Risk prediction method and device for loan business, storage medium and electronic equipment

Abstract

The invention discloses a risk prediction method and device for loan business, a storage medium and electronic equipment. Relates to the field of big data and the field of artificial intelligence, and the method comprises the steps: obtaining the related information of a loan service, the related information comprising at least one of the following information: the investment or financing information of loan funds and the market evaluation information of a loan object; constructing an information cascade graph of the related information based on the related information, and extracting target features from the information cascade graph; inputting the target feature into a target model to obtain a risk increment of the loan service; determining the propagation duration of the related information, and determining a risk assessment value based on the propagation duration and the risk increment; and when the risk assessment value is greater than or equal to the assessment value threshold, determining that the loan business has a risk. Through the method and the device, the problem of relatively low accuracy of risk prediction of the loan business in related technologies is solved.

Description

Technical Field

[0001] The present application relates to the fields of big data and artificial intelligence, and specifically, to a risk prediction method, device, storage medium and electronic device for loan business. Background Art

[0002] Financial risk refers to the possibility of losses suffered by the subject of financial activities due to the uncertainty of various economic variables in the financial market. It is essentially composed of the investment and financing behavior of the subject of the financial market. The subject of the financial risk perception is generated by obtaining risk information in the financial market, which affects the investment and financing behavior of the subject of the financial risk and changes the financial risk. With the development of the Internet, big data has gradually become a necessary production factor in the financial industry, stimulating the emergence of information cascade phenomenon.

[0003] In the related art, the risk prediction method for loan business is to adopt financial risk prediction models, such as financial market liquidity risk prediction models based on time decay effect and financial market liquidity risk prediction models based on financial risk indicators. However, the information propagation in the financial market has domain information characteristics. Taking the financial market liquidity risk as an example, the domain information characteristics of the financial market liquidity risk are mainly manifested in the fact that the propagation of the financial market liquidity risk has domain information characteristics. The financial risk prediction models in the related art cannot capture the accurate domain information characteristics, resulting in a low accuracy rate in predicting financial risks.

[0004] Currently, no effective solution has been proposed to address the problem of low accuracy in risk prediction of loan business in relevant technologies. Summary of the invention

[0005] The main purpose of this application is to provide a risk prediction method, device, storage medium and electronic device for a loan business to solve the problem of low accuracy in risk prediction of loan business in related technologies.

[0006] In order to achieve the above-mentioned purpose, according to one aspect of the present application, a risk prediction method for a loan business is provided. The method comprises: obtaining relevant information of the loan business, wherein the relevant information comprises at least one of the following: investment or financing information of the loan funds and market evaluation information of the loan object; constructing an information cascade graph of the relevant information based on the relevant information, and extracting target features from the information cascade graph, wherein the information cascade graph comprises a plurality of nodes and edges, the nodes are used to characterize the users who receive or transmit the relevant information, and the edges are used to characterize the transmission relationship between the users corresponding to the nodes; inputting the target features into the target model to obtain the risk increment of the loan business, wherein the target model is trained by multiple groups of training samples, each group of training samples comprises historical features and historical risk increments, and the risk increments are used to characterize the risk increase of the loan business caused by the transmission of the relevant information over time; determining the transmission time of the relevant information, and determining the risk assessment value based on the transmission time and the risk increment; and determining that the loan business is at risk when the risk assessment value is greater than or equal to the assessment value threshold.

[0007] Optionally, determining the risk assessment value based on the propagation time and the risk increment includes: determining the time observation window when predicting the loan business risk, splitting the time observation window into N time intervals, where N is a positive integer; determining the discrete variable for each time interval, where the discrete variable is used to characterize the time decay effect within the time interval; determining the sub-risk increment corresponding to each time interval, calculating the product of the sub-risk increment, the interval value of the time interval and the discrete variable of the time interval to obtain the sub-risk assessment value of the time interval; calculating the sum of the sub-risk assessment values ​​of all time intervals and the preset error to obtain the risk assessment value.

[0008] Optionally, determining the sub-risk increment corresponding to each time interval includes: extracting a sub-information cascade graph corresponding to each time interval from the information cascade graph; extracting sub-target features of each sub-information cascade graph from the target features, and inputting the sub-target features into the target model to obtain the sub-risk increment corresponding to the time interval.

[0009] Optionally, extracting target features from the information cascade graph includes: performing sequence sampling on the information cascade graph based on timestamps to obtain a sub-information cascade graph sequence, wherein the timestamp is used to represent the time of the last propagation of relevant information in each time interval, and the sub-information cascade graph sequence contains N sub-information cascade graphs, each sub-information cascade graph corresponding to a time interval; and capturing the features of each sub-information cascade graph from the sub-information cascade graph sequence through a preset graph attention network to obtain the target features.

[0010] Optionally, the features of each sub-information cascade graph are captured from the sub-information cascade graph sequence through a preset graph attention network, and the target features are obtained, including: determining the adjacency matrix of each sub-information cascade graph in the sub-information cascade graph sequence, wherein the rows in the adjacency matrix are used to characterize the node order in the sub-information cascade graph, and the columns in the adjacency matrix are used to characterize the edges connecting each node; for each sub-information cascade graph, determining the feature set of each node in the sub-information cascade graph; determining the adjacent nodes of each node in the sub-information cascade graph based on the adjacency matrix, and for each target node, calculating the attention coefficient based on the feature set of the target node and the feature set of the adjacent nodes of the target node; performing weighted summation on the feature set of the target node based on the attention coefficient to obtain the comprehensive features of the target node; combining the comprehensive features of each node in the sub-information cascade graph into a feature vector, combining the feature vectors of all sub-information cascade graphs to obtain a feature vector set, and determining the feature vector set as the target feature.

[0011] Optionally, the target model is trained in the following manner: obtaining historical risk prediction records of loan businesses, extracting multiple historical related information and the historical risk increment of each related information from the historical risk prediction records; determining the historical characteristics of each historical related information, and determining the historical characteristics and historical risk increment of each historical related information as a group of training samples to obtain multiple groups of training samples; training a preset neural network model with multiple groups of training samples to obtain a target model.

[0012] Optionally, after obtaining the target model, the method also includes: obtaining multiple groups of test samples, inputting each group of test samples into the target model to obtain predicted risk increments, wherein each group of test samples includes sample characteristics and actual risk increments; calculating the standard mean variance based on the predicted risk increments and actual risk increments of each group of test samples to obtain an evaluation value of the target model; when the evaluation value is less than a model evaluation value threshold, updating the target model based on the multiple groups of test samples.

[0013] In order to achieve the above-mentioned purpose, according to another aspect of the present application, a risk prediction device for a loan business is provided. The device includes: an acquisition unit, which is used to acquire relevant information of the loan business, wherein the relevant information includes at least one of the following: investment or financing information of the loan funds and market evaluation information of the loan object; a construction unit, which is used to construct an information cascade graph of the relevant information based on the relevant information, and extract target features from the information cascade graph, wherein the information cascade graph includes multiple nodes and edges, the nodes are used to represent users who receive or transmit relevant information, and the edges are used to represent the transmission relationship between users corresponding to the nodes; an input unit, which is used to input the target feature into a target model to obtain the risk increment of the loan business, wherein the target model is trained by multiple groups of training samples, each group of training samples includes historical features and historical risk increments, and the risk increment is used to represent the risk increase of the loan business caused by the transmission of relevant information over time; a first determination unit, which is used to determine the transmission time of the relevant information, and determine the risk assessment value based on the transmission time and the risk increment; a second determination unit, which is used to determine that the loan business is risky when the risk assessment value is greater than or equal to the assessment value threshold.

[0014] In an embodiment of the present application, relevant information of the loan business is obtained, wherein the relevant information includes at least one of the following: investment or financing information of the loan funds and market evaluation information of the loan object; an information cascade graph of the relevant information is constructed based on the relevant information, and a target feature is extracted from the information cascade graph, wherein the information cascade graph includes a plurality of nodes and edges, the nodes are used to represent users who receive or transmit relevant information, and the edges are used to represent the transmission relationship between users corresponding to the nodes; the target feature is input into a target model to obtain a risk increment of the loan business, wherein the target model is trained by multiple groups of training samples, each group of training samples includes historical features and historical risk increments, and the risk increments are used to represent the transmission of relevant information over time. The method determines the amount of risk increase in the loan business caused by the propagation of relevant information; determines the propagation time of the relevant information, and determines the risk assessment value based on the propagation time and the risk increment; when the risk assessment value is greater than or equal to the assessment value threshold, determines the way in which the loan business is at risk, constructs an information cascade graph for the relevant information of the loan business, extracts the target features from the graph, determines the risk increment based on the target features, and then predicts the risk of the loan business based on the risk increment and the propagation time of the relevant information, thereby achieving the purpose of capturing the accurate domain information characteristics in the relevant information of the loan business, thereby realizing the technical effect of improving the accuracy of risk prediction for the loan business, and then solving the technical problem of low accuracy of risk prediction for the loan business. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] The drawings constituting a part of the present application are used to provide a further understanding of the present application. The illustrative embodiments and descriptions of the present application are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0016] Figure 1 A hardware structure block diagram of a computer terminal (or mobile device) for implementing a risk prediction method for a loan business is shown;

[0017] Figure 2 is a flow chart of a risk prediction method for a loan business provided in an embodiment of the present application;

[0018] Figure 3 is a schematic diagram of a sub-cascade graph sequence sampling structure provided according to an embodiment of the present application;

[0019] Figure 4 is a schematic diagram of a risk prediction device for a loan business provided in an embodiment of the present application;

[0020] Figure 5 It is a structural block diagram of an electronic device according to an embodiment of the present application. DETAILED DESCRIPTION

[0021] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present application.

[0022] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0023] It should be noted that the collected information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for display, data for analysis, etc.) involved in this application are information and data authorized by the user or fully authorized by all parties, and the collection, storage, use, processing, transmission, provision, disclosure and application of relevant data are in compliance with relevant laws, regulations and standards, necessary confidentiality measures are taken, and public order and good customs are not violated, and corresponding operation entrances are provided for users to choose to authorize or refuse. For example, an interface is set up between this system and relevant users or institutions to provide users with corresponding operation entrances for users to choose to agree or refuse the results of automated decision-making; if the user chooses to refuse, the expert decision-making process will be entered.

[0024] Example 1

[0025] According to an embodiment of the present application, a method embodiment for risk prediction of a loan business is also provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0026] The method embodiment provided in the first embodiment of the present application can be executed in a mobile terminal, a computer terminal or a similar computing device. Figure 1 The hardware structure block diagram of a computer terminal (or mobile device) for implementing a risk prediction method for a loan business is shown. Figure 1 As shown, the computer terminal 10 (or mobile device) may include one or more (shown as 102a, 102b, ..., 102n in the figure) processors 102 (the processor 102 may include but is not limited to a processing device such as an MCU (Microcontroller Unit) or an FPGA (Field-Programmable Gate Array), a memory 104 for storing data, and a transmission device 106 for communication functions. In addition, it may also include: a display, an input / output interface (I / O interface), a USB (Universal Serial Bus) port (which may be included as one of the ports of a BUS (Business)), a network interface, a power supply and / or a camera. It will be understood by those skilled in the art that Figure 1 The structure shown is only for illustration and does not limit the structure of the above electronic device. Figure 1 More or fewer components as shown, or with Figure 1 Different configurations are shown.

[0027] It should be noted that the one or more processors 102 and / or other data processing circuits described above may generally be referred to herein as "data processing circuits". The data processing circuits may be embodied in whole or in part as software, hardware, firmware, or any other combination thereof. In addition, the data processing circuit may be a single independent processing module, or may be incorporated in whole or in part into any of the other components in the computer terminal 10 (or mobile device). As described in the embodiments of the present application, the data processing circuit acts as a processor control (e.g., selection of a variable resistor terminal path connected to an interface).

[0028] The memory 104 can be used to store software programs and modules of application software, such as the program instructions / data storage device corresponding to the risk prediction method of the loan business in the embodiment of the present application. The processor 102 executes various functional applications and data processing by running the software programs and modules stored in the memory 104, that is, the risk prediction method of the loan business described above is realized. The memory 104 may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some examples, the memory 104 may further include a memory remotely arranged relative to the processor 102, and these remote memories may be connected to the computer terminal 10 via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0029] The transmission device 106 is used to receive or send data via a network. The specific example of the above network may include a wireless network provided by a communication provider of the computer terminal 10. In one example, the transmission device 106 includes a network adapter (Network Interface Controller, NIC), which can be connected to other network devices through a base station so as to communicate with the Internet. In one example, the transmission device 106 can be a radio frequency (RF) module, which is used to communicate with the Internet wirelessly.

[0030] The display may be, for example, a touch screen liquid crystal display that enables a user to interact with a user interface of the computer terminal 10 (or mobile device).

[0031] Under the above operating environment, this application provides a risk prediction method for loan business. Figure 2 is a flow chart of a risk prediction method for a loan business provided in an embodiment of the present application, such as Figure 2 As shown, the method includes:

[0032] Step S201, obtaining relevant information of the loan business, wherein the relevant information includes at least one of the following: investment or financing information of the loan funds and market evaluation information of the loan object.

[0033] In step S201, the relevant information may include investment or financing information of the loan funds and market assessment information of the loan object. The investment or financing information of the loan funds, such as investment background: understand what kind of investment the loan funds will be used for, including investment type (such as real estate, stocks, bonds, venture capital, etc.), investment scale, expected rate of return, investment period and any relevant market trends or forecasts. Financing purpose: clarify the financing purpose of the loan applicant, such as expanding business, repaying high-interest debts, conducting research and development, acquiring assets, etc. Financial status: assess the financial health of the loan applicant or enterprise, including balance sheet, income statement, cash flow statement, etc. Market conditions: analyze the macroeconomic conditions of the investment market, such as interest rates, inflation, economic growth, etc., as well as market trends in specific areas, such as real estate price trends, stock market fluctuations, etc.

[0034] Market assessment information of loan recipients, such as credit rating: Understand the credit history and credit rating of loan recipients and assess their reliability in repaying loans. Competitive environment: Analyze the competitive landscape of the industry in which the loan recipients are located, understand their market position, market share, status of competitors, and industry development trends. Consumer demand: For loans involving products or services, assess the trend of consumer demand and predict the market's acceptance of products or services invested in loan funds. Technological progress: In science and technology-related fields, it is necessary to assess the impact of technological progress on loan business, such as market opportunities or risks that new technologies may bring.

[0035] Step S202, constructing an information cascade graph of the relevant information based on the relevant information, and extracting target features from the information cascade graph, wherein the information cascade graph includes multiple nodes and edges, the nodes are used to represent users who receive or disseminate relevant information, and the edges are used to represent the dissemination relationship between users corresponding to the nodes.

[0036] In step S202, in the information cascade graph, each node represents a user, who may be a loan applicant, a loan approver, an end user of loan funds, or a market evaluation participant, etc. Nodes may carry attributes such as credit score, loan amount, investment direction, etc. Edges represent the communication relationship between users, such as a loan applicant's request for a loan to a loan approver, a loan approver's feedback on the application, and the flow path of loan funds in the market, etc. Edges may carry a timestamp to reflect the time when the communication action occurs.

[0037] Based on the propagation relationship between users, an information cascade graph is constructed. For example, when a loan application is submitted, the approval process begins, a funding allocation decision is made, or market assessment information is propagated, these events can form a path in the information cascade graph. Each event (such as a loan approval decision) is regarded as a diffusion point in the cascade, connected to subsequent affected events or users.

[0038] Extract static features of each node, such as credit score, financial status, market evaluation score, etc. These features constitute the basic attributes of the node. Extract dynamic features from the information cascade graph, which can include the speed, scale, depth and breadth of the propagation path of information. These features can be extracted using methods such as time series analysis and graph analysis. Model the information cascade graph using graph attention networks or other graph neural network technologies. Capture the strength of the relationship between nodes and assign weights to different propagation paths through the attention mechanism to better understand the complexity of information propagation. Static features and dynamic features are also the target features to be extracted.

[0039] Step S203, input the target features into the target model to obtain the risk increment of the loan business, wherein the target model is trained by multiple groups of training samples, each group of training samples includes historical features and historical risk increments, and the risk increment is used to characterize the increase in risk of the loan business due to the spread of relevant information over time.

[0040] In step S203, historical loan data is collected, including detailed information of loan applicants (such as income, credit score, debt ratio, etc.), market assessment information (such as industry trends, interest rate environment, etc.), and historical characteristics such as historical risk indicators of loan business. At the same time, the historical risk increment is calculated, that is, the change in the risk level of the loan business within a specific time window, which can be achieved by comparing the differences in risk scores or default probabilities of the same loan at different time points.

[0041] Design a supervised learning model, such as a deep neural network or a long short-term memory network. The input of the model is a feature vector set, and the output is a risk increment prediction value. Use the collected training samples (historical features and historical risk increments) to train the model to learn the relationship between the risk increment of the loan business and the propagation of relevant information. The training process includes initializing the model parameters, forward propagation to calculate the prediction value, calculating the loss function, and backpropagating to update the model parameters until the model converges or reaches the preset training rounds.

[0042] The target features (i.e., the features of the current loan business and market assessment information) are input into the trained target model, and the model will output a risk increment prediction value. This prediction value reflects the amount by which the loan business risk may increase in the future based on current information and market conditions.

[0043] For example, the target model may be a multi-layer perceptron, and the target model may be expressed by the following formula:

[0044] △S=MLP(H out );

[0045] Among them, H out is the target feature of the input, MLP is the multi-layer perceptron model, and △S is the increment of the target model output.

[0046] Step S204, determining the propagation time of the relevant information, and determining the risk assessment value based on the propagation time and the risk increment.

[0047] In step S204, the timestamps of relevant information are collected. Based on these timestamps, a time series of information propagation is constructed. The time interval from the information source (such as loan application submission) to the subsequent event (such as loan approval, market trend change) is calculated. That is, the propagation duration.

[0048] Build a mathematical model or algorithm that combines the propagation time and the risk increment to assess the overall risk of the loan business. This can be a weighted average model, where the weights of the propagation time and the risk increment reflect their different contributions to the risk assessment value. Alternatively, design a risk assessment function that takes the propagation time and the risk increment as input and outputs the risk assessment value. For example, an exponential decay function can be used to capture the attenuation effect of the time on the risk impact, while combining the linear or nonlinear effects of the risk increment.

[0049] Step S205: When the risk assessment value is greater than or equal to the assessment value threshold, it is determined that the loan business has risks.

[0050] In step S205, the financial institution sets a risk assessment value threshold according to the risk tolerance. This threshold reflects the highest risk level that the institution can accept. The threshold can also take into account regulatory requirements to ensure that the loan business meets the risk control standards of financial supervision. The risk assessment value is compared with the set assessment value threshold. If the risk assessment value is greater than or equal to the threshold, it means that the risk level of the loan business exceeds the acceptable range of the financial institution. When the risk assessment value exceeds the assessment value threshold, it is confirmed that the loan business has a higher risk. For example, credit risk, market risk or operational risk.

[0051] Based on the risk assessment value, the loan business is divided into different risk levels, such as low risk, medium risk and high risk. For loan businesses with higher risk assessment values, financial institutions can adjust loan interest rates, require more collateral or reduce loan amounts to reduce risks. Conduct more frequent credit assessments and market analysis on high-risk loans to improve early warning capabilities. Reject loans: In the case of extremely high risk assessment values, financial institutions can not approve the loan to avoid potential losses. Based on macroeconomic conditions and market trends, financial institutions can regularly adjust the assessment value threshold to reflect more accurate risk assessment capabilities.

[0052] The risk prediction method for a loan business provided in an embodiment of the present application obtains relevant information about the loan business, wherein the relevant information includes at least one of the following: investment or financing information of the loan funds and market evaluation information of the loan object; constructs an information cascade graph of the relevant information based on the relevant information, and extracts target features from the information cascade graph, wherein the information cascade graph includes multiple nodes and edges, the nodes are used to represent users who receive or transmit relevant information, and the edges are used to represent the transmission relationship between users corresponding to the nodes; the target features are input into a target model to obtain a risk increment for the loan business, wherein the target model is trained by multiple groups of training samples, each group of training samples includes historical features and historical risk increments, and the risk increments are used to represent The amount of risk increase to the loan business caused by the spread of relevant information over time; determining the propagation time of the relevant information, and determining the risk assessment value based on the propagation time and the risk increment; when the risk assessment value is greater than or equal to the assessment value threshold, determining that the loan business is at risk, constructing an information cascade graph for the relevant information of the loan business, extracting target features from the graph, determining the risk increment based on the target features, and then predicting the risk of the loan business based on the risk increment and the propagation time of the relevant information, thereby achieving the purpose of capturing the accurate domain information characteristics in the relevant information of the loan business, thereby realizing the technical effect of improving the accuracy of risk prediction for the loan business, and then solving the technical problem of low accuracy of risk prediction for the loan business.

[0053] In order to determine whether there is a risk in a loan business, it is necessary to calculate a risk assessment value based on the propagation time and the risk increment. Optionally, in the risk prediction method for the loan business provided in the embodiment of the present application, determining the risk assessment value based on the propagation time and the risk increment includes: determining a time observation window when predicting the risk of the loan business, splitting the time observation window into N time intervals, where N is a positive integer; determining a discrete variable for each time interval, where the discrete variable is used to characterize the time attenuation effect within the time interval; determining a sub-risk increment corresponding to each time interval, calculating the product of the sub-risk increment, the interval value of the time interval and the discrete variable of the time interval, and obtaining a sub-risk assessment value for the time interval; calculating the sum of the sub-risk assessment values ​​of all time intervals and a preset error to obtain a risk assessment value.

[0054] In some examples, first, the financial institution needs to set a time observation window, which can be an evaluation period for the loan business, such as a week, a month, or a quarter. The selection of the time observation window can be based on the characteristics of the loan business, the periodicity of market fluctuations, and the availability of data. The time observation window is divided into N consecutive time intervals, where N is a positive integer and can be determined based on business needs and the accuracy of data analysis. For example, if the time observation window is one year and each time interval can be one month, then N = 12. Each time interval corresponds to a discrete variable, which is used to characterize the time decay effect within the time interval. The time decay effect reflects the phenomenon that the loan risk gradually weakens over time. The determination of discrete variables can be based on empirical rules or optimized through data analysis. For example, it can be determined by analyzing the risk change trend at different time points in historical loan data.

[0055] In each time interval, the risk increment is calculated based on the actual performance of the loan business and market assessment information. It can be a change in the probability of default, an adjustment in the credit rating, or any other quantitative indicator that represents a change in risk. For each time interval, the product of the sub-risk increment, the interval value of the time interval (such as days or hours), and the discrete variable of the time interval is calculated. This product reflects the risk increment adjusted for time decay, that is, the sub-risk assessment value in the time interval. The sub-risk assessment values ​​of all time intervals are added to the preset error (a constant used to reflect the uncertainty of the forecast) to obtain the risk assessment value of the entire time observation window.

[0056] For example, suppose the time observation window is [tl, to], use discrete time intervals to approximate the continuous time window, divide the time window into l independent time intervals, and each time window corresponds to a discrete variable λ = {λ△t, △t∈(1, 2, ..., l)} representing the time attenuation effect. Assuming that the corresponding time interval in time tj is △t, its calculation method is as follows:

[0057]

[0058] The calculation of risk assessment value can be expressed by the following formula:

[0059] Risk=∑Risk+λ △t ∑△S+ε

[0060] Among them, Risk is the risk assessment value, λ is the discrete variable, △t That is the time interval, △S is the sub-risk increment within the time interval, and ε is the preset error.

[0061] By calculating the risk assessment value based on the propagation time and the risk increment, this embodiment enables financial institutions to more carefully analyze the risk level of loan business over time, thereby optimizing their risk management strategies while taking into account the decay of time value. It can cope with the dynamic changes of loan risks and improve the timeliness and accuracy of risk management.

[0062] The sub-risk increment can be determined by the sub-information cascade graph. Optionally, in the risk prediction method for the loan business provided in the embodiment of the present application, determining the sub-risk increment corresponding to each time interval includes: extracting the sub-information cascade graph corresponding to each time interval from the information cascade graph; extracting the sub-target features of each sub-information cascade graph from the target features, and inputting the sub-target features into the target model to obtain the sub-risk increment corresponding to the time interval.

[0063] In some examples, the entire time observation window of the information cascade graph is divided into N consecutive time intervals. Each time interval is sufficient to capture the periodic changes in relevant information dissemination and market evaluation, while maintaining sufficient accuracy to assess risk increments. For each time interval, a sub-information cascade graph consisting of all information dissemination events occurring in the time interval is extracted from the information cascade graph. Sub-target features associated with each sub-information cascade graph are extracted from the target features. The sub-target features corresponding to each time interval are input into the target model, and the target model will output the sub-risk increment in each time interval, reflecting the changes in the loan business risk in the time interval.

[0064] By calculating the sub-risk increments, this embodiment allows financial institutions to finely evaluate the changes in the risk of loan business in each time interval based on the sub-information cascade diagram and sub-target characteristics, which helps financial institutions to adjust risk management strategies in real time to cope with dynamic changes in loan business risks. It can more accurately capture the immediate impact of information dissemination and market evaluation on loan business risks, and improve the timeliness and accuracy of risk prediction. At the same time, by independently evaluating each time interval, financial institutions can identify specific stages with higher risks, so as to take more targeted risk management measures.

[0065] After determining the information cascade graph, target features are extracted through a graph attention network. Optionally, in the risk prediction method for loan business provided in an embodiment of the present application, extracting target features from the information cascade graph includes: performing sequence sampling on the information cascade graph based on a timestamp to obtain a sub-information cascade graph sequence, wherein the timestamp is used to characterize the time of the last propagation of relevant information in each time interval, and the sub-information cascade graph sequence includes N sub-information cascade graphs, each sub-information cascade graph corresponding to a time interval; and capturing the features of each sub-information cascade graph from the sub-information cascade graph sequence through a preset graph attention network to obtain the target features.

[0066] In some examples, a time observation window is set, which can cover the entire loan business cycle. Based on the timestamp, the observation window is divided into N non-overlapping time intervals, and each time interval is defined as the time period from the end time of the previous time interval to the next timestamp. Then, the information state when the relevant information was last propagated in each time interval is extracted from the information cascade graph to form N sub-information cascade graphs, forming a sub-information cascade graph sequence.

[0067] The graph attention network automatically assigns weights to the neighbors of each node through the attention mechanism, thereby learning the feature representation of the node. Each sub-information cascade graph is input into the pre-trained graph attention network, and a feature vector is learned for each node in the sub-cascade graph. The vector integrates the properties of the node itself and the role and influence of the node in propagating information in the sub-cascade graph. From the feature vector output by the graph attention network, the representative features of each sub-information cascade graph are extracted, which can be achieved through aggregation functions (such as average, maximum or attention-weighted sum). These representative features reflect the changes in the risk status of the loan business in each time interval and are the basis of the target features.

[0068] The representative features of each sub-information cascade graph are combined with other key factors in the time interval (such as market conditions, policy changes, etc.) to form a comprehensive target feature vector. This vector not only contains the information dissemination characteristics of the loan business, but also integrates the external influencing factors in the time interval, which is an important input for risk assessment.

[0069] For example, Figure 3 is a schematic diagram of a sub-cascade graph sequence sampling structure provided according to an embodiment of the present application, such as Figure 3 As shown, suppose the information cascade graph of message m observed before time t0 is G(t0). As the input of the graph attention network, the information cascade graph can be expressed as gtj=(Utj, Etj, tj), Utj is the node set of the information cascade graph, Etj is the edge set of the information cascade graph, where tj is the timestamp of the last recorded propagation action (tj≤t0), the subject participating in the propagation of m before tj is the node ux∈Utji, Utji is the node set of the sub-information cascade graph of the i-th time interval, the propagation action from ux to uy is the edge (ux, uy)∈Etji, and Etji is the edge set of the sub-information cascade graph of the i-th time interval. Based on the timestamp, gtj is sequentially sampled, and G(t0) can be expressed as a sub-cascade graph sequence Gt0 = {gt1, ..., gtj, ... | tj≤t0}, where any sub-cascade graph is represented by its corresponding adjacency matrix ati (At0 = {at1, ..., atj, ... | tj≤t0}), where the matrix rows correspond to the node order and the matrix columns correspond to the edges connecting the nodes.

[0070] This embodiment extracts target features to obtain a target feature sequence that changes over time, and each feature vector represents the risk status of the loan business within a specific time interval. These features can then be input into the risk prediction model to predict the risk increment of the loan business at different time points, and then evaluate the risk level of the entire loan cycle. Effectively handling the nonlinearity and heterogeneity of information propagation has significant advantages in capturing the dynamic changes of loan risks in the financial market, and improving the accuracy and efficiency of risk prediction.

[0071] Optionally, in the risk prediction method for loan business provided in the embodiment of the present application, the features of each sub-information cascade graph are captured from the sub-information cascade graph sequence through a preset graph attention network, and the target features are obtained, including: determining the adjacency matrix of each sub-information cascade graph in the sub-information cascade graph sequence, wherein the rows in the adjacency matrix are used to characterize the node order in the sub-information cascade graph, and the columns in the adjacency matrix are used to characterize the edges connecting each node; for each sub-information cascade graph, determining the feature set of each node in the sub-information cascade graph; determining the adjacent nodes of each node in the sub-information cascade graph based on the adjacency matrix, and for each target node, calculating the attention coefficient based on the feature set of the target node and the feature set of the adjacent nodes of the target node; performing weighted summation on the feature set of the target node based on the attention coefficient to obtain the comprehensive features of the target node; combining the comprehensive features of each node in the sub-information cascade graph into a feature vector, combining the feature vectors of all sub-information cascade graphs to obtain a feature vector set, and determining the feature vector set as the target feature.

[0072] In some examples, for each sub-information cascade graph (gtj), an adjacency matrix (Atj) is constructed, where the rows of the matrix represent the order of nodes in the sub-information cascade graph, and the columns of the matrix represent the edges connected to each node. For node ux, if it propagates information m before time tj, then there will be non-zero values ​​in its rows, corresponding to the existence of the edge (ux,uy), indicating that the information is propagated from ux to uy. Determine the feature set of each node in each sub-information cascade graph. These features may include basic attributes of the node (such as the credit score of the loan applicant, the loan amount, etc.), the location information of the node in the cascade graph (such as the degree and centrality of the node, etc.), and the time interval information of information propagation between nodes.

[0073] Based on the graph attention network, for each target node ux, the attention coefficient (eij) between it and all adjacent nodes uy is calculated. The feature vectors of ux and uy are concatenated, and then the attention score is calculated through an attention mechanism function (such as a learnable function based on the LeakyReLU (a rectified linear unit function) activation function). The attention coefficient reflects the relative importance of the ux node to the uy node when propagating information. According to the calculated attention coefficient, the feature set of the target node ux is weighted and summed. For example, the attention coefficient is regularized using the Softmax (normalized exponential) function, and then the feature vector of the ux node is weighted according to the obtained attention coefficient to obtain the comprehensive feature vector (ht') of the ux node. The comprehensive feature vectors of each node in the sub-information cascade graph are combined into a feature vector set (H'). This set contains the comprehensive information of all nodes in the information cascade graph after being processed by the graph attention network. It not only pays attention to the characteristics of the node itself, but also incorporates the dynamic weights of the interactions between nodes. The feature vector sets (H') of all sub-information cascade graphs are combined to obtain a comprehensive target feature (Hout). This final feature vector can be generated using a pooling operation (such as global average pooling or global maximum pooling), which integrates the information of all sub-information cascade graphs within the entire observation time window for input into the risk prediction model.

[0074] For example, the evolutionary process of capturing and training cascade structures based on graph attention networks can include:

[0075] 1) The feature vector set of the node mapping is represented as H = {ht1, ..., htj, ...}, htl∈Rf, where f is used to represent the length of the node feature htl, the dimension of the matrix H is [tj, f], and R represents a real number set.

[0076] 2) Determine the adjacent node j of node i according to the adjacency matrix and calculate the attention coefficient:

[0077] eij=a([Whti||Whtj]), j∈Ni;

[0078] Among them, Whti represents mapping the f-dimensional vector hti to the f-dimensional space, Whtj represents mapping the f-dimensional vector htj to the f-dimensional space, and the features of node i and node j after parameter transformation are concatenated to obtain eij.

[0079] 3) After regularization with the Softmax function, the attention score is obtained:

[0080]

[0081] Among them, Ni is the set of neighbor nodes j of node i.

[0082] 4) Sum the feature weights according to the calculated attention coefficient (here we focus more on the single-head attention mechanism):

[0083]

[0084] where h' i It is the comprehensive feature of each node i output by the graph attention network, integrating the information of the neighborhood; σ() is the activation function, which can be a Softmax or Logistic Sigmoid (logistic regression) function. W represents the shared learnable parameters, which is a projection matrix. The convolution is used to simulate the projection transformation of the features of the original node i and its neighbor node j, and the two sets of projection parameters are expanded in dimension through the shared parameters.

[0085] Finally, we can get the feature vector set output by the graph attention layer:

[0086] H'={ht1',...htj',...};

[0087] The output is:

[0088]

[0089] Where N is the number of sub-cascade graphs, and hti' represents the new feature of node i after integrating the neighborhood information.

[0090] This embodiment extracts the feature input required by the deep learning model from the sub-information cascade graph sequence, fully considering the structure of the information propagation network and the temporal dynamic relationship between nodes, which is of great significance for predicting the risk level of loan business over time. It improves the accuracy and interpretability of the prediction.

[0091] In order to predict the risk increment, it is necessary to train the target model. Optionally, in the risk prediction method for the loan business provided in the embodiment of the present application, the target model is trained in the following manner: obtaining the historical risk prediction records of the loan business, extracting multiple historical related information and the historical risk increment of each related information from the historical risk prediction records; determining the historical characteristics of each historical related information, and determining the historical characteristics and historical risk increment of each historical related information as a group of training samples to obtain multiple groups of training samples; training a preset neural network model with multiple groups of training samples to obtain a target model.

[0092] In some examples, historical risk prediction records of loan businesses are collected. These records contain financial institutions' past assessment, monitoring and prediction data on loans, as well as information such as market conditions, economic indicators, and borrowers' credit history related to loans. From the historical risk prediction records, historical relevant information (such as the borrower's financial status, loan amount, market interest rate, etc.) and the corresponding historical risk increment (changes in the risk status of the loan business within a certain time interval) are extracted for each loan business. The historical risk increment can be determined by calculating the difference in risk scores or default probabilities at consecutive time points.

[0093] For each loan business, determine its historical characteristics, which include but are not limited to the borrower's basic information, loan conditions, market environment, etc. Combine the historical characteristics with the corresponding historical risk increments into a set of training samples. For example, if the historical characteristics of a loan business in a certain time interval are F, and the historical risk increment is △S, then the training sample can be expressed as (F, △S). Extract training samples from the historical information of multiple loan businesses to form multiple sets of training samples. Select a preset neural network model (such as a multi-layer perceptron, long short-term memory network, graph neural network, etc.), and input multiple sets of training samples into the model for training. The goal of the model is to learn the mapping relationship from historical characteristics to historical risk increments, so as to predict the future changes in the risk status of the loan business. Through continuous training and adjustment, until the prediction performance of the model reaches the expected standard (such as the loss function converges or reaches a certain prediction accuracy), the target model is obtained.

[0094] This embodiment uses historical data to train a neural network model that can predict the incremental risk of loan business, thereby providing data-driven support for early identification of loan risks and dynamic adjustment of risk strategies.

[0095] In order to ensure the accuracy of the target model's prediction, it is also necessary to update the target model based on the evaluation value of the target model. Optionally, in the risk prediction method for the loan business provided in the embodiment of the present application, after obtaining the target model, the method also includes: obtaining multiple groups of test samples, inputting each group of test samples into the target model to obtain predicted risk increments, wherein each group of test samples includes sample characteristics and actual risk increments; calculating the standard mean variance based on the predicted risk increments and actual risk increments of each group of test samples to obtain the evaluation value of the target model; when the evaluation value is less than the model evaluation value threshold, updating the target model based on the multiple groups of test samples.

[0096] In some examples, one or more groups of test samples are collected, each group of test samples includes sample features (such as historical features of loan business, multi-scale features of information cascade, etc.) and corresponding actual risk increments (i.e., actual changes in loan business risk within a certain time window). The collection of test samples must ensure that they are consistent with the training samples in distribution to avoid overfitting or underfitting of the model. The features of each group of test samples are input into the target model, and the model's learned parameters are used for prediction to obtain the predicted risk increment. The predicted risk increment reflects the change in loan business risk predicted by the target model based on the input features. For each group of test samples, the difference between the predicted risk increment and the actual risk increment is calculated, and then the squares of these differences are averaged and normalized to obtain the standard mean variance.

[0097] The calculated standard mean variance is used as the evaluation indicator of the target model. A lower standard mean variance indicates that the gap between the model's prediction results and the actual risk increment is small, and the model's prediction accuracy is high. The evaluation value threshold is a standard set in advance to determine whether the model's prediction performance reaches the expected minimum error level.

[0098] If the calculated standard mean variance is lower than the set evaluation value threshold, this indicates that the model has achieved an acceptable level of prediction accuracy. However, if the standard mean variance is higher than the threshold, it indicates that the model's prediction performance is poor and requires further tuning or retraining. When the standard mean variance is higher than the threshold, the model can be adjusted or retrained based on feedback information from multiple groups of test samples. For example, strategies such as parameter tuning (such as learning rate, number of hidden layer units, etc.), increasing or decreasing model complexity, and adjusting the data set (such as improving data quality through data enhancement or data cleaning). After adjusting or retraining the target model, it is necessary to use the test samples again to evaluate the model performance and repeat the training until the standard mean variance of the model stabilizes below the evaluation value threshold, indicating that the model's prediction performance meets the expected requirements.

[0099] This embodiment continuously monitors and optimizes the target model to ensure that it maintains high accuracy and reliability in predicting incremental risk of loan business, thereby providing financial institutions with more effective risk management tools.

[0100] It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and that, although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0101] Example 2

[0102] The embodiment of the present application also provides a risk prediction device for a loan business. It should be noted that the risk prediction device for a loan business in the embodiment of the present application can be used to execute the risk prediction method for a loan business provided in the embodiment of the present application. The risk prediction device for a loan business provided in the embodiment of the present application is introduced below.

[0103] According to an embodiment of the present application, a device for implementing the risk prediction method of the above-mentioned loan business is also provided. Figure 4 is a schematic diagram of a risk prediction device for a loan business provided in an embodiment of the present application, such as Figure 4 As shown, the device comprises:

[0104] The acquisition unit 401 is used to acquire relevant information of the loan business, wherein the relevant information includes at least one of the following: investment or financing information of the loan funds and market evaluation information of the loan object;

[0105] A construction unit 402 is used to construct an information cascade graph of the relevant information based on the relevant information, and extract target features from the information cascade graph, wherein the information cascade graph includes a plurality of nodes and edges, the nodes are used to represent users who receive or transmit the relevant information, and the edges are used to represent the transmission relationship between users corresponding to the nodes;

[0106] An input unit 403 is used to input the target feature into the target model to obtain the risk increment of the loan business, wherein the target model is trained by multiple sets of training samples, each set of training samples includes historical features and historical risk increments, and the risk increment is used to characterize the risk increment of the loan business caused by the propagation of relevant information over time;

[0107] A first determining unit 404 is used to determine the propagation time of the relevant information, and determine the risk assessment value based on the propagation time and the risk increment;

[0108] The second determining unit 405 is used to determine that the loan business has risks when the risk assessment value is greater than or equal to the assessment value threshold.

[0109] The risk prediction device for the loan business provided in the embodiment of the present application obtains relevant information of the loan business through an acquisition unit 401, wherein the relevant information includes at least one of the following: investment or financing information of the loan funds and market evaluation information of the loan object; a construction unit 402 constructs an information cascade graph of the relevant information based on the relevant information, and extracts target features from the information cascade graph, wherein the information cascade graph includes multiple nodes and edges, the nodes are used to represent users who receive or transmit relevant information, and the edges are used to represent the transmission relationship between users corresponding to the nodes; an input unit 403 inputs the target features into a target model to obtain a risk increment of the loan business, wherein the target model is trained by multiple groups of training samples, each group of training samples includes historical features and historical risk increments, and the risk increments are used to Characterize the increase in risk to the loan business caused by the spread of relevant information over time; the first determination unit 404 determines the propagation time of the relevant information, and determines the risk assessment value based on the propagation time and the risk increment; the second determination unit 405 determines that the loan business is at risk when the risk assessment value is greater than or equal to the assessment value threshold, constructs an information cascade graph for the relevant information of the loan business, extracts target features from the graph, determines the risk increment based on the target features, and then predicts the risk of the loan business based on the risk increment and the propagation time of the relevant information, thereby achieving the purpose of capturing accurate domain information characteristics in the relevant information of the loan business, thereby realizing the technical effect of improving the accuracy of risk prediction for the loan business, and then solving the technical problem of low accuracy of risk prediction for the loan business.

[0110] Optionally, in the risk prediction device for the loan business provided in the embodiment of the present application, the first determination unit 404 includes: a first determination module, used to determine the time observation window when predicting the risk of the loan business, and split the time observation window into N time intervals, where N is a positive integer; a second determination module, used to determine the discrete variables of each time interval, where the discrete variables are used to characterize the time attenuation effect within the time interval; a third determination module, used to determine the sub-risk increment corresponding to each time interval, calculate the product of the sub-risk increment, the interval value of the time interval and the discrete variable of the time interval, and obtain the sub-risk assessment value of the time interval; a calculation module, used to calculate the sum of the sub-risk assessment values ​​of all time intervals and the preset error to obtain a risk assessment value.

[0111] Optionally, in the risk prediction device for the loan business provided in the embodiment of the present application, the third determination module includes: a first extraction sub-module, used to extract a sub-information cascade graph corresponding to each time interval from the information cascade graph; a second extraction sub-module, used to extract sub-target features of each sub-information cascade graph from the target features, and input the sub-target features into the target model to obtain the sub-risk increment corresponding to the time interval.

[0112] Optionally, in the risk prediction device for the loan business provided in the embodiment of the present application, the construction unit 402 includes: a sampling module, which is used to perform sequence sampling on the information cascade graph based on the timestamp to obtain a sub-information cascade graph sequence, wherein the timestamp is used to represent the time when the relevant information was last propagated in each time interval, and the sub-information cascade graph sequence contains N sub-information cascade graphs, and each sub-information cascade graph corresponds to a time interval; a capture module, which is used to capture the features of each sub-information cascade graph from the sub-information cascade graph sequence through a preset graph attention network to obtain the target feature.

[0113] Optionally, in the risk prediction device for loan business provided in the embodiment of the present application, the capture module includes: a first determination submodule, used to determine the adjacency matrix of each sub-information cascade graph in the sub-information cascade graph sequence, wherein the rows in the adjacency matrix are used to characterize the node order in the sub-information cascade graph, and the columns in the adjacency matrix are used to characterize the edges connecting each node; a second determination submodule, used to determine the feature set of each node in the sub-information cascade graph for each sub-information cascade graph; a third determination submodule, used to determine the adjacent nodes of each node in the sub-information cascade graph based on the adjacency matrix, and for each target node, calculate the attention coefficient based on the feature set of the target node and the feature set of the adjacent nodes of the target node; a summation submodule, used to perform weighted summation of the feature set of the target node based on the attention coefficient to obtain the comprehensive features of the target node; a combination submodule, used to combine the comprehensive features of each node in the sub-information cascade graph into a feature vector, combine the feature vectors of all sub-information cascade graphs to obtain a feature vector set, and determine the feature vector set as the target feature.

[0114] Optionally, in the risk prediction device for the loan business provided in the embodiment of the present application, the device also includes: an extraction unit, used to obtain historical risk prediction records of the loan business, and extract multiple historical related information and the historical risk increment of each related information from the historical risk prediction records; a third determination unit, used to determine the historical characteristics of each historical related information, and determine the historical characteristics and historical risk increment of each historical related information as a group of training samples to obtain multiple groups of training samples; a training unit, used to train a preset neural network model through multiple groups of training samples to obtain a target model.

[0115] Optionally, in the risk prediction device for loan business provided in the embodiment of the present application, the device also includes: a testing unit, used to obtain multiple groups of test samples, input each group of test samples into the target model, and obtain predicted risk increments, wherein each group of test samples includes sample characteristics and actual risk increments; a calculation unit, used to calculate the standard mean variance based on the predicted risk increment and actual risk increment of each group of test samples to obtain an evaluation value of the target model; an updating unit, used to update the target model based on multiple groups of test samples when the evaluation value is less than a model evaluation value threshold.

[0116] It should be noted that the acquisition unit 401, the construction unit 402, the input unit 403, the first determination unit 404 and the second determination unit 405 correspond to steps S201 to S205 in Example 1, and the five units and the corresponding steps implement the same examples and application scenarios, but are not limited to the contents disclosed in the above-mentioned Example 1. It should be noted that the above-mentioned modules or units can be hardware components or software components stored in a memory (e.g., memory 104) and processed by one or more processors (e.g., processors 102a, 102b, ..., 102n), and the above-mentioned modules or units can also be part of the device and can be run in the computer terminal 10 provided in Example 1.

[0117] Example 3

[0118] An embodiment of the present application may provide an electronic device, Figure 5 is a structural block diagram of an electronic device according to an embodiment of the present application. Figure 5 As shown, the electronic device may include: one or more ( Figure 5 (only one is shown) processor 502, memory 504, storage controller, and peripheral interface, wherein the peripheral interface is connected to the radio frequency module, audio module and display.

[0119] Among them, the memory can be used to store software programs and modules, such as program instructions / modules corresponding to the methods and devices in the embodiments of the present application, and the processor executes various functional applications and data processing by running the software programs and modules stored in the memory, that is, realizing the above-mentioned method. The memory may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory may further include a memory remotely arranged relative to the processor, and these remote memories may be connected to the terminal via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0120] The processor can call the information and application stored in the memory through the transmission device to perform the following steps: obtain relevant information of the loan business, wherein the relevant information includes at least one of the following: investment or financing information of the loan funds and market evaluation information of the loan object; construct an information cascade graph of the relevant information based on the relevant information, and extract target features from the information cascade graph, wherein the information cascade graph includes multiple nodes and edges, the nodes are used to represent users who receive or transmit relevant information, and the edges are used to represent the transmission relationship between users corresponding to the nodes; input the target features into the target model to obtain the risk increment of the loan business, wherein the target model is trained by multiple groups of training samples, each group of training samples includes historical features and historical risk increments, and the risk increment is used to represent the increase in risk of the loan business due to the spread of relevant information over time; determine the propagation time of the relevant information, and determine the risk assessment value based on the propagation time and the risk increment; when the risk assessment value is greater than or equal to the assessment value threshold, determine that the loan business is risky.

[0121] The processor can also call the information and application programs stored in the memory through the transmission device to perform the following steps: determine the time observation window when predicting the risk of loan business, split the time observation window into N time intervals, where N is a positive integer; determine the discrete variables for each time interval, where the discrete variables are used to characterize the time decay effect within the time interval; determine the sub-risk increment corresponding to each time interval, calculate the product of the sub-risk increment, the interval value of the time interval and the discrete variable of the time interval, and obtain the sub-risk assessment value of the time interval; calculate the sum of the sub-risk assessment values ​​of all time intervals and the preset error to obtain the risk assessment value.

[0122] The processor can also call the information and applications stored in the memory through the transmission device to perform the following steps: extracting the sub-information cascade graph corresponding to each time interval from the information cascade graph; extracting the sub-target features of each sub-information cascade graph from the target features, and inputting the sub-target features into the target model to obtain the sub-risk increment corresponding to the time interval.

[0123] The processor can also call the information and application programs stored in the memory through the transmission device to perform the following steps: sequentially sample the information cascade graph based on the timestamp to obtain a sub-information cascade graph sequence, wherein the timestamp is used to represent the time when the relevant information was last propagated in each time interval, and the sub-information cascade graph sequence includes N sub-information cascade graphs, and each sub-information cascade graph corresponds to a time interval; and capture the features of each sub-information cascade graph from the sub-information cascade graph sequence through a preset graph attention network to obtain the target feature.

[0124] The processor can also call the information and application stored in the memory through the transmission device to perform the following steps: determine the adjacency matrix of each sub-information cascade graph in the sub-information cascade graph sequence, wherein the rows in the adjacency matrix are used to characterize the node order in the sub-information cascade graph, and the columns in the adjacency matrix are used to characterize the edges connected to each node; for each sub-information cascade graph, determine the feature set of each node in the sub-information cascade graph; determine the adjacent nodes of each node in the sub-information cascade graph based on the adjacency matrix, and for each target node, calculate the attention coefficient based on the feature set of the target node and the feature set of the adjacent nodes of the target node; perform weighted summation of the feature set of the target node based on the attention coefficient to obtain the comprehensive feature of the target node; combine the comprehensive features of each node in the sub-information cascade graph into a feature vector, combine the feature vectors of all sub-information cascade graphs to obtain a feature vector set, and determine the feature vector set as the target feature.

[0125] The processor can also call the information and application programs stored in the memory through the transmission device to perform the following steps: obtain the historical risk prediction records of the loan business, extract multiple historical related information and the historical risk increment of each related information from the historical risk prediction records; determine the historical characteristics of each historical related information, and determine the historical characteristics and historical risk increment of each historical related information as a group of training samples to obtain multiple groups of training samples; train the preset neural network model through the multiple groups of training samples to obtain the target model.

[0126] The processor can also call the information and application programs stored in the memory through the transmission device to perform the following steps: obtain multiple groups of test samples, input each group of test samples into the target model, and obtain predicted risk increments, wherein each group of test samples includes sample characteristics and actual risk increments; calculate the standard mean variance based on the predicted risk increments and actual risk increments of each group of test samples to obtain an evaluation value of the target model; and when the evaluation value is less than the model evaluation value threshold, update the target model based on the multiple groups of test samples.

[0127] An embodiment of the present application provides a method for obtaining relevant information of a loan business, wherein the relevant information includes at least one of the following: investment or financing information of the loan funds and market evaluation information of the loan object; constructing an information cascade graph of the relevant information based on the relevant information, and extracting target features from the information cascade graph, wherein the information cascade graph includes multiple nodes and edges, the nodes are used to represent users who receive or transmit relevant information, and the edges are used to represent the transmission relationship between users corresponding to the nodes; inputting the target features into a target model to obtain a risk increment of the loan business, wherein the target model is trained by multiple groups of training samples, each group of training samples includes historical features and historical risk increments, and the risk increment is used to represent the increase in risk of the loan business due to the spread of relevant information over time; determining the propagation time of the relevant information, and determining a risk assessment value based on the propagation time and the risk increment; and when the risk assessment value is greater than or equal to an assessment value threshold, determining a solution to the risk of the loan business. By constructing an information cascade graph for relevant information of the loan business, extracting target features from the graph, determining the risk increment based on the target features, and then predicting the risk of the loan business based on the risk increment and the propagation time of relevant information, the purpose of capturing accurate domain information characteristics in the relevant information of the loan business is achieved, thereby achieving the technical effect of improving the accuracy of risk prediction for the loan business, and then solving the technical problem of low accuracy of risk prediction for the loan business.

[0128] It can be understood by those skilled in the art that Figure 5 The structure shown is for illustration only, and the electronic device may also be a smart phone, a tablet computer, a PDA, a mobile Internet device (MID), a PAD or other terminal device. Figure 5 The structure of the electronic device is not limited. Figure 5 More or fewer components (such as network interfaces, display devices, etc.) shown in, or having Figure 5 Different configurations are shown.

[0129] A person of ordinary skill in the art can understand that all or part of the steps in the various methods of the above embodiments can be completed by instructing the hardware related to the terminal device through a program, and the program can be stored in a computer-readable storage medium, and the storage medium may include: a flash drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, etc.

[0130] Example 4

[0131] The embodiment of the present application further provides a storage medium. Optionally, in this embodiment, the storage medium can be used to store the program code executed by the risk prediction method for the loan business provided in the first embodiment.

[0132] Optionally, in this embodiment, the above storage medium may be located in any computer terminal in a computer terminal group in a computer network, or in any mobile terminal in a mobile terminal group.

[0133] The present application also provides a computer program product, which, when executed on a data processing device, is suitable for executing the steps of the risk prediction method for a loan business.

[0134] The serial numbers of the above-mentioned embodiments of the present application are for description only and do not represent the advantages or disadvantages of the embodiments.

[0135] In the above embodiments of the present application, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, please refer to the relevant description of other embodiments.

[0136] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. Among them, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of units or modules, which can be electrical or other forms.

[0137] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0138] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0139] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, in essence, or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions to enable a computer device (which can be a personal computer, a server or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), mobile hard disk, disk or optical disk and other media that can store program codes.

[0140] The above is only a preferred implementation of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.

Claims

1. A risk prediction method for a loan business, characterized in that: include: Acquiring relevant information about the loan business, wherein the relevant information includes at least one of the following: investment or financing information of the loan funds and market evaluation information of the loan object; Constructing an information cascade graph of the relevant information based on the relevant information, and extracting target features from the information cascade graph, wherein the information cascade graph includes a plurality of nodes and edges, the nodes are used to represent users who receive or spread the relevant information, and the edges are used to represent the spread relationship between users corresponding to the nodes; Input the target feature into the target model to obtain the risk increment of the loan business, wherein the target model is trained by multiple sets of training samples, each set of training samples includes historical features and historical risk increments, and the risk increment is used to characterize the risk increase of the loan business caused by the relevant information propagating over time; Determine a propagation time of the relevant information, and determine a risk assessment value based on the propagation time and the risk increment; When the risk assessment value is greater than or equal to the assessment value threshold, it is determined that the loan business is risky.

2. The method according to claim 1, characterized in that: Determining the risk assessment value based on the propagation duration and the risk increment includes: Determine a time observation window for predicting the loan business risk, and split the time observation window into N time intervals, where N is a positive integer; Determining a discrete variable for each time interval, wherein the discrete variable is used to characterize a time decay effect within the time interval; Determine the sub-risk increment corresponding to each time interval, calculate the product of the sub-risk increment, the interval value of the time interval and the discrete variable of the time interval, and obtain the sub-risk assessment value of the time interval; The sum of the sub-risk assessment values ​​of all time intervals and the preset error is calculated to obtain the risk assessment value.

3. The method according to claim 2, characterized in that Determine the sub-risk increment corresponding to each time interval including: Extracting a sub-information cascade graph corresponding to each time interval from the information cascade graph; The sub-target features of each sub-information cascade graph are extracted from the target features, and the sub-target features are input into the target model to obtain the sub-risk increment corresponding to the time interval.

4. The method according to claim 1, characterized in that: Extracting target features from the information cascade graph includes: Performing sequence sampling on the information cascade graph based on the timestamp to obtain a sub-information cascade graph sequence, wherein the timestamp is used to represent the time when the relevant information was last propagated in each time interval, and the sub-information cascade graph sequence includes N sub-information cascade graphs, each of which corresponds to a time interval; The features of each sub-information cascade graph are captured from the sub-information cascade graph sequence through a preset graph attention network to obtain the target feature.

5. The method according to claim 4, characterized in that The features of each sub-information cascade graph are captured from the sub-information cascade graph sequence through a preset graph attention network, and the target features obtained include: Determine an adjacency matrix of each sub-information cascade graph in the sub-information cascade graph sequence, wherein the rows in the adjacency matrix are used to represent the order of nodes in the sub-information cascade graph, and the columns in the adjacency matrix are used to represent the edges connected to each node; For each sub-information cascade graph, determining a feature set of each node in the sub-information cascade graph; Determine the adjacent nodes of each node in the sub-information cascade graph based on the adjacency matrix, and for each target node, calculate the attention coefficient based on the feature set of the target node and the feature set of the adjacent nodes of the target node; Performing weighted summation on the feature set of the target node based on the attention coefficient to obtain a comprehensive feature of the target node; The comprehensive features of each node in the sub-information cascade graph are combined into a feature vector, the feature vectors of all sub-information cascade graphs are combined to obtain a feature vector set, and the feature vector set is determined as the target feature.

6. The method according to claim 1, characterized in that The target model is trained in the following way: Obtaining historical risk prediction records of loan businesses, and extracting multiple pieces of historical related information and historical risk increments of each piece of related information from the historical risk prediction records; Determine a historical feature of each piece of historical related information, and determine the historical feature and the historical risk increment of each piece of historical related information as a set of training samples, to obtain multiple sets of training samples; The preset neural network model is trained by the multiple groups of training samples to obtain the target model.

7. The method according to claim 6, characterized in that After obtaining the target model, the method further includes: Acquire multiple groups of test samples, input each group of the test samples into the target model, and obtain predicted risk increments, wherein each group of test samples includes sample characteristics and actual risk increments; Calculating the standard mean variance based on the predicted risk increment and the actual risk increment of each group of test samples to obtain an evaluation value of the target model; When the evaluation value is less than the model evaluation value threshold, the target model is updated based on the multiple groups of test samples.

8. A risk prediction device for loan business, characterized in that: include: An acquisition unit, used to acquire relevant information of the loan business, wherein the relevant information includes at least one of the following: investment or financing information of the loan funds and market evaluation information of the loan object; A construction unit, configured to construct an information cascade graph of the relevant information based on the relevant information, and extract target features from the information cascade graph, wherein the information cascade graph includes a plurality of nodes and edges, the nodes are used to represent users who receive or spread the relevant information, and the edges are used to represent the spread relationship between users corresponding to the nodes; An input unit, used for inputting the target feature into a target model to obtain a risk increment of the loan business, wherein the target model is trained by multiple sets of training samples, each set of training samples includes historical features and historical risk increments, and the risk increment is used to characterize the risk increase of the loan business caused by the relevant information propagating over time; A first determining unit, configured to determine a propagation time of the relevant information, and determine a risk assessment value based on the propagation time and the risk increment; The second determining unit is configured to determine that the loan business has a risk when the risk assessment value is greater than or equal to an assessment value threshold.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium includes a stored executable program, wherein when the executable program is run, the device where the computer-readable storage medium is located is controlled to execute the risk prediction method for the loan business described in any one of claims 1 to 7.

10. An electronic device, characterized in that: include: A memory storing an executable program; A processor is used to run the program, wherein the program, when running, executes the risk prediction method for loan business described in any one of claims 1 to 7.

11. A computer program product comprising computer instructions, characterized in that: When the computer instructions are executed by the processor, the steps of the risk prediction method for loan business described in any one of claims 1 to 7 are implemented.

Images