CN112669143A - Risk assessment method, device and equipment based on associated network and storage medium - Google Patents
Risk assessment method, device and equipment based on associated network and storage medium Download PDFInfo
- Publication number
- CN112669143A CN112669143A CN202110024642.5A CN202110024642A CN112669143A CN 112669143 A CN112669143 A CN 112669143A CN 202110024642 A CN202110024642 A CN 202110024642A CN 112669143 A CN112669143 A CN 112669143A
- Authority
- CN
- China
- Prior art keywords
- target
- community
- node
- network
- order neighbor
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Landscapes
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The method comprises the steps of obtaining a target community of a target node in an association network, wherein the target node is a node corresponding to a target object to be evaluated in the association network, the target community is composed of a plurality of nodes with association degrees larger than a preset value, and obtaining community characteristics of the target community, the community characteristics are determined based on the characteristics of each node in the target community, then inputting the community characteristics and a first characteristic into a risk evaluation model to obtain a risk evaluation result of the target object, the risk evaluation model is obtained based on neural network model training, the first characteristic is determined based on the characteristics of n-order neighbor nodes of the target node, and the risk of the target object is comprehensively and accurately evaluated.
Description
Technical Field
The present application relates to the field of computer technologies, and in particular, to a risk assessment method, apparatus, device, and storage medium based on a correlation network.
Background
At present, the internet and big data are often used for risk assessment of a user who needs to perform a business behavior, so as to avoid the user from bringing a greater risk when performing the business behavior. For example, one, identifying a risk of a user by analyzing characteristics of the user; and secondly, determining that the risk of the user is high when the risk of the user is greater than a preset value in the users having the association relation with the user, and determining that the risk of the user is low when the risk of the user is not greater than the preset value in the users having the association relation with the user.
However, the existing risk assessment methods cannot comprehensively and accurately assess the risk of the user.
Disclosure of Invention
The application provides a risk assessment method, a risk assessment device and a risk assessment storage medium based on a correlation network, which can comprehensively and accurately assess the risk of a target object.
In a first aspect, an embodiment of the present application provides a risk assessment method based on an association network, where the method includes:
acquiring a target community of a target node in an association network, wherein the target node is a node corresponding to a target object to be evaluated in the association network, and the target community consists of a plurality of nodes with association degrees larger than a preset value;
acquiring community characteristics of a target community, wherein the community characteristics are determined based on the characteristics of each node in the target community;
inputting the community characteristics and the first characteristics into a risk evaluation model to obtain a risk evaluation result of the target object, wherein the risk evaluation model is obtained based on neural network model training, the first characteristics are determined based on the characteristics of n-order neighbor nodes of the target node, and n is larger than or equal to 1.
In a second aspect, an embodiment of the present application provides a risk assessment apparatus, including:
the target community comprises a plurality of nodes with the association degree larger than a preset value, and the target community comprises a target node and a target community unit, wherein the target community unit is used for acquiring the target community of the target node in the association network, the target node is a corresponding node of a target object to be evaluated in the association network, and the target community unit is composed of the nodes with the association degree larger than the preset value;
the acquisition unit is further used for acquiring community characteristics of the target community, wherein the community characteristics are determined based on the characteristics of each node in the target community;
and the processing unit is used for inputting the community characteristics and the first characteristics into a risk evaluation model to obtain a risk evaluation result of the target object, the risk evaluation model is obtained based on neural network model training, the first characteristics are determined based on the characteristics of n-order neighbor nodes of the target node, and n is more than or equal to 1.
In a third aspect, an embodiment of the present application provides an electronic device, including: a memory and a processor;
the memory stores computer-executable instructions;
the processor executes the computer-executable instructions stored by the memory, causing the processor to perform the method of the first aspect or embodiments thereof.
In a fourth aspect, an embodiment of the present application provides a storage medium, including: a readable storage medium and a computer program for implementing the method of the first aspect or implementations thereof.
According to the image recognition method and device, the positioning information and the classification information of the input image are recognized in parallel through the image recognition model to obtain the recognition result, the processing efficiency is improved, the storage space is saved through the image recognition model with the multi-task learning mechanism of the shared weight, and the corresponding image processing operation is performed on the image according to the recognition result to ensure that the target object in the image can be recognized accurately.
Drawings
In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings needed to be used in the description of the embodiments or the prior art will be briefly introduced below, and it is obvious that the drawings in the following description are some embodiments of the present application, and for those skilled in the art, other drawings can be obtained according to these drawings without inventive exercise.
Fig. 1 is a schematic diagram of an association network according to an embodiment of the present application;
fig. 2 is a schematic diagram of a network subgraph provided in the embodiment of the present application;
FIG. 3 is a schematic diagram of a community provided by an embodiment of the present application;
fig. 4 is a schematic flowchart of a risk assessment method 400 based on an association network according to an embodiment of the present application;
fig. 5 is a schematic flowchart of a risk assessment method 500 based on an association network according to an embodiment of the present application;
fig. 6 is a schematic flowchart of a risk assessment method 600 based on an association network according to an embodiment of the present application;
fig. 7 is a schematic flow chart of a risk assessment method 700 according to an embodiment of the present application;
fig. 8 is a schematic structural diagram of a risk assessment apparatus 800 according to an embodiment of the present disclosure;
fig. 9 is a schematic hardware structure diagram of an electronic device 900 according to an embodiment of the present application.
Detailed Description
In order to make the objects, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application, and it is obvious that the described embodiments are some embodiments of the present application, but not all embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present application.
In each field, a risk assessment needs to be performed on a user (hereinafter, referred to as a target object) to determine whether the target object carries a greater risk when performing a business action, for example, in a different field, the business action may be a business action in a field of transaction, loan, network security, and the like.
In order to more comprehensively evaluate the risk of the target object, the determination may be made by a user having an association relationship with the target object, such as a1 st order neighbor user directly associated with the target object, and a2 nd order neighbor user (indirectly associated with the target object) associated with the 1 st order neighbor user. For example, whether a blacklist user exists in a1 st order neighbor user or a2 nd order neighbor user, or whether the number of blacklist users is greater than a preset value is determined, and when the blacklist user exists or the number of blacklist users is greater than the preset value, it is determined that the risk of the target object is high.
Therefore, in order to perform comprehensive and accurate risk identification on the target object, the inventive concept of the present application is: and comprehensively judging by combining the characteristics of the target object or the characteristics of the neighbor users of the target object and the community characteristics of the target community in which the target object is located to obtain a comprehensive and accurate risk identification result.
First, technical terms mentioned in the embodiments of the present application will be explained.
And (3) associating the network: the different individual things form a network because of the association. The association network contains two basic elements: nodes and edges.
And (3) node: the vertex in the associated network corresponds to a unique specific or virtual individual in life, and as shown in fig. 1, the node may be a user, a device, a company, a mobile phone call, an account, payment, a mobile hotspot (WIFI), or the like.
Side: the vertices in the network are associated because of the association formed by a relationship. For example, a user has a certain mobile phone number, and an edge is formed between the user and the mobile phone number.
In some embodiments, users may be used as nodes, companies, mobile phone calls, payments, WIFI, and the like are used as sides, users belonging to the same company establish an association relationship with the company as a side, a payer and a payee establish an association relationship with the payment as a side, and users connected to the same WIFI establish a connection relationship, or a company may also be used as a user as a node; in other embodiments, users, companies, mobile phone calls, payments, WIFI, etc. are all used as nodes, and the influence of the characteristics on the target object may not be considered for the nodes of non-users.
And (3) neighbor nodes: nodes connected through edges in the correlation network are adjacent nodes to each other, nodes connected through 1 edge are adjacent nodes of 1 order to each other, nodes connected through 2 edges are adjacent nodes of 2 order to each other, and nodes connected through k edges are adjacent nodes of k order to each other.
Community: in the association network, the nodes can be divided into different communities according to the association degree among the nodes, and the nodes in the same community are connected more closely.
Graph convolution: method for aggregating neighbor node characteristics by using convolution operation
Graphic attention network (graph) mechanism: a neural network learning method aggregates neighbor features by giving different weights to neighbor nodes.
Fig. 2 is a schematic diagram of a network subgraph provided in the embodiment of the present application. It should be understood that the network subgraph is a part of the association network, which includes a target node corresponding to the target object in the association network, and n-th order neighbor nodes of the target node, and it should be understood that the neighbor nodes correspond to neighbor users other than the target object, as shown in fig. 2, the target node 20 is a node corresponding to the target object in the association network, the 1 st order neighbor node 21 is a node directly associated with the target node 20, and the 2 nd order neighbor node 22 is a node indirectly associated with the target node 20 through the 1 st order neighbor node 21.
Fig. 3 is a schematic diagram of a community according to an embodiment of the present application. As shown in fig. 3, the association network 30 may be divided into a plurality of communities, such as community 31, community 32, and community 33, according to the degree of association between nodes. Wherein each community comprises a plurality of nodes, and the community containing the target node is called a target community.
The technical scheme of the embodiment of the application can be applied to various electronic devices. The electronic device may be a terminal device, such as a Mobile Phone (Mobile Phone), a tablet computer (Pad), a computer, and the terminal device may be fixed or Mobile. Alternatively, the electronic device in the embodiment of the present application may also be a server.
The present application is specifically illustrated by the following examples.
Fig. 4 is a flowchart illustrating a risk assessment method 400 based on an association network according to an embodiment of the present disclosure.
In order to comprehensively and accurately perform risk assessment on a target object, the embodiment of the application combines the community characteristics of the community where the target object is located and the first characteristics determined based on the n-order neighbor node of the target object, and obtains a risk assessment result of the target object through a risk assessment model.
As shown in fig. 4, the risk assessment method based on the association network provided in the embodiment of the present application includes:
s401: and acquiring a target community of the target node in the associated network.
The target nodes are corresponding nodes of a target object to be evaluated in the association network, and the target community is composed of a plurality of nodes with association degrees larger than a preset value.
For example, the association network may be grouped in advance, and a plurality of nodes in the association network may be divided to form a plurality of communities. The community containing the target node is the target community.
If the target object already has a corresponding target node in the associated network, namely the target node does not belong to the associated network, acquiring a target community where the target node is located; if the target object does not have a corresponding target node in the association network, namely the target node does not belong to the association network, the target node is created, and a community with the strongest association degree with the target node is determined according to the weight of the association relation between the target node and the 1-order neighbor node of the target node and serves as the target community.
S402: and acquiring community characteristics of the target community.
The community characteristics are determined based on the characteristics of each node in the target community.
The community characteristics may be characteristics in a plurality of evaluation dimensions, for example, in the field of credit loan, the community characteristics may include average age in the community, average income in the community, overdue rate in the community, maximum age in the community, and maximum overdue amount in the community, and it should be understood that the community characteristics may be determined by characteristics of each node in the community.
S403: and inputting the community characteristics and the first characteristics into a risk evaluation model to obtain a risk evaluation result of the target object.
The risk assessment model is obtained based on neural network model training and used for determining a risk assessment result of the target object according to the input community characteristic and the first characteristic.
The first characteristic is determined based on the characteristics of n-order neighbor nodes of a target node, n is larger than or equal to 1, the larger n is, the more comprehensive data coverage is, the higher accuracy is, but the higher complexity and time consumption of operation are, and the size of n can be set by comprehensively processing the speed and the accuracy in an actual application scene.
Optionally, n is equal to 2.
In the step, in order to improve the comprehensiveness and accuracy of the evaluation result, the community characteristics and the first characteristics determined in the step are input into the risk evaluation model together to obtain the risk evaluation result of the target object.
In the embodiment of the application, the target community corresponding to the target object is obtained, the community characteristics of the target community are extracted, the community characteristics and the first characteristics are input into the risk assessment model, the risk assessment result of the target object is obtained, and the target object is subjected to risk assessment together through the community characteristics of the group and the characteristics of the closely related neighbor users, so that the accuracy and the comprehensiveness of the risk assessment are improved.
Fig. 5 is a flowchart illustrating a risk assessment method 500 based on an association network according to an embodiment of the present disclosure.
On the basis of the above embodiment, when the target object does not have a corresponding target node in the association network, the embodiment of the present application determines the target community based on the weights of the association relationships between the target node and a plurality of 1-order neighbor nodes of the target node. As shown in fig. 5:
s501: and summing the weights of the incidence relation between each 1-order neighbor node and the target node in the community containing at least one 1-order neighbor node to obtain the weights of the community and the target node.
S502: and taking the community with the maximum weight value as a target community.
Wherein at least one order-1 neighbor node associated with the target node is divided into one to more communities.
For example, the 1 st order neighbor nodes of the target node a include a1 to a6, a1 belongs to community 1, a2 to a4 belong to community 2, a5 and a6 belong to community 3, the weights of the association relations between the target node a and the a1 to a6 nodes are 1, 2, 3, 4, 5 and 6 in sequence, the association degree between community 1 and target node a can be represented by weight and 1, the association degree between community 2 and target node a can be represented by weight and 10, and the association degree between community 3 and target node a can be represented by weight and 11, so that the community with the largest value of the weight between the community and the target node is community 3, that is, community 3 is the target community.
For example, communities in the associated network may be divided in advance in an offline manner, or divided when a target community is determined, which is not limited in the embodiment of the present application.
Optionally, there are two general ways of community division: 1. according to the associated network structure, the representation of each node about the group is directly found, and a better algorithm is a community discovery algorithm (such as a Louvain algorithm, an infomap algorithm and the like); 2. firstly, performing hidden space mapping on an associated network once, and then searching the similarity of each node in the hidden space, namely an Embedding and clustering (Embedding + clustering) process, wherein the better Embedding method comprises a random walk (randomWalk) algorithm node2vec algorithm, a structure2vec algorithm and the like. The system does not limit the algorithm for dividing the community, and preferably, the community can be divided by using a Louvain algorithm, wherein the Louvain algorithm is self-generating calculation based on modularity. The modularity is shown in equation 1:
wherein m represents the number of edges in the network, k is the node degree, A is the weight of the incidence relation between the adjacent matrix, i.e. the node i and the node j, and the parameter c of the delta functioniIs the community in which the node i is located, cjIs the community in which node j is located, is 1 if (i j) is in a community, and is 0 otherwise.
The community division process comprises the following steps:
1. defining 1 community for each node, then moving the node to the neighbor node community, and deciding which neighbor community the node is merged into by maximizing the amount of change, defined as follows:
wherein k isi，inIs the sum of the weights of the edges between the node i and all the nodes to be moved into the community, and Σ tot is the sum of the weights of the edges of all the nodes in the community to be moved into by the node i, and the above formula can be understood as the first term k in the parenthesesi，inRepresenting the number of connecting edges between the actual node i and the community to be moved into, the second term sigma tot/m is the probability that other nodes are connected with one edge of the community, and sigma tot kiAnd/m is the number of edges connected with the community based on the degree k of the node i under the average condition, if the first term is larger than the second term, the connection degree of the node i and the community exceeds the average prediction, the node is added into the community, otherwise, the node is kept unchanged, namely when the delta Q is larger than 0, the node is merged into the community to be moved into.
2. After the first iteration is finished, each node is divided into a specific community, then the whole community is used as a large node, the large nodes are connected to form a new graph, the connection and the weight among the large nodes are determined through the connection among the communities, then the new graph is subjected to the next-level community division in the process of 1 repeatedly, and the community division process is finished until the moving quantity is less than the preset value or the iteration number reaches the preset number.
Fig. 6 is a flowchart illustrating a risk assessment method 600 based on an association network according to an embodiment of the present application.
On the basis of any of the above embodiments, the embodiment of the present application proposes an implementation manner as shown in fig. 6 for how to obtain the first feature.
S601: and acquiring a network subgraph of the target object in the associated network.
The network subgraph comprises a target node corresponding to a target object and n-order neighbor nodes of the target node.
S602: based on the n-order neighbor nodes, a first feature is obtained.
Illustratively, the first feature includes a neighbor feature and/or a graph convolution feature, where the neighbor feature is a feature of each order neighbor node in n order neighbor nodes of the target node in multiple evaluation dimensions, and the graph convolution feature is a feature obtained by graph convolution of the n order neighbor nodes of the target node.
And when the first feature comprises the neighbor feature, obtaining the feature of each-order neighbor node based on the feature of each neighbor node in the n-order neighbor nodes. For example, the characteristics of each neighboring node, such as age, income, overdue payment, etc., are obtained, and then the characteristics of each order neighboring node, such as average age of 1 order neighbors, average income of 1 order neighbors, overdue rate of 1 order neighbors, maximum age of 2 order neighbors, average income of 2 order neighbors, etc., are determined according to the characteristics of each neighboring node in each order.
When the first characteristic comprises a graph convolution characteristic, based on a network subgraph, carrying out graph convolution on n-order neighbor nodes, for example, carrying out convolution operation on the characteristics of at least one nth-order neighbor node connected with the n-1-order neighbor node aiming at each nth-1-order neighbor node in the n-1-order neighbor nodes to obtain the graph convolution characteristic of the nth-1-order neighbor nodes, combining the graph convolution characteristic into the characteristics of the nth-1-order neighbor nodes, carrying out convolution operation on the characteristics of at least one nth-1-order neighbor node connected with the n-2-order neighbor nodes (including the graph convolution characteristic that the nth-order neighbor node is mapped onto the n-1-order node) aiming at each nth-2-order neighbor node in the n-2-order neighbor nodes to obtain the graph convolution characteristic of the nth-2-order neighbor nodes, and integrating the graph convolution characteristic into the characteristics of the n-2 th-order neighbor node. And repeating the process until the convolution operation is carried out on the characteristics of at least one 1 st-order neighbor node connected with the target node to obtain the graph convolution characteristics of the target node. Optionally, the graph convolution feature may be merged with the feature of the target node as a final graph convolution feature.
Optionally, the graph convolution feature of the target node and the feature of the target node in multiple evaluation dimensions are input into a convolution network model to obtain a graph convolution risk score, the graph convolution risk score is used as a final graph convolution feature, and the convolution network model is obtained by training based on a neural network model in advance.
The embodiment of the application provides the following possible implementation modes for extracting the graph convolution characteristics:
and the nth-order neighbor is transmitted to the (n-1) th-order neighbor through convolution operation, finally collected to the target node, and then graph convolution risk scores are output through the last layer of graph convolution model. The Graph convolution operation mode may be Graph neural Network (GCN) which is an average of all dimension features without weight difference of neighboring nodes, and information transmission of two neighboring layers in the GCN is shown in formula 3
Wherein A and D are an adjacency matrix and a degree matrix. H is a characteristic matrix of each-order node, W is a parameter matrix needing learning, and sigma represents an activation function.
Is a normalized adjacency matrix. The normalized adjacency matrix is multiplied by the feature matrix on the left to obtain the aggregation process of the neighbors, then multiplied by W on the right to obtain the feature summation process, and finally nonlinear transformation is carried out through an activation function. Multiple degrees of neighbor aggregation can be achieved by cascading multiple convolutional layers. Last layer is H(0)I.e. features passed from 1 st order neighbors, H(0)And combining the characteristics with the characteristics of the target node, and inputting the characteristics into a final top-layer model to predict to obtain a risk score.
The Graph convolution operation mode may also be a Graph Attention Network (GAT) which is a weighted summation of features of each dimension with different weights of neighboring nodes, and the GAT assigns different weights, i.e., "Attention coefficients", to different neighboring nodes during neighbor aggregation, so that the features of important neighboring nodes are absorbed after the neighboring nodes are weighted and summed, and noise influence is reduced.
The attention coefficient is obtained by calculating the similarity between the neighbor node and the target node, as shown in formula 4, wherein e isijIs the correlation between the target node i and the neighbor node jThe coefficients of which are such that,
and
is the feature vector and W is the weight matrix. a is the attention calculation function.
There are generally two attention calculation functions as follows:
the first is an addition method: e.g. of the typeij＝LearkyRelu(qi+kj) (5)
Wherein LearkyRelu is an activation function, and qi is used for representing
kjFor representing
The second is a multiplication mode: e.g. of the typeij＝qi·kj (6)
eijAnd obtaining the contribution degree of the neighbor node j to the target node i. Finally, the correlation coefficients e of all the neighbor nodes are normalized to obtain the final attention coefficient, as shown in formula 7:
wherein softmax is a logistic regression function, NiA set of neighbor nodes of i.
The adjacent two orders of information transfer during the GAT convolution is shown as the following equation (8):
fig. 7 is a flowchart illustrating a risk assessment method 700 according to an embodiment of the present disclosure. As shown in fig. 7, in the embodiment of the application, a node and an edge are extracted from a data source to construct an association network, three features of a target node are obtained based on the association network, the first is to obtain a network subgraph of the target node from the association network and extract features of neighbor nodes of the target node, namely neighbor features, the second is to obtain a network subgraph of the target node from the association network and extract graph convolution features of the target node, the third is to perform community division on the association network, match communities where the target node is located and extract community features of the target node, and finally, the neighbor features, the graph convolution features and the community features of the target node are subjected to feature splicing, and the spliced features are input into a risk assessment model to obtain a risk assessment result.
For example, the features input into the risk assessment model may also be any one of, or any combination of, neighbor features, graph convolution features, and community features.
Fig. 8 is a schematic structural diagram of a risk assessment apparatus 800 according to an embodiment of the present application, and as shown in fig. 8, the risk assessment apparatus 800 includes:
an obtaining unit 810, configured to obtain a target community of a target node in an association network, where the target node is a node corresponding to a target object to be evaluated in the association network, and the target community is composed of a plurality of nodes with association degrees greater than a preset value;
the obtaining unit 810 is further configured to obtain a community feature of the target community, where the community feature is determined based on a feature of each node in the target community;
a processing unit 820, configured to input the community feature and a first feature into a risk assessment model to obtain a risk assessment result of the target object, where the risk assessment model is obtained based on neural network model training, the first feature is determined based on a feature of an n-order neighbor node of the target node, and n is greater than or equal to 1.
The risk assessment device 800 provided in this embodiment includes an obtaining unit 810 and a processing unit 820, extracts a community feature of a target community by obtaining the target community corresponding to a target object, and inputs the community feature and a first feature into a risk assessment model to obtain a risk assessment result of the target object, and performs risk assessment on the target object through the community feature of a group and the feature of a closely related neighbor user, so that accuracy and comprehensiveness of the risk assessment are improved.
In one possible design, the obtaining unit 810 is specifically configured to:
determining whether the target node belongs to the associated network;
if the target node belongs to the associated network, acquiring a community containing the target node as a target community;
otherwise, determining the target community based on the weight of the incidence relation between the target node and a plurality of 1-order neighbor nodes of the target node.
In one possible design, the processing unit 820 is specifically configured to:
summing weights of incidence relations between each 1-order neighbor node and the target node in a community containing at least one 1-order neighbor node to obtain the weights of the community and the target node;
and taking the community with the maximum weight value as the target community.
In a specific implementation manner, the obtaining unit 810 is further configured to: acquiring a network subgraph of the target object in the association network, wherein the network subgraph comprises a target node corresponding to the target object and n-order neighbor nodes of the target node;
the processing unit 820 is further configured to obtain the first characteristic based on the n-th order neighbor node.
In a specific implementation, the processing unit 820 is specifically configured to:
and obtaining the characteristics of each-order neighbor node based on the characteristics of each neighbor node in the n-order neighbor nodes.
In a specific implementation, the processing unit 820 is specifically configured to:
performing convolution operation on the characteristics of at least one nth-order neighbor node connected with any nth-1-order neighbor node to obtain the graph convolution characteristics of the nth-1-order neighbor node, and combining the graph convolution characteristics into the characteristics of the nth-1-order neighbor node;
and repeating the process until the convolution operation is carried out on the characteristics of at least one 1 st-order neighbor node connected with the target node to obtain the graph convolution characteristics of the target node.
In a specific implementation, the processing unit 820 is further configured to:
and inputting the graph convolution characteristics of the target node and the characteristics of the target node on a plurality of evaluation dimensions into a convolution network model to obtain a graph convolution risk score, and taking the graph convolution risk score as a final graph convolution characteristic, wherein the convolution network model is obtained by training based on a neural network model in advance.
The risk assessment device provided in this embodiment can be used to implement the method in any of the above embodiments, and its implementation effect is similar to that of the method embodiments, and is not described herein again.
Fig. 9 is a schematic hardware structure diagram of an electronic device 900 according to an embodiment of the present application. As shown in fig. 9, in general, the electronic apparatus 900 includes: a processor 910 and a memory 920.
The processor 910 may include one or more processing cores, such as a 4-core processor, an 8-core processor, and so on. The processor 910 may be implemented in at least one hardware form of a DSP (Digital Signal Processing), an FPGA (Field-Programmable Gate Array), and a PLA (Programmable Logic Array). Processor 910 may also include a main processor and a coprocessor, where the main processor is a processor for Processing data in an awake state, and is also called a Central Processing Unit (CPU); a coprocessor is a low power processor for processing data in a standby state. In some embodiments, the processor 910 may be integrated with a GPU (Graphics Processing Unit), which is responsible for rendering and drawing the content required to be displayed on the display screen. In some embodiments, the processor 910 may further include an AI (Artificial Intelligence) processor for processing computing operations related to machine learning.
Optionally, as shown in fig. 9, the electronic device 900 may further include a transceiver 930, and the processor 910 may control the transceiver 930 to communicate with other devices, and in particular, may transmit information or data to the other devices or receive information or data transmitted by the other devices.
The transceiver 930 may include a transmitter and a receiver, among others. The transceiver 930 may further include one or more antennas.
Optionally, the electronic device 900 may implement corresponding processes in the methods of the embodiments of the present application, and for brevity, details are not described here again.
Those skilled in the art will appreciate that the configuration shown in fig. 9 does not constitute a limitation of the electronic device 900, and may include more or fewer components than those shown, or combine certain components, or employ a different arrangement of components.
Embodiments of the present application also provide a non-transitory computer-readable storage medium, where instructions in the storage medium, when executed by a processor of an electronic device, enable the electronic device to perform the method provided by the above embodiments.
The computer-readable storage medium in this embodiment may be any available medium that can be accessed by a computer or a data storage device such as a server, a data center, etc. that is integrated with one or more available media, and the available media may be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., DVDs), or semiconductor media (e.g., SSDs), etc.
Those of ordinary skill in the art will understand that: all or a portion of the steps of implementing the above-described method embodiments may be performed by hardware associated with program instructions. The program may be stored in a computer-readable storage medium. When executed, the program performs steps comprising the method embodiments described above; and the aforementioned storage medium includes: various media that can store program codes, such as ROM, RAM, magnetic or optical disks.
The embodiment of the present application also provides a computer program product containing instructions, which when run on a computer, causes the computer to execute the method provided by the above embodiment.
It will be understood by those skilled in the art that all or part of the steps for implementing the above embodiments may be implemented by hardware, or may be implemented by a program instructing relevant hardware, where the program may be stored in a computer-readable storage medium, and the above-mentioned storage medium may be a read-only memory, a magnetic disk or an optical disk, etc.
The above description is only exemplary of the present application and should not be taken as limiting the present application, as any modification, equivalent replacement, or improvement made within the spirit and principle of the present application should be included in the protection scope of the present application.
Claims (10)
1. A risk assessment method based on a correlation network is characterized by comprising the following steps:
acquiring a target community of a target node in an association network, wherein the target node is a node corresponding to a target object to be evaluated in the association network, and the target community consists of a plurality of nodes with association degrees larger than a preset value;
acquiring community characteristics of the target community, wherein the community characteristics are determined based on the characteristics of each node in the target community;
inputting the community characteristics and the first characteristics into a risk evaluation model to obtain a risk evaluation result of the target object, wherein the risk evaluation model is obtained based on neural network model training, the first characteristics are determined based on the characteristics of n-order neighbor nodes of the target node, and n is larger than or equal to 1.
2. The method of claim 1, wherein obtaining the target community of the target node in the associative network comprises:
determining whether the target node belongs to the associated network;
if the target node belongs to the associated network, acquiring a community containing the target node as a target community;
otherwise, determining the target community based on the weight of the incidence relation between the target node and a plurality of 1-order neighbor nodes of the target node.
3. The method of claim 2, wherein determining the target community based on weights of incidence relations of the target node and a plurality of 1 st-order neighbor nodes of the target node comprises:
summing weights of incidence relations between each 1-order neighbor node and the target node in a community containing at least one 1-order neighbor node to obtain the weights of the community and the target node;
and taking the community with the maximum weight value as the target community.
4. The method according to any one of claims 1 to 3, further comprising:
acquiring a network subgraph of the target object in the association network, wherein the network subgraph comprises a target node corresponding to the target object and n-order neighbor nodes of the target node;
and obtaining the first characteristic based on the n-order neighbor node.
5. The method of claim 4, wherein if the first feature comprises a neighbor feature that is a feature of each of the n-th order neighbor nodes of the target node in multiple evaluation dimensions, the deriving the first feature based on the n-th order neighbor nodes comprises:
and obtaining the characteristics of each-order neighbor node based on the characteristics of each neighbor node in the n-order neighbor nodes.
6. The method according to claim 4, wherein if the first feature comprises a graph volume feature, the graph volume feature being a feature obtained by graph convolution of an n-th-order neighbor node of the target node, the obtaining the first feature based on the n-th-order neighbor node comprises:
performing convolution operation on the characteristics of at least one nth-order neighbor node connected with any nth-1-order neighbor node to obtain the graph convolution characteristics of the nth-1-order neighbor node, and combining the graph convolution characteristics into the characteristics of the nth-1-order neighbor node;
and repeating the process until the convolution operation is carried out on the characteristics of at least one 1 st-order neighbor node connected with the target node to obtain the graph convolution characteristics of the target node.
7. The method of claim 6, further comprising:
and inputting the graph convolution characteristics of the target node and the characteristics of the target node on a plurality of evaluation dimensions into a convolution network model to obtain a graph convolution risk score, and taking the graph convolution risk score as a final graph convolution characteristic, wherein the convolution network model is obtained by training based on a neural network model in advance.
8. A risk assessment device, comprising:
the device comprises an acquisition unit, a correlation network and a judgment unit, wherein the acquisition unit is used for acquiring a target community of a target node in the correlation network, the target node is a node corresponding to a target object to be evaluated in the correlation network, and the target community consists of a plurality of nodes with correlation degrees larger than a preset value;
the obtaining unit is further configured to obtain a community feature of the target community, where the community feature is determined based on a feature of each node in the target community;
and the processing unit is used for inputting the community characteristics and the first characteristics into a risk evaluation model to obtain a risk evaluation result of the target object, the risk evaluation model is obtained based on neural network model training, the first characteristics are determined based on the characteristics of n-order neighbor nodes of the target node, and n is more than or equal to 1.
9. An electronic device, comprising: a memory and a processor;
the memory stores computer-executable instructions;
the processor executes computer-executable instructions stored by the memory, causing the processor to perform the method of any of claims 1 to 7.
10. A storage medium, comprising:
readable storage medium and computer program for implementing the method according to any of claims 1 to 7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110024642.5A CN112669143A (en) | 2021-01-08 | 2021-01-08 | Risk assessment method, device and equipment based on associated network and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110024642.5A CN112669143A (en) | 2021-01-08 | 2021-01-08 | Risk assessment method, device and equipment based on associated network and storage medium |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN112669143A true CN112669143A (en) | 2021-04-16 |
Family
ID=75413800
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110024642.5A Pending CN112669143A (en) | 2021-01-08 | 2021-01-08 | Risk assessment method, device and equipment based on associated network and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112669143A (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113222609A (en) * | 2021-05-07 | 2021-08-06 | 支付宝(杭州)信息技术有限公司 | Risk identification method and device |
| CN113284030A (en) * | 2021-06-28 | 2021-08-20 | 南京信息工程大学 | Urban traffic network community division method |
| CN113409139A (en) * | 2021-07-27 | 2021-09-17 | 深圳前海微众银行股份有限公司 | Credit risk identification method, apparatus, device, and program |
| CN113744038A (en) * | 2021-09-03 | 2021-12-03 | 上海晓途网络科技有限公司 | Object parameter determination method, device, equipment and storage medium |
| CN113409139B (en) * | 2021-07-27 | 2024-05-28 | 深圳前海微众银行股份有限公司 | Credit risk identification method, apparatus, device and program |
-
2021
- 2021-01-08 CN CN202110024642.5A patent/CN112669143A/en active Pending
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113222609A (en) * | 2021-05-07 | 2021-08-06 | 支付宝(杭州)信息技术有限公司 | Risk identification method and device |
| CN113284030A (en) * | 2021-06-28 | 2021-08-20 | 南京信息工程大学 | Urban traffic network community division method |
| CN113284030B (en) * | 2021-06-28 | 2023-05-23 | 南京信息工程大学 | Urban traffic network community division method |
| CN113409139A (en) * | 2021-07-27 | 2021-09-17 | 深圳前海微众银行股份有限公司 | Credit risk identification method, apparatus, device, and program |
| CN113409139B (en) * | 2021-07-27 | 2024-05-28 | 深圳前海微众银行股份有限公司 | Credit risk identification method, apparatus, device and program |
| CN113744038A (en) * | 2021-09-03 | 2021-12-03 | 上海晓途网络科技有限公司 | Object parameter determination method, device, equipment and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110009174B (en) | Risk recognition model training method and device and server | |
| US20220351049A1 (en) | Method and System for Realizing Machine Learning Modeling Process | |
| CN112669143A (en) | Risk assessment method, device and equipment based on associated network and storage medium | |
| CN111325619A (en) | Credit card fraud detection model updating method and device based on joint learning | |
| CN112085615A (en) | Method and device for training graph neural network | |
| CN111428217B (en) | Fraudulent party identification method, apparatus, electronic device and computer readable storage medium | |
| CN112214775A (en) | Injection type attack method and device for graph data, medium and electronic equipment | |
| CN111340240A (en) | Method and device for realizing automatic machine learning | |
| CN114782161A (en) | Method, device, storage medium and electronic device for identifying risky users | |
| CN111611390B (en) | Data processing method and device | |
| CN115618008A (en) | Account state model construction method and device, computer equipment and storage medium | |
| CN115034379A (en) | Causal relationship determination method and related equipment | |
| CN113191565B (en) | Security prediction method, security prediction device, security prediction medium, and security prediction apparatus | |
| CN113435900A (en) | Transaction risk determination method and device and server | |
| CN116993513A (en) | Financial wind control model interpretation method and device and computer equipment | |
| CN112446777A (en) | Credit evaluation method, device, equipment and storage medium | |
| Kadam et al. | Loan Approval Prediction System using Logistic Regression and CIBIL Score | |
| CN109272398B (en) | Operation request processing system | |
| CN112559640A (en) | Training method and device of atlas characterization system | |
| CN112036418A (en) | Method and device for extracting user features | |
| Roa Ballén | Machine Learning Models and Alternative Data in Credit Scoring: Statistical and Financial impact | |
| CN116611923A (en) | Knowledge graph-based risk data acquisition method, system, equipment and storage medium | |
| CN115713147A (en) | Risk prediction method and device and electronic equipment | |
| CN117951517A (en) | Training method and device of node classification model, electronic equipment and storage medium | |
| CN115859197A (en) | Method and apparatus for identifying abnormal clusters, computing device, storage medium and computer program product |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| CB02 | Change of applicant information | ||
| CB02 | Change of applicant information |
Address after: 401121 b7-7-2, Yuxing Plaza, No.5 Huangyang Road, Yubei District, Chongqing Applicant after: Chongqing duxiaoman Youyang Technology Co.,Ltd. Address before: Room 3075, building 815, Jiayuan district, Shanghai Applicant before: SHANGHAI YOUYANG NEW MEDIA INFORMATION TECHNOLOGY Co.,Ltd. |
|
| TA01 | Transfer of patent application right | ||
| TA01 | Transfer of patent application right |
Effective date of registration: 20211220 Address after: Room 606, 6 / F, building 4, courtyard 10, Xibeiwang Road, Haidian District, Beijing 100085 Applicant after: Du Xiaoman Technology (Beijing) Co.,Ltd. Address before: 401121 b7-7-2, Yuxing Plaza, No.5 Huangyang Road, Yubei District, Chongqing Applicant before: Chongqing duxiaoman Youyang Technology Co.,Ltd. |