CN116307231A

CN116307231A - Method, device, computer equipment and storage medium for out-of-date prediction

Info

Publication number: CN116307231A
Application number: CN202310324299.5A
Authority: CN
Inventors: 马璇; 唐琳娜; 方安; 姚展佳
Original assignee: Industrial and Commercial Bank of China Ltd ICBC
Current assignee: Industrial and Commercial Bank of China Ltd ICBC
Priority date: 2023-03-29
Filing date: 2023-03-29
Publication date: 2023-06-23

Abstract

The application relates to an out-of-life prediction method, an out-of-life prediction device, computer equipment, a storage medium and a computer program product, and relates to the technical field of big data. The method comprises the following steps: acquiring first resource flow data of an object to be predicted in a historical time period; the first resource flow data represents the resource flow condition between the object to be predicted and other objects; constructing a directed relation graph between the object to be predicted and other objects according to the first resource flow data; acquiring second resource flow data of other objects in a historical time period, and acquiring resource flow characteristics of the object to be predicted based on the second resource flow data and the directed relation graph; and acquiring basic information characteristics of the object to be predicted, inputting the resource flow characteristics and the basic information characteristics into a prediction model, and obtaining the out-of-date probability of the resource predicted by the object to be predicted. By adopting the method, the out-of-date prediction result can be obtained more timely and accurately.

Description

Method, device, computer equipment and storage medium for out-of-date prediction

Technical Field

The present application relates to the field of big data technology, and in particular, to an out-of-date prediction method, an apparatus, a computer device, a storage medium, and a computer program product.

Background

With the development of big data technology, the big data technology is widely applied to predictive analysis. Currently, in the process of constructing a prediction model and applying the prediction model, the out-of-date prediction of the pre-support resource uses personal information and credit information of the user as characteristics.

However, the updating of the user personal information and credit information has a certain delay. When the user cannot support the resource on schedule due to emergency, or uses multiple heads to support, borrow new and old methods to deal with the time limit of the resource, the personal information and credit information of the user cannot reflect the conditions or timely, and further the early warning of the overtime risk cannot be accurately made.

Disclosure of Invention

Based on this, it is necessary to provide an over-prediction method, an apparatus, a computer device, a computer readable storage medium and a computer program product for solving the technical problem of inaccurate prediction.

In a first aspect, the present application provides a method of over-prediction. The method comprises the following steps:

acquiring first resource flow data of an object to be predicted in a historical time period; the first resource flow data represents resource flow conditions between the object to be predicted and other objects;

Constructing a directed relation graph between the object to be predicted and the other objects according to the first resource flow data; the directed relation graph comprises a plurality of nodes, each node corresponds to an object, and a directed edge is arranged between two nodes with resource flow relation;

acquiring second resource flow data of the other objects in the historical time period, and acquiring resource flow characteristics of the object to be predicted based on the second resource flow data and the directed relation graph;

and acquiring basic information characteristics of the object to be predicted, and inputting the resource flow characteristics and the basic information characteristics into a prediction model to obtain the out-of-date probability of the object to be predicted for repayment of the pre-supported resource.

In one embodiment, the extracting the resource flow feature of the object to be predicted based on the second resource flow data and the directed graph includes:

extracting object features of the other objects based on the second resource flow data;

based on the directed relation graph, determining structural parameters of the object to be predicted and nodes corresponding to the other objects;

and processing the object characteristics and the structural parameters through a graph convolution neural network to obtain the resource flow characteristics of the object to be predicted.

In one embodiment, the extracting object features of the other objects based on the second resource flow data includes:

normalizing the second resource flow data to obtain normalized second resource flow data;

and processing the normalized second resource flow data through a perceptron to obtain object characteristics of the other objects.

In one embodiment, the determining, based on the directed graph, structural parameters of the object to be predicted and nodes corresponding to the other objects includes:

determining the neighbor number of a first node of the corresponding node of the object to be predicted in the directed relation graph, and taking the neighbor number of the first node as a structural parameter of the corresponding node of the object to be predicted;

and determining the neighbor number of a second node of the other object corresponding nodes in the directed relation graph, and taking the neighbor number of the second node as the structural parameter of the other object corresponding nodes.

In one embodiment, the inputting the resource flow feature and the basic information feature into a prediction model to obtain the out-of-date probability of the object to be predicted to refund the pre-paid resource includes:

Performing relation mapping on the resource flow characteristics and the basic information characteristics to obtain first cross attention characteristics corresponding to the resource flow characteristics and second cross attention characteristics corresponding to the basic information characteristics;

and inputting the first cross attention characteristic and the second cross attention characteristic into a prediction model to obtain the out-of-date probability of the object to be predicted to repay the pre-paid resources.

In one embodiment, the obtaining the basic information feature of the object to be predicted includes:

acquiring basic information of the object to be predicted;

normalizing the basic information to obtain normalized basic information;

and processing the normalized basic information through a deep neural network to obtain the basic information characteristics of the object to be predicted.

In one embodiment, after obtaining the expiration probability of the resource predicted by the target to be predicted for repayment, the method further includes:

acquiring an out-of-period probability threshold;

and determining the object to be predicted as an object with excessive risk when the excessive probability of the object to be predicted repayment of the pre-branched resource exceeds the excessive probability threshold.

In a second aspect, the present application also provides an out-of-life prediction apparatus. The device comprises:

the data acquisition module is used for acquiring first resource flow data of the object to be predicted in a historical time period; the first resource flow data represents resource flow conditions between the object to be predicted and other objects;

the directed graph construction module is used for constructing a directed relationship graph between the object to be predicted and the other objects according to the first resource flow data; the directed relation graph comprises a plurality of nodes, each node corresponds to an object, and a directed edge is arranged between two nodes with resource flow relation;

the feature extraction module is used for acquiring second resource flow data of the other objects in the historical time period and obtaining resource flow features of the object to be predicted based on the second resource flow data and the directed relation graph;

and the out-of-period prediction module is used for acquiring basic information characteristics of the object to be predicted, inputting the resource flow characteristics and the basic information characteristics into a prediction model, and obtaining the out-of-period probability of the object to be predicted for repayment of the pre-supported resource.

In a third aspect, the present application also provides a computer device. The computer device comprises a memory storing a computer program and a processor which when executing the computer program performs the steps of:

In a fourth aspect, the present application also provides a computer-readable storage medium. The computer readable storage medium having stored thereon a computer program which when executed by a processor performs the steps of:

In a fifth aspect, the present application also provides a computer program product. The computer program product comprises a computer program which, when executed by a processor, implements the steps of:

The over-period prediction method, the over-period prediction device, the computer equipment, the storage medium and the computer program product construct a directed relation diagram between the object to be predicted and other objects according to the first resource flow data of the object to be predicted in the historical time period; acquiring second resource flow characteristics of other objects, acquiring the resource flow characteristics of the object to be predicted based on the second resource flow data and the directed relation graph, and reflecting the thought of the characteristics of the object to be predicted by the characteristics of the other objects to acquire the characteristics which have stronger real-time performance and can reflect dynamic information, so that the out-of-date risk can be acquired more timely; and simultaneously, acquiring basic information characteristics of a user to be predicted, and combining the characteristics of resource flow, and inputting the basic information characteristics and the characteristics of two modes into a prediction model together to obtain more accurate out-of-date probability of the target to be predicted for repayment of the pre-supported resources. By the over-period prediction method, not only is the characteristic information with better real-time added, but also the over-period prediction can be completed by combining the basic information characteristics, so that the over-period prediction result can be obtained more timely and accurately.

Drawings

FIG. 1 is a diagram of an application environment for an over-life prediction method in one embodiment;

FIG. 2 is a flow chart of a method of over-prediction in one embodiment;

FIG. 3 is a flow chart of a resource flow feature extraction step in one embodiment;

FIG. 4 is a flowchart illustrating a method of over-prediction in another embodiment;

FIG. 5 is a block diagram of an over-period prediction device in one embodiment;

fig. 6 is an internal structural diagram of a computer device in one embodiment.

Detailed Description

In order to make the objects, technical solutions and advantages of the present application more apparent, the present application will be further described in detail with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are for purposes of illustration only and are not intended to limit the present application.

The method for predicting the out-of-life provided by the embodiment of the application can be applied to an application environment shown in fig. 1. Wherein the terminal 102 communicates with the server 104 via a network. The data storage system may store data that the server 104 needs to process. The data storage system may be integrated on the server 104 or may be located on a cloud or other network server. The server 104 responds to the user demand, acquires the resource flow data of the object to be predicted from the data storage system, constructs a directed relation graph based on the resource flow data, and then extracts resource flow characteristics from the resource flow data and the directed relation graph; and obtaining basic information characteristics of the object to be predicted, and taking the basic information characteristics and the resource flow characteristics together as inputs of a prediction model to obtain the out-of-date probability of the resource reserved by the object to be predicted. The terminal 102 may be, but not limited to, various personal computers, notebook computers, smart phones, tablet computers, internet of things devices, and portable wearable devices, where the internet of things devices may be smart speakers, smart televisions, smart air conditioners, smart vehicle devices, and the like. The portable wearable device may be a smart watch, smart bracelet, headset, or the like. The server 104 may be implemented as a stand-alone server or as a server cluster of multiple servers.

In one embodiment, as shown in FIG. 2, an over-period prediction method is provided, in which the method may be applied to a server 104 as in FIG. 1, the method may include the steps of:

step S201, obtaining first resource flow data of an object to be predicted in a historical time period; the first resource flow data represents the resource flow condition between the object to be predicted and other objects; the history period is a period continuous with the current time.

Wherein the historical time period is a time period continuous with the current time and preceding the current time. Such as the last half year or the last year, etc.

The object to be predicted can be a bank client; the resource flow data may be funds flow data for a bank account of the object to be predicted, which may be generated by transfer, transaction or consumption, among other actions.

The server 104 is responsive to a user-selected object to be predicted, and queries the data storage system for first resource flow data for a historical time period (a particular time period may be selected by the user) for a bank account of the object to be predicted based on an identification of the object to be predicted.

Step S202, constructing a directed relation graph between an object to be predicted and other objects according to first resource flow data; the directed relation graph comprises a plurality of nodes, each node corresponds to an object, and a directed edge is arranged between two nodes with resource flow relation.

Illustratively, the first resource flow data of the object to be predicted includes transfer behavior data related to a personal account, consumption behavior data related to a merchant account, and transaction behavior data related to an investment product or a loan product. The server 104 constructs a directed relation graph between the object to be predicted and other objects by taking one object as a node according to the first resource flow data, wherein the direction of the directed edge identifies the direction of the resource flow, and the first resource flow amount is combined with the behavior type weight corresponding to the resource flow to obtain the edge weight of the directed edge. For example, if the object to be predicted a initiates payment to the merchant B, there is a directed edge between a and B from a to B in the directed graph, and the payment amount is 100, the weight of the consumption behavior is 1.5, and the edge weight of the directed edge of a to B is 150. The directed graph is used as graph data to prepare for extracting the resource flow characteristics.

Step S203, second resource flow data of other objects in the historical time period is obtained, and the resource flow characteristics of the object to be predicted are obtained based on the second resource flow data and the directed relation graph.

The server 104 determines other objects having a transaction relationship with the object to be predicted in a historical time period according to the resource flow data of the object to be predicted, queries second resource flow data of the other objects from the data storage system, and extracts the second resource flow data of the other objects as object features of the other objects through the trained feature extraction model; meanwhile, the server 104 determines structural parameters in the directed graph, including the number of neighbors of each node, as descriptive data of the directed graph according to the constructed directed graph. And obtaining the resource flow characteristics of the object to be predicted based on the object characteristics and the structural parameters.

Step S204, basic information characteristics of the object to be predicted are obtained, the resource flow characteristics and the basic information characteristics are input into a prediction model, and the out-of-date probability of the resource predicted by the object to be predicted is obtained.

The basic information features are extracted according to working information, personal income information, credit information and the like of the object to be predicted.

Wherein the pre-paid resource may be a loan or the like.

Illustratively, the server 104 obtains the basic information characteristics of the object to be predicted in addition to the resource flow characteristics of the object to be predicted. Inputting the basic information features and the resource flow features into a pre-trained prediction model, carrying out feature fusion on the basic information features and the resource flow features in the prediction model, and obtaining the out-of-date probability of the user to be predicted to repay the pre-paid resources according to the fused features.

In the over-period prediction method, a directed relation diagram between the object to be predicted and other objects is constructed according to the first resource flow data of the object to be predicted in the historical time period; acquiring second resource flow characteristics of other objects, acquiring the resource flow characteristics of the object to be predicted based on the second resource flow data and the directed relation graph, and reflecting the thought of the characteristics of the object to be predicted by the characteristics of the other objects to acquire the characteristics which have stronger real-time performance and can reflect dynamic information, so that the out-of-date risk can be acquired more timely; and simultaneously, acquiring basic information characteristics of a user to be predicted, and combining the characteristics of resource flow, and inputting the basic information characteristics and the characteristics of two modes into a prediction model together to obtain more accurate out-of-date probability of the target to be predicted for repayment of the pre-supported resources. By the over-period prediction method, not only is the characteristic information with better real-time added, but also the over-period prediction can be completed by combining the basic information characteristics, so that the over-period prediction result can be obtained more timely and accurately.

In one embodiment, as shown in fig. 3, the step S203 extracts the resource flow characteristics of the object to be predicted based on the resource flow data and the directed graph, which may be further implemented by the following steps:

Step S301 extracts object features of other objects based on the second resource flow data.

Step S302, based on the directed relation diagram, structural parameters of the object to be predicted and nodes corresponding to other objects are determined.

And step S303, processing the object characteristics and the structural parameters through a graph convolution neural network to obtain the resource flow characteristics of the object to be predicted.

Illustratively, each piece of resource flow data of the other object is extracted according to the second resource flow data of the other object, and each piece of resource flow data includes a resource flow amount, a resource flow behavior type (transfer type/transaction type/consumption type, etc.), a resource flow type (in type/out type), etc. Constructing and obtaining the resource flow amount characteristics, the resource flow behavior type characteristics, the resource flow times characteristics and the like corresponding to other objects based on the second resource flow data, and combining the characteristics as object characteristics of the other objects; according to the directed relation graph, considering the respective edge weights of directed edges in the directed relation graph, determining the neighbor number of each node of the directed relation graph, and taking the neighbor number as the structural parameter of the corresponding nodes of the object to be predicted and other objects; and processing the characteristics by combining the object characteristics with the structural parameters through a graph convolution neural network to obtain the resource flow characteristics of the object to be predicted, and representing the resource flow condition of the object to be predicted.

In this embodiment, object features of other objects are extracted through the second resource flow data, and structural parameters of the object to be predicted and nodes corresponding to the other objects are determined through the directed relation graph and are used as inputs of the graph convolution neural network together, so that resource flow features representing resource flow conditions of the object to be predicted are obtained. The resource flow characteristics are used as the characteristics of one mode to finish out-of-period prediction, and the follow-up out-of-period prediction model can predict based on more characteristics, so that the accuracy of out-of-period prediction results is effectively improved.

In one embodiment, the step S301 extracts object features of other objects based on the second resource flow data, and specifically includes: normalizing the second resource flow data to obtain normalized second resource flow data; and processing the normalized second resource flow data through a perceptron to obtain object characteristics of other objects.

Illustratively, normalizing the resource flow amount in the second resource flow data, and then processing the normalized second resource flow amount through a perceptron to obtain the resource flow amount characteristics of other objects; similarly, after normalization processing and perceptron processing, the characteristic resource flow behavior type characteristic and the characteristic resource flow times characteristic are obtained. The above features are combined as object features of other objects.

In this embodiment, the object features reflect not only the resource flow amounts of other objects, but also other resource flow attributes of other objects; the characteristics obtained by the graph convolution neural network have more dimensionality content, and in addition, the accuracy of the out-of-date prediction result can be improved based on the resource flow characteristics of the object to be predicted.

In one embodiment, the step S302 determines structural parameters of the object to be predicted and corresponding nodes of other objects based on the directed graph, which specifically includes: determining the neighbor number of a first node in the directed relation graph of the corresponding node of the object to be predicted, and taking the neighbor number of the first node as the structural parameter of the corresponding node of the object to be predicted; and determining the neighbor number of the second node of the nodes corresponding to other objects in the directed relation graph, and taking the neighbor number of the second node as the structural parameter of the nodes corresponding to other objects.

Illustratively, determining the node neighbor number of each node according to the directed relationship graph; in addition, an adjacency matrix and a degree matrix corresponding to the directed relation graph can be determined based on the directed relation graph, wherein the number of node neighbors of each node is also reflected in the degree matrix. The adjacency matrix is a matrix representing the adjacent relation among all nodes in the directed relation graph, for example, an object A to be predicted is transferred to an object B for a resource 10, the node of the object A to be predicted is denoted as a node 1, the node of the object B is denoted as a node 2, and then the data of the 1 st row and the 2 nd column in the adjacency matrix is denoted as 10; whereas object B has not transitioned into the resource for object a to be predicted, then in the adjacency matrix, row 2, column 1 data should be 0. Further, the degree matrix is a diagonal matrix representing the degree of each node in the directed graph, where the degree is the number of directed edges connected to each node. For example, if the object C goes into the resource to be predicted object a and also obtains the resource from the object a to be predicted, in the directed relationship graph, there are two directed edges between the object a and the object C; the node of object C is denoted as node 3, and in the degree matrix, the data of row 3 and column 3 should be 2.

In this embodiment, the number of node neighbors of each node, that is, the structural parameters of the object to be predicted and the nodes corresponding to other objects, is determined according to the directed graph, and the graph data is converted into digital information, so as to prepare for the follow-up over-prediction by adopting the features related to the directed graph.

In one embodiment, the step S204 inputs the resource flow characteristics and the basic information characteristics into a prediction model to obtain the out-of-date probability of the resource predicted by the target to be predicted for repayment, and specifically includes: performing relation mapping on the resource flow characteristics and the basic information characteristics to obtain first cross attention characteristics corresponding to the resource flow characteristics and second cross attention characteristics corresponding to the basic information characteristics; and inputting the first cross attention characteristic and the second cross attention characteristic into a prediction model to obtain the out-of-date probability of the predicted resource for repayment of the object to be predicted.

Under the cross attention mechanism, the basic information features are transformed through a first transducer model and projected into a feature space of the resource flow features to obtain Query vectors of the basic information features; then, carrying out transformation processing on the resource flow characteristics through a second transducer model to obtain Key vectors of the resource flow characteristics in a characteristic space; then calculating to obtain the similarity (or correlation) between the Query vector of the basic information feature and the Key vector of the resource flow feature; then, carrying out transformation processing on the resource flow characteristics through a third transducer model to obtain Value vectors of the resource flow characteristics in a characteristic space; and finally, combining the similarity with the Value vector of the resource flow feature to obtain a first cross attention feature corresponding to the resource flow feature. The first transducer model, the second transducer model, and the third transducer model were each trained in advance to determine parameters therein. Further, through similar steps, a second cross-attention feature corresponding to the basic information feature is obtained. And finally, inputting the first cross attention characteristic and the second cross attention characteristic into a pre-trained prediction model to obtain the out-of-date probability of the object to be predicted to repay the pre-paid resources.

In this embodiment, by performing a relational mapping on the resource flow feature and the basic information feature, the resource flow feature and the basic information feature are corrected based on the correlation, so as to obtain a corresponding first cross attention feature and second cross attention feature, and the first cross attention feature and the second cross attention feature are used as features of two modes to input into the prediction model, so as to obtain a more accurate over-period prediction result.

In one embodiment, the step S204 obtains basic information features of the object to be predicted, which specifically includes: acquiring basic information of an object to be predicted; normalizing the basic information to obtain normalized basic information; and processing the normalized basic information through the deep neural network to obtain the basic information characteristics of the object to be predicted.

Illustratively, basic information of the object to be predicted is acquired, including work information, income information, credit information, investment product purchase information, and the like of the object to be predicted. If the text type information is used, the text type information is converted into word vectors. And carrying out normalization processing on the basic information, carrying out normalization processing based on the maximum and minimum values if the basic information is the digital information, and carrying out vector normalization directly if the basic information is the vector information. Processing the normalized basic information through a deep neural network to obtain basic information characteristics of an object to be predicted; the number of layers of the deep neural network can be set by a user as required, and parameters in the deep neural network can be determined in advance through training.

In one embodiment, after obtaining the expiration probability of the resource pre-paid for the target to be predicted for repayment in step S204, the method specifically further includes: acquiring an out-of-period probability threshold; when the over-period probability of the object to be predicted repayment pre-branched resources exceeds the over-period probability threshold value, the object to be predicted is determined to be the object with the over-period risk.

The method comprises the steps of obtaining a preset overtime probability threshold value of a user, determining the object to be predicted as an object with overtime risk when the overtime probability of the resource reserved by the object to be predicted is larger than the overtime probability threshold value, informing the user of object information with the overtime risk in a mail mode and the like, enabling the user to conveniently grasp the overtime risk, and further enhancing overtime management.

In another embodiment, as shown in fig. 4, there is provided an over-prediction method, including the steps of:

step S401, acquiring first resource flow data of an object to be predicted in a history period.

Step S402, constructing a directed relation graph between the object to be predicted and other objects according to the first resource flow data.

Step S403, obtaining second resource flow data of other objects in the history period.

Step S404, normalizing the second resource flow data to obtain normalized second resource flow data.

And step S405, processing the normalized second resource flow data through a perceptron to obtain object features of other objects.

Step S406, determining the first node neighbor number and the second node neighbor number of the object to be predicted and the corresponding nodes of other objects in the directed relation graph, and taking the first node neighbor number and the second neighbor node number as the structural parameters of the object to be predicted and the corresponding nodes of other objects.

And S407, processing the object characteristics and the structural parameters through a graph convolution neural network to obtain the resource flow characteristics of the object to be predicted.

Step S408, obtaining basic information of an object to be predicted; and normalizing the basic information to obtain normalized basic information.

And S409, processing the normalized basic information through a deep neural network to obtain basic information characteristics of the object to be predicted.

Step S410, performing relation mapping on the resource flow characteristics and the basic information characteristics to obtain first cross attention characteristics corresponding to the resource flow characteristics and second cross attention characteristics corresponding to the basic information characteristics.

In step S411, the first cross attention feature and the second cross attention feature are input into the prediction model to obtain the out-of-date probability of the object to be predicted to repayment the pre-paid resources.

Step S412, obtaining an out-of-period probability threshold; when the over-period probability of the object to be predicted repayment pre-branched resources exceeds the over-period probability threshold value, the object to be predicted is determined to be the object with the over-period risk.

Illustratively, based on a bank customer to be predicted (i.e., an object to be predicted) selected by a user, first fund flow data (i.e., resource flow data) of a bank account corresponding to the bank customer to be predicted in the last half year (i.e., a historical time period) is queried; the first fund flow data comprises information generated by the actions of transferring, trading, consuming and the like of a bank client to be predicted, including fund flow time, fund flow amount, fund income/income party and the like. And integrating the funds flow data with the same party for entering and party for exiting in the first funds flow data, for example, the customer A transfers three times of funds to B with the amounts of 100, 80 and 50 respectively, and simultaneously transfers two times of funds to A with the amounts of 40 and 30 respectively, so that the funds flow data between A and B are integrated into 230 for transferring A to B and 70 for transferring B to A. And constructing a directed relation diagram taking the bank customer to be predicted as a central node according to the integrated first fund flowing data, describing the fund flowing data by using the directed relation diagram, wherein the directed relation diagram comprises a plurality of nodes, each node corresponds to a fund posting party or a fund issuing party, and a directed edge is arranged between two nodes with the fund flowing relation. And then according to the first fund flow data of the bank clients to be predicted, determining other bank clients which have transaction relations with the bank clients to be predicted in the last half year, and inquiring the second fund flow data of the other bank clients. Normalizing the second fund flowing data, and inputting the normalized second fund flowing data into a perception machine which is trained in advance to obtain object characteristics of other bank clients; meanwhile, determining the neighbor number of each node in the directed relation graph according to the directed relation graph; and then, inputting the object characteristics and the neighbor number of each node into a pre-trained graph convolution neural network, and predicting the resource flow characteristics of the bank clients. In addition, inquiring basic information of a bank customer to be predicted, including resident address, age, gender, expense condition, public accumulation payment condition, financial product purchase condition and the like, normalizing digital class information in the basic information by using a maximum and minimum value, and directly normalizing vector class information; and then, processing the normalized basic information through a pre-trained deep neural network to obtain the basic information characteristics of the object to be predicted. Then, based on a cross attention mechanism, the features (the resource flow features and the basic information features) of the two modes are mapped to obtain the interdependence relationship between the features of the two modes, so that the importance degree of the features of the two modes is corrected to obtain a first cross attention feature and a second cross attention feature. The first cross attention feature and the second cross attention feature are input into a pre-trained prediction model to obtain the overtime probability of the bank client to be predicted to pay back the fund (such as loan) pre-paid. And determining the overtime risk of the bank user to be predicted according to the overtime probability.

In the embodiment, the fund flow characteristics of customer transfer, transaction, consumption and the like are extracted through the graph convolution neural network, and are combined with the basic information characteristics of the customer, so that not only is the basic information of self credit and income of the customer considered, but also the fund flow condition of the customer in the latest time period is considered; the customer basic information features are extracted through the deep neural network, and the fund flow features of the customers and the importance degrees corresponding to the basic information features are corrected according to the cross attention mechanism, so that the customers can be more accurately and timely helped to master the excessive risk of the loans of the customers, and further risk management is enhanced.

In order to facilitate understanding of the embodiments of the present application by those skilled in the art, the present application will be described below with a specific example of an over-prediction method. For convenience of understanding, in this example, the network layer is described according to a single-layer network model, and in practical application, the network layer number is a super parameter and can be adjusted by a user. The method specifically comprises the following steps:

step one, constructing a client directed relation graph according to data such as transfer transaction records, consumption records and the like of a client to be predicted in the last half year, and extracting a characteristic vector of the client fund flow by using a graph convolution neural network:

(1) Determining the transaction object of the customer to be predicted in the last half year, and normalizing the transfer, transaction and transaction amount recorded in the last half year of the transaction object to obtain U _{attr_trans} 。

(2) Construction of transaction amount feature vector U of transaction object _{attr_feat} Formula (wherein W and B are model parameters, updated by model training after random initialization, σ is an activation function):

U _{attr_feat} ＝σ(W _{attr_trans} *U _{attr_trans} +B _{attr_trans} )

(3) Constructing object feature vector U of transaction object at corresponding node of directed graph _{node_feat} Formula (U) _embedding Representing other feature vectors for the transaction object than the transaction amount feature, the specific feature extraction may be done by a trained perceptron):

U _{node_feat} ＝U _embedding +U _{attr_feat}

(4) Computing user characteristics U of clients to be predicted using a graph-convolution neural network _trans The formula (wherein W and B are model parameters, after random initialization, the model is trained and updated, sigma is an activation function, and deg (U (i)) represents the neighbor number of a client i to be predicted in a directed relation diagram, and N _i Representing all neighbor sets (i.e., transaction objects) of the client i to be predicted in the directed graph, each transaction objectDenoted by j and is used to indicate the number,

representation (adjacency matrix + identity matrix of directed graph),>

representation->

Degree matrix of->

Representation matrix->

Data in row j column i):

step two, basic information such as customer resident addresses, ages, sexes, balance conditions, public accumulation and payment conditions, financial product purchase conditions and the like is used, and feature vectors of the basic information of the customer are extracted through a deep neural network after normalization:

Normalization operation is carried out on each basic information (in order to facilitate better learning of the deep neural network) so as to obtain the original basic information characteristics U _{base_src} Extracting feature vectors of customer basic information by using a deep neural network, wherein a formula (W and B are model parameters, and after random initialization, updating by model training, and sigma is an activation function):

U _base ＝σ(W _{base_src} *U _{base_src} +B _{base_src} )

step three, combining U by using a cross-attention mechanism _trans And U _base (wherein W and B are modesType parameters, updated by model training after random initialization):

(1) calculation is based on U _trans Cross attention results ATTN of (c) _{trans_base} ：

(1) U is transformed by linear transformation _base Projected to U _trans The characteristic space is located to obtain a Query vector Q _{trans_base}

Q _{trans_base} ＝W _{trans_base} *U _base +B _{trans_base}

(2) Construction of U by linear transformation _trans Key vector K in feature space _trans

K _trans ＝W _{trans_k} *U _trans +B _{trans_k}

(3) Calculate Q _{trans_base} And K _trans Similarity sim of (2) _{trans_base} (with vector cosine similarity,

each vector can be considered as a whole to calculate

(4) Construction of U by linear transformation _trans Value vector V in feature space _trans

V _trans ＝W _{trans_v} *U _trans +B _{trans_v}

(5) Combining similarity sim _{trans_base} And Value vector V _trans Calculation is based on U _trans Cross attention results ATTN of (c) _{trans_base}

ATTN _{trans_base} ＝sim _{trans_base} *V _trans

(2) Calculation is based on U _base Cross attention results ATTN of (c) _{base_trans} ：

(1) U is transformed by linear transformation _trans Projected to U _base The characteristic space is located to obtain a Query vector Q _{base_trans}

Q _{base_trans} ＝W _{base_trans} *U _trans +B _{base_trans}

(2) Construction of U by linear transformation _base Key vector K in feature space _base

K _base ＝W _{base_k} *U _base +B _{base_k}

(3) Calculate Q _{base_trans} And K _base Similarity sim of (2) _{base_trans} (with vector cosine similarity, each vector can be considered as a whole to calculate)

(4) Construction of U by linear transformation _base Value vector V in feature space _base

V _base ＝W _{base_v} *U _base +B _{base_v}

(5) Combining similarity sim _{base_trans} And Value vector V _base Calculation is based on U _base Cross attention results ATTN of (c) _{base_trans}

ATTN _{base_trans} ＝sim _{base_trans} *V _base

Step four, using the cross attention feature of the step three, inputting a perceptron to classify, predicting an excess probability value P of a client loan, and adopting a formula (wherein W and B are model parameters, and after random initialization, updating through model training, sigma is an activation function, and concat is a feature splicing function)

/>

And fifthly, taking the client with P larger than the preset overtime probability threshold as a risk client according to the overtime probability value P of the loan obtained in the step four, so that a bank can master the overtime risk of the client loan more timely, and further, the risk management is enhanced.

In the example, the fund flow characteristics of customer transfer, transaction, consumption and the like are extracted through the graph convolutional neural network, and are combined with the basic information characteristics of the customer, so that not only is the basic information of self credit and income of the customer considered, but also the fund flow condition of the customer in the latest time period is considered, and in addition, the fund flow characteristics are determined by the idea of grouping by people based on the object characteristics of other objects; the customer basic information features are extracted through the deep neural network, and the fund flow features of the customers and the importance degrees corresponding to the basic information features are corrected according to the cross attention mechanism, so that the bank can be more accurately and timely helped to master the excessive risk of the loan of the customers, and further risk management is enhanced.

It should be understood that, although the steps in the flowcharts related to the embodiments described above are sequentially shown as indicated by arrows, these steps are not necessarily sequentially performed in the order indicated by the arrows. The steps are not strictly limited to the order of execution unless explicitly recited herein, and the steps may be executed in other orders. Moreover, at least some of the steps in the flowcharts described in the above embodiments may include a plurality of steps or a plurality of stages, which are not necessarily performed at the same time, but may be performed at different times, and the order of the steps or stages is not necessarily performed sequentially, but may be performed alternately or alternately with at least some of the other steps or stages.

Based on the same inventive concept, the embodiment of the application also provides an over-period prediction device for realizing the above-mentioned over-period prediction method. The implementation of the solution provided by the apparatus is similar to the implementation described in the above method, so the specific limitation in the embodiments of the over-prediction apparatus provided below may be referred to the limitation of the over-prediction method hereinabove, and will not be repeated here.

In one embodiment, as shown in fig. 5, there is provided an over-period prediction apparatus including: a data acquisition module 501, a directed graph construction module 502, a feature extraction module 503, and an out-of-date prediction module 504, wherein:

a data obtaining module 501, configured to obtain first resource flow data of an object to be predicted in a historical time period; the first resource flow data represents the resource flow condition between the object to be predicted and other objects;

the directed graph construction module 502 is configured to construct a directed graph between the object to be predicted and other objects according to the first resource flow data; the directed relation graph comprises a plurality of nodes, each node corresponds to an object, and a directed edge is arranged between two nodes with resource flow relation;

the feature extraction module 503 acquires second resource flow data of other objects in a historical time period, and extracts resource flow features of the object to be predicted based on the second resource flow data and the directed relation graph;

and the over-period prediction module 504 is configured to obtain basic information features of the object to be predicted, input the resource flow features and the basic information features into the prediction model, and obtain an over-period probability of the object to be predicted to repayment the pre-supported resource.

In one embodiment, the feature extraction module 503 is further configured to extract object features of other objects based on the second resource flow data; based on the directed relation diagram, determining structural parameters of the object to be predicted and nodes corresponding to other objects; and processing the object characteristics and the structural parameters through the graph convolution neural network to obtain the resource flow characteristics of the object to be predicted.

In one embodiment, the feature extraction module 503 is further configured to normalize the second resource flow data to obtain normalized second resource flow data; and processing the normalized second resource flow data through a perceptron to obtain object characteristics of other objects.

In an embodiment, the feature extraction module 503 is further configured to determine a number of neighbors of the first node in the directed graph, where the number of neighbors of the first node is used as a structural parameter of the corresponding node of the object to be predicted; and determining the neighbor number of the second node of the nodes corresponding to other objects in the directed relation graph, and taking the neighbor number of the second node as the structural parameter of the nodes corresponding to other objects.

In one embodiment, the above out-of-life prediction module 504 is further configured to perform a relationship mapping on the resource flow feature and the basic information feature to obtain a first cross attention feature corresponding to the resource flow feature and a second cross attention feature corresponding to the basic information feature; and inputting the first cross attention characteristic and the second cross attention characteristic into a prediction model to obtain the out-of-date probability of the predicted resource for repayment of the object to be predicted.

In one embodiment, the above-mentioned over-period prediction module 504 is further configured to obtain an over-period probability threshold; when the over-period probability of the object to be predicted repayment pre-branched resources exceeds the over-period probability threshold value, the object to be predicted is determined to be the object with the over-period risk.

In one embodiment, the over-period prediction apparatus further includes a basic information acquisition model for acquiring basic information of an object to be predicted; normalizing the basic information to obtain normalized basic information; and processing the normalized basic information through the deep neural network to obtain the basic information characteristics of the object to be predicted.

The above-described individual modules in the over-period prediction apparatus may be implemented in whole or in part by software, hardware, or a combination thereof. The above modules may be embedded in hardware or may be independent of a processor in the computer device, or may be stored in software in a memory in the computer device, so that the processor may call and execute operations corresponding to the above modules.

In one embodiment, a computer device is provided, which may be a server, the internal structure of which may be as shown in fig. 6. The computer device includes a processor, a memory, an Input/Output interface (I/O) and a communication interface. The processor, the memory and the input/output interface are connected through a system bus, and the communication interface is connected to the system bus through the input/output interface. Wherein the processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The database of the computer device is used for storing resource flow data and basic information of the object to be predicted. The input/output interface of the computer device is used to exchange information between the processor and the external device. The communication interface of the computer device is used for communicating with an external terminal through a network connection. The computer program is executed by a processor to implement a method of over-prediction.

It will be appreciated by those skilled in the art that the structure shown in fig. 6 is merely a block diagram of some of the structures associated with the present application and is not limiting of the computer device to which the present application may be applied, and that a particular computer device may include more or fewer components than shown, or may combine certain components, or have a different arrangement of components.

In one embodiment, a computer device is provided comprising a memory and a processor, the memory having stored therein a computer program, the processor when executing the computer program performing the steps of:

acquiring first resource flow data of an object to be predicted in a historical time period; the first resource flow data represents the resource flow condition between the object to be predicted and other objects;

constructing a directed relation graph between the object to be predicted and other objects according to the first resource flow data; the directed relation graph comprises a plurality of nodes, each node corresponds to an object, and a directed edge is arranged between two nodes with resource flow relation;

acquiring second resource flow data of other objects in a historical time period, and acquiring resource flow characteristics of the object to be predicted based on the second resource flow data and the directed relation graph;

And acquiring basic information characteristics of the object to be predicted, inputting the resource flow characteristics and the basic information characteristics into a prediction model, and obtaining the out-of-date probability of the resource predicted by the object to be predicted.

In an embodiment, there is also provided a computer device comprising a memory and a processor, the memory having stored therein a computer program, the processor implementing the steps of the method embodiments described above when the computer program is executed.

In one embodiment, a computer readable storage medium is provided having a computer program stored thereon, which when executed by a processor, performs the steps of:

In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored which, when executed by a processor, carries out the steps of the method embodiments described above.

In one embodiment, a computer program product is provided comprising a computer program which, when executed by a processor, performs the steps of:

In an embodiment, a computer program product is provided, comprising a computer program which, when executed by a processor, implements the steps of the method embodiments described above.

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

Those skilled in the art will appreciate that implementing all or part of the above described methods may be accomplished by way of a computer program stored on a non-transitory computer readable storage medium, which when executed, may comprise the steps of the embodiments of the methods described above. Any reference to memory, database, or other medium used in the various embodiments provided herein may include at least one of non-volatile and volatile memory. The nonvolatile Memory may include Read-Only Memory (ROM), magnetic tape, floppy disk, flash Memory, optical Memory, high density embedded nonvolatile Memory, resistive random access Memory (ReRAM), magnetic random access Memory (Magnetoresistive Random Access Memory, MRAM), ferroelectric Memory (Ferroelectric Random Access Memory, FRAM), phase change Memory (Phase Change Memory, PCM), graphene Memory, and the like. Volatile memory can include random access memory (Random Access Memory, RAM) or external cache memory, and the like. By way of illustration, and not limitation, RAM can be in the form of a variety of forms, such as static random access memory (Static Random Access Memory, SRAM) or dynamic random access memory (Dynamic Random Access Memory, DRAM), and the like. The databases referred to in the various embodiments provided herein may include at least one of relational databases and non-relational databases. The non-relational database may include, but is not limited to, a blockchain-based distributed database, and the like. The processors referred to in the embodiments provided herein may be general purpose processors, central processing units, graphics processors, digital signal processors, programmable logic units, quantum computing-based data processing logic units, etc., without being limited thereto.

The technical features of the above embodiments may be arbitrarily combined, and all possible combinations of the technical features in the above embodiments are not described for brevity of description, however, as long as there is no contradiction between the combinations of the technical features, they should be considered as the scope of the description.

The above examples only represent a few embodiments of the present application, which are described in more detail and are not to be construed as limiting the scope of the present application. It should be noted that it would be apparent to those skilled in the art that various modifications and improvements could be made without departing from the spirit of the present application, which would be within the scope of the present application. Accordingly, the scope of protection of the present application shall be subject to the appended claims.

Claims

1. A method of over-prediction, the method comprising:

2. The method of claim 1, wherein the extracting the resource flow characteristics of the object to be predicted based on the second resource flow data and the directed graph comprises:

3. The method of claim 2, wherein extracting object features of the other objects based on the second resource flow data comprises:

4. The method according to claim 2, wherein the determining structural parameters of the object to be predicted and the corresponding nodes of the other objects based on the directed graph comprises:

5. The method of claim 1, wherein the inputting the resource flow characteristics and the base information characteristics into a predictive model yields an out-of-date probability of the object to be predicted to refund the pre-paid resources, comprising:

6. The method according to claim 1, wherein said obtaining basic information features of the object to be predicted comprises:

acquiring basic information of the object to be predicted;

normalizing the basic information to obtain normalized basic information;

7. The method of claim 1, further comprising, after said deriving an excess probability that said object to be predicted refunds a pre-paid resource:

acquiring an out-of-period probability threshold;

8. An overrun prediction device, the device comprising:

9. A computer device comprising a memory and a processor, the memory storing a computer program, characterized in that the processor implements the steps of the method of any of claims 1 to 7 when the computer program is executed.

10. A computer readable storage medium, on which a computer program is stored, characterized in that the computer program, when being executed by a processor, implements the steps of the method of any of claims 1 to 7.

11. A computer program product comprising a computer program, characterized in that the computer program, when being executed by a processor, implements the steps of the method of any of claims 1 to 7.