CN117196808A

CN117196808A - Mobility risk prediction method and related device for peer business

Info

Publication number: CN117196808A
Application number: CN202311151111.8A
Authority: CN
Inventors: 陈灿然; 敖倩; 付小奇; 徐一茗; 汪雅丽
Original assignee: Bank of China Ltd
Current assignee: Bank of China Ltd
Priority date: 2023-09-07
Filing date: 2023-09-07
Publication date: 2023-12-08

Abstract

The application discloses a mobility risk prediction method and a related device for peer-to-peer business, which can be applied to the field of artificial intelligence, the field of big data or the field of finance. Firstly, the similarity between target normalized data corresponding to a reference user and reference normalized data corresponding to a business user is calculated to analyze whether the reference user similar to the business user exists or not. When the similarity is smaller than a preset similarity threshold, the fact that the reference user does not exist is indicated, a risk prediction model is called to conduct risk prediction on the target normalized data, a liquidity risk analysis result is obtained, and liquidity risk prediction of the business of the same industry is achieved. In addition, the risk prediction model is obtained based on training data, and during model training, the overfitting of the model can be relieved, the generalization capability of the model is improved, and the accuracy of risk prediction of the risk prediction model is further improved by configuring the main task sample and the auxiliary task sample.

Description

Mobility risk prediction method and related device for peer business

Technical Field

The application relates to the field of mobility risk prediction, in particular to a mobility risk prediction method and a related device for peer business.

Background

Liquidity risk refers to the risk that commercial banks, while having the ability to clear, cannot timely obtain sufficient funds or cannot timely obtain sufficient funds at reasonable cost to deal with the asset growth or pay for due liabilities. The generation of liquidity risks besides the possible imperfect liquidity plan of the commercial bank, the management defects in the risk fields of credit, market, operation and the like can also lead to insufficient liquidity of the commercial bank, even cause risk diffusion, and cause liquidity difficulty and even risk infection of the whole financial system. Thus, liquidity risk is one of the important risks faced by commercial banks.

Among the mobility risks, mobility risks of the business of the same industry are serious. If the possible mobility risk is predicted in time before the mobility risk of the business of the same industry occurs, the risk can be controlled and processed in advance. How to predict mobility risk of the peer business is a technical problem that needs to be solved by those skilled in the art.

Disclosure of Invention

In view of the above, the present invention discloses a method and a related device for predicting mobility risk of peer business, so as to solve the problem of urgent need for mobility risk prediction of peer business.

In order to solve the technical problems, the invention adopts the following technical scheme:

a mobility risk prediction method of a peer business comprises the following steps:

acquiring service information of business users of the same industry, and determining risk prediction key information based on the service information;

normalizing the risk prediction key information to obtain target normalized data; the target normalized data includes: attribute normalization data and behavior normalization data;

acquiring reference normalized data corresponding to a reference user of the same-industry service user, and calculating the similarity between the target normalized data and the reference normalized data;

if the similarity is smaller than a preset similarity threshold, a risk prediction model is called to perform risk prediction on the target normalized data, and a liquidity risk analysis result of the peer business users is obtained;

the risk prediction model is obtained based on training data; the training data comprises risk prediction attribute information samples and risk prediction behavior information samples; the risk prediction attribute information sample is configured as a primary task sample when training the risk prediction model, and the risk prediction behavior information sample is configured as a secondary task sample when training the risk prediction model.

Optionally, the service information includes user attribute information and user behavior information;

based on the business information, determining risk prediction key information comprises the following steps:

determining a space vector corresponding to the user attribute information and taking the space vector as a first space vector;

determining a space vector corresponding to the user behavior information and taking the space vector as a second space vector;

and taking the first space vector and the second space vector as the risk prediction key information.

Optionally, normalizing the risk prediction key information to obtain target normalized data, including:

acquiring an index quantization rule;

performing index quantization processing on the first space vector based on the index quantization rule to obtain first index quantized data, and performing index quantization processing on the second space vector to obtain second index quantized data;

determining attribute normalization data corresponding to the first index quantization data and behavior normalization data corresponding to the second index quantization data based on a normalization configuration rule;

and taking the attribute normalization data and the behavior normalization data as target normalization data.

Optionally, calculating the similarity between the target normalized data and the reference normalized data includes:

and determining the similarity between the target normalized data and the reference normalized data by adopting a K-means algorithm.

Optionally, invoking a risk prediction model to perform risk prediction on the target normalized data to obtain a liquidity risk analysis result of the peer business user, including:

configuring the attribute standardization data as primary task data and the behavior standardization data as secondary task data;

and inputting the main task data and the auxiliary task data into the risk prediction model to obtain a liquidity risk analysis result of the peer business user.

Optionally, the training process of the risk prediction model includes:

acquiring training data; the training data comprises risk prediction attribute information samples and risk prediction behavior information samples; the risk prediction attribute information sample is configured as a main task sample when the risk prediction model is trained, and the risk prediction behavior information sample is configured as an auxiliary task sample when the risk prediction model is trained;

and training the risk prediction model by adopting a multi-task learning mode and using the training data until a preset training stopping condition is reached.

Optionally, if at least one of the similarities is not less than a preset similarity threshold, the method further includes:

screening out the target reference users with the maximum similarity from all the reference users with the similarity not smaller than the preset similarity threshold;

and taking the liquidity risk analysis result of the target reference user as the liquidity risk analysis result of the peer business user.

A liquidity risk prediction apparatus for a peer service, comprising:

the information determining module is used for acquiring service information of the same-service users and determining risk prediction key information based on the service information;

the data processing module is used for carrying out standardization processing on the risk prediction key information to obtain target standardization data; the target normalized data includes: attribute normalization data and behavior normalization data;

the similarity calculation module is used for acquiring reference normalized data corresponding to the reference users of the peer business users and calculating the similarity between the target normalized data and the reference normalized data;

the risk analysis module is used for calling a risk prediction model to perform risk prediction on the target normalized data if the similarity is smaller than a preset similarity threshold value, so as to obtain a liquidity risk analysis result of the peer business users;

An electronic device comprising a memory and a processor;

the memory is used for storing at least one instruction;

the processor is configured to execute the at least one instruction to implement the liquidity risk prediction method described above.

A computer readable storage medium storing at least one instruction that when executed by a processor implements the liquidity risk prediction method described above.

Compared with the prior art, the invention has the following beneficial effects:

according to the technical scheme, the invention provides a mobility risk prediction method and a related device for the same business. In the invention, firstly, the similarity between the target normalized data corresponding to the reference user and the reference normalized data corresponding to the same-business user is calculated to analyze whether the reference user similar to the same-business user exists or not. When the similarity is smaller than a preset similarity threshold, the fact that the reference user similar to the same-business user does not exist is indicated, and at the moment, a risk prediction model is called to conduct risk prediction on the target normalized data, so that a liquidity risk analysis result of the same-business user is obtained, and liquidity risk prediction of the same-business is achieved. In addition, the risk prediction model is obtained based on training data, and during model training, the overfitting of the model can be relieved, the generalization capability of the model is improved, and the accuracy of risk prediction of the risk prediction model is further improved by configuring the main task sample and the auxiliary task sample.

Drawings

In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings that are required in the embodiments or the description of the prior art will be briefly described, and it is obvious that the drawings in the following description are only embodiments of the present invention, and that other drawings can be obtained according to the disclosed drawings without inventive effort for a person skilled in the art.

Fig. 1 is a flow chart of a method for predicting mobility risk of a peer business according to an embodiment of the present invention;

FIG. 2 is a schematic structural diagram of a risk prediction model according to an embodiment of the present invention;

FIG. 3 is a flowchart of a method for obtaining target normalized data according to an embodiment of the present invention;

fig. 4 is a schematic structural diagram of a mobility risk prediction apparatus for peer-to-peer business according to an embodiment of the present invention;

fig. 5 is a schematic structural diagram of an electronic device according to an embodiment of the present invention.

Detailed Description

The following description of the embodiments of the present invention will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present invention, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.

The technical terms used in the present invention will now be explained so that those skilled in the art can more clearly understand the present invention.

Big data: the data set with large scale which is greatly beyond the capability range of the traditional database software tool in the aspects of acquisition, storage, management and analysis has four large characteristics of massive data scale, rapid data circulation, various data types and low value density.

Multitasking learning (Multi-task learning): training data is derived from multiple tasks in the same domain, and task information and relationships between the multiple tasks are incorporated into the learning process. Focusing on only a single model may promote potential information for a target task, and by sharing parameters between different tasks to some extent, better results may be achieved. The performance of part of the tasks as primary tasks (performance) is primarily considered, the other tasks being auxiliary tasks (auxliarytask) which help improve the effect of the model fitting. In addition, the over fitting of the model can be relieved to a certain extent, and the generalization capability of the model is improved.

Business of the same industry: the method is characterized by comprising the following steps of carrying out various businesses which are developed by other financial institutions and take the same-industry funds as the core, wherein the business comprises the same-industry borrowing, the same-industry storage, the same-industry buying and selling (selling and buying) and the like, and the business investment business of the corresponding-money investment.

Risk infection: risk infection is the phenomenon that one or more banks risk going to a large number of other banks or the entire financial system.

Liquidity risk refers to the risk that commercial banks, while having the ability to clear, cannot timely obtain sufficient funds or cannot timely obtain sufficient funds at reasonable cost to deal with the asset growth or pay for due liabilities.

The importance of mobility risk management is gradually becoming a common knowledge of industry and supervision, mobility risks are found to have the characteristics of relevance, interconversion, transmission and coupling in research, and risk propagation channels are more complex, so that the conditions of cross-market and cross-field are increasingly prominent. Liquidity risks are more complex and extensive in terms of reasons than credit risks, market risks and operational risks, and are generally considered to be a comprehensive risk. The mobility risk is generated, and besides the possible imperfect mobility plan of the commercial bank, the management defects in the risk fields of credit, market, operation and the like can also lead to insufficient mobility of the commercial bank, even cause risk diffusion, so that mobility difficulty and even risk infection occur in the whole financial system. Thus, liquidity risk is one of the important risks faced by commercial banks.

At present, when the business bank predicts the liquidity risk, a small number of static management indexes are determined in advance, a calculation formula corresponding to the static management indexes is determined, data are manually obtained, an index value of the static management indexes is calculated based on the calculation formula, and a liquidity risk prediction result is obtained based on index value analysis.

The mobility risk prediction mode wastes manpower and is poor in user experience. The static management index only reflects the mobility risk condition of a bank in a certain historical time, belongs to post-detection, is not comprehensive, and is easy to occur with good index but large risk.

In order to solve the problems of manpower waste, incapability of timely predicting liquidity risks and incomplete indexes caused by manual analysis of liquidity risk prediction, the inventor discovers that various data of peer business users can influence a final liquidity risk result, so that all data which can influence the liquidity risk result can be included for analyzing liquidity risks, and comprehensiveness of liquidity risk analysis can be ensured.

In addition, the method and the system can predict the liquidity risk based on the current data of the business users of the same industry and reflect the liquidity risk in the future of the bank.

In addition, aiming at the problem of wasting manpower, a program corresponding to the mobility risk prediction method of the business in the same industry can be configured, and the program is operated on the electronic equipment to automatically perform mobility risk prediction, so that manpower is liberated, and the problem of low prediction accuracy caused by human subjectivity can be avoided.

In addition, in order to improve the prediction accuracy, the model obtained by training the training data can be used for carrying out liquidity risk prediction, and the accuracy of carrying out liquidity risk prediction by using the model is higher as the model is obtained by training a large amount of training data.

In order to improve the prediction efficiency, the method and the system firstly analyze whether the reference user similar to the business users of the same industry exists, and if so, take the liquidity risk analysis result of the reference user as the liquidity risk analysis result of the business users of the same industry. If not, the model is called for liquidity risk analysis.

Therefore, the embodiment of the invention provides a mobility risk prediction method and a related device for the same business. In the invention, firstly, the similarity between the target normalized data corresponding to the reference user and the reference normalized data corresponding to the same-business user is calculated to analyze whether the reference user similar to the same-business user exists or not. When the similarity is smaller than a preset similarity threshold, the fact that the reference user similar to the same-business user does not exist is indicated, and at the moment, a risk prediction model is called to conduct risk prediction on the target normalized data, so that a liquidity risk analysis result of the same-business user is obtained, and liquidity risk prediction of the same-business is achieved. In addition, the risk prediction model is obtained based on training data, and during model training, the overfitting of the model can be relieved, the generalization capability of the model is improved, and the accuracy of risk prediction of the risk prediction model is further improved by configuring the main task sample and the auxiliary task sample.

It should be noted that the mobility risk prediction method and the related device for the peer business provided by the invention can be used in the artificial intelligence field, the big data field or the financial field. The foregoing is merely exemplary, and the application fields of the mobility risk prediction method and the related device for the peer service provided by the present invention are not limited.

Based on the above, referring to fig. 1, a method for predicting mobility risk of a peer service may include:

s11, acquiring service information of the same-service users, and determining risk prediction key information based on the service information.

In practical application, the peer service user refers to a user configured with peer service, and in general, if there is peer service between two banks, the peer service user refers to the two banks.

The business information of the business users in the same industry refers to business data of business in the same industry (such as business lending, business storage, buying and selling (selling and buying) and the like) of the users in the same industry, and the business information of the business users in the same industry can be obtained from a database of a bank, and generally, the business information comprises user attribute information and user behavior information.

In this embodiment, the user attribute information and the user behavior information may be collected and integrated and transmitted to the electronic device executing the mobility risk prediction method of the peer service in this embodiment, or may be stored in the background system, where the subsequent electronic device obtains the data from the background system, and in addition, when the electronic device is a device configured with the background system, the device directly obtains the data from the data stored in the device. The background system can screen out useless data and repeated data and integrate effective data.

The user attribute information refers to business static data, which can also be called business content data, and is data not affected by user behaviors, such as basic information data including liquidity asset amount, liquidity liability amount, liquidity proportion, real receipts, asset basic structures (such as cash balance, deposit central bank deposit balance, other accounts payable balance, fixed asset equity, intangible asset balance and the like), liability conditions (such as deposit balances, unit deposit balances and the proportion of deposit occupied by the unit deposit balances, interest payable, stakeholder payable, other accounts payable and the like), withdrawal and risk compensation conditions (such as poor loan balance, poor yield, loan loss special preparation gold balance, capital net amount, weighted risk asset amount, capital sufficiency and the like), and personalized information data including risk preference. Wherein the risk preferences may e.g. be biased towards robustness, low risk etc., the risk preferences may be indicated with corresponding numerical identifiers, e.g. robustness with 1, low risk with 2 etc.

The user behavior information refers to business dynamic data, which can also be called business behavior data, and is data influenced by user behaviors, such as historical business data and historical homonymy product conditions. Historical business data such as: each business borrowing business and business deposit business (like business lending, business deposit, etc.), transaction flow information (such as transaction amount, transaction currency, transaction opponent certificate information, transaction channel, actual interest rate, country, region, administrative division of the transaction opponent location and registration place, transaction direction, transaction use, transaction date, cash transfer identification, etc.). Historical homography product conditions are as follows: the same industry stock issuing number, ticket face value, credit risk rating, actual interest rate, reference interest rate, the same industry stock investment number, balance, credit risk rating, actual interest rate, reference interest rate, and the like.

In this embodiment, the user behavior information is introduced for mobility risk prediction, and compared with the manner of using only static data to perform static management index calculation in the prior art, the influence of user behavior on mobility risk can be considered, the consideration factors are more comprehensive, and the accuracy of mobility risk prediction is further improved.

It should be noted that, the user information (including but not limited to user equipment information, user personal information, etc.) and the data (including but not limited to data for analysis, stored data, presented data, etc.) related to the present application are information and data authorized by the user or sufficiently authorized by each party, and the collection, use and processing of the related data need to comply with the related laws and regulations and standards of the related country and region.

After the service information is acquired, risk prediction key information is determined based on the service information.

Wherein the risk prediction key information is service information expressed in a vector form.

Specifically, the process of determining risk prediction key information is as follows:

1) And determining a space vector corresponding to the user attribute information and taking the space vector as a first space vector.

In this embodiment, the user attribute information is mapped to a multidimensional space to obtain a first space vector.

The user attribute information is exemplified by a part of the content. If the user attribute information includes liquidity balance, liquidity liability balance, liquidity proportion, real receipts, asset infrastructure (such as cash balance, deposit central bank funds balance, other accounts payable balance, fixed equity, intangible asset balance).

In this embodiment, the asset infrastructure includes five contents, which are then divided into one dimension for analysis.

The user attribute information includes nine information, namely, liquidity asset total, liquidity liability total, liquidity ratio, real receipts capital, cash balance, deposit central bank funds balance, other accounts payable balance, fixed asset equity, intangible asset balance.

If the liquidity balance sum is 10000, the liquidity liability sum is 5000, the liquidity proportion is 0.5, the real receipt capital is 3000, the cash balance is 2000, the deposit balance of the central bank is 1000, the other accounts receivable balances are 3000, the fixed asset net value is 1000, and the intangible asset balance is 1000.

When space mapping is performed, the numerical value of each dimension in the user attribute information is directly used as a vector dimension value mapped to the multidimensional space, namely, the space vector corresponding to the user attribute information is:

(10000,5000,0.5,3000,2000,1000,3000,1000,1000)

This space vector is the first space vector.

It should be noted that if the content of the data in a certain dimension is not a number but a text, the content is converted into a number identifier corresponding to the text, for example, the low risk is converted into 2.

2) And determining a space vector corresponding to the user behavior information and taking the space vector as a second space vector.

The process of determining the spatial vector corresponding to the user behavior information is similar to the process of determining the spatial vector corresponding to the user attribute information, please refer to the above description.

3) And taking the first space vector and the second space vector as the risk prediction key information.

Specifically, the risk prediction key information includes a first spatial vector and the second spatial vector, and in the risk prediction key information, the first spatial vector and the second spatial vector are used as two independent vectors, that is, vector addition and other operations are not performed on the first spatial vector and the second spatial vector.

In this embodiment, the user attribute information and the user behavior information are converted into space vectors, so that data which can be processed by a subsequent risk analysis model can be obtained, and subsequent risk analysis prediction is facilitated.

And S12, carrying out standardization processing on the risk prediction key information to obtain target standardization data.

Wherein the target normalized data comprises: attribute normalization data and behavior normalization data.

Specifically, the attribute normalization data is data obtained by normalizing the first space vector, and the behavior normalization data is data obtained by normalizing the second space vector.

In this embodiment, the normalization process refers to performing quantization processing on the first spatial vector and the second spatial vector, and converting the quantized result into a corresponding vector representation after the quantization processing.

By the normalization processing, the user data is quantized, and the complexity of subsequent data processing can be reduced.

For example, with the first space vector:

(10000,5000,0.5,3000,2000,1000,3000,1000,1000) by way of example.

The current data value of the first space vector is larger, the complexity is higher during the subsequent processing, at this time, the first dimension 10000 of the first space vector is taken as an example, the dimension represents the total amount of the liquidity assets, the total amount of the liquidity assets is set to be marked with 1 when not less than 100000 and marked with 0 when less than 100000, and the total amount of the liquidity assets 10000 can be adjusted to be 1 so as to reduce the subsequent data calculation complexity.

In this embodiment, the risk prediction key information may be normalized, for example, the liquidity asset total amount 10000 is adjusted to 1, because when the liquidity asset total amount is greater than 100000, the asset total amount may be considered to be larger, and no matter what the asset total amount is, the user belongs to the user with the larger asset total amount, so that a simple number may be used to represent larger data.

S13, obtaining reference normalized data corresponding to the reference users of the same-industry service users, and calculating the similarity between the target normalized data and the reference normalized data.

In this embodiment, the reference user refers to a user who has the same business as the business user and has the liquidity risk analysis result. For example, the peer business users are middle and construction lines, and the reference users can be middle and agriculture lines, middle and industry businesses, middle and rural credit agencies, etc.

When the reference users are determined, big data analysis can be adopted to analyze and obtain the reference users of the same business. The number of reference users may or may not be one, or may not be plural.

In this embodiment, the reference user is configured, because if the peer service user in this embodiment is similar to the reference user, the liquidity risk of the peer service user and the reference user are similar, and the liquidity risk analysis result of the reference user can be directly used as the liquidity risk analysis result of the peer service user, without calling the risk analysis model to perform liquidity risk analysis of the peer service user, so that the liquidity risk analysis flow is simplified, and the efficiency is improved.

If the reference users similar to the business users of the same industry do not exist, the business users of the same industry need to be subjected to liquidity risk analysis independently, and a risk analysis model is called to analyze liquidity risks of the business users of the same industry.

And when determining whether similar reference users exist, acquiring reference normalized data of the reference users, wherein the reference normalized data is similar to the target normalized data and also comprises attribute normalized data and behavior normalized data.

And when the similarity between the target normalized data and the reference normalized data is calculated, a K-means algorithm can be adopted to determine the similarity between the target normalized data and the reference normalized data.

Specifically, a K-means algorithm may be used to calculate the similarity of the attribute normalized data and the similarity of the behavior normalized data, and then the two similarities are weighted to obtain the final similarity.

Or combining the attribute standardization data and the behavior standardization data, and then calculating the similarity of the combined data, wherein the specific implementation mode can be set according to the actual scene.

In the embodiment, the K-means algorithm is adopted because the K-means algorithm can find K partitions which minimize the square error function value, the effect is good, and the total clustering time complexity can be reduced.

S14, judging whether the similarity is smaller than a preset similarity threshold value or not; if yes, go to step S15.

In this embodiment, when the number of reference users is one or more, the similarity of each reference user is obtained, and then compared with a preset similarity threshold. The preset similarity threshold is preset by a technician and is used for analyzing whether the similar reference users exist or not.

If each similarity is smaller than the preset similarity threshold, it indicates that there is no reference user similar to the business users, and step S15 is executed at this time.

If at least one similarity is not smaller than the preset similarity threshold, the existence of the reference users similar to the business users in the same industry is indicated. At this time, the reference user with the largest similarity among all the reference users with the similarity not smaller than the preset similarity threshold value can be selected and used as the target reference user, and then the liquidity risk analysis result of the target reference user is used as the liquidity risk analysis result of the peer business user.

Specifically, the similarity can be sequenced from a large order to a small order based on the numerical value of the similarity, after the sequencing is finished, the reference user corresponding to the maximum similarity is found out, the user is the user most similar to the business users of the same industry, the mobility risk of the user is the most similar, the mobility risk analysis result of the user is used as the mobility risk analysis result of the business users of the same industry, the mobility risk of the business users of the same industry is determined without calling a risk prediction model, the calculation resource for calling the analysis of the risk prediction model is saved, the waiting time of the user is also saved, and the mobility risk can be rapidly predicted.

It should be noted that when similar users exist, the peer service users and the similar users in the embodiment may be used as a type of peer service users, and risk prediction category labels may be marked for the type of peer service users.

S15, a risk prediction model is called to conduct risk prediction on the target normalized data, and a liquidity risk analysis result of the business users in the same industry is obtained.

Specifically, if the similarity is smaller than the preset similarity threshold, it is indicated that there is no reference user similar to the business users, and then a risk prediction model is required to be called to perform risk prediction on the target normalized data, so as to obtain a liquidity risk analysis result of the business users.

The risk prediction model in this embodiment may be a machine learning model or a deep learning model, where the machine learning model or the deep learning model in this embodiment adopts a multi-task learning mode, and the input of the model is the target normalized data, specifically, the attribute normalized data and the behavior normalized data.

In the multi-task learning mode, the attribute standardization data is static data which exists stably for the business users of the same industry, but the behavior standardization data is behavior data of the business users of the same industry, is continuously changed along with the change of the behavior of the users, belongs to dynamic data and has sporasiveness, so in the embodiment, the attribute standardization data is used as main task data, and the behavior standardization data is used as auxiliary task data.

Through the multi-task learning mode in this embodiment, the performance of the primary task is primarily considered, and through the primary task, the required features can be learned, and the secondary task can help improve the effect of model fitting. In addition, the auxiliary task can alleviate the overfitting of the model to a certain extent, and improves the generalization capability of the model.

The architecture of the risk prediction model adopting the multitask learning mode may refer to fig. 2, and it should be noted that fig. 2 is only a simple example of a model structure, which is used to explain the operation logic of the model, and the specific model structure needs to be configured according to the actual scenario.

The risk prediction model is trained based on a large amount of training data, and the specific training data comprises:

risk prediction attribute information samples and risk prediction behavior information samples.

The risk prediction attribute information sample refers to the attribute normalization data, and the risk prediction behavior information sample refers to the behavior normalization data.

In practical application, service information samples of a plurality of peer service users can be obtained in advance, and then the steps S11-S12 are executed to obtain corresponding attribute standardization data and behavior standardization data, wherein the attribute standardization data is used as a risk prediction attribute information sample, and the behavior standardization data is used as a risk prediction behavior information sample.

Because a large number of business information samples of the business users of the same industry are collected, a large number of risk prediction attribute information samples and risk prediction behavior information samples are used for model training.

When training the risk prediction model, firstly acquiring the training data, then adopting a multi-task learning mode, and training the risk prediction model by using the training data until reaching the preset training stopping condition.

In specific training, the risk prediction attribute information sample is configured as a main task sample when training the risk prediction model, and the risk prediction behavior information sample is configured as an auxiliary task sample when training the risk prediction model.

In detail, taking a risk prediction model as a deep learning model as an example, the risk prediction attribute information sample and the risk prediction behavior information sample respectively form two independent training targets, and the risk prediction attribute information sample is taken as a main task sample when the risk prediction model is trained, namely, risk prediction attribute analysis is taken as a main task. The risk prediction behavior information sample is used as an auxiliary task sample when the risk prediction model is trained, namely, the risk prediction behavior is analyzed to be used as an auxiliary task. And respectively establishing respective loss functions for the main task and the auxiliary task, taking the first layers of the deep neural network as a shared layer as supervision and guidance for a training model, sharing and expressing risk prediction attribute information samples and risk prediction behavior information samples, and jointly updating parameters according to gradients calculated by the two tasks in the stage of a gradient descent algorithm. The network is split at the last full-connection layer, and parameters corresponding to the loss are independently learned, so that fitting is better focused, and a mobility risk image of a representative peer business user is formed, wherein the mobility risk image is a final mobility risk analysis result.

In the training process of the model, if the preset training stopping condition is reached, the training is stopped, and if the loss function is smaller than the preset loss function value, or the training frequency reaches the set maximum training frequency, etc.

In addition, if the feature of the risk prediction behavior is to be mainly learned, the risk prediction behavior information sample may be used as a main task sample, and the risk prediction attribute information sample may be used as an auxiliary task sample.

In addition, the main task and the auxiliary task are not distinguished, the risk prediction attribute information sample and the risk prediction behavior information sample can be directly combined to obtain an information sample, and the information sample is used for directly training a model, so that the model does not use a multi-task learning mode any more.

In the embodiment, the risk prediction model in the multi-task learning mode is adopted, so that key features of the attribute standardization data can be learned, and the behavior standardization data can be used for avoiding overfitting, thereby improving the accuracy of model prediction.

After training to obtain a risk prediction model, performing risk prediction on the target normalized data by calling the risk prediction model, and when obtaining a liquidity risk analysis result of the peer business user, firstly configuring the attribute normalized data as main task data and configuring the behavior normalized data as auxiliary task data. And in the specific configuration, the configuration can be performed on a parameter configuration interface of the risk prediction model.

And then inputting the main task data and the auxiliary task data into the risk prediction model to obtain a liquidity risk analysis result of the peer business user. And displaying the liquidity risk analysis result on a business interface, and prompting the liquidity risk of the business users of the same industry.

In this embodiment, the liquidity risk analysis result may be a risk analysis result of each dimension in the business information, such as a large interest rate risk, a low liability risk, and the like. The liquidity risk analysis result is a liquidity risk portrait, and can represent the risk condition of the user. The liquidity risk analysis result can be predicted through the liquidity risk prediction model in the embodiment, the prediction accuracy can be improved, and further, peer business users with liquidity risks can be found in time, and liquidity risks are avoided in advance.

In another implementation manner of this embodiment, in order to verify the accuracy of the risk prediction model, when at least one similarity is not less than a preset similarity threshold, the liquidity risk analysis result of the target reference user is used as the liquidity risk analysis result of the peer business user, and then the target normalized data is input into the risk prediction model, so as to obtain a final liquidity risk analysis result.

And comparing the liquidity risk analysis result obtained through the risk prediction model with the liquidity risk analysis result of the target reference user, and if the liquidity risk analysis result is the same as the liquidity risk analysis result of the target reference user, indicating that the accuracy of the risk prediction model is higher. If the two models are different, the accuracy of the risk prediction model is not high, and retraining is needed.

At this time, the target normalized data is used as new training data, and is input into the risk prediction model, the risk prediction model is retrained, and the risk prediction model is continuously perfected through continuous training and optimization, so that the accuracy of the model is improved.

In this embodiment, first, by calculating the similarity between the target normalized data corresponding to the reference user and the reference normalized data corresponding to the peer service user, whether the reference user similar to the peer service user exists is analyzed. When the similarity is smaller than a preset similarity threshold, the fact that the reference user similar to the same-business user does not exist is indicated, and at the moment, a risk prediction model is called to conduct risk prediction on the target normalized data, so that a liquidity risk analysis result of the same-business user is obtained, and liquidity risk prediction of the same-business is achieved. In addition, the risk prediction model is obtained based on training data, and during model training, the overfitting of the model can be relieved, the generalization capability of the model is improved, and the accuracy of risk prediction of the risk prediction model is further improved by configuring the main task sample and the auxiliary task sample.

The above embodiment refers to normalizing the risk prediction key information to obtain target normalized data, and a specific implementation process will be described. Referring to fig. 3, "normalizing the risk prediction key information to obtain target normalized data" may include:

s21, acquiring an index quantization rule.

In practical applications, the index quantization rule specifies the standard of index quantization of each dimension (also referred to as index) in the business information, for example, the set liquidity asset total is higher when not less than 100000, the liquidity asset total is lower when less than 100000, the liquidity proportion is larger when not less than 0.5, the liquidity proportion is smaller when less than 0.5, and the like of other dimensions.

S22, performing index quantization processing on the first space vector based on the index quantization rule to obtain first index quantized data, and performing index quantization processing on the second space vector to obtain second index quantized data.

Specifically, since the first space vector and the second space vector are space vector representations of each dimension in the above-described service information, the index normalization can be performed for each dimension in the space vectors with reference to the index quantization standard of each dimension in the index quantization rule.

Such as (10000,5000,0.5,3000,2000,1000,3000,1000,1000) is adjusted to (higher, lower, higher, centered, lower, higher, lower, centered, lower) so that the traffic information has a quantization criterion.

S23, determining attribute normalization data corresponding to the first index quantization data and behavior normalization data corresponding to the second index quantization data based on a normalization configuration rule.

Specifically, since the model cannot recognize (higher, lower, higher, centered, lower, higher, lower, centered, lower) such text input, it is necessary to convert the text to a corresponding numerical value for model recognition and processing.

In this embodiment, the normalized configuration rule configures the correspondence between the text and the numerical value, for example, the higher is adjusted to 1, the lower is adjusted to 0, and the middle is adjusted to 2. The (higher, lower, higher, centered, lower, higher, lower, centered, lower) vector in this embodiment can be adjusted to (1,0,1,2,0,1,0,2,0), i.e., the vector with larger value is converted to the vector with smaller value by quantization and normalization, so as to reduce the complexity of the subsequent processing analysis.

It should be noted that, the normalized process in this embodiment is a labeling process, and the forming model can identify the processed basic label data.

S24, taking the attribute normalization data and the behavior normalization data as target normalization data.

Specifically, in the target normalized data, the attribute normalized data and the behavior normalized data exist as two independent data, and the calculation operations such as data addition and subtraction are not performed.

In this embodiment, the vector with a larger value in this embodiment is converted into the vector with a smaller value through quantization and normalization processing, so that the complexity of subsequent processing analysis is reduced.

Optionally, on the basis of the foregoing embodiment of the mobility risk prediction method of the peer service, another embodiment of the present invention provides a mobility risk prediction apparatus of the peer service, referring to fig. 4, may include:

the information determining module 11 is configured to obtain service information of a peer service user, and determine risk prediction key information based on the service information;

the data processing module 12 is configured to normalize the risk prediction key information to obtain target normalized data; the target normalized data includes: attribute normalization data and behavior normalization data;

the similarity calculation module 13 is configured to obtain reference normalized data corresponding to a reference user of the peer service user, and calculate similarity between the target normalized data and the reference normalized data;

The risk analysis module 14 is configured to invoke a risk prediction model to perform risk prediction on the target normalized data if the similarity is smaller than a preset similarity threshold value, so as to obtain a liquidity risk analysis result of the peer business user;

Further, the service information comprises user attribute information and user behavior information;

the information determining module 11 is configured to, when determining risk prediction key information based on the service information, specifically:

and determining a space vector corresponding to the user attribute information, determining a space vector corresponding to the user behavior information as a first space vector, and taking the first space vector and the second space vector as the risk prediction key information as a second space vector.

Further, the data processing module 12 is specifically configured to:

Acquiring an index quantization rule;

Further, the similarity calculation module 13 is specifically configured to:

Further, the risk analysis module 14 is specifically configured to:

and configuring the attribute standardization data as main task data, configuring the behavior standardization data as auxiliary task data, and inputting the main task data and the auxiliary task data into the risk prediction model to obtain a liquidity risk analysis result of the peer business users.

Further, the model training module is used for:

Further, the risk processing module is further included for:

and screening out target reference users with maximum similarity from all reference users with the similarity not smaller than the preset similarity threshold, and taking the liquidity risk analysis result of the target reference users as the liquidity risk analysis result of the peer business users.

It should be noted that, in the specific working process of each module in this embodiment, please refer to the corresponding description in the above embodiment, and no further description is given here.

Optionally, on the basis of the embodiment of the mobility risk prediction method and device of the peer service, another embodiment of the present invention provides an electronic device, where the electronic device includes a memory and a processor;

the memory is used for storing at least one instruction;

As shown in fig. 5, the present invention further provides an electronic device, which may include: a processor 1 and a memory 2;

wherein the processor 1 and the memory 2 complete communication with each other through the communication bus 3;

a processor 1 for executing at least one instruction.

And a memory 2 for storing at least one instruction, wherein the instruction is stored in a computer program in the memory 2, and the computer program is a program corresponding to the fluidity risk prediction method.

The processor 1 may be a central processing unit CPU or a specific integrated circuit ASIC (Application Specific Integrated Circuit) or one or more integrated circuits configured to implement embodiments of the present invention.

The memory 2 may comprise a high-speed RAM memory or may further comprise a non-volatile memory (non-volatile memory), such as at least one disk memory.

The processor executes at least one instruction to implement the mobility risk prediction method.

Optionally, on the basis of the foregoing embodiments of the mobility risk prediction method and apparatus for a peer service, another embodiment of the present invention provides a computer readable storage medium, where the computer readable storage medium stores at least one instruction, and the at least one instruction when executed by a processor implements the mobility risk prediction method described above.

Finally, it is further noted that relational terms such as first and second, and the like are used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Moreover, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one … …" does not exclude the presence of other like elements in a process, method, article, or apparatus that comprises the element.

In the present specification, each embodiment is described in a progressive manner, and each embodiment is mainly described in a different point from other embodiments, and identical and similar parts between the embodiments are all enough to refer to each other.

The previous description of the disclosed embodiments is provided to enable any person skilled in the art to make or use the present invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other embodiments without departing from the spirit or scope of the invention. Thus, the present invention is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. The method for predicting the mobility risk of the business of the same industry is characterized by comprising the following steps:

2. The liquidity risk prediction method of claim 1, wherein the business information comprises user attribute information and user behavior information;

3. The liquidity risk prediction method according to claim 2, wherein the normalizing the risk prediction key information to obtain target normalized data comprises:

acquiring an index quantization rule;

4. The liquidity risk prediction method of claim 1, wherein calculating the similarity of the target normalized data and the reference normalized data comprises:

5. The liquidity risk prediction method according to claim 1, wherein the step of calling a risk prediction model to perform risk prediction on the target normalized data to obtain a liquidity risk analysis result of the peer business user comprises the following steps:

6. The liquidity risk prediction method of claim 1, wherein the training process of the risk prediction model comprises:

7. The method of claim 1, further comprising, if at least one of the similarities is not less than a predetermined similarity threshold:

8. A mobility risk prediction apparatus for a peer service, comprising:

9. An electronic device comprising a memory and a processor;

the memory is used for storing at least one instruction;

the processor is configured to execute the at least one instruction to implement the liquidity risk prediction method of any one of claims 1-7.

10. A computer readable storage medium storing at least one instruction that when executed by a processor implements the liquidity risk prediction method of any one of claims 1-7.