preset financial system, which includes both numerical and textual data. A feature extraction model is used to extract features from the numerical data and text data of the preset financial system, respectively, to obtain feature vectors for the numerical data and text data of the preset financial system. The feature vectors of the numerical data and text data of the preset financial system are concatenated to obtain the global feature vector of the preset financial system. The global feature vector of the preset financial system is then input into the causal inference model, which generates the causal strength of each preset financial institution. A causal hypergraph of the preset financial system is then constructed based on the causal strength of each preset financial institution. Message passing operations are performed on the causal hypergraph of the preset financial system to generate the risk transmission characteristics of the preset financial system. The risk transmission characteristics of the preset financial system, the true risk level of the preset financial system, and the true risk contribution information of the preset financial system are combined into a sample. Different samples are combined into a training set, which is used to train the risk prediction model. During the training process, the average prediction loss of the risk prediction model is generated through the average prediction loss model, the independence constraint loss of the risk prediction model is generated through the constraint loss model, the sparsity constraint loss of the risk prediction model is generated through the sparsity model, and the regularization loss of the risk prediction model is generated through the regularization function. Based on the average prediction loss, the independence constraint loss, the sparsity constraint loss, the regularization loss, and the total loss model, the total loss of the risk prediction model is generated. When the total loss is less than a preset value, the risk prediction model is trained and the trained risk prediction model is saved. The trained risk prediction model generates the risk level and risk contribution information of the target financial system. Based on the risk level and risk contribution information of the target financial system, the risk sources of the target financial system are identified.
2. The risk measure method of claim 1, wherein, The process involves generating the risk level and risk contribution information of the target financial system through a trained risk prediction model, and identifying risk sources within the target financial system based on these information, including: Acquire multimodal data of the target financial system, which includes both numerical and textual data of the target financial system. A feature extraction model is used to extract features from the numerical data and text data of the target financial system, respectively, to obtain the feature vectors of the numerical data and text data of the target financial system. The feature vectors of the numerical data and text data of the target financial system are then concatenated to obtain the global feature vector of the target financial system. The global feature vector of the target financial system is input into the causal inference model. The causal strength of each target financial institution is generated through the causal inference model. The causal hypergraph of the target financial system is constructed through the causal strength of each target financial institution. Message passing operation is performed on the causal hypergraph of the target financial system to generate the risk transmission characteristics of the target financial system. The risk transmission characteristics of the target financial system are input into the trained risk prediction model. The trained risk prediction model generates the current risk level and current risk contribution information of the target financial system. The current risk contribution information of the target financial system includes the risk contribution value corresponding to each target financial institution. When the current risk level of the target financial system is greater than the preset level, the risk contribution value of each target financial institution is sorted, and the target financial institution with the largest risk contribution value is selected as the risk source of the target financial system.
3. The risk measure method of claim 1, wherein, The real risk contribution information of the preset financial system includes the risk contribution value corresponding to each preset financial institution; The numerical data of the preset financial system includes the interbank lending rate and the debt ratio of each preset financial institution. The text data of the preset financial system includes the news, announcements and research report summaries of each preset financial institution. The preset financial institutions are the financial institutions in the preset financial system. The numerical data of the target financial system includes the interbank lending rate and the debt ratio of each target financial institution. The text data of the target financial system includes the news, announcements, and research report summaries of each target financial institution. The target financial institutions are the financial institutions in the target financial system.
4. The risk measure method of claim 1, wherein, The The average prediction loss model is defined as follows: ； an average prediction loss representing the risk prediction model; denotes the total number of quantiles sampled by the risk prediction model in each training iteration; representing a first sampled quantile level value; representing a quantile level value; Ytrepresents the true variable at time point t; denotes the prediction value output by the risk prediction model at the t-th time point.
5. The risk measure method of claim 1, wherein, The The constrained loss model is defined as follows: ， ； an independence constraint loss representing a risk prediction model for encouraging decoupling of structural contagion characterization from macro-environmental features; represents a statistical function; represents the invariant features extracted at the t-th time point; represents a linear projection function; represents a conditional feature at the t-th time point; denotes the interference characteristic at the t-th time point; The sparse model is defined as follows: ； a sparsity constraint loss representing a risk prediction model, for controlling complexity of the risk prediction model to prevent overfitting; denotes the interaction component at the t-th time point at the quantile level value; denotes the quantile level value; denotes the L1 norm.
6. The risk measure method of claim 1, wherein, The total loss model is defined as follows: ； total loss representing the risk prediction model; independence constraint loss representing the risk prediction model; a sparsity constraint loss representing a risk prediction model; regularization term loss representing a risk prediction model; , , are respectively a first weight parameter, a second weight parameter, a third weight parameter, and the first weight parameter, the second weight parameter, the third weight parameter are respectively used for controlling the relative importance degrees of the independence constraint loss of the risk prediction model, the sparsity constraint loss of the risk prediction model, and the regular term loss of the risk prediction model in the total loss model.
7. A risk measurement device based on multimodal data and causal hypergraphs, characterized in that, Applied to electronic devices, including: The acquisition module is used to acquire multimodal data of a preset financial system, which includes numerical data and text data of the preset financial system. The determination module is used to extract features from the numerical data and text data of the preset financial system using a feature extraction model, respectively, to obtain feature vectors for the numerical data and text data of the preset financial system. The feature vectors of the numerical data and text data of the preset financial system are concatenated to obtain the global feature vector of the preset financial system. The global feature vector of the preset financial system is then input into the causal inference model, which generates the causal strength of each preset financial institution. The causal hypergraph of the preset financial system is then constructed based on the causal strength of each preset financial institution. The generation module is used to perform message passing operations on the causal hypergraph of the preset financial system, generate the risk transmission characteristics of the preset financial system, and combine the risk transmission characteristics of the preset financial system, the real risk level of the preset financial system, and the real risk contribution information of the preset financial system into a sample. The training module is used to form a training set from different samples and use the training set to train the risk prediction model. During the training process, the average prediction loss of the risk prediction model is generated through the average prediction loss model, the independence constraint loss of the risk prediction model is generated through the constraint loss model, the sparsity constraint loss of the risk prediction model is generated through the sparsity model, and the regularization loss of the risk prediction model is generated through the regularization function. Based on the average prediction loss, independence constraint loss, sparsity constraint loss, regularization loss, and total loss model of the risk prediction model, the total loss of the risk prediction model is generated. When the total loss is less than a preset value, the risk prediction model is trained and the trained risk prediction model is saved. The identification module is used to generate the risk level and risk contribution information of the target financial system through the trained risk prediction model, and to identify the risk sources of the target financial system based on the risk level and risk contribution information of the target financial system.
8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the risk measurement method as described in any one of claims 1 to 6.