CN112905340B

CN112905340B - System resource allocation method, device and equipment

Info

Publication number: CN112905340B
Application number: CN202110170693.9A
Authority: CN
Inventors: 李瑾瑜; 裴洪斌; 马超; 宋虎
Original assignee: Industrial and Commercial Bank of China Ltd ICBC
Current assignee: Industrial and Commercial Bank of China Ltd ICBC
Priority date: 2021-02-08
Filing date: 2021-02-08
Publication date: 2024-07-02
Anticipated expiration: 2041-02-08
Also published as: CN112905340A

Abstract

The embodiment of the specification relates to the technical field of artificial intelligence, and discloses a system resource allocation method, a device and equipment, wherein the method comprises the following steps: the risk event data are collected, a risk relation network map is constructed based on the collected relation data, sub-graph segmentation of abnormal nodes is carried out on the risk relation network map, and the influence calculation of the propagation degree of the abnormal event is carried out based on the segmented abnormal node sub-graph. And combining the non-graph influence score characteristics and the network structure graph characteristics, training and constructing a risk prediction model, performing risk prediction on the target user based on the trained risk prediction model, and distributing resources of the target user based on a risk prediction result. The accuracy of the risk prediction model is improved, and further, the accuracy of risk prediction is improved. The whole process is highly abstract, weakly associated with business scenes, and the abnormal event predictions of different risk types can be applicable to the flow line, so that the applicability is wide.

Description

System resource allocation method, device and equipment

Technical Field

The present disclosure relates to the field of artificial intelligence technologies, and in particular, to a system resource allocation method, apparatus, and device.

Background

With the rapid development of big data service platform technology, financial resource service types and selectable service channels are more and more diversified and more convenient, and risk prediction of users is more and more important for financial institutions. For example, on-line loan business which is more convenient for some service channels, because of relatively less manual intervention, if the risk prediction of the user is not accurate enough, larger loss can be brought to the financial institution.

The current commonly used user risk prediction method is mainly a classification method of an intelligent learning model, modeling is carried out based on data of known client risks, and a new sample is used for carrying out user risk prediction by utilizing a model obtained through training so as to determine the risk of a user. However, the existing model establishment has high requirements on the professional knowledge of modeling staff, and needs to master the professional knowledge in different fields, and the related staff are as follows: modeling personnel, business specialists, developers, operation and maintenance personnel, and the like. The means and the accuracy of the model established by different modeling personnel may be different, so that the accuracy of the risk prediction of the user is affected, the accuracy of resources allocated to the corresponding user is further affected, and the use experience of the user is reduced.

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

Disclosure of Invention

An object of an embodiment of the present disclosure is to provide a system resource allocation method, apparatus, and device, which improve accuracy of risk prediction, so as to adjust system resources of a user in time, and improve system performance.

In one aspect, an embodiment of the present disclosure provides a system resource allocation method, applied to a server, where the method includes:

Collecting a risk prediction sample set for representing risk characteristics of a user, wherein the risk prediction sample set comprises relation data of a plurality of prediction sample events and basic risk characteristics corresponding to each prediction sample event;

Constructing a risk relation network map of the predicted sample event based on the relation data, wherein the risk relation network map comprises nodes and edges connected with the nodes, the nodes are event or relation parties associated with the predicted sample event, and the edges are used for representing the relation among the nodes;

Performing abnormal node subgraph segmentation on the risk relation network graph to obtain risk relation subgraphs corresponding to abnormal nodes in the risk relation network graph;

calculating event influence scores of all event nodes and relation influence scores of all relation nodes in all risk relation subgraphs;

Extracting influence score characteristics and network structure characteristics of each node in each risk relationship sub-graph based on event influence scores of event nodes and relationship influence scores of relationship nodes corresponding to each risk relationship sub-graph;

And carrying out model training on the risk prediction model by utilizing at least one of the basic risk feature, the influence score feature and the network structure feature to obtain a trained risk prediction model, so as to carry out risk prediction on a target user by utilizing the trained risk prediction model, and distributing system resources to the target user based on the risk prediction result.

Further, the model training the risk prediction model by using at least one of the basic risk feature, the impact score feature and the network structure feature comprises:

performing feature combination on the basic risk feature, the influence score feature and the network structure feature to obtain a plurality of combined features;

Inputting each combined feature into a feature screening model, taking risk labels corresponding to each predicted sample event as output of the feature screening model, and scoring the features in each combined feature by using the feature screening model to obtain importance scores corresponding to each feature;

Taking the feature with the importance score larger than a preset threshold value as a risk prediction target feature;

And carrying out model training on the risk prediction model by utilizing the risk prediction target characteristics.

Further, the performing abnormal node subgraph segmentation on the risk relation network graph includes:

Determining the range of an abnormal node subgraph corresponding to the abnormal node by taking the abnormal node in the risk relation network graph as a starting point according to a subgraph segmentation condition, and segmenting the abnormal node subgraph corresponding to the abnormal node from the risk relation network graph based on the determined range, wherein the subgraph segmentation condition comprises at least one of the following components:

maximum order of associated nodes of abnormal nodes in the abnormal node subgraph;

the maximum number of nodes in the abnormal node subgraph;

Minimum degree of associated nodes of abnormal nodes in the abnormal node subgraph;

The relationship between nodes in the abnormal node subgraph is a specified relationship type.

Further, the calculating the event influence score of each event node and the relationship influence score of each relationship node in each risk relationship sub-graph includes:

according to the types of the edges connected with each node in each risk relation sub-graph, calculating node connection relation values among each node in each risk relation sub-graph, wherein the node connection relation values are used for representing the tightness degree of the connection relation among each node;

Calculating the degree of each node in each risk relation subgraph according to the weight of the edge connected with each node in each risk relation subgraph;

And calculating the event influence score of each event node and the relationship influence score of each relationship node in each risk relationship sub-graph according to the node connection relationship value and the degree of each node.

Further, the method includes calculating an event impact score for each event node and a relationship impact score for each relationship node using the following formula:

Wherein, Representing event impact scores corresponding to the ith event node, s _i representing the ith event node, alpha epsilon [0,1], beta epsilon [0,1] being a hyper-parameter, n _R representing the total number of relationship nodes in the risk relationship graph,Representing a node connection relationship value between the ith event node and the jth relationship node,Representing the degree of the ith event node,Representing the degree of the j-th relationship node,Representing a relationship influence score corresponding to a j-th relationship node, r _i representing a j-th relationship node,Representing the initial value of the event impact score corresponding to the ith event node, n _S representing the total number of event nodes in the risk relationship graph,And the initial value of the relationship influence score corresponding to the j-th relationship node is represented.

Further, the calculating a node connection relationship value between each node in each risk relationship sub-graph according to the type of the edge connected to each node in each risk relationship sub-graph includes:

acquiring the weight of the edge connected with each node according to the type of the edge connected with each node in each risk relation sub-graph;

Based on the weight of the edge connected with each node, calculating the node connection relation value between each node in each risk relation sub-graph according to the following formula:

Where s _i represents the ith event node, r _i represents the jth relationship node, Representing a node connection relationship value between the ith event node and the jth relationship node,Representing the weight of the edge between event node s _i and relationship node r _j.

Further, the method further comprises:

After a trained risk prediction model is obtained, the trained risk prediction model is released to be online, and the calling times and the running state of the risk prediction model are recorded;

providing interpretation information for a prediction result of the risk prediction model according to an interpretation request submitted by a user, and acquiring feedback information of the user on the interpretation information;

And maintaining the risk prediction model based on the calling times and the running state, and optimizing the risk prediction model based on the feedback information.

Further, the risk prediction model includes: an event state prediction model and an abnormal event prediction model;

when the trained risk prediction model is used for carrying out risk prediction on a target user, the trained event state prediction model and the trained abnormal event prediction model are used for carrying out risk prediction on the target user, and the risk prediction result of the target user is comprehensively obtained based on the results output by the event state prediction model and the abnormal event prediction model.

In another aspect, the present specification provides a system resource allocation apparatus, the apparatus comprising:

the data acquisition module is used for acquiring a risk prediction sample set for representing risk characteristics of a user, wherein the risk prediction sample set comprises relationship data of a plurality of prediction sample events and basic risk characteristics corresponding to each prediction sample event;

The network map drawing module is used for constructing a risk relation network map of the predicted sample event based on the relation data, wherein the risk relation network map comprises nodes and edges connected with the nodes, the nodes are events or relational parties associated with the predicted sample event, and the edges are used for representing the relation among the nodes;

The abnormal sub-graph segmentation module is used for carrying out abnormal node sub-graph segmentation on the risk relation network graph to obtain risk relation sub-graphs corresponding to each abnormal node in the risk relation network graph;

The node influence analysis module is used for calculating event influence scores of all event nodes in all risk relation subgraphs and relation influence scores of all relation nodes;

The feature extraction module is used for extracting the influence score features and the network structure features of the predicted sample event based on the event influence scores of the event nodes corresponding to each risk relation subgraph and the relation influence scores of the relation nodes;

and the resource allocation module is used for carrying out model training on the risk prediction model by utilizing at least one of the basic risk characteristic, the influence score characteristic and the network structure characteristic to obtain a trained risk prediction model so as to carry out risk prediction on a target user by utilizing the trained risk prediction model, so as to allocate system resources to the target user based on the risk prediction result.

In yet another aspect, embodiments of the present disclosure provide a system resource allocation device, applied to a server, where the device includes at least one processor and a memory for storing processor-executable instructions, and where the instructions, when executed by the processor, implement a system resource allocation method including the above.

According to the system resource allocation method, the device and the equipment provided by the specification, the risk relation network map is constructed based on the acquired relation data by acquiring the data of the risk event, then the sub-graph segmentation of the abnormal nodes is carried out on the risk relation network map, and the influence calculation of the propagation degree of the abnormal event is carried out based on the segmented abnormal node sub-graph. And combining the non-graph influence score characteristics and the network structure graph characteristics, training and constructing a risk prediction model, performing risk prediction on the target user based on the trained risk prediction model, and distributing resources of the target user based on a risk prediction result. The accuracy of the risk prediction model is improved, and further, the accuracy of risk prediction is improved. The whole process is highly abstract, is weakly associated with a business scene, can be suitable for the production line of different risk types of abnormal event prediction, has wide applicability, can complete the creation of a risk prediction model without having very deep expert knowledge, reduces the complexity of model creation and reduces the time of model creation.

Drawings

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

Fig. 1 is a schematic flow chart of an embodiment of a system resource allocation method provided in an embodiment of the present disclosure;

FIG. 2 is a schematic diagram of a risk relationship network graph in one embodiment of the present disclosure;

FIG. 3 is a schematic representation of the order of associated nodes in one embodiment of the present disclosure;

FIG. 4 is a schematic diagram of a system resource allocation device according to one embodiment of the present disclosure;

Fig. 5 is a block diagram of a hardware configuration of a system resource allocation server in one embodiment of the present specification.

Detailed Description

In order to make the technical solutions in the present specification better understood by those skilled in the art, the technical solutions in the embodiments of the present specification will be clearly and completely described below with reference to the drawings in the embodiments of the present specification, and it is obvious that the described embodiments are only some embodiments of the present specification, not all embodiments. All other embodiments, which can be made by one of ordinary skill in the art without undue burden from the present disclosure, are intended to be within the scope of the present disclosure.

In one scenario example provided in the embodiments of the present disclosure, the system resource allocation method may be applied to a device that performs system resource allocation, where the device may include one server, or may include a server cluster formed by a plurality of servers. For the target user, the server can extract feature data of various information of the target user as feature data of the target user, and then, the risk prediction is carried out on the target user by utilizing a preconfigured algorithm or model and the like to obtain a risk prediction result of the target user so as to adjust the resources of the target user based on the risk prediction result. The resource may include, for example, a data resource such as a service, product, etc. offered or recommended to the user, such as for a loan business scenario, the resource may be the amount of the loan allocated to the target user. For the cloud platform data service business scenario, the resources can be system data resources allocated to the target user and the like. Through accurately identifying the risk of the user, the resources can be more accurately and reasonably allocated to the user, the experience of the user is improved, and the resource loss of the mechanism is effectively reduced.

Fig. 1 is a flowchart of an embodiment of a system resource allocation method provided in an embodiment of the present disclosure. Although the present description provides methods and apparatus structures as shown in the examples or figures described below, more or fewer steps or modular units may be included in the methods or apparatus, either conventionally or without inventive effort. In the steps or the structures where there is no necessary causal relationship logically, the execution order of the steps or the module structure of the apparatus is not limited to the execution order or the module structure shown in the embodiments or the drawings of the present specification. In actual implementation, the apparatus, the server, or the end product of the method or the module structure of (a) may execute sequentially or in parallel (e.g., in a parallel processor or a multi-threaded processing environment, or even in an implementation environment including distributed processing, server clusters) according to an embodiment or a method or a module structure shown in the drawings.

In a specific embodiment, as shown in fig. 1, in one embodiment of the system resource allocation method provided in the present disclosure, the method may be applied to a server, a computer, a smart phone, a tablet computer, and other devices, and the method may include the following steps:

Step 102, collecting a risk prediction sample set for representing risk characteristics of a user, wherein the risk prediction sample set comprises relation data of a plurality of prediction sample events and basic risk characteristics corresponding to each prediction sample event.

In a specific implementation process, according to the type of an event requiring risk prediction, acquiring the relationship data of a predicted sample event of a corresponding type, and performing data processing such as feature extraction on the acquired relationship data to acquire basic risk features corresponding to each predicted sample event. For example: the type of time to be predicted may be determined first, such as: loans, credit cards, insurance, etc. these may be of the default or fraudulent event type, with the event being the center, and the direct parties to the event being found. For example, the relationship data related to a loan event may include a borrower, a guarantor, and the like. The obtained predicted sample event has an abnormal event with risk, a normal event without risk, and an unknown event without risk, and the predicted sample time can be marked based on the risk state of the event. Labeling can be divided into three categories: abnormal part, normal part, unknown. An abnormal part refers to an event in which an abnormal situation occurs, such as default, fraud, etc., a normal part is an event in which a user determines that there is no problem, and it is unknown that the user does not confirm whether an abnormality occurs. Feature extraction is performed based on the relation data of each predicted sample event, and basic risk features of the predicted sample time are obtained, for example: and extracting the basic characteristics of the event related to each predicted sample event and the relation party to generate different broad tables. Taking a fraud event as an example, the basic risk features of event extraction, such as the amount of event fraud, the length of time from application passage until fraud occurs, the number of relational parties involved in the fraud. Basic risk features extracted by the relationship party, such as legal responsibilities assumed by the relationship party, fraudulent events related to the relationship party, and events related to the relationship party. The basic risk profile may be denoted as x ^owner.

Step 104, constructing a risk relation network map of the predicted sample event based on the relation data, wherein the risk relation network map comprises nodes and edges connected with the nodes, the nodes are events or relational parties associated with the predicted sample event, and the edges are used for representing the relation among the nodes.

In a specific implementation process, a relationship network structure can be formed by utilizing modes such as a graph database and the like based on the relationship data corresponding to each predicted sample event, and a risk relationship network map is constructed. When building a relational network structure, the event is required to be connected with a relational party only, and the event is not directly connected with the event. FIG. 2 is a schematic diagram of a risk relationship network map in one embodiment of the present disclosure, where the risk relationship network map includes nodes and edges connecting the nodes, and the nodes may be events associated with predicted sample events, such as: loans, repayment, etc., and the relationship involved in the event are as follows: borrowers, insurers, etc., different types of nodes may be represented using different icons, edges between nodes may represent relationships between nodes, and there may be connected edges between nodes in the relationship. As shown in fig. 2, in the embodiment of the present specification, an event node may be defined as s _i,i∈{1,...,n_S, a relationship node as r _j,j∈{1,...,n_R,Are edges between nodes.

The risk relation network map may be drawn corresponding to one predicted sample event, or a risk relation network map may be drawn together with the predicted sample event having a relation, or a risk relation network map may be drawn together with all the predicted sample events, which may be specifically set according to actual needs, and embodiments of the present disclosure are not limited specifically.

And 106, carrying out abnormal node subgraph segmentation on the risk relation network graph to obtain risk relation subgraphs corresponding to the abnormal nodes in the risk relation network graph.

In a specific implementation process, because the original relationship network may be very huge, in the embodiment of the present disclosure, the subgraph may be segmented according to the influence range of a single event, in the embodiment of the present disclosure, mainly using an abnormal node as a center, and segmenting nodes and edges closely related to the abnormal node from the whole risk relationship network map, so as to obtain a risk relationship subgraph corresponding to each abnormal node. The abnormal nodes are abnormal event nodes generally, whether each event node in the risk relation network map is abnormal or not can be determined according to the real state of the predicted sample event, and the sub-graph range is determined from event points marked as abnormal pieces, and each abnormal point is used for segmenting one sub-graph.

In some embodiments of the present disclosure, an abnormal node in the risk relationship network map may be used as a starting point, a range of an abnormal node sub-graph corresponding to the abnormal node may be determined according to a sub-graph segmentation condition, and the abnormal node sub-graph corresponding to the abnormal node may be segmented from the risk relationship network map based on the determined range, where the sub-graph segmentation condition includes at least one of the following:

the maximum number of nodes in the abnormal node subgraph;

In a specific implementation process, in the embodiment of the present disclosure, multiple sub-graph splitting modes may be provided, where the first is: the maximum order of the associated node associated with the abnormal node is taken as a segmentation condition, namely, how many steps are associated from the abnormal node, and in the implementation of the specification, the system defaults to 6 steps and the default value of the order supports user adjustment. Wherein, the associated node can be understood as a node directly or indirectly connected with the abnormal node, and the order of the associated node can be understood as the number of edges between the associated node and the abnormal node. Fig. 3 is a schematic order diagram of an association node in an embodiment of the present disclosure, as shown in fig. 3, with a relationship node r1 as a starting point, the nodes s1, s2, s3 are all first-order association nodes of the relationship node r1, that is, the order between the nodes s1, s2, s3 and the node r1 is 1, and r2, r3 are second-order association nodes of the relationship node r1, that is, the order between the nodes r2, r3 and the node r1 is 2.

The first slicing mode may also be set as a distance between outliers, or may also be understood as setting an order spaced from outliers. Such as: the system defaults that the distance order between two abnormal nodes exceeds 6 steps (namely point 1 (event) -side 1-point 2 (relational party) -side 2-point 3 (event) -side 3-point 4 (relational party) -side 4-point 5 (event) -side 5-point 6 (relational party) -side 6-point 7 (event)), no obvious influence is considered between the two abnormal nodes, sub-graph segmentation is carried out at the side 6, and user adjustment is supported by the distance order defaults.

The second segmentation mode is: and setting the maximum node number of each abnormal node subgraph. Such as: the system defaults to 200 points. The quantity default value supports user adjustment. The method comprises the steps of taking an abnormal node in a risk relation network map as a starting point, dividing a target node and a corresponding edge in the risk relation network map into risk relation subgraphs corresponding to the abnormal node, wherein the target node is directly or indirectly connected with the abnormal node, and the number of nodes in the divided risk relation subgraphs is smaller than the maximum number of nodes.

The third segmentation mode is: and (5) setting the density. The lowest degree threshold, i.e., the minimum degree, of the nodes in the abnormal node subgraph may be set. Points above a certain degree are included in the abnormal node subgraph. Such as: the lowest threshold of the default event node of the system is 3, the lowest threshold of the relation node is 5, and the degree default value supports user adjustment. The nodes with degrees larger than the minimum degrees and the corresponding edges of the nodes are segmented into risk relation subgraphs corresponding to the abnormal nodes by taking the abnormal nodes in the risk relation network map as a starting point, and if a plurality of points which do not meet the degree requirements exist between the nodes with the two degrees meeting the requirements, the nodes which do not meet the degree requirements can be pruned, and only the nodes which meet the degree requirements are reserved. The degree of a node may be calculated based on the weight of the first order associated node's edges, such as: the accumulated value of the weights of the first order associated node's edges of a node may be taken as the degree of that node.

The fourth middle segmentation mode is: only event-relational parties of a specific relational type are selected as the same abnormal node subgraph. Such as: the system default relationship type is the event direct principal, and the relationship default supports user adjustment. For example: and dividing the nodes with the relationship of the abnormal nodes as the designated relationship types and the corresponding edges into risk relationship subgraphs corresponding to the abnormal nodes by taking the abnormal nodes in the risk relationship network map as a starting point.

One or more types of nodes which have great influence on the abnormal event can be selected from the four segmentation modes to segment the risk relation network map, so that the abnormal node relation network is simplified, and the nodes which have great influence on the abnormal event are reserved, thereby improving the efficiency and the accuracy of risk prediction.

And 108, calculating event influence scores of all event nodes and relationship influence scores of all relationship nodes in each risk relationship subgraph.

In a specific implementation process, after the risk relation subgraphs corresponding to the abnormal nodes are segmented, event influence scores of the event nodes and relation influence scores of the relation nodes can be calculated according to the connection modes of the nodes and the edges in the risk relation subgraphs. The influence score may represent the influence degree of the abnormal node on the event node or the relationship node in the risk relationship subgraph, may be calculated based on the tightness degree of the connection relationship between the node and the abnormal node, and the specific calculation mode may be selected according to actual needs.

In some embodiments of the present disclosure, the calculating the event impact score of each event node and the relationship impact score of each relationship node in each risk relationship subgraph includes:

In a specific implementation process, node connection relation values between all nodes in each risk relation sub-graph can be calculated according to the types of edges connected with all nodes in the risk relation sub-graph, namely the relation types among the nodes, wherein the node connection relation values are used for representing the tightness degree of the connection relation between all the nodes.

In some embodiments of the present disclosure, the weights of the edges connected to each node may be obtained according to the types of the edges connected to each node in each risk relationship subgraph;

Where s _i represents the ith event node, r _i represents the jth relationship node, Representing a node connection relationship value between the ith event node and the jth relationship node,Representing the weight of the edge between event node s _i and relationship node r _j. Wherein,That is, the weight of the edge can be set according to the actual application scenario, for example: the event is a loan, the relational party is a borrower, a guarantee party, a mortgage party, a stakeholder, i.e. a loan-borrower, a loan-guarantee party, a lender-mortgage party, a lender-stakeholder, and if the importance of the guarantee relationship is considered to be higher than the investment relationship of the stakeholder, the weight of the guarantee relationship can be set to 0.7, and the investment relationship is set to 0.3.

The node connection matrix between the event node and the relationship node can be defined based on the calculated node connection relationship value:

And then calculating the degree of each node in each risk relation sub-graph based on the weight of the edge connected with each node in each risk relation sub-graph, for example: definition of the definition For the degree of the event node s _i,Is the degree of the relation node r _j. The degree is calculated according to the weights of the first-order associated points and edges. Taking the event node s _i as an example,Representing a first order associated node of event node s _i.

Based on the calculated degrees of the nodes, a degree diagonal matrix of the event and the relation party can be obtained: d _S,D_R. Wherein the method comprises the steps of(D _S)_ii represents the ith row and ith column elements of matrix D _S, (D _R)_jj represents the jth column and jth row elements of matrix D _R), with the other elements of the matrix being 0.

Based on the calculated node connection relation value and the degree of the node, the event influence score of each event node and the relation influence score of each relation node in each risk relation sub-graph can be calculated, the functional relation between the influence score of the node and the node connection relation value and the degree of the node can be analyzed through a mathematical simulation or model training or expert experience and other modes, and further, based on the calculated node connection relation value and the degree of the node, the influence score of each node in each risk relation sub-graph is calculated.

In some embodiments of the present description, the following formulas may be used to calculate the event impact score for each event node and the relationship impact score for each relationship node:

Wherein, The method is characterized in that the method comprises the steps of representing event influence scores corresponding to an ith event node, s _i represents the ith event node, alpha epsilon [0,1], beta epsilon [0,1] are super-parameters, can represent weight distribution of network structures and initial state influence scores, and can be set to be a default value alpha=0.85 and beta=1. n _R represents the total number of relationship nodes in the risk relationship graph,Representing a node connection relationship value between the ith event node and the jth relationship node,Representing the degree of the ith event node,Representing the degree of the j-th relationship node,Representing a relationship influence score corresponding to a j-th relationship node, r _i representing a j-th relationship node,Representing the initial value of the event impact score corresponding to the ith event node, n _S representing the total number of event nodes in the risk relationship graph,And the initial value of the relationship influence score corresponding to the j-th relationship node is represented.

The left part of the plus sign in the above formula may represent the influence of the network structure, and the right part may represent the influence degree of the initial state. Wherein, the impact score of the initial state can be expressed as:

All event nodes and relationship nodes in each risk relationship sub-graph can be represented in a matrix form based on the calculated influence scores of the nodes:

Wherein, D _S、D_R represents diagonal matrix of event nodes and relationship nodes, respectively, and W represents node connection matrix between nodes in the risk relationship subgraph.

The solving process comprises the following steps:

and selecting a risk relation subgraph. For each subgraph:

input: setting alpha and beta super-parameter values; initialization of

The calculation process comprises the following steps:

And (3) calculating:

Iterative calculation:

Until convergence, an event impact score for each point within the subgraph is obtained M _S、M_R.

In the embodiment of the specification, by calculating the network propagation influence score of the abnormal part, the score can be used for generating derivative characteristic variables through four-rule operation and the like. The common abnormal part influence infection is to calculate the probability that events around the network structure of the abnormal part also deteriorate into the abnormal part, but the method needs to calculate the probability of abnormal part conduction, needs to set a probability model and accumulate historical data to calculate the conduction probability, and after the probability is calculated, the probability values of the adjacent points of the abnormal part are difficult to accumulate and the like to carry out four arithmetic operations, so that the four arithmetic results lack of statistical significance, and are unfavorable for the design of derivative characteristic variables. In the embodiment of the specification, the influence score of the abnormal part on the surrounding relation network event is calculated, and the score can be subjected to four arithmetic operations to obtain different statistical indexes with statistical significance. Further, if the conduction probability is calculated, it is difficult to judge the conduction direction when both the correlation points are abnormal pieces. However, in the embodiment of the present disclosure, when calculating the influence score, the calculation principle is to influence the degree of influence of the abnormality of the origin on other points in the network structure, regardless of the states of the surrounding points, and the states of the other points do not influence the propagation of the influence degree.

And 110, extracting influence score characteristics and network structure characteristics of each node in the risk relation subgraph based on the event influence scores of the event nodes and the relation influence scores of the relation nodes corresponding to each risk relation subgraph.

In a specific implementation process, after event influence scores of event nodes and relationship influence scores of relationship nodes of each risk relationship sub-graph are calculated, feature extraction can be performed on each risk relationship sub-graph based on the influence scores of each node, and influence score features and network structure features of predicted sample events are extracted. Wherein the influence score feature may represent a feature extracted based on the influence score, such as may include: an influence score of the node; the K-order fractional number of the influence score of the neighboring node under the N-order moment of the node (for example, the numerical value on 25% of the fractional points after the influence score of the neighboring node of the first-order moment of the node is ordered from big to small); the mean value/standard deviation of the influence scores of the neighbor nodes under the N-order moment of the node. The system may default to n= {1,2}, k= {0,0.25,0.5,0.75,1}, default values support user adjustment. The impact score feature may be denoted as x ^score.

The network structural features may represent feature variables extracted by the network nodes based on the risk relationship subgraph, such as may include: the node N order; the node degree; the node N-order moment is the abnormal part duty ratio (for example, the point first-order moment neighbor node is the total line/first order of the abnormal part); the normal piece duty ratio under the N-order moment of the node; whether an abnormal part exists under the N-order moment of the node. The system may default to n= {1,2}, with default values supporting user adjustment. The network architecture feature may be denoted as x ^graph.

And 112, performing model training on the risk prediction model by using at least one of the basic risk feature, the influence score feature and the network structure feature to obtain a trained risk prediction model, and performing risk prediction on a target user by using the trained risk prediction model to allocate system resources to the target user based on the risk prediction result.

In a specific implementation process, after extracting the influence score features and the network structure features of each node in each risk relation sub-graph, the risk prediction model can be trained by combining the basic risk features extracted in the step 102. The risk prediction model can be trained by adopting a supervised mode, the system can also be provided with a model algorithm library, and the model algorithm library can comprise risk prediction models constructed by different algorithms, such as: logistic regression models, ligthGBM (LIGHT GRADIENT Boosting Machine) models, and the like. A user can select a model of a specified type from the algorithm library, and meanwhile, the user is supported to upload a model algorithm package according to a fixed template format, select a custom algorithm and perform model training by using the extracted features.

The user may choose to perform model training on one or more risk prediction models, and after generating a trained risk prediction model, obtain ROC (Receiver Operating Characteristic, receiver operating characteristic curve) values, AR (Accuracy Ratio) values, KS (Kolmogorov-Smirnov) value evaluation values of each model by default. A default may be set to filter the models in ROC, AR, KS order, leaving the models with high evaluation values. The default value supports user adjustment, the user can adjust model evaluation indexes and model screening conditions, the user can manually adjust other models, a user-defined model feature table, a path and a table name stored in a model result table, and the running frequency of the models.

After the risk prediction model is trained, the risk prediction can be carried out on the target user by using the trained risk prediction model, the risk prediction result of the target user is obtained, the prediction results of the multiple risk prediction models can be compared, and the abnormal probability predicted by the multiple risk prediction models is taken as the prediction result of the event. The target user may be a user, or may be an event corresponding to the target user. And performing resource allocation to the target user based on the risk prediction result of the target user. The resources may include, for example, data resources such as services, products, etc. provided or recommended to the user. For example, in a loan business scenario, a user with a higher risk level may be offered a smaller amount of loan or refused to be offered a loan; and for users with lower risk levels, more loan amount can be increased. Or the amount of resources allocated to the user at each stage may be dynamically adjusted based on the risk prediction result of the user.

When risk prediction is performed on a target user, relationship data corresponding to the target user may be extracted, and the relationship data is input into a risk prediction system constructed based on the method provided in the embodiment of the present specification, so that the relationship data of the target user is processed by adopting the method in the embodiment, for example: and extracting characteristics of the relation data, drawing a risk relation network map based on the relation data of the target user, and carrying out sub-graph segmentation of the abnormal nodes. And then calculating and extracting features based on the abnormal node subgraphs, and inputting the extracted features into a trained risk prediction model to obtain a risk prediction result of the target user.

Further, in some embodiments of the present description, the risk prediction model may include an event state prediction model and an abnormal event prediction model, and the event state prediction model may predict a state of an event such as: the event state is abnormal, normal or unknown, and an abnormal event prediction model can be used to predict the probability of an abnormal event. And when the risk prediction is carried out on the target user, carrying out the risk prediction by adopting two risk prediction models together, and synthesizing the output results of the two models to obtain the risk prediction result of the target user. Such as: comparing the predicted result of the same event as the predicted result of the probability of the event state/abnormal piece, taking a higher value as the predicted result of the event, such as: if the event state prediction model predicts that the state of the event A is abnormal with the probability of 70%, the abnormal event prediction model predicts that the abnormal state of the event A is 10%, and then the predicted result of the event state prediction model is taken as the predicted result of the event A.

According to the embodiment of the specification, a training mode of double models is adopted, states of partial events cannot be obtained in practice, if the states are set to be normal or abnormal, certain influence is brought to model training, prediction of event states/abnormal part probability is split, synchronous calculation is carried out by using different models, and prediction results of the events are comprehensively given according to results of the double models.

According to the system resource allocation method provided by the implementation of the specification, the risk relation network map is constructed based on the acquired relation data by acquiring the data of the risk event, then the sub-graph segmentation of the abnormal nodes is carried out on the risk relation network map, and the influence calculation of the propagation degree of the abnormal event is carried out based on the segmented abnormal node sub-graph. And combining the non-graph influence score characteristics and the network structure graph characteristics, training and constructing a risk prediction model, performing risk prediction on the target user based on the trained risk prediction model, and distributing resources of the target user based on a risk prediction result. The accuracy of the risk prediction model is improved, and further, the accuracy of risk prediction is improved. The whole process is highly abstract, is weakly associated with a business scene, can be suitable for the production line of different risk types of abnormal event prediction, has wide applicability, can complete the creation of a risk prediction model without having very deep expert knowledge, reduces the complexity of model creation and reduces the time of model creation.

In some embodiments of the present disclosure, the training the risk prediction model using at least one of the basic risk feature, the impact score feature, and the network structure feature includes:

In a specific implementation process, after the influence score features and the network structure features corresponding to each node are extracted based on the risk relation subgraph, the extracted features can be screened, and features with larger influence on a risk prediction result can be screened. In the embodiment of the present disclosure, an initial feature variable set may be first constructed based on the extracted basic risk feature, the impact score feature, and the network structural feature, for example:

For the event state prediction model: d ^known{x,y},x＝{x^owner,x^score,x^graph},y＝y^known;

For the abnormal event prediction model: d ^fraud{x,y},x＝{x^owner,x^score,x^graph},y＝y^fraud.

Wherein x ^owner is a basic risk feature, x ^score is an influence score feature, x ^graph is a network structure feature, y ^known and y ^fraud are risk labels in an event state prediction model and an abnormal event prediction model, wherein:

Based on the initial set of feature variables, the features are combined, such as: in some embodiments of the present disclosure, six combinations of three modes may be employed:

Mode one: only one type of characteristic variable in x ^owner,x^score,x^graph is selected.

Combination one ：D^known{x^owner,y^known},D^known{x^score,y^known},D^known{x^graph,y^known}

Combination II ：D^fraud{x^owner,y^fraud},D^fraud{x^score,y^fraud},D^fraud{x^graph,y^fraud}

Mode two: x ^owner,x^score,x^graph are combined two by two.

And (3) combining three: d ^known{x^owner,x^score,y^known},D^known{x^owner,x^graph,y^known },

D^known{x^score,x^graph,y^known}

Combination four: d ^fraud{x^owner,x^graph,y^fraud},D^fraud{x^score,x^graph,y^fraud },

D^known{x^score,x^graph,y^known}

Mode three: x ^owner,x^score,x^graph all choices.

And (5) combining: d ^known{x^owner,x^score,x^graph,y^known }

And (3) combining six: d ^fraud{x^owner,x^score,x^graph,y^fraud }

Variable screening is performed on the combined features in different modes, and the variable screening modes can select feature screening models, such as: a tree model is used. For each combination, inputting all feature variables under the combination as a feature screening model, wherein risk labels of the predicted sample time corresponding to the feature variables are as follows: whether the characteristic is abnormal, whether the risk exists or not, and the like are output as a characteristic screening model, and each characteristic is scored to generate an importance score of a characteristic variable. Ranking the importance scores from high to low, a higher importance score indicates a higher degree of influence of the feature variable on the final result. The importance scores are larger than a preset threshold value, namely the feature variables are arranged in TopN, the system defaults to the first 40% of all the variables, and the default supports user adjustment. And taking the screened characteristics as predicted target characteristics, and performing model training on the risk prediction model by using the target predicted characteristics.

Feature screening can be carried out on the feature combinations of the three different modes, a best combination mode is selected, and features with importance scores larger than a preset threshold value are selected from the best combination mode to serve as prediction target features.

By combining the extracted features in different modes and screening the features in different combination modes, the accuracy of risk prediction features is improved, and the accuracy of risk prediction results is further improved.

In some embodiments of the present description, the method further comprises:

In a specific implementation process, after the risk prediction model is trained, the risk prediction model can be issued and put on line after confirmation of a user, and the system background can be linked with the version deployment and installation device. All information of the online operation of the model, including characteristic variables and model parameters, can periodically operate the model according to the set frequency, save the generated result to a result table saving path in the model training process in the foregoing embodiment, save the characteristic variables of the score generated in the foregoing embodiment, and save the characteristic variables of the network to a characteristic table saving path in the foregoing embodiment.

The online risk prediction model may include two models-model-predicted event states, i.e., whether the state of the predicted event will be explicit, { normal, abnormal, { unknown }. Model two predicts event exception status, i.e. if the predicted event will be abnormal, { abnormal piece }, { normal piece, unknown }.

The system records the times of model call, the running state of the model, the times of success and failure, and monitors the operation and maintenance of the model. And supports updating models in the foreground, failed task rescaling, etc.

The prediction result of the model can provide interpretable explanation, namely, the characteristic variable with higher influence degree on the result and the influence direction thereof are interpreted. The user can select the prediction record to be interpreted, the system gives visual analysis interpretation information to the prediction record, the interpretation information can comprise the influence degree and the influence direction of the characteristic variable on the prediction result, the user can confirm whether to approve the analysis one by one to the given characteristic variable, the confirmation result is returned to the system, and the feedback information of the user is obtained. When the model is automatically trained next time, the weights of variables which are not widely considered to be significant by users are reduced by the feature variable screening ring nodes, and the weights of the variables screened into the model are reduced. And (3) feeding back whether the model prediction result is consistent with the actual situation or not in the system by the user, feeding back the labeling result to the system, and updating the labeling data when the system is trained next time. Based on feedback information of the prediction result of the model and the interpretation information of the prediction result, the model is optimized to improve the accuracy of the prediction result of the model.

In the embodiment of the specification, the online monitoring feedback flow of model research and development is unified and fused, and general model result feedback mainly collects the result of model prediction. Although various four-rule operations can be performed on different characteristic variables through system automation to generate derivative characteristic variables, the variables do not necessarily have business meanings, and the referencefor carrying out business analysis on users is low. According to the embodiment of the specification, the communication channel between the modeling personnel and the first-line business users is opened by collecting the feedback of the users to the characteristic variables.

In the present specification, each embodiment of the method is described in a progressive manner, and the same and similar parts of each embodiment are referred to each other, and each embodiment mainly describes differences from other embodiments. Reference is made to the description of parts of the method embodiments where relevant.

Based on the above system resource allocation method, one or more embodiments of the present disclosure further provide an apparatus for system resource allocation. The apparatus may include a system (including a distributed system), software (applications), modules, components, servers, clients, etc. that employ the methodologies of the embodiments of the present specification, in combination with the necessary apparatus to implement the hardware. Based on the same innovative concepts, embodiments of the present description provide for devices in one or more embodiments, such as the following embodiments. Because the implementation schemes and methods of the device for solving the problems are similar, the implementation of the device in the embodiments of the present disclosure may refer to the implementation of the foregoing method, and the repetition is omitted. As used below, the term "unit" or "module" may be a combination of software and/or hardware that implements the intended function. While the systems, apparatus described in the following embodiments are preferably implemented in software, implementation in hardware, or a combination of software and hardware, is also possible and contemplated.

Fig. 4 is a schematic structural diagram of a system resource allocation device according to an embodiment of the present disclosure, as shown in fig. 4, where the system resource allocation device provided in some embodiments of the present disclosure may be applied to a server in the foregoing embodiment, and may specifically include:

A data collection module 41, configured to collect a risk prediction sample set having risk characteristics for characterizing a user, where the risk prediction sample set includes relationship data of a plurality of predicted sample events and basic risk characteristics corresponding to each predicted sample event;

a network map drawing module 42, configured to construct a risk relationship network map of the predicted sample event based on the relationship data, where the risk relationship map includes nodes and edges connected to each node, the nodes are event or relational parties associated with the predicted sample event, and the edges are used to characterize a relationship between each node;

The abnormal sub-graph segmentation module 43 is configured to perform abnormal node sub-graph segmentation on the risk relationship network graph, and obtain risk relationship sub-graphs corresponding to each abnormal node in the risk relationship network graph;

The node influence analysis module 44 is configured to calculate an event influence score of each event node and a relationship influence score of each relationship node in each risk relationship subgraph;

the feature extraction module 45 is configured to extract an impact score feature and a network structure feature of the predicted sample event based on an event impact score of an event node corresponding to each risk relationship sub-graph and a relationship impact score of a relationship node;

The resource allocation module 46 is configured to perform model training on the risk prediction model by using at least one of the basic risk feature, the impact score feature, and the network structure feature, and obtain a trained risk prediction model, so as to perform risk prediction on a target user by using the trained risk prediction model, so as to allocate system resources to the target user based on the risk prediction result.

According to the system resource allocation device provided by the embodiment of the specification, the risk relation network map is constructed based on the acquired relation data by acquiring the data of the risk event, then the sub-graph segmentation of the abnormal nodes is carried out on the risk relation network map, and the influence calculation of the propagation degree of the abnormal event is carried out based on the segmented abnormal node sub-graph. And by combining the non-graph influence score characteristics and the network structure graph characteristics, the training construction of the risk prediction model is carried out, so that the accuracy of the risk prediction model is improved, and the accuracy of risk prediction is further improved. The whole process is highly abstract, is weakly associated with a business scene, can be suitable for the production line of different risk types of abnormal event prediction, has wide applicability, can complete the creation of a risk prediction model without having very deep expert knowledge, reduces the complexity of model creation and reduces the time of model creation.

It should be noted that the above-mentioned device may further include other embodiments according to the description of the corresponding method embodiment. Specific implementation manner may refer to the description of the corresponding method embodiments, which is not described herein in detail.

The embodiments of the present disclosure also provide a system resource allocation device, applied to a server, where the device includes at least one processor and a memory for storing instructions executable by the processor, and the instructions when executed by the processor implement a system resource allocation method including the foregoing embodiments, for example:

It should be noted that the above description of the apparatus according to the method embodiment may also include other implementations. Specific implementation may refer to descriptions of related method embodiments, which are not described herein in detail.

The method or apparatus of the above embodiments provided in the present specification may implement service logic by a computer program and be recorded on a storage medium, and the storage medium may be read and executed by a computer to implement the effects of the schemes described in the embodiments of the present specification.

The method embodiments provided in the embodiments of the present specification may be performed in a mobile terminal, a computer terminal, a server, or similar computing device. Taking the example of running on a server, fig. 5 is a block diagram of a hardware structure of a system resource allocation server in one embodiment of the present specification, and the computer terminal may be the system resource allocation server or the system resource allocation processing device in the above embodiment. The server 10 as shown in fig. 5 may include one or more (only one is shown in the figure) processors 100 (the processors 100 may include, but are not limited to, a microprocessor MCU or a processing device such as a programmable logic device FPGA), a nonvolatile memory 200 for storing data, and a transmission module 300 for communication functions. It will be appreciated by those of ordinary skill in the art that the configuration shown in fig. 5 is merely illustrative and is not intended to limit the configuration of the electronic device described above. For example, server 10 may also include more or fewer components than shown in FIG. 5, for example, may also include other processing hardware such as a database or multi-level cache, a GPU, or have a different configuration than that shown in FIG. 5.

The nonvolatile memory 200 may be used to store software programs and modules of application software, such as program instructions/modules corresponding to the driving data processing method in the embodiment of the present disclosure, and the processor 100 executes the software programs and modules stored in the nonvolatile memory 200, thereby executing various functional applications and resource data updates. The non-volatile memory 200 may include high speed random access memory, and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid state memory. In some examples, the non-volatile memory 200 may further include memory located remotely from the processor 100, which may be connected to the computer terminal via a network. Examples of such networks include, but are not limited to, the internet, intranets, office and networks, mobile communication networks, and combinations thereof.

The transmission module 300 is used to receive or transmit data via a network. Specific examples of the network described above may include a wireless network provided by a communication provider of a computer terminal. In one example, the transmission module 300 includes a network adapter (Network Interface Controller, NIC) that can connect to other network devices through a base station to communicate with the internet. In one example, the transmission module 300 may be a Radio Frequency (RF) module for communicating with the internet wirelessly.

The foregoing describes specific embodiments of the present disclosure. Other embodiments are within the scope of the following claims. In some cases, the actions or steps recited in the claims can be performed in a different order than in the embodiments and still achieve desirable results. In addition, the processes depicted in the accompanying figures do not necessarily require the particular order shown, or sequential order, to achieve desirable results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

The method and apparatus for allocating system resources provided in the embodiments of the present disclosure may be implemented in a computer by executing corresponding program instructions by a processor, for example, implemented on a PC side using the c++ language of a windows operating system, implemented on a linux system, or implemented on an intelligent terminal using, for example, android, iOS system programming languages, and implemented on a processing logic of a quantum computer.

In this specification, each embodiment is described in a progressive manner, and identical and similar parts of each embodiment are referred to each other, and each embodiment is mainly described in a different manner from other embodiments. In particular, for a hardware + program class embodiment, the description is relatively simple as it is substantially similar to the method embodiment, and reference is made to the partial description of the method embodiment where relevant.

Although one or more embodiments of the present description provide method operational steps as embodiments or flowcharts, more or fewer operational steps may be included based on conventional or non-inventive approaches. The order of steps recited in the embodiments is merely one way of performing the order of steps and does not represent a unique order of execution. In actual implementation of the apparatus or the terminal product, the methods illustrated in the embodiments or the drawings may be performed sequentially or in parallel (e.g., in a parallel processor or a multi-threaded processing environment, or even in a distributed resource data update environment). The terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, it is not excluded that additional identical or equivalent elements may be present in a process, method, article, or apparatus that comprises an element. The terms first, second, etc. are used to denote a name, but not any particular order.

For convenience of description, the above devices are described as being functionally divided into various modules, respectively. Of course, when one or more of the present description is implemented, the functions of each module may be implemented in the same piece or pieces of software and/or hardware, or a module that implements the same function may be implemented by a plurality of sub-modules or a combination of sub-units, or the like. The above-described apparatus embodiments are merely illustrative, e.g., the division of elements is merely a logical functional division, and there may be additional divisions in actual implementation, e.g., multiple elements or components may be combined or integrated into another system, or some features may be omitted, or not performed. Alternatively, the coupling or direct coupling or communication connection shown or discussed with each other may be an indirect coupling or communication connection via some interfaces, devices or units, which may be in electrical, mechanical or other form.

In this specification, each embodiment is described in a progressive manner, and identical and similar parts of each embodiment are referred to each other, and each embodiment is mainly described in a different manner from other embodiments. In particular, for system embodiments, the description is relatively simple as it is substantially similar to method embodiments, and reference is made to the section of the method embodiments where relevant. In the description of the present specification, a description referring to terms "one embodiment," "some embodiments," "examples," "specific examples," or "some examples," etc., means that a particular feature, structure, material, or characteristic described in connection with the embodiment or example is included in at least one embodiment or example of the present specification. In this specification, schematic representations of the above terms are not necessarily directed to the same embodiment or example. Furthermore, the particular features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples. Furthermore, the different embodiments or examples described in this specification and the features of the different embodiments or examples may be combined and combined by those skilled in the art without contradiction.

The foregoing is merely an example of one or more embodiments of the present specification and is not intended to limit the one or more embodiments of the present specification. Various modifications and alterations to one or more embodiments of this description will be apparent to those skilled in the art. Any modification, equivalent replacement, improvement, or the like, which is within the spirit and principles of the present specification, should be included in the scope of the claims.

Claims

1. A method for allocating system resources, the method comprising:

model training is carried out on the risk prediction model by utilizing at least one of the basic risk feature, the influence score feature and the network structure feature to obtain a trained risk prediction model, so that risk prediction is carried out on a target user by utilizing the trained risk prediction model, and system resources are distributed to the target user based on the risk prediction result;

The method for calculating the event influence score of each event node and the relationship influence score of each relationship node in each risk relationship subgraph comprises the following steps: according to the types of the edges connected with each node in each risk relation sub-graph, calculating node connection relation values among each node in each risk relation sub-graph, wherein the node connection relation values are used for representing the tightness degree of the connection relation among each node; calculating the degree of each node in each risk relation subgraph according to the weight of the edge connected with each node in each risk relation subgraph; and calculating the event influence score of each event node and the relationship influence score of each relationship node in each risk relationship sub-graph according to the node connection relationship value and the degree of each node.

2. The method of claim 1, wherein the model training of the risk prediction model using at least one of the base risk feature, the impact score feature, and the network structure feature comprises:

3. The method of claim 1, wherein said performing an outlier node sub-graph cut on the risk relationship network graph comprises:

the maximum number of nodes in the abnormal node subgraph;

4. The method of claim 1, comprising calculating an event impact score for each event node and a relationship impact score for each relationship node using the formula:

Wherein, Representing event impact scores corresponding to the ith event node, s _i representing the ith event node, alpha epsilon [0,1], beta epsilon [0,1] being a hyper-parameter, n _R representing the total number of relationship nodes in the risk relationship graph,Representing a node connection relationship value between the ith event node and the jth relationship node,Representing the degree of the ith event node,Representing the degree of the j-th relationship node,Represents a relationship influence score corresponding to a j-th relationship node, ri represents a j-th relationship node,Representing the initial value of the event impact score corresponding to the ith event node, n _S representing the total number of event nodes in the risk relationship graph,And the initial value of the relationship influence score corresponding to the j-th relationship node is represented.

5. The method of claim 1, wherein calculating the node connection relationship values between the nodes in each risk relationship graph according to the types of the edges connected by the nodes in each risk relationship graph, comprises:

6. The method of claim 1, wherein the method further comprises:

7. The method of claim 1, wherein the risk prediction model comprises: an event state prediction model and an abnormal event prediction model;

8. A system resource allocation apparatus, the apparatus comprising:

The resource allocation module is used for carrying out model training on the risk prediction model by utilizing at least one of the basic risk characteristics, the influence score characteristics and the network structure characteristics to obtain a trained risk prediction model, carrying out risk prediction on a target user by utilizing the trained risk prediction model, and allocating system resources to the target user based on the risk prediction result;

The node influence analysis module is specifically configured to calculate node connection relation values between nodes in each risk relation sub-graph according to types of edges connected by each node in each risk relation sub-graph, where the node connection relation values are used to represent tightness of connection relations between the nodes; calculating the degree of each node in each risk relation subgraph according to the weight of the edge connected with each node in each risk relation subgraph; and calculating the event influence score of each event node and the relationship influence score of each relationship node in each risk relationship sub-graph according to the node connection relationship value and the degree of each node.

9. A system resource allocation device, applied to a server, the device comprising at least one processor and a memory for storing processor executable instructions which when executed by the processor implement steps comprising the method of any of the preceding claims 1-7.