CN115719275A

CN115719275A - Risk monitoring method and device based on transaction data and electronic equipment

Info

Publication number: CN115719275A
Application number: CN202211511535.6A
Authority: CN
Inventors: 陈镇发; 吕涛; 类维彬; 陈树勇
Original assignee: Industrial and Commercial Bank of China Ltd ICBC
Current assignee: Industrial and Commercial Bank of China Ltd ICBC
Priority date: 2022-11-29
Filing date: 2022-11-29
Publication date: 2023-02-28

Abstract

The invention discloses a risk monitoring method and device based on transaction data and electronic equipment, and relates to the field of artificial intelligence, wherein the method comprises the following steps: acquiring transaction data, wherein the transaction data is generated when a target customer uses a financial exchange rate type transaction system in the current time period; extracting transaction detail data and customer operation data related to each transaction product in the transaction data; inputting the transaction detail data and the customer operation data into a target model, and receiving a transaction behavior predicted value output by the target model, wherein at least a time decay function is fused into the target model, and the transaction behavior predicted value is used for representing the predicted occurrence value of the target customer on various transaction behaviors; and outputting alarm information under the condition that the transaction behavior estimated value exceeds the risk neutral index threshold value. The invention solves the technical problem that the risk monitoring method in the related technology can only realize the post-transaction statistics and can not predict the risk condition of the transaction behavior.

Description

Risk monitoring method and device based on transaction data and electronic equipment

Technical Field

The invention relates to the field of artificial intelligence, in particular to a risk monitoring method and device based on transaction data and electronic equipment.

Background

In the related art, in order to implement enterprise risk monitoring and early warning, a statistical report or a statistical batch method is generally used, and a related statistical threshold is manually set, so that whether enterprise risks conform to risk neutrality or not is identified afterwards under a risk neutrality rule which is manually set, and then early warning and intervention are performed afterwards. However, the early warning method has the following significant disadvantages: the lack of intelligence, only making post-transaction statistics, the inability to predict customer transaction behavior, the inability to identify risks ahead of time and to make prospective guidance to customers ahead of time, makes the market risk of the enterprise likely to translate into a financial institution's credit risk.

And another risk monitoring and early warning mode comprises the following steps: in the collaborative filtering algorithm, due to the existence of data sparseness and cold start problems and the limited information carried by the client behavior numerical value, and as the data scale is increasingly huge, the client scoring matrix is increasingly sparse, so that the prediction accuracy of the risk early warning method based on collaborative filtering needs to be improved.

In view of the above problems, no effective solution has been proposed.

Disclosure of Invention

The embodiment of the invention provides a risk monitoring method and device based on transaction data and electronic equipment, and aims to at least solve the technical problem that the risk monitoring method in the related technology can only realize transaction post statistics and cannot predict the risk condition of transaction behaviors.

According to an aspect of an embodiment of the present invention, there is provided a risk monitoring method based on transaction data, including: acquiring transaction data, wherein the transaction data is generated when a target customer uses a financial exchange rate type transaction system in the current time period, and the transaction data at least comprises the following steps: trading the product; extracting transaction detail data and customer operation data in the transaction data, wherein the transaction detail data at least comprises: a transaction time; inputting the transaction detail data and the customer operation data into an object model, and receiving a transaction behavior pre-evaluation value output by the object model, wherein at least a time attenuation function is integrated into the object model, the time attenuation function is used for representing the bias heat of the target customer to various transaction behaviors along with the advance of the transaction time, and the transaction behavior pre-evaluation value is used for representing a predicted occurrence value of the target customer to various transaction behaviors; and outputting alarm information under the condition that the transaction behavior predicted value exceeds a risk neutral index threshold value.

Optionally, the target model is trained by acquiring a plurality of historical transaction detail data and a plurality of historical operation data of all customers, and extracting a transaction data matrix in each historical transaction detail data and an operation data matrix in each historical operation data, wherein the transaction data matrix comprises: historical trading products, historical trading behaviors, historical trading volume, historical trading times and historical trading time periods, wherein the operation data matrix comprises the following components: the historical trading products, the historical trading behaviors, and the historical trading times; extracting the maximum value and the minimum value of the historical trading behavior of each customer in the historical trading period; based on the maximum numerical value and the minimum numerical value, adopting a dimensionless quantitative processing interval scaling strategy to carry out standardized processing on the transaction data matrix and the operation data matrix to obtain a historical transaction behavior occurrence numerical matrix; and constructing an initial behavior prediction model, and training the initial behavior prediction model by adopting the historical transaction behavior occurrence value matrix to obtain the target model.

Optionally, the step of training the initial behavior prediction model by using the historical transaction behavior occurrence value matrix to obtain the target model includes: decomposing the historical transaction behavior occurrence value matrix into a user hidden factor matrix and a transaction behavior hidden factor matrix by adopting a preset singular value decomposition algorithm; acquiring a user offset item, a transaction behavior offset item and an occurrence value average item from the historical transaction behavior occurrence value matrix; and integrating the user bias item, the transaction behavior bias item, the occurrence value average item, the time attenuation function, the transposition of the user hidden factor matrix and the transaction behavior hidden factor matrix into the initial behavior pre-estimation model to train and obtain the target model.

Optionally, obtaining the target model after training includes: carrying out initialization assignment on the user hidden factor matrix and the transaction behavior hidden factor matrix, and estimating the occurrence value of the specified transaction behavior which is not generated by the client by adopting the initial behavior estimation model to obtain a behavior initial estimation value; calculating an error value between the behavior initial estimated value and a real occurrence value of the specified transaction behavior; calculating a sum of squares of errors for all customers associated with the specified transaction behavior based on the error value for each customer; and training the initial behavior prediction model by adopting a gradient descent method so as to reduce the sum of squares of errors to a minimum value and obtain the target model.

Optionally, obtaining a user bias term from the historical transaction behavior occurrence value matrix includes: extracting the total number of customers from the historical transaction behavior occurrence value matrix; extracting a set of all transaction behaviors of each client from the historical transaction behavior occurrence value matrix; calculating the sum of values generated by all the clients conducting the set of transaction behaviors; calculating a user bias item of the customer based on the total number of the customers and the sum of values generated by the customer in the set of all transaction behaviors, wherein the user bias item is used for analyzing the bias heat of the customer on various transaction behaviors when operating different types of transaction products.

Optionally, obtaining a transaction behavior bias term from the historical transaction behavior occurrence value matrix includes: extracting the total number of the transaction behaviors from the historical transaction behavior occurrence value matrix; extracting each transaction behavior from the historical transaction behavior occurrence value matrix to carry out a set of all transaction behaviors; calculating the sum of numerical values generated by the set of all the transaction behaviors carried out by each transaction behavior; calculating a transaction behavior bias item of the customer based on the total number of transaction behaviors and the sum of values generated by all transaction behaviors performed by each transaction behavior, wherein the transaction behavior bias item is used for analyzing the bias heat of the customer on transaction products indicated by different transaction behaviors.

Optionally, when obtaining the occurrence value average item from the historical transaction behavior occurrence value matrix, the method includes: extracting the total number of matrix sets from the historical transaction behavior occurrence value matrix; extracting the sum of all occurrence values from the historical transaction behavior occurrence value matrix; and calculating the average value item of the occurrence values based on the total number of the matrix sets and the sum of all the occurrence values.

Optionally, before inputting the transaction detail data and the customer operation data into the goal model, the method further comprises: configuring all the transaction behaviors to be estimated; configuring time attenuation functions corresponding to product types of different transaction products, wherein the time attenuation functions are determined based on the absolute value of the difference between a time attenuation coefficient and product operation time, and the absolute value of the difference between the product operation time refers to the absolute value of the difference between the latest historical transaction time point and the current time point when the transaction action occurs; configuring the risk neutral indicator threshold.

According to another aspect of the embodiments of the present invention, there is also provided a risk monitoring method and apparatus based on transaction data, including: the acquisition unit is used for acquiring transaction data, wherein the transaction data is generated when a target customer uses a financial exchange rate type transaction system in the current time period, and the transaction data at least comprises the following steps: trading the product; an extracting unit, configured to extract transaction detail data and customer operation data in the transaction data, where the transaction detail data is associated with each transaction product, and the transaction detail data at least includes: a transaction time; the input unit is used for inputting the transaction detail data and the customer operation data into an object model and receiving a transaction behavior pre-evaluation value output by the object model, wherein at least a time attenuation function is blended into the object model, the time attenuation function is used for representing the bias heat of the target customer to various transaction behaviors along with the advance of the transaction time, and the transaction behavior pre-evaluation value is used for representing the predicted occurrence value of the target customer to various transaction behaviors; and the output unit is used for outputting alarm information under the condition that the transaction behavior estimated value exceeds a risk neutral index threshold value.

Optionally, the risk monitoring method device based on transaction data further includes: the system comprises a first acquisition subunit, a second acquisition subunit and a third acquisition subunit, wherein the first acquisition subunit is used for acquiring a plurality of historical transaction detail data and a plurality of historical operation data of all clients, and extracting a transaction data matrix in each historical transaction detail data and an operation data matrix in each historical operation data, and the transaction data matrix comprises: historical trading products, historical trading behaviors, historical trading volume, historical trading times and historical trading time periods, wherein the operation data matrix comprises: the historical transaction product, the historical transaction behavior, the historical transaction times; the first extraction subunit is used for extracting the maximum value and the minimum value of the historical trading behavior of each customer in the historical trading time period; the first processing subunit is used for processing the transaction data matrix and the operation data matrix in a standardized manner by adopting a dimensionless quantitative processing interval scaling strategy based on the maximum numerical value and the minimum numerical value to obtain a historical transaction behavior occurrence numerical value matrix; the first construction subunit is used for constructing an initial behavior estimation model, and training the initial behavior estimation model by adopting the historical transaction behavior occurrence value matrix to obtain the target model.

Optionally, the first building subunit comprises: the first decomposition module is used for decomposing the historical transaction behavior occurrence value matrix into a user hidden factor matrix and a transaction behavior hidden factor matrix by adopting a preset singular value decomposition algorithm; the first acquisition module is used for acquiring a user offset item, a transaction behavior offset item and an occurrence value average item from the historical transaction behavior occurrence value matrix; the first merging module is configured to merge the user bias item, the transaction behavior bias item, the occurrence value averaging item, the time decay function, the transpose of the user hidden factor matrix, and the transaction behavior hidden factor matrix into the initial behavior pre-estimation model to obtain the target model through training.

Optionally, the first merging module includes: the first estimation submodule is used for carrying out initialization assignment on the user hidden factor matrix and the transaction behavior hidden factor matrix and estimating the occurrence value of the appointed transaction behavior which is not generated by the client by adopting the initial behavior estimation model to obtain a behavior initial estimation value; a first calculation submodule, configured to calculate an error value between the initial predicted value of the action and a true occurrence value of the specified transaction action; a second calculation submodule for calculating a sum of squares of errors associated with the specified transaction activities for all customers based on the error value for each customer; and the first training submodule is used for training the initial behavior prediction model by adopting a gradient descent method so as to reduce the sum of squares of errors to a minimum value and obtain the target model.

Optionally, the first obtaining module includes: the first extraction submodule is used for extracting the total number of customers from the historical trading behavior occurrence numerical matrix; the second extraction submodule is used for extracting a set of all trading behaviors of each client from the historical trading behavior occurrence value matrix; the third calculation submodule is used for calculating the sum of numerical values generated by all sets of transaction behaviors of each client; and the fourth calculation submodule is used for calculating a user bias item of the client based on the total number of the client and the sum of values generated by the client when the client conducts the set of all transaction behaviors, wherein the user bias item is used for analyzing the bias heat of the client on various transaction behaviors when the client operates different types of transaction products.

Optionally, the first obtaining module further includes: the third extraction submodule is used for extracting the total number of the transaction behaviors from the historical transaction behavior occurrence numerical matrix; the fourth extraction submodule is used for extracting each transaction behavior from the historical transaction behavior occurrence value matrix to carry out a set of all transaction behaviors; the fifth calculation submodule is used for calculating the sum of numerical values generated by the set of all the transaction behaviors carried out by each transaction behavior; and the sixth calculation submodule is used for calculating a transaction behavior bias item of the customer based on the total number of the transaction behaviors and the sum of values generated by all transaction behaviors in a set during each transaction behavior, wherein the transaction behavior bias item is used for analyzing the bias heat of the customer on transaction products indicated by different transaction behaviors.

Optionally, the first obtaining module further includes: a fifth extraction submodule, configured to extract a total number of matrix sets from the historical transaction behavior occurrence value matrix; a sixth extraction submodule, configured to extract a sum of all occurrence values from the historical transaction behavior occurrence value matrix; and the seventh calculation submodule is used for calculating the average value item of the occurrence values based on the total quantity of the matrix set and the sum of all the occurrence values.

Optionally, the input unit includes: the first configuration subunit is used for configuring all the transaction behaviors to be estimated; the second configuration subunit is used for configuring a time attenuation function corresponding to product types of different transaction products, wherein the time attenuation function is determined based on an absolute value of a difference between a time attenuation coefficient and product operation time, and the absolute value of the difference between the product operation time refers to an absolute value of a difference between a latest historical transaction time point and a current time point when the transaction action occurs; a third configuration subunit for configuring the risk neutral indicator threshold.

According to another aspect of the embodiments of the present invention, there is also provided a computer-readable storage medium, which includes a stored computer program, wherein when the computer program runs, the computer-readable storage medium is controlled to execute any one of the above risk monitoring methods based on transaction data.

According to another aspect of embodiments of the present invention, there is also provided an electronic device, comprising one or more processors and a memory for storing one or more programs, wherein the one or more programs, when executed by the one or more processors, cause the one or more processors to implement any one of the above-mentioned methods for risk monitoring based on transaction data.

In the present disclosure, the following steps are employed: acquiring transaction data, wherein the transaction data is generated when a target customer uses the financial exchange rate type transaction system in the current time period, and the transaction data at least comprises the following steps: the transaction data at least comprises the following transaction detail data: and transaction time, inputting transaction detail data and customer operation data into a target model, and receiving a transaction behavior pre-evaluation value output by the target model, wherein at least a time attenuation function is integrated into the target model, the time attenuation function is used for representing the bias heat of the target customer to various transaction behaviors along with the advance of the transaction time, the transaction behavior pre-evaluation value is used for representing the predicted occurrence value of the target customer to various transaction behaviors, and alarm information is output under the condition that the transaction behavior pre-evaluation value exceeds a risk neutral index threshold value.

According to the method and the system, the transaction behavior risk is predicted based on transaction data and transaction time, intelligent service is provided for a risk monitoring system, advance prediction and warning of transaction behavior are achieved, risks are recognized in advance, prospective guidance for customers is made, market risks of the customers are prevented from being converted into credit risks of financial institutions, and the technical problem that risk conditions of transaction behavior cannot be predicted due to the fact that the risk monitoring method in the related technology can only achieve transaction statistics after the transaction is finished is solved.

In the method, the characteristic extraction index and the prediction index can be set and adjusted according to actual business needs, so that a monitoring method and a monitoring system which can be adjusted according to actual needs, are intelligent, can predict future transaction behaviors of clients and can perform risk early warning are constructed.

In the disclosure, a settable time decay function is introduced to be combined into an SVD (Singular Value Decomposition) algorithm, so as to improve the accuracy of risk prediction.

Drawings

The accompanying drawings, which are included to provide a further understanding of the invention and are incorporated in and constitute a part of this application, illustrate embodiment(s) of the invention and together with the description serve to explain the invention without limiting the invention. In the drawings:

FIG. 1 is a flow diagram of an alternative transaction data based risk monitoring method according to an embodiment of the present invention;

FIG. 2 is a flow diagram of an alternative method of constructing a risk monitoring model according to an embodiment of the invention;

FIG. 3 is an alternative risk monitoring system architecture diagram based on transactional data in accordance with an embodiment of the invention;

FIG. 4 is a schematic diagram of an alternative transaction data based risk monitoring device according to an embodiment of the present invention;

fig. 5 is a block diagram of a hardware structure of an electronic device (or mobile device) of a risk monitoring method based on transaction data according to an embodiment of the present invention.

Detailed Description

In order to make the technical solutions of the present invention better understood, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.

It should be noted that the terms "first," "second," and the like in the description and claims of the present invention and in the drawings described above are used for distinguishing between similar elements and not necessarily for describing a particular sequential or chronological order. It is to be understood that the data so used is interchangeable under appropriate circumstances such that the embodiments of the invention described herein are capable of operation in sequences other than those illustrated or described herein. Furthermore, the terms "comprises," "comprising," and "having," and any variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, system, article, or apparatus that comprises a list of steps or elements is not necessarily limited to those steps or elements expressly listed, but may include other steps or elements not expressly listed or inherent to such process, method, article, or apparatus.

It should be noted that the risk monitoring method and apparatus based on transaction data in the present disclosure may be used in the artificial intelligence field for performing risk monitoring and early warning on transaction behaviors, and may also be used in any field except the artificial intelligence field for performing risk monitoring and early warning on transaction behaviors, and the application fields of the risk monitoring method and apparatus based on transaction data in the present disclosure are not limited.

It should be noted that the relevant information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for presentation, analyzed data, etc.) referred to in the present disclosure are information and data authorized by the user or sufficiently authorized by each party. For example, an interface is provided between the system and the relevant user or organization, before obtaining the relevant information, an obtaining request needs to be sent to the user or organization through the interface, and after receiving the consent information fed back by the user or organization, the relevant information is obtained.

In the actual operation process of an enterprise, contract signing, payment receipt and foreign currency fund exchange often do not occur simultaneously, a period of several months or even longer may be needed, the fluctuation of the exchange rate can affect the operation result of a client, and exchange rate risks are easily generated, so that a financial institution is needed to monitor and early warn the operation state and the exchange rate risks of each client.

The invention needs to realize the risk early warning of financial institutions to each client and provide risk neutral assessment service, wherein risk neutral refers to that clients (enterprises or individuals) bring exchange rate fluctuation into daily financial decisions, focus on main business and reduce the negative influence of the exchange rate fluctuation on main business and enterprise finance as much as possible so as to realize budget achievement, improve operation predictability, manage investment risk and other main business targets. When the financial institution carries out exchange rate risk management product marketing, the financial institution needs to use the compliance operation as a root point to realize risk neutral monitoring, and meanwhile, continuous monitoring and dynamic management are well done in the service life, so that the self risk of the financial institution caused by the speculative behavior of a client is avoided.

It should be noted that the exchange rate risk exposure sources in the following embodiments of the present invention include, but are not limited to: the balance of export money collection, import money payment, foreign currency deposit and other assets, foreign currency loan/bond and other debts, overseas investment, statement conversion and the like.

Meanwhile, a Latent semantic Model is introduced in the invention, neutral monitoring and early warning of foreign exchange risks of customers are realized through the Latent semantic Model integrated with a time context, wherein the Latent semantic Model (Latent Factor Model) is a recommendation Model with better prediction effect and larger application scene, and as the name suggests, the Latent semantic Model is the core of the Model and is specifically embodied in that the Model can map customers and transaction products into a shared Latent feature vector space (Latent vector space) according to a scoring matrix between the customers and the transaction products, the customers and the transaction behaviors are respectively represented by using the Latent feature vector (Latent vector), then the transaction behaviors of the customers are modeled on the basis of the Latent feature vector, so that the prediction of the transaction behaviors is completed, and finally, the system can perform early warning according to the occurrence value of the predicted transaction behaviors and the business needs. On the basis of time context information integrated with transaction behaviors of customers, the business can maintain the correlation degree of the time information on a system (for foreign exchange market quotations, the quotation change speed is very high, and the time is an important influence factor for monitoring and forecasting the foreign exchange risk neutral), so that the monitoring and early warning operation is more flexible, operable and accurate.

The method can be applied to various risk monitoring systems/equipment/devices/products (such as various mobile terminals, PC terminals and financial institutions APP), the risk of the transaction behavior is predicted based on transaction data and transaction time, intelligent service is provided for the risk monitoring system, advance prediction and warning of the transaction behavior are realized, the risk is recognized in advance, and prospective guidance for the client is made, so that the problems that the market risk of the client is converted into the credit risk of the financial institutions and the like are avoided.

The present invention will be described in detail with reference to examples.

Example one

In accordance with an embodiment of the present invention, there is provided an embodiment of a method for risk monitoring based on transactional data, it is noted that the steps illustrated in the flowchart of the drawings may be performed in a computer system such as a set of computer-executable instructions and that, although a logical order is illustrated in the flowchart, in some cases the steps illustrated or described may be performed in an order different than here.

Fig. 1 is a flow chart of an alternative risk monitoring method based on transaction data according to an embodiment of the present invention, as shown in fig. 1, the method includes the following steps:

step S101, transaction data is obtained, wherein the transaction data is generated when a target customer uses a financial exchange rate transaction system in the current time period, and the transaction data at least comprises the following steps: trading the product;

step S102, transaction detail data and customer operation data which are related to each transaction product in the transaction data are extracted, wherein the transaction detail data at least comprise: a transaction time;

step S103, inputting the transaction detail data and the customer operation data into a target model, and receiving a transaction behavior pre-evaluation value output by the target model, wherein at least a time attenuation function is fused into the target model, the time attenuation function is used for representing the bias heat of the target customer to various transaction behaviors along with the advance of transaction time, and the transaction behavior pre-evaluation value is used for representing the predicted occurrence value of the target customer to various transaction behaviors;

and step S104, outputting alarm information under the condition that the predicted value of the transaction behavior exceeds the risk neutral index threshold value.

Through the steps, transaction data are firstly obtained, wherein the transaction data are generated when the target customer uses the financial exchange rate transaction system in the current time period, and the transaction data at least comprise the following steps: the transaction data at least comprises the following transaction detail data: and (3) transaction time, inputting transaction detail data and customer operation data into a target model, and receiving a transaction behavior pre-evaluation value output by the target model, wherein at least a time attenuation function is fused into the target model, the time attenuation function is used for representing the bias heat of the target customer to various transaction behaviors along with the advance of the transaction time, the transaction behavior pre-evaluation value is used for representing the predicted occurrence value of the target customer to various transaction behaviors, and finally, alarm information is output under the condition that the transaction behavior pre-evaluation value exceeds a risk neutral index threshold value. In the embodiment, the risk of the transaction behavior is predicted based on the transaction data and the transaction time, intelligent service is provided for a risk monitoring system, advance prediction and warning of the transaction behavior are realized, the risk is recognized in advance, and prospective guidance is made for a client, so that the market risk of the client is prevented from being converted into the credit risk of a financial institution, and the technical problem that the risk condition of the transaction behavior cannot be predicted because the risk monitoring method in the related technology can only realize statistics after the transaction is solved.

The following is a detailed description of the above embodiments.

In the embodiment of the invention, before inputting the transaction detail data and the customer operation data into the target model, the method comprises the following steps: configuring all transaction behaviors to be estimated; configuring time attenuation functions corresponding to product types of different transaction products, wherein the time attenuation functions are determined based on absolute values of differences between time attenuation coefficients and product operation times, and the absolute value of the difference between the product operation times refers to the absolute value of a difference value between a latest historical transaction time point and a current time point when a transaction action occurs; and configuring a risk neutral index threshold value.

It should be noted that, in order to monitor and predict the risk, a transaction behavior relationship between the customer and a specific transaction product needs to be established first, and a transaction behavior occurrence value matrix is established for subsequent feature extraction. In this embodiment, when configuring all transaction behaviors to be estimated, a service person sets a series of transaction behaviors of a transaction product related to risk neutrality or capable of performing risk neutrality determination according to experience and a current product tradable type, wherein dimensions such as a transaction product (type), a transaction contract element, a statistical element and the like are provided through a system page for the service person to set up in a combined manner, and the transaction contract requirements include but are not limited to a transaction direction, a currency pair, a delivery mode and the like, for example, the following transaction behavior indexes are set: the amount of forward sale remittance dollar remittance/remittance transaction (amount), the forward sale remittance harbor coin remittance/remittance hanging order booking stroke number, the forward sale remittance Euro bill hanging order remittance/remittance cancelling stroke number, the forward sale remittance dollar remittance/remittance contract balancing stroke number, and the like.

In this embodiment, a time decay function (denoted by f (Δ t) in this embodiment) is configured to be a parameter that represents a change in a transaction behavior occurrence value of a transaction product according to a change between an occurrence time and a current time, and a smaller difference between an occurrence time of a latest transaction (for example, a forward sale/settlement/receipt reservation) in the transaction behavior of the transaction product and the current time is, a higher popularity of the market quotation representing the current time is, and a larger value of the transaction behavior in the future is considered to be. For example, if the domestic export transaction at the current time is high in popularity and rapid in growth, the client order contract can estimate that the foreign exchange income increases in a future period of time, the enterprise client can perform the hedging transaction behavior related to the closing of the sales remittance in order to avoid the risk of exchange rate, and the difference between the transaction behavior occurrence time of the current specific product and the current time is smaller.

The time decay function is formulated as follows:

in this embodiment, a time attenuation coefficient (represented by α in this embodiment) needs to be set, and the time attenuation coefficient can be set according to a transaction behavior indication set and current market judgment, and the system provides a function of recording the coefficient; and delta t is the absolute value of the difference between the latest transaction time and the current time of the transaction behavior of the transaction product. The time decay function formula uses an exponential formula to express that the heat of the transaction behavior gradually decays in an exponential form along with the time, wherein alpha is greater than 0,f (0) =1 represents the initial heat of the transaction behavior and decreases in an exponential form along with the time.

It should be noted that, when setting the threshold of the risk neutral indicator (product, contract trading element), the set threshold may be set according to the types of percentage, 0-1 intermediate value, etc., and the threshold parameter may be set to a positive percentage value, for example, the transaction amount of the derivative is greater than the historical average variation value, the reverse leveling frequency is greater than the historical average variation value, and the exhibition frequency is less than 30% of the historical average variation value (in the embodiment of the present invention, the threshold of the risk neutral indicator is not specifically limited, and may be set according to actual application requirements), for example, the number of reverse leveling times of forward dollar settlement trading of the enterprise customer is set to exceed 30% of the historical variation.

The system provides products, transaction directions, transaction types, stroke numbers, transaction amounts and operation actions for business personnel to set with threshold value combinations.

It should be noted that after the transaction behavior, the time decay function, and the risk neutral index threshold are configured, a risk monitoring model needs to be constructed before the user transaction behavior is predicted and alarmed.

Fig. 2 is a flowchart of an alternative method for constructing a risk monitoring model according to an embodiment of the present invention, and as shown in fig. 2, the method for constructing a risk monitoring model includes the following steps:

step S201, obtaining a plurality of historical transaction detail data and a plurality of historical operation data of all customers, and extracting a transaction data matrix in each historical transaction detail data and an operation data matrix in each historical operation data, where the transaction data matrix includes: historical trading products, historical trading behaviors, historical trading volume, historical trading times and historical trading time periods, wherein the operation data matrix comprises: historical transaction products, historical transaction behaviors and historical transaction times;

it should be noted that, when constructing the risk monitoring model, a large amount of historical data is required as a basis for model construction. The method comprises the steps of relying on the existing transaction system, recording transaction detail data of all client exchange rate type product transactions and operation data of clients in a client transaction process, synchronizing the transaction detail data and the operation data to an offline training system in real time when a model is built, calculating data of the clients in transaction behavior indexes set by businesses through the transaction detail data, and carrying out data standardization processing.

Specifically, when the risk monitoring model is constructed, the transaction behavior, the number of times of transaction behavior (product introduction clicking), the duration of transaction behavior (market browsing), and the like of a specific transaction product in a certain historical time period can be counted.

Step S202, extracting the maximum value and the minimum value of the historical trading behavior of each client in the historical trading time period;

step S203, based on the maximum value and the minimum value, adopting a dimensionless quantitative processing interval scaling strategy to carry out standardized processing on a transaction data matrix and an operation data matrix to obtain a historical transaction behavior occurrence value matrix;

and S204, constructing an initial behavior estimation model, and training the initial behavior estimation model by adopting a historical transaction behavior occurrence value matrix to obtain a target model.

It should be noted that when a numerical matrix of historical transaction behaviors of a user is obtained, taking statistics of forward sales/remittance transaction amounts (amounts) of each client in the last quarter as an example, a client U1 transacts 200 ten thousand dollars in the last quarter for forward sales/remittance dollar/remittance transaction amount A1 and 100 transactions in the last quarter for forward sales/remittance dollar/remittance amount A2, introduces an introduction A3 to forward sales/remittance dollar/remittance products that has been clicked 500 times in the last quarter, and browses A4 to forward sales/remittance dollar product quotations for 500 seconds; the enterprise client U2 transacts the forward selling remittance dollar and remittance transaction amount A1 of 800 ten thousand dollars in the quarter, transacts the forward selling remittance dollar and remittance stroke number A2 in the quarter, clicks 0 time in the quarter for introducing the forward selling remittance dollar and remittance product A3, and browses the forward selling remittance dollar product A4 for 100 seconds in the quarter; the enterprise client U3 transacts 0 ten thousand dollars in the quarter for the forward-term accumulation dollar accumulation transaction amount A1, transacts 0 dollars in the quarter for the forward-term accumulation dollar accumulation transaction amount A2, clicks 10 times in the quarter for the introduction A3 of the forward-term accumulation dollar accumulation product, and browses A4 for the forward-term accumulation dollar product for 10000 seconds. From this, a transaction data matrix and an operational data matrix are derived, which are shown below by table 1:

TABLE 1 transaction data and operation matrix

	A1	A2	A3	A4	……
						U1	200	100	500	500
U2	800	2	0	100
						U3	0	0	10	10000
……

Then, a dimensionless quantization processing interval scaling strategy is used for standardizing the parameter value a in the matrix, and the specific standardization formula is as follows:

a＝(a′-min)/(max-min)；

wherein, a' represents a numerical value generated by a user operating a certain transaction behavior in a historical time period, max represents a maximum numerical value of the transaction behavior, and min represents a minimum numerical value of the transaction behavior.

Thereby obtaining the historical transaction behavior occurrence value matrix R of the user _ua The matrix is shown below by table 2:

TABLE 2 transaction behavior occurrence matrix

	A1	A2	A3	A4	……
						U1	0.25	1	1	0.04
U2	1	0.02	0	0
						U3	0	0	0.02	1
……

In the embodiment of the invention, the step of training the initial behavior estimation model by adopting the historical transaction behavior occurrence value matrix to obtain the target model comprises the following steps: decomposing a historical transaction behavior occurrence value matrix into a user hidden factor matrix and a transaction behavior hidden factor matrix by adopting a preset singular value decomposition algorithm; acquiring a user offset item, a transaction behavior offset item and an occurrence value average item from a historical transaction behavior occurrence value matrix; and (3) fusing a user bias item, a transaction behavior bias item, a generation value average item, a time attenuation function, the transposition of a user hidden factor matrix and a transaction behavior hidden factor matrix into the initial behavior prediction model to train and obtain a target model.

It should be noted that, in this embodiment, a Singular Value Decomposition (SVD) algorithm is adopted, and a trading behavior occurrence value matrix calculation method based on a user bias term, a trading behavior bias term, an occurrence value average term, and an integration time decay function is trained by a random gradient descent method to obtain a user hidden factor matrix and a trading behavior hidden factor matrix.

In the embodiment of the present invention, when obtaining the user bias item from the historical transaction behavior occurrence value matrix, the method includes: extracting the total number of customers from the historical transaction behavior occurrence value matrix; extracting a set of all transaction behaviors of each client from the historical transaction behavior occurrence value matrix; calculating the sum of values generated by all the clients conducting the set of transaction behaviors; and calculating a user bias item of the client based on the total number of the clients and the sum of values generated by the clients in the set of all transaction behaviors, wherein the user bias item is used for analyzing the bias heat of the clients on various transaction behaviors when operating different types of transaction products.

It should be noted that, different customers may have a phenomenon biased to a certain type of transaction behavior due to different business background directions, so that in this embodiment, a user bias item needs to be determined, and the probability that the customer generates the type of transaction behavior for a certain transaction product is quickly evaluated through the user bias item. In order to obtain a more accurate prediction result, the embodiment reflects the deviation phenomenon in the training model, and the user biases the item b _u To describe, the bias term is expressed as follows:

wherein R is _u Represents the set of all transactions performed by enterprise customer u, sum (R) _u ) Represents the sum of the values, len (R), that occur for the set of all transactions conducted by the Enterprise client u _u ) Represents R _u Number of sets (corresponding to the total number of customers mentioned above).

In the embodiment of the present invention, when obtaining the transaction behavior bias item from the historical transaction behavior occurrence value matrix, the method includes: extracting the total number of the transaction behaviors from the historical transaction behavior occurrence value matrix; extracting each transaction behavior from the historical transaction behavior occurrence value matrix to carry out a set of all transaction behaviors; calculating the sum of values generated by the set of all transaction behaviors carried out by each transaction behavior; and calculating a transaction behavior bias item of the customer based on the total number of the transaction behaviors and the sum of values generated by all transaction behaviors in a set, wherein the transaction behavior bias item is used for analyzing the bias heat of the customer on transaction products indicated by different transaction behaviors.

It should be noted that, for different transaction behaviors, the popularity is different in the financial market according to reasons such as current affairs hotspots, social production, and the like, so that in this embodiment, a transaction behavior bias item needs to be determined, and the probability that the customer generates various transaction behaviors for different transaction products is evaluated through the bias item. In order to obtain more accurate prediction results, the biased phenomenon is embodied in a training model, and the embodiment uses the transaction behavior bias term b _a To describe, the bias term is expressed as follows:

wherein R is _a Represents that the transaction action a carries out the set of all transaction actions, sum (R) _a ) The sum of the values, len (R), representing the occurrence of a transaction action a on the set of business clients _a ) Represents R _a Number of sets (corresponding to the total number of transactions described above).

In the embodiment of the present invention, when obtaining the occurrence value average item from the historical transaction behavior occurrence value matrix, the method includes: extracting the total quantity of the matrix set from the historical transaction behavior occurrence value matrix; extracting the sum of all occurrence values from the historical transaction behavior occurrence value matrix; and calculating the average value item of the occurrence values based on the total quantity of the matrix set and the sum of all the occurrence values.

It should be noted that, in order to obtain a more accurate prediction result, an average value is calculated from all values in the transaction behavior occurrence value matrix of all enterprises in the training data set, so as to obtain an occurrence value average term μ, where the occurrence value average term is expressed as follows:

wherein R is _ua Set of value matrices, sum (R), representing the occurrence of all user transactions _ua ) Representing the sum of all occurrences in a matrix of occurrences of all user transactions, len (R) _ua ) Represents R _ua Number of sets (corresponding to the total number of matrix sets described above).

The user transaction behavior occurrence value matrix R _ua Expressed as the product of two matrices, called the user implicit factor matrix p _u Transaction behavior implicit factor matrix q _a 。

It should be noted that, after obtaining the user bias term, the transaction behavior bias term, the occurrence value average term, the user hidden factor matrix and the transaction behavior hidden factor matrix, the method for predicting the transaction behavior occurrence value matrix is modeled based on a Singular Value Decomposition (SVD) algorithm:

r _ua ′＝μ+b _u +b _a +p _u ^T ·q _a ·f(Δt)；

wherein mu represents the occurrence number average value item; b _u A user bias term; b _a A transaction behavior bias term; p is a radical of _u ^T Matrix p representing implicit factors of users _u Transposing; q. q.s _a Representing a transaction behavior implicit factor matrix; f (Δ t) represents a time decay function; r is _ua ' means user u dealPredicted occurrence value of the easy action a.

In the embodiment of the present invention, when the target model is obtained by training, the method includes: carrying out initialization assignment on the user hidden factor matrix and the transaction behavior hidden factor matrix, and estimating the occurrence value of the appointed transaction behavior which is not generated by the client by adopting an initial behavior estimation model to obtain a behavior initial estimated value; calculating an error value between the initial predicted value of the behavior and a real occurrence value of the designated transaction behavior; calculating the sum of the squares of the errors of all customer associated specified transaction behaviors based on the error value of each customer; and training the initial behavior prediction model by adopting a gradient descent method so as to reduce the sum of squares of errors to a minimum value and obtain a target model.

From the feature extraction, it can be known if the matrix R _ua The value of row a in u is 0, indicating that the transaction action a has not occurred for customer u. During training, a factor matrix p is hidden for a user _u Implicit factor matrix q with transaction behavior _a Performing initialization assignment to predict the occurrence value of transaction behavior a which has not occurred to customer u by using a user transaction behavior occurrence value prediction formula, wherein the prediction value is r _ua ′。

Suppose that the actual value is known as the occurrence of a transaction action a by a customer u, and is r _ua (matrix R) _ua The values of the row u and column a) in the middle, the error between the actual value and the predicted value can be expressed as: e.g. of a cylinder _ua ＝r _ua -r _ua ' (corresponding to the error between the initial predicted value of the calculated action and the actual occurrence of the designated trading action as described above), and thus the total sum of squared errors can be calculated as: SSE = ∑ Σ _u,a e _ua ² . The user implicit factor matrix p is then used as long as SSE is minimized by training _u Implicit factor matrix q with transaction behavior _a The closest true values can be fit and the SSE minimized in this example using the usual gradient descent training. Training to obtain user hidden factor matrix p of fitting true value _u Implicit factor matrix q with transaction behavior _a And therefore, the occurrence value of the transaction behavior a which is not occurred by the client u is predicted and calculated as follows:

R _ua ′＝μ+b _u +b _a +p _u ^T ·q _a ·f(Δt)，

wherein R is _ua ' indication: and (4) when the SSE is reduced to the minimum, the predicted value of the occurrence value of the transaction behavior a which has not occurred to the client u.

Step S101, transaction data is obtained, wherein the transaction data is generated when the target customer uses the financial exchange rate transaction system in the current time period, and the transaction data at least comprises the following steps: and (5) trading the product.

It should be noted that, the embodiment of the present invention uses a risk monitoring system as an implementation subject, and the risk monitoring system can perform risk neutral prediction and alarm based on user transaction data and transaction time.

Wherein the transaction products are stored in a financial exchange rate type transaction system, the transaction products including but not limited to: the currency of the specified currency is selected from the group consisting of an immediate sale of the foreign exchange, a forward sale of the foreign exchange, a lost sale of the foreign exchange of the specified currency and a lost sale of the foreign exchange of the specified currency.

Step S102, transaction detail data and customer operation data which are related to each transaction product in the transaction data are extracted, wherein the transaction detail data at least comprise: the transaction time.

It should be noted that, because the market quotation of the foreign exchange has extremely fast change, and time is an important influence factor for the risk prediction system, the embodiment of the invention introduces the time context information of the transaction behavior into the risk monitoring system, so that the correlation degree of the time information can be maintained on the system, and the monitoring and early warning operation can be operated, and is more flexible and accurate.

Step S103, inputting the transaction detail data and the customer operation data into a target model, and receiving a transaction behavior pre-evaluation value output by the target model, wherein at least a time attenuation function is fused into the target model, the time attenuation function is used for representing the bias heat of the target customer to various transaction behaviors along with the advance of transaction time, and the transaction behavior pre-evaluation value is used for representing the predicted occurrence value of the target customer to various transaction behaviors.

It should be noted that after the transaction detail data and the customer operation data are obtained, the transaction detail data and the customer operation data are input into a risk prediction model which is constructed in advance, and a transaction behavior prediction value is obtained through calculation, that is, a transaction behavior occurrence value which has not occurred to a customer is predicted.

And step S104, outputting alarm information under the condition that the transaction behavior estimated value exceeds the risk neutral index threshold value.

It should be noted that, if the obtained predicted value of the transaction behavior exceeds the set threshold, the system performs a reminder and sends a mail notification and a short message notification at the same time, and the service staff performs further investigation online according to the notification.

By the embodiment, the transaction behavior risk is predicted based on the transaction data and the transaction time, intelligent service is provided for the risk monitoring system, advance prediction and warning of transaction behavior are realized, risks are recognized in advance, prospective guidance for customers is made, market risks of enterprises are prevented from being converted into credit risks of financial institutions, and the technical problem that risk conditions of transaction behavior cannot be predicted because the risk monitoring method in the related technology can only realize transaction statistics afterwards is solved.

The invention is described below in connection with an alternative embodiment.

Example two

The embodiment provides a risk monitoring system based on transaction data, and each implementation module included in the risk monitoring system based on transaction data corresponds to each implementation step in the first embodiment.

Fig. 3 is an architecture diagram of an optional risk monitoring system based on transaction data according to an embodiment of the present invention, and as shown in fig. 3, the risk monitoring system based on transaction data includes four main bodies, namely, an exchange rate transaction system, an offline training and prediction module, a service setting module, and a risk early warning module.

The exchange rate type transaction system comprises transaction products and transaction types, wherein the transaction products comprise but are not limited to: the trade types include but are not limited to, spot-sale remittance, forward-sale remittance, lost-sale remittance, foreign-currency lost-currency of a specified currency, foreign-currency options of a specified currency, foreign-currency lost-currency of a specified currency, and the trade types include: real-time, hang list, cancel, delivery, reverse level, and exhibition period.

The business setting module is used for business personnel to predict and set a predicted transaction behavior index, a time attenuation coefficient and a risk neutral index and a threshold value, the business personnel sets a transaction behavior index related to risk neutral, and the transaction behavior is defined as a transaction behavior index formed by combining product, transaction contract factors and statistical factor dimensions; the business personnel sets the coefficients in the training module and the prediction module, the coefficient is defined as a coefficient which is more than 0 and less than or equal to 1, and the business can be set according to the time relation degree of the set trading behavior index, the current market environment and the trading behavior index; setting a risk neutral indicator (product, deal contract element) threshold, for example, may set the number of forward dollar forward trade reverse runs of an enterprise customer to be more than 30% more than historical variation.

The off-line training and predicting module is used for training the risk predicting module and predicting the transaction behavior, wherein the off-line training comprises a feature extraction and training model, the feature extraction is that a program carries out operation processing according to all transaction detail data of all customers, extracts the transaction detail data and operation data, and constructs a user transaction behavior occurrence value matrix; the training model is mainly used for off-line learning training according to a data matrix processed by the feature extraction module and data set by a service, and the module decomposes the matrix through a random gradient descent method based on a trading behavior occurrence numerical matrix calculation method of a user bias item, a trading behavior bias item, an occurrence numerical average item and an integration time decay function which are configured in advance, and trains to obtain a user hidden factor matrix and a trading behavior hidden factor matrix; and the prediction module is used for calculating according to the model trained by learning by the system so as to predict the numerical value of the transaction behavior which does not occur to the user.

The risk early warning module is used for calculating the specific predicted quantity of the transaction behavior indexes, carrying out system reminding, mail notification and short message notification, sending the messages to business personnel, and carrying out offline investigation according to the notification by the business personnel.

Based on the risk monitoring system, the risk monitoring comprises the following steps:

the method comprises the following steps: and business personnel set transaction behavior indexes, time attenuation functions and risk neutral early warning thresholds through a business setting module.

In order to monitor and predict risks, a transaction behavior relationship between a client and a specific transaction product needs to be established first, and a transaction behavior occurrence value matrix is established for subsequent feature extraction. In this embodiment, when configuring all transaction behaviors to be estimated, a service person sets a series of transaction behaviors of a transaction product related to risk neutrality or capable of performing risk neutrality determination according to experience and a current product tradable type, wherein dimensions such as a transaction product (type), a transaction contract element, a statistical element and the like are provided through a system page for the service person to set up in a combined manner, and the transaction contract requirements include but are not limited to a transaction direction, a currency pair, a delivery mode and the like, for example, the following transaction behavior indexes are set: the amount of forward selling money dollar junction/sale transaction (amount), the number of forward selling money harbor junction/sale bill hanging order making strokes, the number of forward selling money euro bill hanging order junction/sale bill withdrawing strokes, the number of forward selling money dollar junction/sale bill reversing strokes, the number of forward selling money dollar junction/sale bill balance strokes, the number of forward selling money dollar junction/sale bill spread strokes, etc.

For a specific product transaction, the smaller the difference between the occurrence time of the latest transaction (such as forward sale and remittance of a pending statement) and the current time, the more popular the market conditions representing the current time, and the larger the value of the transaction occurring in the future. For example, if the degree of heat of export transactions in china at the current time is high and the increase is rapid, the enterprise order contract may estimate that the foreign exchange income increases in a future period of time, the enterprise client may perform the hedging transaction behavior related to the closing of the sales remittance in order to avoid the risk of exchange rate, the smaller the difference between the occurrence time of the transaction behavior of the current specific product and the current time is, and based on the above background, a time decay function (denoted by f (Δ t) in this embodiment) is configured to represent a parameter that the occurrence value of the transaction behavior of the transaction product changes with the change between the occurrence time and the current time.

The time decay function is formulated as follows:

in this embodiment, a time attenuation coefficient (represented by α in this embodiment) needs to be set, and the setting can be performed according to the transaction behavior indication set and the current market quotation judgment, and the system provides an entry coefficient function; and delta t is the absolute value of the difference between the latest transaction time and the current time of the transaction behavior of the transaction product. The time decay function formula uses an exponential formula to express that the heat of the transaction behavior gradually decays in an exponential form along with the time, wherein alpha is greater than 0,f (0) =1 represents the initial heat of the transaction behavior and decreases in an exponential form along with the time.

When the threshold of the risk neutral index (product and contract element) is set, the set threshold can be set according to the types of percentage, 0-1 intermediate value and the like, and the threshold parameter can be set to be a positive percentage value, for example, the number of derivative transactions is less than 30% of the historical average variation value, the number of reverse closing times is less than the historical average variation value, and the number of exhibition times is less than the historical average variation value. For example, the number of reverse closing of forward dollar settlement transactions of the enterprise customer may be set to exceed 30% compared with the historical variation (the risk neutral index threshold is not specifically limited in the embodiment of the present invention, and may be set according to the actual application requirements).

The system provides products, transaction directions, transaction types, stroke numbers, transaction amounts and operation data for business personnel to set with threshold value combinations.

Step two: feature extraction

When a risk monitoring model is constructed, a large amount of historical data is required to be used as a model construction basis. The method comprises the steps of recording transaction detail data of all enterprise customer exchange rate product transactions and operation data of customers in the customer transaction process depending on the existing transaction system, synchronizing the transaction detail data and the operation data to an offline training system in real time when a model is built, calculating data of enterprise customers in transaction behavior indexes set by businesses through the transaction detail data, and carrying out data standardization processing.

Specifically, when the risk monitoring model is constructed, the transaction of the transaction behavior of a specific product within a certain period of time, the number of transaction behaviors, the number of times of the transaction behavior (product introduction click), the duration of the transaction behavior (market browsing), and the like can be counted.

For example, when constructing the risk monitoring model, taking statistics of the forward-selling/remitting transaction amount (amount) of each enterprise client in the quarter as an example, the enterprise client U1 transacts 200 ten thousand dollars in the quarter for the forward-selling/remitting dollar remitting transaction amount A1, transacts 100 dollars in the quarter for the forward-selling/remitting dollar remitting transaction amount A2, introduces a description of the forward-selling/remitting dollar remitting product A3, clicks 500 times in the quarter, and browses the forward-selling/remitting dollar product market quotation A4 for 500 seconds; the enterprise client U2 transacts the amount of money of 800 ten thousand dollars in the quarter for the forward-term settlement and remittance dollar amount A1, transacts 2 times in the quarter for the forward-term settlement and remittance dollar amount A2, introduces A3 to the forward-term settlement and remittance dollar product, clicks 0 time in the quarter, and browses A4 for the forward-term settlement and remittance dollar product for 100 seconds in the quarter; the enterprise client U3 deals the forward-selling remittance dollar remittance transaction amount A1 with 0 ten thousand dollars in the quarter, deals the forward-selling remittance dollar remittance transaction amount A2 with 0 dollars in the quarter, introduces the forward-selling remittance dollar remittance product A3 clicked 10 times in the quarter, browses the forward-selling remittance dollar product A4 for 10000 seconds in the quarter, and firstly obtains transaction data and an operation matrix.

a＝(a′-min)/(max-min)；

wherein, a' represents a numerical value generated by a user operating a certain transaction behavior in a historical time period, max represents a maximum numerical value of the transaction behavior, min represents a minimum numerical value of the transaction behavior, and therefore a user transaction behavior occurrence numerical value matrix R is obtained _ua 。

Step three: model training

In the embodiment, a Singular Value Decomposition (SVD) algorithm is adopted, a trading behavior occurrence numerical matrix calculation method based on a user bias term, a trading behavior bias term, an occurrence numerical average term and an integration time attenuation function is adopted, and a decomposed user hidden factor matrix and a trading behavior hidden factor matrix are obtained through training by a random gradient descent method.

Firstly, user bias items need to be acquired, and different enterprise customers have a phenomenon of being biased to certain transaction behaviors due to different business background directions. For example, an enterprise client who only conducts export transaction has a future collection and is willing to avoid the risk of exchange rate through hedging, and is biased to have more transaction behaviors similar to forward settlement transaction to avoid the risk of exchange rate; the enterprise client only doing import trading has the future to pay money and is willing to avoid the exchange rate risk through the hedging, and is biased to have more trading behaviors similar to buying the expanding option, and the enterprise can lock the worst exchange rate of future money buying at the execution price level of the option by buying the foreign exchange expanding option, and enjoys the benefits brought by the possible increasing value of the exchange rate of the designated currency while avoiding the risk of losing the designated currency.

For the above reasons, the present embodiment needs to determine a user bias item, by which the probability of the transaction behavior of the customer on a certain transaction product is quickly evaluated. In order to obtain a more accurate prediction result, the embodiment reflects the deviation phenomenon in the training model, and the user biases the item b _u To describe, the bias term is expressed as follows:

For different transaction behaviors, the popularity is different in the financial market according to reasons such as current affair hotspots, social production and the like, so that in the embodiment, a transaction behavior bias item needs to be determined, and the probability that the client generates various transaction behaviors for different transaction products is evaluated through the bias item. To obtain more accurate prediction results, this biased phenomenon is characterizedIn the exercise model, the embodiment biases the term b with the transaction behavior _a To describe, the bias term is expressed as follows:

wherein R is _a Represents that the transaction action a carries out the set of all transaction actions, sum (R) _a ) The sum of the values, len (R), that occur to represent that the transaction action a is proceeding with the set of business customers _a ) Represents R _a Number of sets (corresponding to the total number of transactions described above).

In order to obtain a more accurate prediction result, calculating an average value of all values in the transaction behavior occurrence value matrix of all enterprises in the training data set to obtain an occurrence value average value item mu, wherein the occurrence value average value item is represented as follows:

wherein R is _ua Set of value matrices, sum (R), representing the occurrence of all user transactions _ua ) Representing the sum of all occurrence values in a matrix of occurrence values for all user transactions Len (R) _ua ) Represents R _ua Number of sets (corresponding to the total number of matrix sets described above).

Generating a numerical matrix R of user transaction behaviors _ua Expressed as the product of two matrices, called the user implicit factor matrix p _u Transaction behavior implicit factor matrix q _a 。

After obtaining a user bias item, a transaction behavior bias item, an occurrence numerical value average item, a user hidden factor matrix and a transaction behavior hidden factor matrix, modeling a transaction behavior occurrence numerical value matrix prediction method based on a Singular Value Decomposition (SVD) algorithm:

r _ua ′＝μ+b _u +b _a +p _u ^T ·q _a ·f(Δt)；

wherein mu represents the occurrence number average value item; b _u A user bias term; b _a A transaction behavior bias term; p is a radical of _u ^T Matrix p representing implicit factors of users _u Transposing; q. q.s _a Representing a transaction behavior implicit factor matrix; f (Δ t) represents a time decay function; r is _ua ' represents the predicted occurrence of the transaction activity a by the user u.

From the feature extraction, if the matrix R is _ua The value of row a in u is 0, indicating that the transaction action a has not occurred for customer u. During training, a factor matrix p is hidden for a user _u Implicit factor matrix q with transaction behavior _a Carrying out initialization assignment, namely predicting the occurrence value of the transaction behavior a which has not occurred to the client u of the enterprise by using a user transaction behavior occurrence value prediction formula, wherein the predicted value is r _ua ′。

Suppose that the actual value is known as the occurrence of a transaction a by a customer u, and is r _ua (matrix R) _ua The values of the u middle row and a column), the error of the real value from the predicted value can be expressed as: e.g. of the type _ua ＝r _ua -r _ua ' (corresponding to the error between the initial predicted value of the calculated action and the actual occurrence of the designated trading action as described above), and thus the total sum of squared errors can be calculated as: SSE = ∑ Σ _u,a e _ua ² . The user implicit factor matrix p is then used as long as SSE is minimized by training _u Implicit factor matrix q with transaction behavior _a The closest true values can be fit and the SSE minimized in this example using the usual gradient descent training. Training to obtain user hidden factor matrix p of fitting true value _u Implicit factor matrix q with transaction behavior _a And therefore, the occurrence value of the transaction behavior a which is not occurred by the client u is predicted and calculated as follows:

R _ua ′＝μ+b _u +b _a +p _u ^T ·q _a ·f(Δt)，

wherein R is _ua ' indication: and when the SSE is reduced to the minimum, the predicted value of the occurrence value of the transaction action a which does not occur to the client u. .

Step four, risk neutral early warning

And calculating according to the risk neutral indexes set by the service personnel and the transaction behavior occurrence numerical values which have not occurred to the client to obtain the specific number of the transaction behavior indexes which have not occurred to the client. If the risk neutral index threshold value is that the forward dollar settlement transaction reverse leveling times of the enterprise client exceed 30% of historical variation, historical data comparison is carried out according to the predicted times, if the risk neutral index threshold value exceeds a preset threshold value, the system carries out system reminding and sends mail notification and short message notification at the same time, and business personnel carry out offline investigation according to the notification.

In the embodiment of the invention, service personnel can set the characteristic extraction index and the prediction index according to needs, and combine the time decay function into the SVD algorithm to carry out accurate prediction and alarm on transaction behaviors, so that the problem that the current risk neutral management lacks intellectualization is solved.

The invention is described below in connection with an alternative embodiment.

EXAMPLE III

The present embodiment provides a risk monitoring device based on transaction data, where each implementation unit included in the risk monitoring device based on transaction data corresponds to each implementation step in the first embodiment.

Fig. 4 is a schematic diagram of an alternative transaction data based risk monitoring device according to an embodiment of the present invention, as shown in fig. 4, the transaction data based risk monitoring device includes: an acquisition unit 41, an extraction unit 42, an input unit 43, an output unit 44, wherein,

an obtaining unit 41, configured to obtain transaction data, where the transaction data is data generated when a target customer uses a financial exchange rate transaction system in a current time period, and the transaction data at least includes: trading the product;

an extracting unit 42, configured to extract transaction detail data and customer operation data associated with each transaction product in the transaction data, where the transaction detail data at least includes: a transaction time;

the input unit 43 is configured to input the transaction detail data and the customer operation data into the target model, and receive a transaction behavior pre-evaluation value output by the target model, where at least a time decay function is incorporated into the target model, the time decay function is used to characterize a bias heat of the target customer to various transaction behaviors along with the progress of transaction time, and the transaction behavior pre-evaluation value is used to characterize a predicted occurrence value of the target customer to various transaction behaviors;

and the output unit 44 is used for outputting alarm information under the condition that the predicted value of the transaction behavior exceeds the risk neutral index threshold value.

The risk monitoring device based on transaction data obtains transaction data through the obtaining unit 41, where the transaction data is data generated when the target customer uses the financial exchange rate transaction system in the current time period, and the transaction data at least includes: trading the product; transaction detail data and customer operation data associated with each transaction product in the transaction data are extracted by the extraction unit 42, wherein the transaction detail data at least comprises: a transaction time; inputting the transaction detail data and the customer operation data into a target model through an input unit 43, and receiving a transaction behavior pre-evaluation value output by the target model, wherein at least a time attenuation function is integrated into the target model, the time attenuation function is used for representing the bias heat of the target customer to various transaction behaviors along with the advance of transaction time, and the transaction behavior pre-evaluation value is used for representing the predicted occurrence value of the target customer to various transaction behaviors; in case the predicted value of the transaction behaviour exceeds the risk neutral indicator threshold, an alarm message is output via the output unit 44.

In the embodiment, the risk of the transaction behavior is predicted based on the transaction data and the transaction time, intelligent service is provided for a risk monitoring system, advance prediction and warning of the transaction behavior are realized, the risk is recognized in advance, and prospective guidance is made for a client, so that the market risk of an enterprise is prevented from being converted into the credit risk of a financial institution, and the technical problem that the risk condition of the transaction behavior cannot be predicted because the risk monitoring method in the related technology can only realize statistics after the transaction is solved.

Optionally, the risk monitoring method device based on transaction data further includes: the first acquisition subunit is used for acquiring a plurality of historical transaction detail data and a plurality of historical operation data of all clients, and extracting a transaction data matrix in each historical transaction detail data and an operation data matrix in each historical operation data, wherein the transaction data matrix comprises: historical trading products, historical trading behaviors, historical trading volume, historical trading times and historical trading time periods, wherein the operation data matrix comprises the following components: historical transaction products, historical transaction behaviors and historical transaction times; the first extraction subunit is used for extracting the maximum value and the minimum value of the historical trading behavior of each client in the historical trading time period; the first processing subunit is used for processing the transaction data matrix and the operation data matrix in a standardized manner by adopting a dimensionless quantitative processing interval scaling strategy based on the maximum numerical value and the minimum numerical value to obtain a historical transaction behavior occurrence numerical matrix; the first construction subunit is used for constructing an initial behavior estimation model, and training the initial behavior estimation model by adopting a historical transaction behavior occurrence numerical matrix to obtain a target model.

Optionally, the first building subunit comprises: the first decomposition module is used for decomposing the historical transaction behavior occurrence value matrix into a user hidden factor matrix and a transaction behavior hidden factor matrix by adopting a preset singular value decomposition algorithm; the first acquisition module is used for acquiring a user offset item, a transaction behavior offset item and an occurrence value average item from a historical transaction behavior occurrence value matrix; the first merging module is used for merging the user bias item, the transaction behavior bias item, the occurrence value mean item, the time attenuation function, the transposition of the user hidden factor matrix and the transaction behavior hidden factor matrix into the initial behavior pre-estimation model so as to train and obtain the target model.

Optionally, the first merging module includes: the first estimation submodule is used for carrying out initialization assignment on the user hidden factor matrix and the transaction behavior hidden factor matrix and estimating the occurrence value of the appointed transaction behavior which is not generated by the client by adopting an initial behavior estimation model to obtain a behavior initial estimation value; the first calculation submodule is used for calculating an error value between the action initial estimated value and a real occurrence value of the specified transaction action; the second calculation submodule is used for calculating the error square sum of all the customer-associated specified transaction behaviors based on the error value of each customer; and the first training submodule is used for training the initial behavior prediction model by adopting a gradient descent method so as to reduce the sum of squares of errors to a minimum value and obtain a target model.

Optionally, the first obtaining module includes: the first extraction submodule is used for extracting the total number of customers from the historical transaction behavior occurrence numerical matrix; the second extraction submodule is used for extracting a set of all trading behaviors of each client from the historical trading behavior occurrence value matrix; the third calculation submodule is used for calculating the sum of numerical values generated by all sets of transaction behaviors of each client; and the fourth calculation submodule is used for calculating a user bias item of the client based on the total number of the client and the sum of values generated by the client when the client conducts the set of all transaction behaviors, wherein the user bias item is used for analyzing the bias heat of the client on various transaction behaviors when the client operates different types of transaction products.

Optionally, the first obtaining module further includes: the third extraction submodule is used for extracting the total number of the transaction behaviors from the historical transaction behavior occurrence numerical matrix; the fourth extraction submodule is used for extracting each transaction behavior from the historical transaction behavior occurrence value matrix to carry out a set of all transaction behaviors; the fifth calculation sub-module is used for calculating the sum of numerical values generated by all the trading behaviors in the set of the trading behaviors; and the sixth calculating submodule is used for calculating a transaction behavior bias item of the customer based on the total number of the transaction behaviors and the sum of values generated by all transaction behaviors in a set during each transaction behavior, wherein the transaction behavior bias item is used for analyzing the bias heat of the customer on transaction products indicated by different transaction behaviors.

Optionally, the first obtaining module further includes: the fifth extraction submodule is used for extracting the total quantity of the matrix set from the historical transaction behavior occurrence numerical matrix; the sixth extraction submodule is used for extracting the sum of all occurrence values from the historical transaction behavior occurrence value matrix; and the seventh calculation submodule is used for calculating the average value item of the occurrence values based on the total quantity of the matrix set and the sum of all the occurrence values.

Optionally, the input unit includes: the first configuration subunit is used for configuring all transaction behaviors to be estimated; the second configuration subunit is used for configuring a time attenuation function corresponding to the product types of different transaction products, wherein the time attenuation function is determined based on the absolute value of the difference between the time attenuation coefficient and the product operation time, and the absolute value of the difference between the product operation time refers to the absolute value of the difference between the latest historical transaction time point and the current time point when the transaction action occurs; and the third configuration subunit is used for configuring a risk neutral index threshold value.

The risk monitoring device based on transaction data may further include a processor and a memory, where the acquiring unit 41, the extracting unit 42, the input unit 43, the output unit 44, and the like are stored in the memory as program units, and the processor executes the program units stored in the memory to implement corresponding functions.

The processor comprises a kernel, and the kernel calls a corresponding program unit from the memory. The kernel can be set to be one or more, risk monitoring and alarming are carried out on the transaction behaviors by adjusting kernel parameters, and the risk of the transaction behaviors is predicted based on transaction data and transaction time.

The memory may include volatile memory in a computer readable medium, random Access Memory (RAM) and/or nonvolatile memory such as Read Only Memory (ROM) or flash memory (flash RAM), and the memory includes at least one memory chip.

According to another aspect of the embodiments of the present invention, there is also provided a computer-readable storage medium, which includes a stored computer program, wherein when the computer program runs, the apparatus where the computer-readable storage medium is located is controlled to execute any one of the above methods for risk monitoring based on transaction data.

According to another aspect of embodiments of the present invention, there is also provided an electronic device, including one or more processors and a memory for storing one or more programs, wherein the one or more programs, when executed by the one or more processors, cause the one or more processors to implement any one of the above-mentioned risk monitoring methods based on transaction data.

The present application also provides a computer program product adapted to perform a program for initializing the following method steps when executed on a data processing device: acquiring transaction data, wherein the transaction data is generated when a target customer uses the financial exchange rate type transaction system in the current time period, and the transaction data at least comprises the following steps: trading the product; extracting transaction detail data and customer operation data which are related to each transaction product in the transaction data, wherein the transaction detail data at least comprises the following steps: a transaction time; inputting the transaction detail data and the customer operation data into a target model, and receiving a transaction behavior pre-evaluation value output by the target model, wherein at least a time attenuation function is fused into the target model, the time attenuation function is used for representing the bias heat of the target customer to various transaction behaviors along with the advance of transaction time, and the transaction behavior pre-evaluation value is used for representing the predicted occurrence value of the target customer to various transaction behaviors; and outputting alarm information under the condition that the transaction behavior estimated value exceeds the risk neutral index threshold value.

Fig. 5 is a block diagram of a hardware structure of an electronic device (or mobile device) of a risk monitoring method based on transaction data according to an embodiment of the present invention. As shown in fig. 5, the electronic device may include one or more processors 502 (shown with 502a, 502b, … …,502 n), the processors 502 (the processors 502 may include, but are not limited to, a processing device such as a microprocessor MCU or a programmable logic device FPGA), and a memory 504 for storing data. Besides, the method can also comprise the following steps: a display, an input/output interface (I/O interface), a Universal Serial Bus (USB) port (which may be included as one of the ports of the I/O interface), a network interface, a keyboard, a power supply, and/or a camera. It will be understood by those skilled in the art that the structure shown in fig. 5 is only an illustration and is not intended to limit the structure of the electronic device. For example, the electronic device may also include more or fewer components than shown in FIG. 5, or have a different configuration than shown in FIG. 5.

The above-mentioned serial numbers of the embodiments of the present invention are only for description, and do not represent the advantages and disadvantages of the embodiments.

In the above embodiments of the present invention, the descriptions of the respective embodiments have respective emphasis, and for parts that are not described in detail in a certain embodiment, reference may be made to related descriptions of other embodiments.

In the embodiments provided in the present application, it should be understood that the disclosed technology can be implemented in other ways. The above-described apparatus embodiments are merely illustrative, and for example, a division of a unit may be a logical division, and in actual implementation, there may be another division, for example, multiple units or components may be combined or may be integrated into another system, or some features may be omitted, or may not be executed. In addition, the shown or discussed mutual coupling or direct coupling or communication connection may be an indirect coupling or communication connection through some interfaces, units or modules, and may be in an electrical or other form.

The units described as separate parts may or may not be physically separate, and parts displayed as units may or may not be physical units, may be located in one position, or may be distributed on a plurality of units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of the embodiment.

In addition, functional units in the embodiments of the present invention may be integrated into one processing unit, or each unit may exist alone physically, or two or more units are integrated into one unit. The integrated unit can be realized in a form of hardware, and can also be realized in a form of a software functional unit.

The integrated unit, if implemented in the form of a software functional unit and sold or used as a stand-alone product, may be stored in a computer readable storage medium. Based on such understanding, the technical solution of the present invention may be embodied in the form of a software product, which is stored in a storage medium and includes instructions for causing a computer device (which may be a personal computer, a server, or a network device) to execute all or part of the steps of the method according to the embodiments of the present invention. And the aforementioned storage medium includes: a U-disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), a removable hard disk, a magnetic or optical disk, and other various media capable of storing program codes.

The foregoing is only a preferred embodiment of the present invention, and it should be noted that it is obvious to those skilled in the art that various modifications and improvements can be made without departing from the principle of the present invention, and these modifications and improvements should also be considered as the protection scope of the present invention.

Claims

1. A risk monitoring method based on transaction data, comprising:

acquiring transaction data, wherein the transaction data is generated when a target customer uses a financial exchange rate type transaction system in the current time period, and the transaction data at least comprises the following steps: trading the product;

extracting transaction detail data and customer operation data in the transaction data, wherein the transaction detail data at least comprises: a transaction time;

inputting the transaction detail data and the customer operation data into an object model, and receiving a transaction behavior pre-evaluation value output by the object model, wherein at least a time attenuation function is integrated into the object model, the time attenuation function is used for representing the bias heat of the target customer to various transaction behaviors along with the advance of the transaction time, and the transaction behavior pre-evaluation value is used for representing a predicted occurrence value of the target customer to various transaction behaviors;

and outputting alarm information under the condition that the transaction behavior estimated value exceeds a risk neutral index threshold value.

2. The risk monitoring method of claim 1, wherein the target model is trained by:

acquiring a plurality of historical transaction detail data and a plurality of historical operation data of all customers, and extracting a transaction data matrix in each historical transaction detail data and an operation data matrix in each historical operation data, wherein the transaction data matrix comprises: historical trading products, historical trading behaviors, historical trading volume, historical trading times and historical trading time periods, wherein the operation data matrix comprises: the historical transaction product, the historical transaction behavior, the historical transaction times;

extracting a maximum value and a minimum value of the historical trading behavior of each customer in the historical trading time period;

based on the maximum numerical value and the minimum numerical value, adopting a dimensionless quantitative processing interval scaling strategy to carry out standardized processing on the transaction data matrix and the operation data matrix to obtain a historical transaction behavior occurrence numerical value matrix;

and constructing an initial behavior prediction model, and training the initial behavior prediction model by adopting the historical transaction behavior occurrence value matrix to obtain the target model.

3. The risk monitoring method according to claim 2, wherein the step of training the initial behavior prediction model by using the historical transaction behavior occurrence value matrix to obtain the target model comprises:

decomposing the historical transaction behavior occurrence value matrix into a user hidden factor matrix and a transaction behavior hidden factor matrix by adopting a preset singular value decomposition algorithm;

acquiring a user offset item, a transaction behavior offset item and an occurrence value average item from the historical transaction behavior occurrence value matrix;

and fusing the user bias item, the transaction behavior bias item, the occurrence value average item, the time attenuation function, the transposition of the user hidden factor matrix and the transaction behavior hidden factor matrix into the initial behavior pre-estimation model to train and obtain the target model.

4. The risk monitoring method of claim 3, wherein training the target model comprises:

carrying out initialization assignment on the user hidden factor matrix and the transaction behavior hidden factor matrix, and estimating the occurrence value of the specified transaction behavior which is not generated by the client by adopting the initial behavior estimation model to obtain a behavior initial estimation value;

calculating an error value between the behavior initial estimated value and a real occurrence value of the specified transaction behavior;

calculating a sum of squares of errors for all customers associated with the specified transaction behavior based on the error value for each customer;

and training the initial behavior prediction model by adopting a gradient descent method so as to reduce the sum of squares of errors to a minimum value and obtain the target model.

5. The risk monitoring method of claim 3, wherein obtaining a user bias term from the matrix of historical transaction behavior occurrence values comprises:

extracting the total number of customers from the historical transaction behavior occurrence value matrix;

extracting a set of all transaction behaviors of each client from the historical transaction behavior occurrence value matrix;

calculating the sum of values generated by all the clients conducting the set of transaction behaviors;

calculating a user bias item of the customer based on the total number of the customers and the sum of values generated by the customer in the set of all transaction behaviors, wherein the user bias item is used for analyzing the bias heat of the customer on various transaction behaviors when operating different types of transaction products.

6. The risk monitoring method of claim 3, wherein obtaining a transaction activity bias from the matrix of historical transaction activity occurrence values comprises:

extracting the total number of the transaction behaviors from the historical transaction behavior occurrence value matrix;

extracting each transaction behavior from the historical transaction behavior occurrence value matrix to carry out a set of all transaction behaviors;

calculating the sum of values generated by the set of all transaction behaviors carried out by each transaction behavior;

calculating a transaction behavior bias item of the customer based on the total number of transaction behaviors and the sum of values generated by all transaction behaviors performed by each transaction behavior, wherein the transaction behavior bias item is used for analyzing the bias heat of the customer on transaction products indicated by different transaction behaviors.

7. The risk monitoring method of claim 3, wherein obtaining an occurrence value mean from the historical trading activity occurrence value matrix comprises:

extracting the total number of matrix sets from the historical transaction behavior occurrence value matrix;

extracting the sum of all occurrence values from the historical transaction behavior occurrence value matrix;

and calculating the average value item of the occurrence values based on the total number of the matrix sets and the sum of all the occurrence values.

8. The risk monitoring method according to any one of claims 1 to 7, wherein prior to inputting the transaction detail data and the customer operation data into a goal model, the method further comprises:

configuring all transaction behaviors to be estimated;

configuring time attenuation functions corresponding to product types of different transaction products, wherein the time attenuation functions are determined based on the absolute value of the difference between a time attenuation coefficient and product operation time, and the absolute value of the difference between the product operation time refers to the absolute value of the difference between the latest historical transaction time point and the current time point when the transaction action occurs;

configuring the risk neutral indicator threshold.

9. A risk monitoring device based on transaction data, comprising:

the acquisition unit is used for acquiring transaction data, wherein the transaction data is generated when a target customer uses a financial exchange rate type transaction system in the current time period, and the transaction data at least comprises the following steps: trading the product;

an extracting unit, configured to extract transaction detail data and customer operation data in the transaction data, where the transaction detail data is associated with each transaction product, and the transaction detail data at least includes: a transaction time;

the input unit is used for inputting the transaction detail data and the customer operation data into an object model and receiving a transaction behavior pre-evaluation value output by the object model, wherein at least a time attenuation function is blended into the object model, the time attenuation function is used for representing the bias heat of the target customer to various transaction behaviors along with the advance of the transaction time, and the transaction behavior pre-evaluation value is used for representing the predicted occurrence value of the target customer to various transaction behaviors;

and the output unit is used for outputting alarm information under the condition that the transaction behavior estimated value exceeds a risk neutral index threshold value.

10. A computer-readable storage medium, comprising a stored computer program, wherein the computer program, when executed, controls an apparatus in which the computer-readable storage medium is located to perform the risk monitoring method based on transaction data according to any one of claims 1 to 8.

11. An electronic device comprising one or more processors and memory storing one or more programs, wherein the one or more programs, when executed by the one or more processors, cause the one or more processors to implement the transaction data based risk monitoring method of any one of claims 1-8.