CN115809905A

CN115809905A - Object credibility assessment method and device and related products

Info

Publication number: CN115809905A
Application number: CN202111069861.1A
Authority: CN
Inventors: 刘全赟
Original assignee: Tenpay Payment Technology Co Ltd
Current assignee: Tenpay Payment Technology Co Ltd
Priority date: 2021-09-13
Filing date: 2021-09-13
Publication date: 2023-03-17

Abstract

The embodiment of the application discloses an object reliability assessment method and device and a related product. The application relates to the technical field of machine learning. In the method, the characteristics of the related object of the object to be evaluated are obtained firstly, wherein the characteristics comprise historical investigation characteristics and/or transaction characteristics. And then, performing feature aggregation on the object to be evaluated and the associated object based on the type of the association relationship between the object to be evaluated and the associated object to obtain the aggregated features of the object to be evaluated. In the scheme, the incidence relations of different types can have different influences on the evaluation of the object reliability, and the differentiation and the effective utilization of different incidence relation types among the objects are realized. Therefore, the method is different from the undifferentiated application mode of the incidence relation among the objects in the prior art, the credibility of the object to be evaluated is evaluated according to the aggregated characteristics of the object to be evaluated, and a more accurate object credibility evaluation result can be obtained.

Description

Object credibility assessment method and device and related products

Technical Field

The present application relates to the field of reliability assessment technologies, and in particular, to a method and an apparatus for assessing object reliability, and a related product.

Background

With the rise of electronic commerce, more and more objects are selected to reside in an electronic commerce platform, and commodities are sold on the platform. In order to ensure the healthy operation of the e-commerce platform, the object needs to be evaluated for credibility.

In the prior art, poor records of objects are generally used for reliability evaluation. In order to more effectively suppress the adverse effect of an object with low credibility, it is necessary to evaluate the credibility of a newly-resident object in the object residing stage. Confidence assessment is difficult because newly-populated objects initiate fewer transactions, often lacking in poor records. One solution is to evaluate the confidence level of a newly enlistened object by bad records of already enlistened objects associated with the newly enlistened object. However, in the object reliability evaluation, only the existence of the association relationship between the objects is considered, and the types of different association relationships cannot be distinguished and effectively utilized, so that the accuracy of the reliability evaluation result of the object is insufficient.

Disclosure of Invention

The embodiment of the application provides an object reliability assessment method, an object reliability assessment device and a related product, so that the accuracy of object reliability assessment is improved.

In view of this, a first aspect of the present application provides an object reliability assessment method, including:

obtaining the characteristics of the associated object of the object to be evaluated; the characteristics comprise historical investigation characteristics and/or transaction characteristics; the associated object and the object to be evaluated are positioned in a target platform;

performing feature aggregation on the object to be evaluated and the associated object based on the type of the association relationship between the object to be evaluated and the associated object to obtain the aggregated features of the object to be evaluated;

and evaluating the credibility of the object to be evaluated according to the aggregated features.

A second aspect of the present application provides an object reliability evaluating apparatus, including:

the characteristic acquisition unit is used for acquiring the characteristics of the associated object of the object to be evaluated; the characteristics comprise historical investigation characteristics and/or transaction characteristics; the associated object and the object to be evaluated are positioned in a target platform;

the characteristic aggregation unit is used for carrying out characteristic aggregation on the object to be evaluated and the associated object based on the type of the association relation between the object to be evaluated and the associated object to obtain the aggregated characteristics of the object to be evaluated;

and the object reliability evaluation unit is used for evaluating the reliability of the object to be evaluated according to the aggregated features.

A third aspect of the present application provides an object reliability assessment apparatus, including a processor and a memory:

the memory is used for storing the program codes and transmitting the program codes to the processor;

the processor is adapted to perform the steps of the object trustworthiness assessment method as provided above in the first aspect, in accordance with instructions in the program code.

A fourth aspect of the present application provides a computer-readable storage medium for storing program code for performing the method for object reliability assessment provided by the first aspect.

According to the technical scheme, the embodiment of the application has the following advantages:

the embodiment of the application provides an object credibility assessment method. In the method, the characteristics of the related object of the object to be evaluated are obtained firstly, wherein the characteristics comprise historical investigation characteristics and/or transaction characteristics. If the associated object has bad records, the record can be represented by a history checking characteristic and/or a transaction characteristic. And then, performing feature aggregation on the object to be evaluated and the associated object based on the type of the association relationship between the object to be evaluated and the associated object to obtain the aggregated features of the object to be evaluated. In the scheme, the incidence relations of different types can have different influences on the evaluation of the object reliability, for example, the incidence relation of a key type plays a more important role in feature aggregation, so that the differentiation and the effective utilization of different incidence relation types among the objects are realized. Therefore, the method is different from the undifferentiated application mode of the incidence relation among the objects in the prior art, the credibility of the object to be evaluated is evaluated according to the aggregated characteristics of the object to be evaluated, and a more accurate object credibility evaluation result can be obtained.

Drawings

Fig. 1 is a flowchart of an object credibility assessment method according to an embodiment of the present application;

fig. 2 is a flowchart of another object credibility assessment method according to an embodiment of the present application;

FIG. 3 is a heterogeneous subgraph corresponding to an object to be evaluated, which is extracted when the target order is 2;

FIG. 4 is a schematic illustration of the formation of a processed heterogeneous subgraph for FIG. 3;

FIG. 5 is a diagram illustrating the identification of meta-paths from a heterogeneous subgraph;

FIG. 6 is a flow diagram of node level aggregation of features;

FIG. 7 is a flow diagram of semantic level aggregation of features;

FIG. 8 is a schematic diagram of node-level aggregation and semantic-level aggregation of node features;

FIG. 9 is a schematic node diagram of a heterogeneous graph neural network participating in aggregation layer by layer during node feature aggregation of a second-order heterogeneous subgraph;

FIG. 10 is a schematic diagram of training a confidence evaluation model;

fig. 11 is a schematic structural diagram of an object reliability assessment apparatus according to an embodiment of the present application;

fig. 12 is a schematic structural diagram of a server for evaluating the reliability of an object according to an embodiment of the present application;

fig. 13 is a schematic structural diagram of a terminal device for evaluating reliability of an object according to an embodiment of the present application.

Detailed Description

As can be seen from the foregoing description, in the prior art, after the existence of the association relationship between the objects is known, the object reliability evaluation is directly made based on the existing association relationship. Since different association relationship types are not distinguished during evaluation, the association relationship types cannot be effectively applied during evaluation of object credibility. As an example, the relationship between the new resident object and the resident object is often not unique, the new resident object A and the resident object B have the same contact mailbox, and the new resident object A and the resident object C have the same enterprise unified credit code. In the prior art, when the reliability of a new stationing object is evaluated by applying a homogeneous graph neural network, because only one node type and one relation type are available in the homogeneous graph, different types of relations between the new stationing object and the stationed object cannot be distinguished and effectively utilized. In the scheme, different types of association relations are treated in the same way. For example, it is believed that the populated object B and the populated object C have an equal impact on assessing the trustworthiness of the new populated object A. Thus, the reliability evaluation result is not accurate enough.

Aiming at the problems, the application provides a novel object credibility assessment method, a novel object credibility assessment device and a related product for improving the accuracy of an object credibility assessment result. According to the technical scheme, the specific incidence relation types of the object to be evaluated and the incidence object are utilized, and the characteristics of the object are aggregated based on the type of the incidence relation. And finally, evaluating the credibility of the object to be evaluated according to the aggregated features. The scheme takes the type of the incidence relation between the objects as the basis of the aggregation object characteristics and is a new strategy for evaluating the reliability of the objects. Because influence of different types of association relations on credibility may be different, the credibility level of the object can be more accurately evaluated by applying the scheme to the object association relation types in a differentiated manner.

The above-mentioned object reliability evaluation method may be applied to a processing device having an object reliability evaluation function, for example, a terminal device or a server. The method is independently executed by the terminal equipment or the server, can also be applied to a network scene of communication between the terminal equipment and the server, and is operated by the cooperation of the terminal equipment and the server. The terminal device may be a mobile phone, a desktop computer, a Personal Digital Assistant (PDA for short), a tablet computer, or the like. The server may be understood as an application server, or may also be a Web server, and in actual deployment, the server may be an independent physical server, or may be a server cluster or a distributed system formed by a plurality of physical servers. The terminal and the server may be directly or indirectly connected through wired or wireless communication, and the application is not limited herein.

In addition, the present application relates to Artificial Intelligence (AI) technology. Artificial intelligence is a theory, method, technique and application system that uses a digital computer or a machine controlled by a digital computer to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use the knowledge to obtain the best results. In other words, artificial intelligence is a comprehensive technique of computer science that attempts to understand the essence of intelligence and produce a new intelligent machine that can react in a manner similar to human intelligence. Artificial intelligence is the research of the design principle and the realization method of various intelligent machines, so that the machines have the functions of perception, reasoning and decision making.

The artificial intelligence technology is a comprehensive subject and relates to the field of extensive technology, namely the technology of a hardware level and the technology of a software level. The artificial intelligence base technologies generally include technologies such as sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, big data processing technologies, operation/interaction systems, mechatronics, and the like. The artificial intelligence software technology mainly comprises a computer vision technology, a voice processing technology, a natural language processing technology, machine learning/deep learning, automatic driving, intelligent traffic and the like.

Natural Language Processing (NLP) is an important direction in the fields of computer science and artificial intelligence. It studies various theories and methods that enable efficient communication between humans and computers using natural language. Natural language processing is a science integrating linguistics, computer science and mathematics. Therefore, the research in this field will involve natural language, i.e. the language that people use everyday, so it is closely related to the research of linguistics. Natural language processing techniques typically include text processing, semantic understanding, machine translation, robotic question and answer, knowledge mapping, and the like.

Machine Learning (ML) is a multi-domain cross subject, and relates to multi-domain subjects such as probability theory, statistics, approximation theory, convex analysis, algorithm complexity theory and the like. The method specially studies how a computer simulates or realizes the learning behavior of human beings so as to acquire new knowledge or skills and reorganize the existing knowledge structure to continuously improve the performance of the computer. Machine learning is the core of artificial intelligence, is the fundamental approach for computers to have intelligence, and is applied to all fields of artificial intelligence. Machine learning and deep learning generally include techniques such as artificial neural networks, belief networks, reinforcement learning, transfer learning, inductive learning, and formula learning.

In the embodiment of the present application, the processing device may perform feature aggregation by the above techniques such as natural language processing, machine learning, and the like. For convenience of understanding, the object reliability assessment method provided by the embodiments of the present application is described below with reference to the embodiments and the accompanying drawings.

Fig. 1 is a flowchart of an object reliability evaluation method according to an embodiment of the present application. As shown in fig. 1, the method comprises:

s101: obtaining the characteristics of the associated object of the object to be evaluated; the features include historical review features and/or transaction features.

The associated object and the object to be evaluated reside on the same platform, referred to herein as the target platform. The associated object of the object to be evaluated may be determined by the registration information. When an object is hosted by a platform, its registration information is typically provided to the platform. In some possible implementation scenarios, such as a financial transaction scenario, the object referred to by the present scenario may be a merchant. If the merchant is not trusted, the risk is brought to the transaction on the platform, such as the loss of economic interest of the purchaser. When the associated object and the object to be evaluated are merchants, the registration information includes, but is not limited to, at least one of the following types: contact mobile phone number, legal person ID card, bank card number, enterprise unified credit code, shareholder representative, merchant full name or contact mailbox. For example, when a certain object and the object to be evaluated are the same in the same type of registration information, it can be determined that the object and the object to be evaluated have an association relationship of this type. Therefore, as an example, the association relationship between the object to be evaluated and the associated object may be embodied as the same registration information.

For example, when the contact mobile phone number of a certain object is the same as the contact mobile phone number of the object to be evaluated, it can be determined that the certain object is the associated object of the object to be evaluated. The type of the association relationship between the two is as follows: the contact mobile phone number.

For another example, when the enterprise unified credit code of a certain object is the same as the enterprise unified credit code of the object to be evaluated, it can be determined that the certain object is the associated object of the object to be evaluated. The type of the association relationship between the two is as follows: the enterprise unifies the credit codes.

For objects residing on the target platform, features are provided, including but not limited to transaction features, historical review features, and the like. For an object that is newly launched into the platform, the transaction characteristics may be null because the time of its launch is short, the amount of transactions that occur is small or no transactions occur. In addition, if the object has not been investigated, its historical investigation characteristics may also be null.

The historical investigation characteristics comprise positive investigation result characteristics or negative investigation result characteristics, and the transaction characteristics comprise positive transaction result characteristics or negative transaction result characteristics; both the negative findings characteristic and the negative transaction results characteristic point to be untrustworthy. As an example, the history checking feature may be a blacked out record, a label, and the like. For example, if an object is blacklisted by the platform, the historical search features are negative search result features. On the contrary, if the object is never blacked or marked with a black label, the historical investigation feature is a positive investigation result feature. The transaction characteristics of the object may include gender ratio, average age, number of blocked strokes, etc. of the transaction user. Taking the number of interception strokes as an example, interception indicates that the transaction was not successfully performed, which for transactions that the platform actively intercepts may be due to active interception performed by the object's credit level or complaint volume. Therefore, in an alternative implementation manner, if the number of the intercepted strokes exceeds the preset number, the transaction feature of the number of the intercepted strokes is a negative transaction result feature; otherwise, if the number of the intercepted strokes does not exceed the preset number, the transaction characteristic is a positive transaction result characteristic. Of course, the preset number may be set according to actual requirements, and is not limited numerically here.

The historical investigation characteristics and the transaction characteristics can be represented by positive or negative, and can also be represented by visual numerical values. The above description is presented as an example form of a history review feature and a transaction feature. The characteristics of the object are not limited to the history check characteristics and the transaction characteristics. For example, the information of the industry and commerce of the object, such as the scale of the object registration company, the registered funds, the established market, etc., can be included.

S102: and performing feature aggregation on the object to be evaluated and the associated object based on the type of the association relationship between the object to be evaluated and the associated object to obtain the aggregated features of the object to be evaluated.

In this embodiment of the present application, S102 specifically refers to guiding specific execution of feature aggregation by using an association relationship type between an object to be evaluated and an associated object as a basis for aggregating features. In S102, feature aggregation is performed by depending on the association relationship type, so that the feature aggregation has certain policy.

An alternative implementation of this step is described below in terms of a heterogeneous graph. As described above, there is only one node type and one relationship type in the homogenous graph, and the different association relationship types cannot be distinguished and effectively utilized. Heterogeneous graphs support different types of relationships between nodes. Therefore, the characteristic of the heterogeneous graph can be skillfully utilized in the embodiment of the application to aggregate the object characteristics in the heterogeneous graph. In the heterogeneous graph, objects are used as nodes, incidence relations among the objects are used as edges, and the feature of the objects is represented by the feature vector of each node. In the heterogeneous graph, the heterogeneous graph may be divided into a plurality of meta-paths according to the type of edges between nodes, that is, the type of association relationship between objects, node-level aggregation may be performed inside the meta-paths, and semantic-level aggregation may be performed on different meta-paths. And finally, taking the semantic level aggregation result of the node characteristics representing the object to be evaluated as the aggregated characteristics of the object to be evaluated. The feature aggregation of the nodes in the heterogeneous graph will be explained in detail in conjunction with the embodiments later.

S103: and evaluating the credibility of the object to be evaluated according to the aggregated features.

The aggregated features of the objects to be evaluated are obtained by performing feature aggregation according to the types of the association relations among the objects, and the fact that the association relations of different types have different influences on the reliability of the objects to be evaluated is considered, for example, the association relations of key types play a more important role in feature aggregation, so that the differentiation and the effective utilization of different types of association relations among the objects are realized. Therefore, different from the undifferentiated application mode of the prior art for the incidence relation between the objects, the reliability evaluation is performed on the object to be evaluated according to the aggregated characteristics of the object to be evaluated, so that the value of the characteristics of the objects with different incidence relations when the reliability evaluation is performed on the object to be evaluated can be more seriously considered, and a more accurate reliability evaluation result is obtained.

The following describes an implementation of implementing node feature aggregation based on a heterogeneous graph with reference to a flowchart shown in fig. 2. Fig. 2 is a flowchart of another object reliability assessment method according to an embodiment of the present application. The method shown in fig. 2 comprises:

s201: and determining an object to be evaluated.

And when a new object is placed in the target platform, the target platform adds the new object as a new node into the heterogeneous graph of the placed object of the target platform. Edges between the new node and the old node are constructed according to the incidence relation. That is, the objects are represented by nodes in the heterogeneous graph of the embedded objects, the nodes record the characteristics of the objects, the association relationship between the nodes is represented by edges, and the types of the edges correspond to the types of the association relationship. For example, if the new node has a contact phone number in common with an old node, an edge of the type of the contact phone number is constructed between the new node and the old node.

Before evaluating the object to be evaluated, the object to be evaluated is determined first. As an example, the node representing the object to be evaluated may be determined in the heterogeneous graph of the residing object according to all registration information (e.g., a contact mobile phone number, a legal identity card, a bank card number, an enterprise unified credit code, a shareholder representative, a merchant full name, and a contact mailbox) of the object to be evaluated, the residence time of the target platform, and the like.

S202: and extracting a heterogeneous subgraph corresponding to the object to be evaluated from the heterogeneous graph of the resident object of the target platform.

In order to improve the evaluation efficiency of the object to be evaluated and reduce the interference of the non-associated object on the evaluation of the credibility of the object to be evaluated, a heterogeneous graph corresponding to the object to be evaluated is extracted from the whole heterogeneous graph of the resident object in the step, and for the convenience of distinguishing, the heterogeneous graph is called a heterogeneous subgraph. The heterogeneous subgraph includes the nodes of the object to be evaluated (i.e., the nodes representing the object to be evaluated), the nodes of the associated object (i.e., the nodes representing the associated object), and the edges between the nodes in the heterogeneous subgraph. The associated object and the object to be evaluated are located in the target platform, and the locating time of the associated object is earlier than that of the object to be evaluated. When the credibility of the new resident object is evaluated, the characteristic of the associated object with the resident time earlier than that of the new resident object has availability, so that the associated object of the object to be evaluated can be screened out according to the resident time of the object. Other objects with the parking time synchronized with the object to be evaluated or later than the parking time of the object to be evaluated are not considered.

The heterogeneous subgraph can be selected according to the target order N. The target order N represents the maximum order of the associated object node used for expecting to evaluate the reliability of the object to be evaluated. According to the target order N, extracting an N-order heterogeneous subgraph corresponding to an object to be evaluated from a heterogeneous graph of a stationing object of a target platform; the associated object nodes in the N-order heterogeneous subgraph comprise associated object nodes with the order less than or equal to N.

For example, N =1, indicating that 1 order of associated object nodes need to be included in the heterogeneous subgraph; n =2, which indicates that the heterogeneous subgraph needs to contain 1-order associated object nodes and 2-order associated object nodes; n =3, which indicates that the heterogeneous subgraph needs to include 1-order associated object nodes, 2-order associated object nodes and 3-order associated object nodes. The 1-order associated object node is a node which is directly connected with an object node to be evaluated through an edge in the heterogeneous graph of the embedded object, the 2-order associated object node is a node which is directly connected with the 1-order associated object node through an edge except the object node to be evaluated, the 3-order associated object node is a node which is directly connected with the 2-order associated object node through an edge except the 1-order associated object node, and the like.

Fig. 3 is a heterogeneous subgraph corresponding to an object to be evaluated, which is extracted when the target order is 2. As shown in fig. 3, a node marked with 0 is an object node to be evaluated, a node marked with 1 is a 1-order associated object node, and a node marked with 2 is a 2-order associated object node. In the heterogeneous subgraph shown in fig. 3, different 3 types of edges are shown in different lineation modes, including 3 types of mobile phone numbers, bank cards and identification cards.

S203: and backtracking the characteristics of the nodes in the heterogeneous subgraph to the residence time of the object to be evaluated.

The technical scheme provided by the embodiment of the application is mainly used for evaluating the credibility of the late-entry object to be evaluated through the characteristics of the early-entry associated object. The time for extracting the heterogeneous subgraph is possibly later than the entrance time of the object to be evaluated, so that partial characteristics of the nodes in the heterogeneous subgraph are possibly established after the entrance time of the object to be evaluated. In order to ensure the accuracy of the evaluation prediction, the characteristics of the remaining associated objects are traced back according to the residence time of the object to be evaluated in the step, that is, the characteristics of each associated object need to be traced back to the residence time of the object to be evaluated on the target platform, and the characteristics generated after the residence time of the object to be evaluated are removed.

In the embodiment of the present application, a reliability evaluation model is trained in advance. When feature aggregation is needed, feature aggregation is carried out on the object node to be evaluated and the associated object node in the heterogeneous subgraph through the reliability evaluation model based on the type of the edge in the heterogeneous subgraph, and the aggregated feature of the object node to be evaluated is obtained. The credibility evaluation model comprises a heterogeneous map neural network. The training mode of the reliability evaluation model will be described in detail later, and the application of the reliability evaluation model is described through S204 to S206.

In order to ensure that the reliability evaluation model normally processes the input heterogeneous subgraph, in an optional implementation manner, before executing S204, the heterogeneous subgraph may be further processed to convert an undirected graph (i.e., the heterogeneous subgraph) into a directed graph through processing, so as to facilitate processing by the heterogeneous graph neural network. The specific processing method is as follows:

1) And adding self-loop edges to the nodes in the heterogeneous subgraph. By adding the self-loop edge, the information of the node can be obtained by each convolution.

2) The self-looping edge is assigned the type of each edge in the heterogeneous subgraph. That is, a self-loop is added to each association of each node. Edges between nodes are only of a particular type, while self-looping edges have each type. Fig. 4 is a schematic illustration of the formation of a processed heterogeneous subgraph for fig. 3. As shown in fig. 4, a ring-shaped self-ring edge is added to each node.

3) And converting the edges in the heterogeneous subgraph into two directed edges. It can be readily seen in fig. 3 that the edges between the nodes do not point in the direction. In contrast to fig. 4, the edges between nodes are pointed by arrows, which indicates that the undirected edges are converted into directed edges.

4) And deleting the directed edges in the heterogeneous subgraph starting from the nodes of the objects to be evaluated. As shown in fig. 4, by deleting the directed edge starting from the object node to be evaluated, only the directional arrow pointing to the edge constructed by the object node to be evaluated exists.

Because the object node to be evaluated does not contain attribute information, such as transaction characteristics or historical investigation characteristics, no matter all-zero characteristic filling or mean characteristic filling is used, certain influence on information extraction and importance coefficient calculation of a neighbor node (namely a 1-order associated object node) cannot be avoided. Therefore, the structure is directly changed, all edges which take the object node to be evaluated as the starting point are directly deleted except the ring edge, and the influence of the object node to be evaluated can be ignored when the characteristics of the associated object node are updated.

Obtaining a processed heterogeneous subgraph through the operations of the steps 1) to 4). The heterogeneous subgraph processed in S204 hereinafter may be specifically the heterogeneous subgraph processed through the above steps.

S204: a heterogeneous subgraph is identified as a plurality of meta-paths by different types of edges.

In the embodiment of the present application, the meta path (meta path) is identified based on the type of the edge. For example, a heterogeneous subgraph contains 3 types of edges, where a first type of edge is used to determine a first meta-path, a second type of edge is used to determine a second meta-path, and a third type of edge is used to determine a third meta-path. FIG. 5 is a diagram illustrating the recognition of meta-paths from a heterogeneous subgraph.

Based on the determination of the plurality of meta-paths in S204, the node features are then aggregated. In S205 to S206 described later, the nodes are not particularly specified, and the following operation is performed for each node in the heterogeneous subgraph. Wherein, S205 introduces node-level aggregation of features, and the node-level aggregation occurs inside the meta-path; s206 introduces semantic-level aggregation of features, which occurs between meta-paths.

S205: and performing feature aggregation between nodes in the meta-path to obtain a first aggregated feature of the nodes in the meta-path.

FIG. 6 is a flow diagram of node-level aggregation of features. Referring to fig. 6, the node-level feature aggregation procedure includes:

s2051: and acquiring a normalized first aggregation weight of the target neighbor node to the target node in the meta-path based on the characteristics of the target node, the characteristics of the target neighbor node and the characteristics of the non-target neighbor node.

For a more accurate and clear description, and for ease of understanding, the concept of a target node is used herein. The target node is any node in the meta-path, and the target neighbor node is any neighbor node connected with the target node through an edge in the meta-path to which the target node belongs. The non-target neighbor node refers to other neighbor nodes of the target node except the target neighbor node in the meta-path to which the target node belongs.

Suppose that the meta-path Φ includes nodes i, j, etc., where the node i is regarded as the target node and the node j is regarded as a target neighbor node of the target node i. The feature vectors of node i and node j are denoted as h respectively _i 'and h' _j . Target neighbor node jFirst aggregation weight for target node i

Representing the importance of the feature of node j to node i. It is to be understood that the first aggregate weight

The higher the feature representing node j is for node i the higher the importance. In order to make it easier for different neighbor nodes to compare the first aggregation weight of the target node, the first aggregation weight needs to be normalized in this step. The implementation of this step is described below in conjunction with equation (1).

Combining equation (1), the first aggregation weight can be weighted by softmax function in this step

And (6) carrying out normalization.

Is the normalized first aggregate weight. h is a total of _i '||h' _j Indicating that the characteristics of the target node and the target neighbor node are to be spliced. a is _Φ For the linear transformation matrix corresponding to the meta-path phi,

is a linear transformation matrix a _Φ The inverse matrix of (c).

Characteristic h of splicing _i '||h' _j And converting into a form suitable for the meta-path phi. σ () represents an activation function.

Set of all neighbor nodes representing target node i on meta-path ΦThe node j and other non-target nodes are contained in the union. k represents a set

Any one of the neighboring nodes. h' _k Is the feature vector of node k.

S2052: and performing linear transformation on the characteristics of the target node by using the normalized first aggregation weight of the target node by the target neighbor node in the meta-path to obtain a first linear transformation result of the characteristics of the target node by the target neighbor node.

In connection with the example above, the result of the first linear transformation of the characteristics of the target neighbor node j to the target node i is represented as:

s2053: and accumulating the first linear transformation results of the target node by each neighbor node of the target node in the meta path to obtain an accumulation result.

The accumulated result is expressed as:

wherein

Represents the set of all neighbor nodes of the target node i on the meta-path Φ.

S2054: and carrying out nonlinear transformation on the accumulated result through a nonlinear activation function to obtain a first aggregated feature of the target node in the meta-path.

First aggregated characteristics of target node i in meta-path phi obtained through nonlinear transformation

The expression of (a) is as follows:

the first aggregated characteristic of the target node in the meta-path is the result of the target node undergoing node-level aggregation in the meta-path. The operations of S2051 to S2054 may be performed on each node of each meta-path, and a result of node-level aggregation of the characteristics thereof is obtained.

After the feature node level aggregation of the above S2051 to S2054, the following flow of semantic level aggregation of features is entered.

S206: and for the same node, respectively carrying out feature aggregation among the meta-paths based on the first aggregated features aggregated in different meta-paths to obtain a second aggregated feature of the same node.

FIG. 7 is a flow diagram of semantic level aggregation of features. As shown in fig. 7, the flow of semantic level aggregation includes:

s2061: the first aggregated feature of the nodes in the meta-path is transformed into a weight scalar.

S2062: an average of the weight scalars of all nodes in the meta-path is obtained as the second aggregate weight.

In the formula (3), the first and second groups of the compound,

and a weight scalar quantity obtained by the first feature transformation after aggregation of the nodes in the meta path is represented. Wherein the content of the first and second substances,

representing a first aggregated feature of a target node i in a meta-path Φ

b denotes an intercept, and W denotes a weighted linear transformation matrix. The vector used to convert a vector into a scalar in a heterogeneous graph neural network is represented as vector q, the inverse of which is denoted as q ^T 。

Equation (3) is an expression for averaging the scalar weights of all nodes in the meta-path Φ. Wherein V representsThe meta-path Φ is the set of all nodes, | V | is the total number of nodes in the set. Second polymerization weight ω _Φi And representing the weight corresponding to the meta-path phi when the node i is subjected to feature semantic level aggregation.

S2063: and obtaining the normalized second aggregation weight of the target meta-path according to the second aggregation weight of the target meta-path and the second aggregation weights of other meta-paths except the target meta-path.

The target meta-path is any meta-path in a plurality of meta-paths of the heterogeneous subgraph, and phi represents the target meta-path. In order to facilitate semantic level aggregation of node features, a second aggregation weight obtained at S2062 is further required

And (6) carrying out normalization. Normalized second aggregation weights

Is a probability, the expression is as follows:

in combination with equation (4), this normalization is performed by the softmax function. In formula (4), P represents the total number of heterogeneous sub-primitive paths.

A second aggregation weight representing a feature semantic level aggregation for the target node i, the target meta-path Φ. If the values of phi are different from 1 to P, then

And respectively representing the feature semantic level aggregation of the target node i and the second aggregation weights of other meta-paths except the target meta-path.

S2064: and performing linear transformation on the first aggregated features of the nodes in the target element path by using the normalized second aggregation weight of the target element path to obtain a second linear transformation result of the target element path on the features of the nodes.

Taking the target meta-path as the meta-path Φ and the node i as an example, the second linear transformation result of the target meta-path for the characteristics of the node is represented as:

wherein

Represents the first aggregated feature of node i in meta-path phi,

a normalized second aggregation weight representing the target meta-path Φ.

S2065: and accumulating the second linear transformation results of the characteristics of the same node by different element paths to obtain a second aggregated characteristic of the same node.

Taking node i as an example, for node i, the second aggregated feature Z is _i The expression is as follows:

in equation (5), P represents the total number of heterogeneous sub-primitive paths.

And the second aggregated feature of the node is the feature of the node after node level aggregation in the meta-path and semantic level aggregation between the meta-paths. Compared with the node-level aggregation, the method is equivalent to that each node updates the original characteristics with the second aggregated characteristics.

Fig. 8 is a schematic diagram of node-level aggregation and semantic-level aggregation of node features. As shown in fig. 8, the heterogeneous subgraph is divided into three meta-paths, and node-level aggregation is performed inside each meta-path, so as to obtain a first aggregated feature, which is denoted by Z in fig. 8. After the node-level aggregation is performed, semantic-level aggregation between meta-paths is performed on the node features, and specifically, normalized second aggregation weights are used to process the first aggregated features, which are denoted by β in fig. 8. Finally, the second aggregated feature is obtained, and the one-round aggregation of the node features by the model is completed.

In the above-described steps of the embodiment, a single node is taken as an example to illustrate the overall process of feature aggregation. If the target order is 1, the aggregation operation is only needed to be executed on the object node to be evaluated, and the object node to be evaluated only has 1-order associated object nodes. And if the target order is an integer larger than 1, the aggregation of the nodes needs to be realized layer by layer through a multilayer heterogeneous graph neural network of the credibility assessment model.

For an example, the target order is 2, and the number of layers of the heterogeneous map neural network of the reliability evaluation model is matched with the target order, that is, the model includes two layers of heterogeneous map neural networks. In the first-layer heterogeneous graph neural network, feature updating needs to be performed on object nodes to be evaluated and 1-order associated object nodes (namely, original features are updated through features obtained through node-level aggregation and semantic-level aggregation). In the second-layer heterogeneous graph neural network, feature updating is only needed to be carried out on the object node to be evaluated.

As another example, the target order is 3, i.e. the associated object nodes contained in the heterogeneous subgraph relate to orders 1, 2 and 3. In the first-layer heterogeneous graph neural network, feature updating needs to be carried out on the object nodes to be evaluated, the 1-order associated object nodes and the 2-order associated object nodes. In the second-layer heterogeneous graph neural network, feature updating needs to be carried out on the object node to be evaluated and the 1 st order associated object node. In the third-layer heterogeneous graph neural network, only the characteristic updating is needed to be carried out on the nodes of the objects to be evaluated.

As can be seen from the above example, the updated node degree is gradually reduced during processing of each layer network. The implementation process of the whole credibility assessment model after processing and acquiring the aggregated characteristics of the nodes of the object to be assessed is described in a layering manner as follows:

and for the mth-layer heterogeneous graph neural network (m is an integer from 1 to N-1), performing feature aggregation on the mth-layer heterogeneous graph neural network based on the types of edges between the object node to be evaluated and the 1 st-order to N-m +1 st-order associated object nodes in the N-order heterogeneous subgraph to update the N-order heterogeneous subgraph. That is, when the feature aggregation is performed in the mth layer heterogeneous graph neural network, the features of the 1 st order to N-m +1 st order associated object nodes are used. The node feature update category for the N-th order heterogeneous subgraph can only involve the object node to be evaluated and the 1 to N-m order associated object nodes, because the features of the N-m +1 order associated object nodes are not used in the next layer of network processing.

And performing feature aggregation on the object node to be evaluated through the N-layer heterogeneous graph neural network based on the type of the edge between the object node to be evaluated and the 1-order associated object node in the N-order heterogeneous subgraph updated by the N-1-layer heterogeneous graph neural network to obtain the aggregated feature of the object node to be evaluated through the N-layer heterogeneous graph neural network. That is, when feature aggregation is performed in the nth-layer heterogeneous graph neural network, features of 1 st-order related object nodes are used. In the last layer of network, the update of the feature aggregation result may only focus on the object node to be evaluated.

Fig. 9 is a schematic node diagram of a heterogeneous graph neural network participating in aggregation layer by layer during node feature aggregation of a second-order heterogeneous subgraph. As shown in fig. 9, in the first-level heterogeneous graph neural network, the feature of the associated object node of order 2 affects the update of the feature of the associated object node of order 1. In the second-layer heterogeneous graph neural network, only the characteristics of 1-order associated object nodes influence the updating of the characteristics of the object nodes to be evaluated.

S207: and evaluating the credibility of the object to be evaluated according to the aggregated features.

And S206, obtaining the aggregated features of the object nodes to be evaluated as feature vectors with fixed dimensions until the end. In an alternative implementation manner, the probability value of the object node to be evaluated, which is a risk object (i.e. an object with low credibility), may be obtained through a multi-layer fully-connected network according to the aggregated characteristics of the object node to be evaluated. The probability value represents the risk level of the object to be evaluated, and the higher the probability is, the higher the risk level is, and the lower the credibility is; the lower the probability, the lower the risk level, and the higher the confidence.

The above description may implement the credibility assessment of the object to be assessed through the credibility assessment model. The following describes a training process of the reliability evaluation model with reference to the drawings. FIG. 10 is a diagram of training a confidence evaluation model. As shown in fig. 10, a heterogeneous subgraph corresponding to a sample object is obtained first. Sample objects refer to objects for which confidence needs to be assessed during training. Both the data of the sample object and its heterogeneous subgraphs can be obtained from historical data.

In the embodiment of the application, the platform where the sample object resides is constructed with a heterogeneous graph of the resident object sample. Similar to the heterogeneous graph of the resident objects, the features of the objects resident on the platform are listed as nodes in the heterogeneous graph. The heterogeneous graph of the embedded object sample represents the object by nodes, the nodes record the characteristics of the object, the edges represent the association relationship between the nodes, and the types of the edges correspond to the types of the association relationship.

In order to train the model, a heterogeneous subgraph corresponding to the sample object needs to be extracted from the heterogeneous graph of the embedded object sample. The heterogeneous subgraph corresponding to the sample object comprises sample object nodes, associated object nodes of the sample object nodes and edges among the nodes in the heterogeneous subgraph corresponding to the sample object; the entry time of the associated object of the sample object is earlier than the entry time of the sample object.

And performing feature aggregation on the sample object nodes and the associated object nodes of the sample object nodes in the heterogeneous subgraph corresponding to the sample object through the model to be trained based on the type of the edges in the heterogeneous subgraph corresponding to the sample object, so as to obtain the aggregated features of the sample object nodes. And the number of layers of the heterogeneous graph neural network set in the model to be trained is matched with the maximum order of the associated object node of the yelling in the heterogeneous subgraph corresponding to the sample object.

And then, carrying out credibility evaluation on the sample object according to the aggregated characteristics of the sample object nodes to obtain a credibility evaluation result. For example, the probability that the sample object is an untrusted object (risk object) is obtained.

Each sample object is provided with a confidence label. Taking the object as an example of a merchant, the credibility label indicates the real situation that the sample merchant becomes a black merchant in the future (a merchant with very high risk and hindering the platform from normal transaction, for example, a credit has a very large problem or a service has a very large problem, etc.). For example, a confidence label of 1 indicates that it is a black business, and a confidence label of 0 indicates that it belongs to a white business (no characteristics that point to is not trusted exist).

Since the model has obtained the reliability evaluation result in the training process, the coefficients of the heterogeneous graph neural network in the model to be trained can be adjusted according to the difference between the reliability evaluation result of the sample object and the reliability label of the sample object until the training cutoff condition is met, and the reliability evaluation model is obtained.

The training cutoff condition may be that the number of times of model training iteration reaches a preset threshold, or that the loss value is smaller than a preset value. The loss function may specifically be a function of a gap between the credibility assessment result for the sample object and the credibility label for the sample object. Higher loss values indicate greater gaps.

By applying the object credibility assessment method provided by the technical scheme, after a new object is parked and basic information is submitted, the associated heterogeneous subgraph and the feature vector of the associated object in the subgraph can be quickly obtained in the graph according to the basic information, and the information is used for prediction through a trained model to obtain a credibility assessment result. In an alternative implementation, an object whose confidence level is indicated to be low in the confidence level evaluation result may be rejected from successfully residing in the platform.

In the object parking stage, the amount of information available is small compared to the transaction stage, and it is difficult to develop an effective policy using information other than the penalty records associated with the object. And only by the penalty record, the strategy is difficult to cover a large number of untrusted objects associated with the resident objects when the resident objects exist, and further the resident of the untrusted objects cannot be timely and effectively controlled.

According to the scheme, a heterogeneous graph of the newly embedded object is constructed, a heterogeneous graph neural network model is trained by extracting heterogeneous subgraphs of the newly embedded object on the graph, characteristics of different paths associated to the object are extracted, therefore, comprehensive judgment is conducted on the credibility of the newly embedded object by utilizing transaction information, punishment information, company main body information and the like of multi-level associated objects, and finally, punishment or troubleshooting decision is assisted. According to the scheme, the information in the heterogeneous subgraphs can be effectively aggregated by using the multi-stage association information, different association relations are distinguished, and weights are automatically distributed to the different association relations. Compared with a general method for converting the object into the structured data, the method has the advantages that the accuracy of the credibility evaluation is improved to a certain extent, the malicious registered object can be detected more accurately finally, the manual auditing cost of the object during parking is saved, and the credibility is prevented in advance.

Based on the object reliability evaluation method provided by the foregoing embodiment, correspondingly, the application further provides an object reliability evaluation device. The following describes an implementation of the object reliability evaluation device with reference to the embodiments and the drawings.

Fig. 11 is a schematic structural diagram of an object reliability evaluation apparatus provided in the application embodiment. The object reliability evaluation device 110 shown in fig. 11 includes:

a feature obtaining unit 111, configured to obtain a feature of an object to be evaluated, which is related to the object; the characteristics comprise historical troubleshooting characteristics and/or transaction characteristics; the associated object and the object to be evaluated are positioned in a target platform;

the feature aggregation unit 112 is configured to perform feature aggregation on the object to be evaluated and the associated object based on the type of the association relationship between the object to be evaluated and the associated object, and obtain an aggregated feature of the object to be evaluated;

and the object reliability evaluation unit 113 is configured to perform reliability evaluation on the object to be evaluated according to the aggregated features.

Optionally, the feature acquiring unit 111 includes:

the device comprises a to-be-evaluated object determining unit, a to-be-evaluated object determining unit and a to-be-evaluated object determining unit, wherein the to-be-evaluated object determining unit is used for determining the to-be-evaluated object; the associated object and the object to be evaluated are placed in the target platform, and the placement time of the associated object is earlier than that of the object to be evaluated;

the heterogeneous subgraph extraction unit is used for extracting a heterogeneous subgraph corresponding to an object to be evaluated from a heterogeneous graph of the object to be parked of the target platform; representing the object by nodes in the heterogeneous graph of the embedded object, recording the characteristics of the object by the nodes, representing the incidence relation between the nodes by edges, wherein the type of the edges corresponds to the type of the incidence relation; the heterogeneous subgraph comprises object nodes to be evaluated, associated object nodes and edges among the nodes in the heterogeneous subgraph;

and the characteristic backtracking unit is used for backtracking the characteristics of the nodes in the heterogeneous subgraph to the residence time of the object to be evaluated.

The feature aggregation unit 112 is specifically configured to perform feature aggregation on the object node to be evaluated and the associated object node in the heterogeneous subgraph based on the type of the edge in the heterogeneous subgraph through the reliability evaluation model, so as to obtain an aggregated feature of the object node to be evaluated.

Optionally, the feature aggregation unit 112 includes:

a meta path identifying unit for identifying the heterogeneous subgraph as a plurality of meta paths by different types of edges; the meta path comprises nodes with the same type of edges;

a node-level feature aggregation unit 112, configured to perform feature aggregation between nodes in the meta-path to obtain a first aggregated feature of the nodes in the meta-path;

and a semantic level feature aggregation unit 112, configured to perform feature aggregation between meta-paths on the same node based on the first aggregated features aggregated in different meta-paths, respectively, to obtain a second aggregated feature of the same node.

Optionally, the node-level feature aggregation unit 112 includes:

the first normalization subunit is used for acquiring a normalized first aggregation weight of the target neighbor node on the basis of the characteristics of the target node, the characteristics of the target neighbor node and the characteristics of the non-target neighbor node in the meta-path; the target node is any node in the meta-path, the target neighbor node is any neighbor node connected with the target node through an edge in the meta-path, and the non-target neighbor node is other neighbor nodes of the target node except the target neighbor node in the meta-path;

the first transformation subunit is used for performing linear transformation on the characteristics of the target node by using the normalized first aggregation weight of the target neighbor node to the target node in the meta-path to obtain a first linear transformation result of the characteristics of the target neighbor node to the target node;

the first accumulation unit is used for accumulating the first linear transformation result of each neighbor node of the target node in the meta-path to the target node to obtain an accumulation result;

and the second transformation subunit is used for carrying out nonlinear transformation on the accumulated result through a nonlinear activation function to obtain a first post-aggregation characteristic of the target node in the meta-path.

Optionally, the semantic level feature aggregation unit 112 includes:

a third transformation subunit, configured to transform the first aggregated features of the nodes in the meta-path into a weight scalar;

the average operation subunit is used for obtaining the average value of the weight scalars of all the nodes in the element path as a second aggregation weight;

the second normalization subunit is configured to obtain a normalized second aggregation weight of the target meta-path according to the second aggregation weight of the target meta-path and the second aggregation weights of other meta-paths other than the target meta-path; the target meta-path is any meta-path in a plurality of meta-paths of the heterogeneous subgraphs;

the fourth transformation subunit is configured to perform linear transformation on the first aggregated features of the nodes in the target meta-path by using the normalized second aggregation weight of the target meta-path, so as to obtain a second linear transformation result of the target meta-path on the features of the nodes;

and the second accumulation subunit is used for accumulating the second linear transformation results of the characteristics of the same node by different element paths to obtain a second aggregated characteristic of the same node.

Optionally, the object credibility evaluating apparatus 110 may further include:

the order determining unit is used for determining a target order N, and the target order N is a positive integer larger than 1;

the heterogeneous subgraph extraction unit is specifically used for extracting an N-order heterogeneous subgraph corresponding to an object to be evaluated from a heterogeneous graph of the object to be assessed according to the target order N; the associated object nodes in the N-order heterogeneous subgraph comprise associated object nodes with the order less than or equal to N.

Optionally, the credibility evaluation model comprises N layers of heterogeneous graph neural networks, and the heterogeneous graph neural networks from the 1 st layer to the N th layer operate layer by layer;

the feature aggregation unit 112 is configured to:

performing feature aggregation on the m-th layer heterogeneous graph neural network based on the types of edges between object nodes to be evaluated and associated object nodes from 1 order to N-m +1 order in the N-order heterogeneous subgraphs to update the N-order heterogeneous subgraphs; m is an integer of 1 to N-1;

and performing feature aggregation on the object node to be evaluated through the N-layer heterogeneous graph neural network based on the type of the edge between the object node to be evaluated and the 1-order associated object node in the N-order heterogeneous subgraph updated by the N-1-layer heterogeneous graph neural network to obtain the aggregated feature of the object node to be evaluated through the N-layer heterogeneous graph neural network.

Optionally, the object credibility assessment apparatus 110 further includes:

the heterogeneous subgraph processing unit is used for adding a self-loop edge to the nodes in the heterogeneous subgraph and endowing the self-loop edge with the type of each edge in the heterogeneous subgraph before feature aggregation is carried out on the object node to be evaluated and the associated object node in the heterogeneous subgraph on the basis of the types of the edges in the heterogeneous subgraph through the credibility evaluation model; and converting the edges in the heterogeneous subgraph into two directed edges to obtain the processed heterogeneous subgraph.

Optionally, the heterogeneous subgraph processing unit is further configured to delete directed edges in the heterogeneous subgraph starting from the node of the object to be evaluated.

Alternatively, the reliability evaluation model is obtained by training a model training unit in the lower subject reliability evaluation device 110. The model training unit includes:

the training graph acquisition unit is used for extracting heterogeneous subgraphs corresponding to the sample objects from the heterogeneous graph of the entrance object sample; representing the object by using nodes in the heterogeneous graph of the embedded object sample, recording the characteristics of the object by using the nodes, representing the incidence relation between the nodes by using edges, wherein the type of the edges corresponds to the type of the incidence relation; the heterogeneous subgraph corresponding to the sample object comprises sample object nodes, associated object nodes of the sample object nodes and edges among the nodes in the heterogeneous subgraph corresponding to the sample object; the residence time of the associated object of the sample object is earlier than the residence time of the sample object;

the model processing unit is used for carrying out feature aggregation on sample object nodes and associated object nodes of the sample object nodes in the heterogeneous subgraph corresponding to the sample object through the model to be trained based on the type of the edges in the heterogeneous subgraph corresponding to the sample object, so as to obtain the aggregated features of the sample object nodes;

the evaluation unit is used for evaluating the credibility of the sample object according to the aggregated characteristics of the sample object nodes to obtain a credibility evaluation result;

and the coefficient adjusting unit is used for adjusting the coefficient of the heterogeneous graph neural network in the model to be trained according to the difference between the reliability evaluation result of the sample object and the reliability label of the sample object until the training cut-off condition is met, and obtaining the reliability evaluation model.

Fig. 12 is a schematic diagram of a server 900 according to an embodiment of the present application, where the server 900 may have a relatively large difference due to different configurations or performances, and may include one or more Central Processing Units (CPUs) 922 (e.g., one or more processors) and a memory 932, and one or more storage media 930 (e.g., one or more mass storage devices) for storing applications 942 or data 944. Memory 932 and storage media 930 can be, among other things, transient storage or persistent storage. The program stored on the storage medium 930 may include one or more modules (not shown), each of which may include a series of instruction operations for the server. Still further, a central processor 922 may be provided in communication with the storage medium 930 to execute a series of instruction operations in the storage medium 930 on the server 900.

The server 900 may also include one or more power supplies 926, one or more wired or wireless network interfaces 950, one or more input-output interfaces 958, and/or one or more operating systems 941, such as Windows Server, mac OS XTM, unixTM, linuxTM, freeBSDTM, and so forth.

The steps performed by the server in the above embodiment may be based on the server structure shown in fig. 12.

The CPU 922 is configured to execute the following steps:

Another object reliability evaluation device is provided in the embodiment of the present application, as shown in fig. 13, for convenience of description, only a part related to the embodiment of the present application is shown, and details of the specific technology are not disclosed, please refer to the method part in the embodiment of the present application. The terminal may be any terminal device including a mobile phone, a tablet computer, a Personal Digital Assistant (Personal Digital Assistant, PDA for short), a Point of sale terminal (POS for short), a vehicle-mounted computer, and the like, taking the terminal as the mobile phone:

fig. 13 is a block diagram illustrating a partial structure of a mobile phone related to a terminal according to an embodiment of the present disclosure. Referring to fig. 13, the handset includes: radio Frequency (RF) circuit 1010, memory 1020, input unit 1030, display unit 1040, sensor 1050, audio circuit 1060, wireless fidelity (WiFi) module 1070, processor 1080, and power source 1090. Those skilled in the art will appreciate that the handset configuration shown in fig. 13 is not intended to be limiting and may include more or fewer components than those shown, or some components may be combined, or a different arrangement of components.

The following describes each component of the mobile phone in detail with reference to fig. 13:

RF circuit 1010 may be used for receiving and transmitting signals during information transmission and reception or during a call, and in particular, for processing downlink information of a base station after receiving the downlink information to processor 1080; in addition, data for designing uplink is transmitted to the base station. In general, RF circuit 1010 includes, but is not limited to, an antenna, at least one Amplifier, a transceiver, a coupler, a Low Noise Amplifier (Low Noise Amplifier; LNA), a duplexer, and the like. In addition, the RF circuitry 1010 may also communicate with networks and other devices via wireless communications. The wireless communication may use any communication standard or protocol, including but not limited to Global System for Mobile communication (GSM), general Packet Radio Service (GPRS), code Division Multiple Access (CDMA), wideband Code Division Multiple Access (WCDMA), long Term Evolution (LTE), email), short Message Service (SMS), etc.

The memory 1020 can be used for storing software programs and modules, and the processor 1080 executes various functional applications and data processing of the mobile phone by operating the software programs and modules stored in the memory 1020. The memory 1020 may mainly include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application program required for at least one function (such as a sound playing function, an image playing function, etc.), and the like; the storage data area may store data (such as audio data, a phonebook, etc.) created according to the use of the cellular phone, etc. Further, the memory 1020 may include high speed random access memory and may also include non-volatile memory, such as at least one magnetic disk storage device, flash memory device, or other volatile solid state storage device.

The input unit 1030 may be used to receive input numeric or character information and generate key signal inputs related to user settings and function control of the cellular phone. Specifically, the input unit 1030 may include a touch panel 1031 and other input devices 1032. The touch panel 1031, also called a touch screen, may collect a touch operation performed by a user on or near the touch panel 1031 (e.g., an operation performed by a user on or near the touch panel 1031 using any suitable object or accessory such as a finger, a stylus, etc.) and drive the corresponding connection device according to a predetermined program. Alternatively, the touch panel 1031 may include two parts, a touch detection device and a touch controller. The touch detection device detects the touch direction of a user, detects a signal brought by touch operation and transmits the signal to the touch controller; the touch controller receives touch information from the touch sensing device, converts the touch information into touch point coordinates, sends the touch point coordinates to the processor 1080, and can receive and execute commands sent by the processor 1080. In addition, the touch panel 1031 may be implemented by various types such as a resistive type, a capacitive type, an infrared ray, and a surface acoustic wave. The input unit 1030 may include other input devices 1032 in addition to the touch panel 1031. In particular, other input devices 1032 may include, but are not limited to, one or more of a physical keyboard, function keys (e.g., volume control keys, switch keys, etc.), a track ball, a mouse, a joystick, and the like.

The display unit 1040 may be used to display information input by a user or information provided to the user and various menus of the cellular phone. The Display unit 1040 may include a Display panel 1041, and optionally, the Display panel 1041 may be configured by using a Liquid Crystal Display (LCD), an Organic Light-Emitting Diode (OLED), or the like. Further, the touch panel 1031 can cover the display panel 1041, and when the touch panel 1031 detects a touch operation on or near the touch panel 1031, the touch operation is transmitted to the processor 1080 to determine the type of the touch event, and then the processor 1080 provides a corresponding visual output on the display panel 1041 according to the type of the touch event. Although in fig. 13, the touch panel 1031 and the display panel 1041 are two separate components to implement the input and output functions of the mobile phone, in some embodiments, the touch panel 1031 and the display panel 1041 may be integrated to implement the input and output functions of the mobile phone.

The handset may also include at least one sensor 1050, such as a light sensor, motion sensor, and other sensors. Specifically, the light sensor may include an ambient light sensor and a proximity sensor, wherein the ambient light sensor may adjust the brightness of the display panel 1041 according to the brightness of ambient light, and the proximity sensor may turn off the display panel 1041 and/or the backlight when the mobile phone moves to the ear. As one of the motion sensors, the accelerometer sensor can detect the magnitude of acceleration in each direction (generally, three axes), can detect the magnitude and direction of gravity when stationary, and can be used for applications of recognizing the gesture of the mobile phone (such as horizontal and vertical screen switching, related games, magnetometer gesture calibration), vibration recognition related functions (such as pedometer and tapping), and the like; as for other sensors such as a gyroscope, a barometer, a hygrometer, a thermometer, and an infrared sensor, which can be configured on the mobile phone, further description is omitted here.

Audio circuitry 1060, speaker 1061, microphone 1062 may provide an audio interface between the user and the handset. The audio circuit 1060 can transmit the electrical signal converted from the received audio data to the speaker 1061, and the electrical signal is converted into a sound signal by the speaker 1061 and output; on the other hand, the microphone 1062 converts the collected sound signal into an electrical signal, which is received by the audio circuit 1060 and converted into audio data, which is then processed by the audio data output processor 1080 and then sent to, for example, another cellular phone via the RF circuit 1010, or output to the memory 1020 for further processing.

WiFi belongs to short-distance wireless transmission technology, and the mobile phone can help the user to send and receive e-mail, browse web pages, access streaming media, etc. through the WiFi module 1070, which provides wireless broadband internet access for the user. Although fig. 13 shows the WiFi module 1070, it is understood that it does not belong to the essential constitution of the handset, and may be omitted entirely as needed within the scope not changing the essence of the invention.

The processor 1080 is a control center of the mobile phone, connects various parts of the whole mobile phone by using various interfaces and lines, and executes various functions of the mobile phone and processes data by operating or executing software programs and/or modules stored in the memory 1020 and calling data stored in the memory 1020, thereby integrally monitoring the mobile phone. Optionally, processor 1080 may include one or more processing units; preferably, the processor 1080 may integrate an application processor, which primarily handles operating systems, user interfaces, application programs, etc., and a modem processor, which primarily handles wireless communications. It is to be appreciated that the modem processor described above may not be integrated into processor 1080.

The handset also includes a power supply 1090 (e.g., a battery) for powering the various components, which may preferably be logically coupled to the processor 1080 via a power management system that may be used to manage charging, discharging, and power consumption.

Although not shown, the mobile phone may further include a camera, a bluetooth module, etc., which will not be described herein.

In the embodiment of the present application, the processor 1080 included in the terminal further has the following functions:

obtaining the characteristics of the associated object of the object to be evaluated; the characteristics comprise historical troubleshooting characteristics and/or transaction characteristics; the associated object and the object to be evaluated are positioned in a target platform;

performing feature aggregation on the object to be evaluated and the associated object based on the type of the association relationship between the object to be evaluated and the associated object to obtain aggregated features of the object to be evaluated;

An embodiment of the present application further provides a computer-readable storage medium, configured to store a program code, where the program code is configured to execute any one implementation of the method for evaluating object credibility described in the foregoing embodiments.

The present application further provides a computer program product including instructions, which when run on a computer, causes the computer to execute any one of the embodiments of the object credibility assessment method described in the foregoing embodiments.

It can be clearly understood by those skilled in the art that, for convenience and simplicity of description, the specific working processes of the above-described systems, apparatuses and units may refer to the corresponding processes in the foregoing method embodiments, and are not described herein again.

In the several embodiments provided in the present application, it should be understood that the disclosed system, apparatus and method may be implemented in other ways. For example, the above-described apparatus embodiments are merely illustrative, and for example, the division of the units is only one type of logical functional division, and other divisions may be realized in practice, for example, multiple units or components may be combined or integrated into another system, or some features may be omitted, or not executed. In addition, the shown or discussed mutual coupling or direct coupling or communication connection may be an indirect coupling or communication connection through some interfaces, devices or units, and may be in an electrical, mechanical or other form.

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

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

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

The above embodiments are only used for illustrating the technical solutions of the present application, and not for limiting the same; although the present application has been described in detail with reference to the foregoing embodiments, it should be understood by those of ordinary skill in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some technical features may be equivalently replaced; and such modifications or substitutions do not depart from the spirit and scope of the corresponding technical solutions in the embodiments of the present application.

Claims

1. An object credibility assessment method is characterized by comprising the following steps:

2. The method of claim 1, wherein obtaining the characteristics of the associated object of the object to be evaluated comprises:

determining the object to be evaluated; the residence time of the associated object is earlier than that of the object to be evaluated;

extracting a heterogeneous subgraph corresponding to the object to be evaluated from the heterogeneous graph of the object to be resident of the target platform; representing the objects by nodes in the heterogeneous graph of the embedded objects, recording the characteristics of the objects by the nodes, representing the incidence relation among the nodes by edges, wherein the types of the edges correspond to the types of the incidence relation; the heterogeneous subgraph comprises object nodes to be evaluated, associated object nodes and edges among the nodes in the heterogeneous subgraph;

backtracking the characteristics of the nodes in the heterogeneous subgraph to the residence time of the object to be evaluated;

the feature aggregation is performed on the object to be evaluated and the associated object based on the association relationship type of the object to be evaluated and the associated object, so as to obtain the aggregated features of the object to be evaluated, and the method includes:

and performing feature aggregation on the object node to be evaluated and the associated object node in the heterogeneous subgraph based on the type of the edge in the heterogeneous subgraph through a reliability evaluation model to obtain the aggregated feature of the object node to be evaluated.

3. The method of claim 2, wherein the obtaining the aggregated features of the object nodes to be evaluated by performing feature aggregation on the object nodes to be evaluated and the associated object nodes in the heterogeneous subgraph based on the types of the edges in the heterogeneous subgraph through a credibility assessment model comprises:

identifying the heterogeneous subgraph as a plurality of meta-paths by different types of edges; the meta-path comprises nodes with the same type of edges;

performing feature aggregation between nodes in the meta-path to obtain a first aggregated feature of the nodes in the meta-path;

and for the same node, respectively carrying out feature aggregation among the meta-paths based on the first aggregated features aggregated in different meta-paths to obtain a second aggregated feature of the same node.

4. The method according to claim 3, wherein the performing feature aggregation between nodes in the meta-path to obtain a first aggregated feature of a node in the meta-path comprises:

obtaining a normalized first aggregation weight of the target neighbor node for the target node based on characteristics of the target node, characteristics of a target neighbor node, and characteristics of a non-target neighbor node in the meta-path; the target node is any node in the meta-path, the target neighbor node is any neighbor node connected with the target node through an edge in the meta-path, and the non-target neighbor node is other neighbor nodes of the target node except the target neighbor node in the meta-path;

performing linear transformation on the characteristics of the target node by using the normalized first aggregation weight of the target node by the target neighbor node in the meta-path to obtain a first linear transformation result of the characteristics of the target node by the target neighbor node;

accumulating first linear transformation results of each neighbor node of the target node in the meta-path to the target node to obtain an accumulation result;

and carrying out nonlinear transformation on the accumulated result through a nonlinear activation function to obtain a first post-aggregation characteristic of the target node in the meta-path.

5. The method according to claim 3, wherein the performing, for the same node, feature aggregation between meta-paths based on first aggregated features aggregated in different meta-paths respectively to obtain second aggregated features of the same node comprises:

transforming the first aggregated features of the nodes in the meta-path into a weight scalar;

obtaining an average value of the weight scalars of all nodes in the meta-path as a second aggregation weight;

obtaining a normalized second aggregation weight of the target meta-path according to the second aggregation weight of the target meta-path and second aggregation weights of other meta-paths except the target meta-path; the target meta-path is any one meta-path in a plurality of meta-paths of the heterogeneous subgraph;

performing linear transformation on the first aggregated features of the nodes in the target element path by using the normalized second aggregation weight of the target element path to obtain a second linear transformation result of the target element path on the features of the nodes;

and accumulating the second linear transformation results of the characteristics of the same node by different element paths to obtain a second aggregated characteristic of the same node.

6. The method of claim 2, further comprising:

determining a target order N, wherein the target order N is a positive integer greater than 1;

the extracting of the heterogeneous subgraph corresponding to the object to be evaluated from the heterogeneous graph of the resident object of the target platform comprises the following steps:

extracting N-order heterogeneous subgraphs corresponding to the object to be evaluated from the heterogeneous graph of the object to be resided of the target platform according to the target order N; the associated object nodes in the N-order heterogeneous subgraph comprise associated object nodes with an order less than or equal to N.

7. The method of claim 6, wherein the credibility assessment model comprises N layers of heterogeneous graph neural networks, operating layer by layer from layer 1 to layer N of heterogeneous graph neural networks;

the obtaining the aggregated features of the object nodes to be evaluated by performing feature aggregation on the object nodes to be evaluated and the associated object nodes in the heterogeneous subgraph through the credibility evaluation model based on the types of the edges in the heterogeneous subgraph comprises the following steps:

performing feature aggregation on the basis of the types of edges between object nodes to be evaluated and associated object nodes from 1 order to N-m +1 order in the N-order heterogeneous subgraph through an m-th layer heterogeneous graph neural network to update the N-order heterogeneous subgraph; m is an integer of 1 to N-1;

and performing feature aggregation on the object nodes to be evaluated through the N-th layer heterogeneous graph neural network based on the types of edges between the object nodes to be evaluated and the 1-th order associated object nodes in the N-order heterogeneous subgraph updated by the N-1-th layer heterogeneous graph neural network to obtain the aggregated features of the object nodes to be evaluated through the N-th layer heterogeneous graph neural network.

8. The method of claim 2, further comprising, before the feature aggregating, by the credibility assessment model, the object node to be assessed and the associated object node in the heterogeneous subgraph based on the type of the edge in the heterogeneous subgraph, further comprising:

adding a self-loop edge to nodes in the heterogeneous subgraph, and giving the type of each edge in the heterogeneous subgraph to the self-loop edge;

and converting the edges in the heterogeneous subgraph into two directed edges to obtain the processed heterogeneous subgraph.

9. The method of claim 8, further comprising, after the converting the edges in the heterogeneous subgraph into two directed edges:

and deleting the directed edge in the heterogeneous subgraph by taking the node of the object to be evaluated as the start.

10. The method of any of claims 2-9, wherein the credibility assessment model is trained by:

extracting heterogeneous subgraphs corresponding to the sample objects from the heterogeneous graph of the embedded object sample; representing the objects by nodes in the heterogeneous graph of the embedded object sample, recording the characteristics of the objects by the nodes, representing the incidence relation among the nodes by edges, wherein the types of the edges correspond to the types of the incidence relation; the heterogeneous subgraph corresponding to the sample object comprises sample object nodes, associated object nodes of the sample object nodes and edges among the nodes in the heterogeneous subgraph corresponding to the sample object; the residence time of the associated object of the sample object is earlier than the residence time of the sample object;

performing feature aggregation on the sample object nodes and the associated object nodes of the sample object nodes in the heterogeneous subgraph corresponding to the sample object through a model to be trained based on the type of edges in the heterogeneous subgraph corresponding to the sample object to obtain aggregated features of the sample object nodes;

carrying out credibility evaluation on the sample object according to the aggregated characteristics of the sample object nodes to obtain a credibility evaluation result;

and adjusting the coefficients of the heterogeneous graph neural network in the model to be trained according to the difference between the reliability evaluation result of the sample object and the reliability label of the sample object until a training cut-off condition is met, and obtaining the reliability evaluation model.

11. The method according to any one of claims 1 to 9, wherein the object to be evaluated and the associated object are both merchants, and the association relationship between the object to be evaluated and the associated object is specifically the same registration information, and the registration information includes any one of the following types:

contact mobile phone number, legal person ID card, bank card number, enterprise unified credit code, shareholder representative, merchant full name or contact mailbox.

12. The method according to any one of claims 1-9, wherein the historical review features comprise positive review result features or negative review result features, and the transaction features comprise positive transaction result features or negative transaction result features; both the negative vetting result characteristic and the negative transaction result characteristic point to be untrustworthy.

13. An object credibility assessment apparatus, comprising:

the characteristic acquisition unit is used for acquiring the characteristics of the associated object of the object to be evaluated; the characteristics comprise historical troubleshooting characteristics and/or transaction characteristics; the associated object and the object to be evaluated are resident in a target platform;

the feature aggregation unit is used for performing feature aggregation on the object to be evaluated and the associated object based on the type of the association relationship between the object to be evaluated and the associated object to obtain the aggregated features of the object to be evaluated;

and the object credibility evaluation unit is used for evaluating the credibility of the object to be evaluated according to the aggregated features.

14. A computer device, the device comprising a processor and a memory:

the memory is used for storing program codes and transmitting the program codes to the processor;

the processor is configured to execute the subject credibility assessment method of any one of claims 1-12 according to instructions in the program code.

15. A computer-readable storage medium for storing program code for performing the subject credibility assessment method of any one of claims 1-12.