A method, apparatus and related products for evaluating the credibility of an object.

Abstract

This application discloses a method, apparatus, and related products for evaluating object credibility. This application relates to the field of machine learning technology. The method first obtains the characteristics of the associated objects of the object to be evaluated, including historical investigation characteristics and / or transaction characteristics. Then, based on the type of association between the object to be evaluated and its associated objects, feature aggregation is performed on the object to be evaluated and its associated objects to obtain aggregated features of the object to be evaluated. In this solution, different types of association relationships can have different influences on evaluating object credibility, achieving the distinction and effective utilization of different types of association relationships between objects. Therefore, unlike existing technologies that indiscriminately apply association relationships between objects, this application evaluates the credibility of the object to be evaluated based on its aggregated features, resulting in more accurate object credibility evaluation results.

Description

Technical Field

[0001] This application relates to the field of credibility assessment technology, and in particular to a method, apparatus and related products for assessing the credibility of an object. Background Technology

[0002] With the rise of e-commerce, more and more businesses are choosing to join e-commerce platforms and sell their goods. To ensure the healthy operation of these platforms, it is necessary to conduct credibility assessments on the businesses they target.

[0003] In existing technologies, credibility assessment typically utilizes a target's negative records. To more effectively mitigate the adverse effects of low-credibility targets, it's necessary to assess the credibility of newly added targets during the onboarding phase. Since newly added targets initiate fewer transactions and often lack negative records, credibility assessment is difficult. One solution is to assess the credibility level of a new target by examining the negative records of existing targets associated with it. However, this approach only considers the existence of relationships between targets, failing to differentiate and effectively utilize different types of relationships, resulting in insufficient accuracy in the credibility assessment results. Summary of the Invention

[0004] This application provides an object credibility assessment method, apparatus, and related products to improve the accuracy of object credibility assessment.

[0005] In view of this, the first aspect of this application provides a method for evaluating the credibility of an object, the method comprising:

[0006] Obtain the characteristics of the related objects of the object to be evaluated; the characteristics include historical investigation characteristics and / or transaction characteristics; the related objects and the object to be evaluated are registered on the target platform;

[0007] Based on the type of association between the object to be evaluated and the associated objects, feature aggregation is performed on the object to be evaluated and the associated objects to obtain the aggregated features of the object to be evaluated.

[0008] The credibility of the object to be evaluated is assessed based on the aggregated features.

[0009] A second aspect of this application provides an object credibility assessment apparatus, the apparatus comprising:

[0010] The feature acquisition unit is used to obtain the features of the related objects of the object to be evaluated; the features include historical investigation features and / or transaction features; the related objects and the object to be evaluated are registered on the target platform;

[0011] The feature aggregation unit is used to aggregate features of the object to be evaluated and its associated objects based on the type of association between the object to be evaluated and its associated objects, and obtain the aggregated features of the object to be evaluated.

[0012] The object credibility assessment unit is used to assess the credibility of the object to be assessed based on the aggregated features.

[0013] A third aspect of this application provides an object credibility assessment device, the device including a processor and a memory:

[0014] The memory is used to store program code and transfer the program code to the processor;

[0015] The processor is used to execute the steps of the object trustworthiness assessment method provided in the first aspect above, according to the instructions in the program code.

[0016] The fourth aspect of this application provides a computer-readable storage medium for storing program code for performing the object trustworthiness assessment method provided in the first aspect above.

[0017] As can be seen from the above technical solutions, the embodiments of this application have the following advantages:

[0018] This application provides a method for assessing the credibility of an object. The method first obtains the characteristics of the associated objects of the object to be assessed, including historical investigation characteristics and / or transaction characteristics. If the associated objects have negative records, this can be reflected through historical investigation characteristics and / or transaction characteristics. Next, based on the type of association between the object to be assessed and the associated objects, feature aggregation is performed on the object to be assessed and the associated objects to obtain the aggregated characteristics of the object to be assessed. In this solution, different types of association relationships can have different influences on assessing the credibility of an object. For example, key types of association relationships play a more important role in feature aggregation, thereby achieving the differentiation and effective utilization of different types of association relationships between objects. Therefore, unlike existing technologies that indiscriminately apply association relationships between objects, this application assesses the credibility of the object to be assessed based on the aggregated characteristics of the object to be assessed, resulting in a more accurate object credibility assessment. Attached Figure Description

[0019] Figure 1 A flowchart illustrating an object credibility assessment method provided in this application embodiment;

[0020] Figure 2 A flowchart illustrating another object credibility assessment method provided in this application embodiment;

[0021] Figure 3 The heterosubgraph corresponding to the object to be evaluated is extracted when the target order is 2;

[0022] Figure 4 In response to Figure 3 A schematic diagram of the formation of the processed heteroproton diagram;

[0023] Figure 5 A schematic diagram for identifying multiple metapaths from a heterosubgraph;

[0024] Figure 6 A flowchart illustrating node-level aggregation based on features;

[0025] Figure 7 A flowchart illustrating the semantic-level aggregation of features;

[0026] Figure 8 This is a schematic diagram illustrating node-level and semantic-level aggregation of node features;

[0027] Figure 9 A schematic diagram of nodes participating in the aggregation of node features in a heterogeneous graph layer by layer in a neural network.

[0028] Figure 10 A schematic diagram for training a credibility assessment model;

[0029] Figure 11 A schematic diagram of the structure of an object credibility assessment device provided in an embodiment of this application;

[0030] Figure 12 A schematic diagram of the structure of a server for evaluating the trustworthiness of an object, provided as an embodiment of this application;

[0031] Figure 13 This is a schematic diagram of the structure of a terminal device for evaluating the credibility of an object, provided as an embodiment of this application. Detailed Implementation

[0032] As discussed earlier, existing technologies, upon learning of relationships between objects, directly assess object credibility based on these relationships. Because different types of relationships are not differentiated during the assessment, the types of relationships are not effectively utilized in evaluating object credibility. For example, the relationship between a new entrant and an existing entrant is often not singular; new entrant A and existing entrant B may share the same contact email address, and new entrant A and existing entrant C may share the same unified corporate credit code. When existing technologies use homogeneous graph neural networks to assess the credibility of new entrants, the homogeneous graph only has one type of node and relationship, making it impossible to distinguish and effectively utilize different types of relationships between new entrants and existing entrants. These solutions treat different types of relationships "equally." For example, they assume that existing entrants B and C have equal influence on assessing the credibility of new entrant A. This leads to insufficient accuracy in the credibility assessment results.

[0033] To address the above issues, this application provides a novel object credibility assessment method, apparatus, and related products to improve the accuracy of object credibility assessment results. The technical solution of this application utilizes the specific relationship types between the object to be assessed and its associated objects, aggregating object features based on the type of relationship. Finally, the credibility of the object to be assessed is evaluated based on the aggregated features. This solution uses the type of relationship between objects as the basis for aggregating object features, representing a novel strategy for assessing object credibility. Since different types of relationships may have varying impacts on credibility, this solution, through differentiated application of object relationship types, can more accurately assess the credibility level of objects.

[0034] The object credibility assessment method mentioned above can be applied to processing devices that have object credibility assessment capabilities, such as terminal devices or servers. This method can be executed independently by the terminal device or server, or it can be applied to network scenarios where the terminal device and server communicate, operating in cooperation. The terminal device can be a mobile phone, desktop computer, personal digital assistant (PDA), tablet computer, etc. The server can be understood as an application server or a web server. In actual deployment, the server can be a standalone physical server, a server cluster consisting of multiple physical servers, or a distributed system. The terminal and server can be directly or indirectly connected via wired or wireless communication; this application does not impose any restrictions on this connection.

[0035] Furthermore, this application also relates to Artificial Intelligence (AI) technology. Artificial intelligence is the theory, methods, technology, and application systems that utilize digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to obtain optimal results. In other words, artificial intelligence is a comprehensive technology within computer science that attempts to understand the essence of intelligence and produce a new kind of intelligent machine capable of reacting in a manner similar to human intelligence. Artificial intelligence also studies the design principles and implementation methods of various intelligent machines, enabling them to possess the functions of perception, reasoning, and decision-making.

[0036] Artificial intelligence (AI) is a comprehensive discipline encompassing a wide range of fields, including both hardware and software technologies. Fundamental AI technologies generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing, operating / interactive systems, and mechatronics. AI software technologies primarily include computer vision, speech processing, natural language processing, as well as machine learning / deep learning, autonomous driving, and intelligent transportation.

[0037] Natural Language Processing (NLP) is an important field within computer science and artificial intelligence. It studies the theories and methods for enabling effective communication between humans and computers using natural language. NLP is a science that integrates linguistics, computer science, and mathematics. Therefore, research in this field involves natural language—the language people use in daily life—and thus it has a close relationship with linguistic research. NLP techniques typically include text processing, semantic understanding, machine translation, question answering, and knowledge graphs.

[0038] Machine Learning (ML) is a multidisciplinary field involving probability theory, statistics, approximation theory, convex analysis, and algorithm complexity theory. It specifically studies how computers can simulate or implement human learning behavior to acquire new knowledge or skills and reorganize existing knowledge structures to continuously improve their performance. Machine learning is the core of artificial intelligence and the fundamental way to endow computers with intelligence; its applications span all areas of artificial intelligence. Machine learning and deep learning typically include techniques such as artificial neural networks, belief networks, reinforcement learning, transfer learning, inductive learning, and instructional learning.

[0039] In the embodiments of this application, the processing device can perform feature aggregation using the above-mentioned natural language processing, machine learning, and other technologies. For ease of understanding, the object credibility evaluation method provided in the embodiments of this application will be described below with reference to the accompanying drawings.

[0040] Figure 1 This is a flowchart illustrating an object credibility assessment method provided in an embodiment of this application. Figure 1 As shown, the method includes:

[0041] S101: Obtain the characteristics of the related objects of the object to be evaluated; the characteristics include historical investigation characteristics and / or transaction characteristics.

[0042] The related object and the object to be evaluated are registered on the same platform, referred to here as the target platform. The related object of the object to be evaluated can be determined through registration information. When an object registers on the platform, it typically provides its registration information. In some possible implementation scenarios, such as financial transactions, the object referred to in this solution can be a merchant. If the merchant is untrustworthy, it is highly likely to pose risks to transactions on the platform, such as damage to the buyer's economic rights. When both the related 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 person's mobile phone number, legal representative's ID card, bank card number, unified corporate credit code, shareholder representative, full name of the merchant, or contact email address. For example, when an object shares the same type of registration information as the object to be evaluated, it can be determined that there is a relationship of this type between the two objects. Therefore, as an example, the relationship between the object to be evaluated and the related object can specifically be reflected in the same registration information.

[0043] For example, if the contact phone number of an object is the same as the contact phone number of the object to be evaluated, then it can be determined that the object is an associated object of the object to be evaluated. The type of association between the two is: contact phone number.

[0044] For example, if the unified corporate credit code of an object is the same as that of the object to be evaluated, then it can be determined that the object is a related object of the object to be evaluated. The type of relationship between the two is: unified corporate credit code.

[0045] For entities joining a target platform, they possess certain characteristics, including but not limited to transaction characteristics and historical screening characteristics. For newly joined entities, due to their short time on the platform and minimal or no transaction volume, their transaction characteristics may be empty. Furthermore, if no screening has been conducted on the entity, its historical screening characteristics may also be empty.

[0046] Historical screening characteristics include both positive and negative screening results, while transaction characteristics include both positive and negative transaction results. Both negative screening and negative transaction results indicate untrustworthiness. For example, historical screening characteristics can include blacklisting records and tags. For instance, if an object has been blacklisted by the platform, its historical screening characteristics are negative. Conversely, if an object has never been blacklisted or tagged, its historical screening characteristics are positive. Transaction characteristics can include the gender ratio of users, average age, and number of blocked transactions. Taking the number of blocked transactions as an example, blocking indicates that a transaction failed. For transactions actively blocked by the platform, this may be due to the object's credit level or the number of complaints. Therefore, in one optional implementation, if the number of blocked transactions exceeds a preset number, this transaction characteristic is a negative transaction result; conversely, if the number of blocked transactions does not exceed the preset number, this transaction characteristic is a positive transaction result. Of course, the preset number can be set according to actual needs; no numerical limit is imposed here.

[0047] Historical screening characteristics and transaction characteristics can be reflected not only positively or negatively, but also represented by intuitive numerical values. The above introduction is only an example of historical screening characteristics and transaction characteristics. The characteristics of an object are not limited to historical screening characteristics and transaction characteristics. For example, they can also include the object's business registration information, such as the size of the registered company, registered capital, and market establishment details.

[0048] S102: Based on the type of association between the object to be evaluated and the associated objects, perform feature aggregation on the object to be evaluated and the associated objects to obtain the aggregated features of the object to be evaluated.

[0049] In this embodiment, S102 specifically refers to using the relationship type between the object to be evaluated and related objects as the basis for feature aggregation, guiding the specific execution of feature aggregation. In S102, feature aggregation is performed by relying on the relationship type, giving feature aggregation a certain strategic dimension.

[0050] The following describes an optional implementation of this step using heterogeneous graphs. As mentioned earlier, homogeneous graphs have only one type of node and one type of relationship, making it impossible to distinguish and effectively utilize different relationship types. Heterogeneous graphs, however, support different relationship types between nodes. Therefore, this embodiment cleverly utilizes this characteristic of heterogeneous graphs to aggregate object features. In a heterogeneous graph, objects are nodes, and the relationships between objects are edges. The feature vector of each node represents the features of the object. In a heterogeneous graph, it can first be divided into multiple meta-paths according to the type of edges between nodes, i.e., the relationship type between objects. Node-level aggregation is performed within each meta-path, and semantic-level aggregation is performed on different meta-paths. Finally, the semantic-level aggregation result representing the node features of the object to be evaluated is used as the aggregated feature of the object to be evaluated. The following sections will provide a detailed explanation of the feature aggregation of nodes in heterogeneous graphs with reference to embodiments.

[0051] S103: Conduct a credibility assessment of the object to be evaluated based on the aggregated features.

[0052] The aggregated features of the object to be evaluated are obtained by aggregating features based on the types of relationships between objects. This takes into account that different types of relationships can have varying impacts on the credibility of the evaluated object; for example, key types of relationships play a more significant role in feature aggregation, thus enabling the differentiation and effective utilization of different types of relationships between objects. Therefore, unlike existing technologies that indiscriminately apply relationships between objects, this application evaluates the credibility of the object to be evaluated based on its aggregated features. This approach more effectively considers the value of features from objects with different relationships in the credibility evaluation, resulting in a more accurate credibility assessment.

[0053] The following is combined Figure 2 The flowchart shown illustrates the implementation method of node feature aggregation based on heterogeneous graphs. Figure 2 A flowchart illustrating another object credibility assessment method provided in this application embodiment. Figure 2 The methods shown include:

[0054] S201: Identify the object to be evaluated.

[0055] Whenever a new object joins the target platform, the platform adds it as a new node to its heterogeneous graph of joined objects. The edges between new and old nodes are constructed based on their relationships. In other words, the heterogeneous graph represents objects as nodes, records the object's characteristics, and represents the relationships between nodes as edges, with the edge type corresponding to the type of the relationship. For example, if a new node and an old node share a common contact phone number, then an edge of type "contact phone number" is constructed between the new node and the old node.

[0056] Before evaluating the object to be evaluated, the object must first be identified. As an example, nodes representing the object to be evaluated can be identified in the heterogeneous graph of the objects based on all the registration information of the object to be evaluated (such as contact person's mobile phone number, legal representative's ID card, bank card number, unified credit code of the enterprise, shareholder representative, full name of the merchant and contact email), and the time of entry on the target platform.

[0057] S202: Extract the heterogeneous subgraph corresponding to the object to be evaluated from the heterogeneous graph of the objects on the target platform.

[0058] To improve the efficiency of evaluating the object under evaluation and reduce the interference of non-related objects in assessing the credibility of the object under evaluation, this step extracts the heterogeneous graph corresponding to the object under evaluation from the entire heterogeneous graph of the inbound objects. For ease of distinction, this heterogeneous graph is referred to as a heterogeneous subgraph. The heterogeneous subgraph includes the node of the object under evaluation (i.e., the node representing the object under evaluation), the nodes of related objects (i.e., the nodes representing related objects), and the edges between the nodes in the heterogeneous subgraph. Both the related objects and the object under evaluation are registered on the target platform, and the registration time of the related objects is earlier than that of the object under evaluation. When evaluating the credibility of a newly registered object, the characteristics of related objects whose registration time is earlier than that of the new object are usable. Therefore, related objects of the object under evaluation can be screened based on their registration time. Other objects whose registration time is the same as or later than that of the object under evaluation are not considered.

[0059] The heterogeneous subgraph can be selected based on the target order N. Here, the target order N represents the maximum order of the associated object nodes expected to be used in assessing the credibility of the object to be evaluated. Specifically, this step can involve extracting the N-order heterogeneous subgraph corresponding to the object to be evaluated from the heterogeneous graph of the objects on the target platform, based on the target order N; the associated object nodes in the N-order heterogeneous subgraph include associated object nodes with an order less than or equal to N.

[0060] For example, N=1 indicates that the heterogeneous subgraph needs to contain first-order related object nodes; N=2 indicates that the heterogeneous subgraph needs to contain both first-order and second-order related object nodes; N=3 indicates that the heterogeneous subgraph needs to contain first-order, second-order, and third-order related object nodes. Here, a first-order related object node is a node in the heterogeneous subgraph that is directly connected to the node to be evaluated via an edge; a second-order related object node is a node other than the node to be evaluated that is directly connected to a first-order related object node via an edge; a third-order related object node is a node other than a first-order related object node that is directly connected to a second-order related object node via an edge, and so on.

[0061] Figure 3 This is the heterosubgraph corresponding to the object to be evaluated extracted when the target order is 2. For example... Figure 3 As shown, nodes marked with 0 are the nodes to be evaluated, nodes marked with 1 are first-order related nodes, and nodes marked with 2 are second-order related nodes. Figure 3 The heterogeneous subgraph also shows three different types of edges through different line drawing methods, including mobile phone numbers, bank cards, and ID cards.

[0062] S203: Backtrack the features of nodes in the heterogeneous subgraph to the entry time of the object to be evaluated.

[0063] The technical solution provided in this application mainly assesses the credibility of later-entered objects by evaluating the features of earlier-entered related objects. The extraction of the heterogeneous subgraph may occur later than the entry time of the object to be evaluated, resulting in some features of nodes in the heterogeneous subgraph being established after the entry time of the object to be evaluated. To ensure the accuracy of the evaluation prediction, this step backtracks the features of the remaining related objects according to the entry time of the object to be evaluated. That is, the features of each related object need to be traced back to the entry time of the object to be evaluated on the target platform, eliminating features generated after the entry time of the object to be evaluated.

[0064] In this embodiment, a credibility assessment model is pre-trained. When feature aggregation is required, the credibility assessment model aggregates features of the node to be evaluated and its associated nodes in the heterogeneous subgraph based on the edge type, obtaining the aggregated features of the node to be evaluated. The credibility assessment model includes a heterogeneous graph neural network. The training method of the credibility assessment model will be described in detail later; the application of the credibility assessment model is first introduced through S204~S206.

[0065] To ensure the credibility evaluation model can properly process the input heterogeneous subgraph, in one optional implementation, the heterogeneous subgraph can be processed before executing S204 to convert the undirected graph (i.e., heterogeneous subgraph) into a directed graph, facilitating processing by the heterogeneous graph neural network. The specific processing method is as follows:

[0066] 1) Add self-loop edges to nodes in the heterogeneous subgraph. By adding self-loop edges, each convolution can obtain information about the node itself.

[0067] 2) Assign a type to each edge in the heterogeneous subgraph to self-loop edges. This means adding a self-loop to each association of every node. Edges between nodes have only a specific type, while self-loop edges have every possible type. Figure 4 In response to Figure 3 A schematic diagram of the processed heteroproton diagram. (e.g.) Figure 4 As shown, a loop-shaped self-loop edge has been added to each node.

[0068] 3) Convert the edges in the heterogeneous subgraph into two directed edges. Figure 3 It's easy to see that the edges between nodes have no direction. And in contrast... Figure 4 Then, the edges between nodes will have arrows pointing to them, indicating that undirected edges have been transformed into directed edges.

[0069] 4) Delete directed edges in the heterogeneous subgraph that originate from the node to be evaluated. For example... Figure 4 As shown, by deleting the directed edges that start from the node to be evaluated, only the directional arrows pointing to the node to be evaluated remain for the edges constructed from the nodes to be evaluated.

[0070] Since the nodes to be evaluated do not contain attribute information, such as transaction features or historical investigation features, filling them with either all-zero features or mean features will inevitably affect the information extraction and importance coefficient calculation of neighboring nodes (i.e., first-order related object nodes). Therefore, we directly change the structure here by deleting all edges starting from the node to be evaluated, except for the self-loop edges, to ensure that the influence of the node to be evaluated can be ignored when updating the features of related object nodes.

[0071] By performing steps 1) to 4) above, the processed heteroproton map is obtained. The heteroproton map processed by S204 in the following text can specifically be the heteroproton map processed through the above steps.

[0072] S204: Identify heterogeneous subgraphs as multiple metapaths by different edge types.

[0073] In this embodiment, the type of edge is used to identify the meta path. For example, a heterosubgraph contains three types of edges, where the first type of edge is used to determine the first meta path, the second type of edge is used to determine the second meta path, and the third type of edge is used to determine the third meta path. Figure 5 This is a schematic diagram for identifying multiple metapaths from a heterosubgraph.

[0074] Based on the multiple meta-paths identified in S204, the next step is to aggregate node features. In S205 and S206, which are described later, "node" is not a specific term; the following operations are performed on each node in the heterogeneous subgraph. S205 introduces node-level feature aggregation, which occurs within a meta-path; S206 introduces semantic-level feature aggregation, which occurs between meta-paths.

[0075] S205: Perform feature aggregation between nodes in the metapath to obtain the first aggregated features of the nodes in the metapath.

[0076] Figure 6 A flowchart illustrating node-level aggregation based on features. See also... Figure 6 The node-level feature aggregation process includes:

[0077] S2051: Based on the characteristics of the target node, the characteristics of the target neighbor nodes, and the characteristics of the non-target neighbor nodes in the metapath, obtain the normalized first aggregation weight of the target neighbor nodes relative to the target node.

[0078] To describe this more accurately and clearly, and for ease of understanding, the concept of a target node is used here. A target node is any node in the metapath, and a target neighbor node is any neighbor node in the metapath to which the target node belongs that is connected to the target node through an edge. Non-target neighbor nodes are any other neighbor nodes of the target node in the metapath to which the target node belongs, excluding the target neighbor nodes.

[0079] Assuming for the metapath This includes nodes such as node i and node j, where node i is considered the target node, and node j is considered a target neighbor node of target node i. The feature vectors of node i and node j are respectively represented as... and The first aggregate weight of target neighbor node j with respect to target node i. , representing the importance of node j's features to node i. This can be understood as the first aggregation weight. The higher the value, the more important the features of node j are to node i. To make it easier to compare the first aggregate weights of different neighboring nodes for the target node, the first aggregate weights need to be normalized in this step. The implementation of this step is described below with reference to formula (1).

[0080] Formula (1)

[0081] As can be seen from formula (1), the first aggregation weight can be adjusted using the softmax function in this step. Normalize. The first aggregation weight is the normalized weight. This indicates that the features of the target node and its neighboring nodes should be concatenated. Metapath The corresponding linear transformation matrix, The linear transformation matrix The inverse matrix. Features that can be spliced Transform to a suitable metapath In the form of. () represents the activation function. Metapath The set of all neighboring nodes of the target node i, including node j and other non-target nodes. k represents the set... Any neighboring node in the array. Let be the feature vector of node k.

[0082] S2052: Use the normalized first aggregation weights of the target neighbor nodes in the metapath to perform a linear transformation on the features of the target node, and obtain the first linear transformation result of the target neighbor nodes on the features of the target node.

[0083] Based on the example above, the first linear transformation result of the features of target neighbor node j on target node i can be expressed as: .

[0084] S2053: Accumulate the first linear transformation results of each neighboring node of the target node in the metapath to obtain the accumulated result.

[0085] The cumulative result is represented as: .in Metapath The set of all neighboring nodes of the target node i.

[0086] S2054: The first aggregated feature of the target node in the metapath is obtained by performing a nonlinear transformation on the accumulated result through a nonlinear activation function.

[0087] Metapath obtained through nonlinear transformation The first aggregated feature of target node i The expression is as follows:

[0088] Formula (2)

[0089] The first aggregated feature of the target node in the metapath is the result of node-level aggregation of the target node in the metapath. For each node in each metapath, the above operations S2051~S2054 can be performed to obtain the node-level aggregation result of its features.

[0090] After the feature node-level aggregation of S2051~S2054 above, we will now proceed to the semantic-level aggregation of features.

[0091] S206: For the same node, perform feature aggregation between meta-paths based on the first aggregated features aggregated from different meta-paths to obtain the second aggregated features of the same node.

[0092] Figure 7 This is a flowchart illustrating the semantic-level aggregation of features. For example... Figure 7 As shown, the semantic-level aggregation process includes:

[0093] S2061: Transform the first aggregated feature of a node in the metapath into a weight scalar.

[0094] S2062: Obtain the average of the weight scalars of all nodes in the metapath as the second aggregation weight.

[0095] Formula (3)

[0096] In formula (3), This represents the weight scalar obtained from the first aggregation and feature transformation of nodes in the meta-path. Wherein, Metapath The first aggregated feature of target node i , Indicates the intercept. This represents the linear transformation matrix of the weights. In heterogeneous graphical neural networks, the vector used to convert a vector into a scalar is represented as a vector. Its reverse is written as .

[0097] Formula (3) is the metapath The expression is used to average the weight scalars of all nodes. Metapath The set of all nodes This represents the total number of nodes in the set. Second aggregation weight. This represents the metapath for node i during feature semantic-level aggregation. The corresponding weights.

[0098] S2063: Obtain the normalized second aggregation weight of the target meta-path based on the second aggregation weight of the target meta-path and the second aggregation weights of other meta-paths besides the target meta-path.

[0099] The target meta-path is any one of the multiple meta-paths in a heterosubgraph, assuming... This represents the target meta-path. To facilitate semantic-level aggregation of node features, the second aggregation weights obtained in S2062 also need to be adjusted. Normalization is performed. The second aggregation weights are then normalized. The probability is expressed as follows:

[0100] Formula (4)

[0101] As can be seen from formula (4), this normalization operation is achieved through the softmax function. In formula (4), This represents the total number of heteroproton primitive paths. This indicates that feature semantic-level aggregation is performed on target node i, and the target meta-path is... The second aggregation weight. In 1 to When the values ​​in are different, then These represent the second aggregation weights for feature semantic-level aggregation of target node i and other meta-paths besides the target meta-path, respectively.

[0102] S2064: Using the normalized second aggregation weights of the target meta-path, perform a linear transformation on the first aggregated features of the nodes in the target meta-path to obtain the second linear transformation result of the target meta-path on the features of the nodes.

[0103] Using the target metapath as the metapath Taking node i as an example, the second linear transformation result of the target metapath with respect to the features of the node is expressed as: .in Metapath The first aggregated feature of node i Indicates the target metapath The normalized second aggregation weight.

[0104] S2065: The second linear transformation results of the features of the same node by different meta-paths are accumulated to obtain the second aggregated features of the same node.

[0105] Taking node i as an example, for node i, its second aggregated feature The expression is as follows:

[0106] Formula (5)

[0107] In formula (5), This represents the total number of heteroproton primitive paths.

[0108] The second aggregated feature of a node is the feature obtained after node-level aggregation within the metapath and semantic-level aggregation between metapaths. Compared to the feature before node-level aggregation, it is equivalent to updating the original feature with the second aggregated feature of each node.

[0109] Figure 8 This is a schematic diagram illustrating node-level and semantic-level aggregation of node features. For example... Figure 8 As shown, the heterosubgraph is divided into three meta-paths. Node-level aggregation is performed within each meta-path to obtain the first aggregated features. Figure 8 Z represents the node. After node-level aggregation, semantic-level aggregation between meta-paths is performed on the node features. Specifically, normalized second aggregation weights are used to process the features after the first aggregation. Figure 8 China and Israel This indicates that the second aggregated feature was finally obtained, completing one round of aggregation of node features by the model.

[0110] In the above-described embodiments, a single node was used as an example to illustrate the entire process of feature aggregation. If the target order is 1, the above aggregation operation only needs to be performed on the node to be evaluated, because the node to be evaluated only has 1st-order associated node. However, if the target order is an integer greater than 1, the aggregation of nodes needs to be implemented layer by layer through a multi-layer heterogeneous graph neural network of the credibility evaluation model.

[0111] For example, if the target order is 2, the number of layers in the heterogeneous graph neural network of the credibility assessment model matches the target order, meaning the model includes two layers of heterogeneous graph neural networks. In the first layer of the heterogeneous graph neural network, feature updates are needed for the node to be evaluated and the first-order related node (i.e., updating the original features using features obtained through node-level aggregation and semantic-level aggregation). In the second layer of the heterogeneous graph neural network, only the feature updates are needed for the node to be evaluated.

[0112] To illustrate further, let's consider a target order of 3, meaning the heterogeneous subgraph contains associated object nodes of orders 1, 2, and 3. In the first layer of the heterogeneous graph neural network, feature updates are needed for the node to be evaluated, the 1st-order associated object nodes, and the 2nd-order associated object nodes. In the second layer, feature updates are needed for the node to be evaluated and the 1st-order associated object nodes. In the third layer, only the node to be evaluated needs feature updates.

[0113] As illustrated in the example above, the order of the updated nodes gradually decreases during processing at each layer of the network. The following section describes the implementation process of the entire credibility evaluation model in processing the aggregated features of the nodes to be evaluated:

[0114] For the m-th layer of the heterogeneous graph neural network (where m is an integer from 1 to N-1), the m-th layer performs feature aggregation based on the types of edges between the evaluation node and the associated object nodes of orders 1 to N-m+1 in the N-order heterogeneous subgraph to update the N-order heterogeneous subgraph. In other words, the features of the associated object nodes of orders 1 to N-m+1 are used during feature aggregation in the m-th layer. The scope of node feature updates for the N-order heterogeneous subgraph can be limited to the evaluation node and the associated object nodes of orders 1 to Nm, because the features of the associated object nodes of orders N-m+1 are no longer used in the next layer of the network.

[0115] The Nth layer of the heterogeneous graph neural network aggregates features of the node to be evaluated based on the types of edges between the node to be evaluated and the first-order associated node in the Nth-order heterogeneous subgraph updated by the (N-1)th layer of the heterogeneous graph neural network. This results in the aggregated features of the node to be evaluated after passing through the Nth layer of the heterogeneous graph neural network. In other words, the features of the first-order associated node are used during feature aggregation in the Nth layer of the heterogeneous graph neural network. In the final layer, the update of the feature aggregation results can focus solely on the node to be evaluated.

[0116] Figure 9 This is a schematic diagram illustrating the layer-by-layer participation of a heterogeneous graph neural network in the aggregation of node features for a second-order heterogeneous subgraph. (Example:) Figure 9 As shown, in the first layer of the heterogeneous graph neural network, the features of second-order related object nodes affect the update of the features of first-order related object nodes. However, in the second layer of the heterogeneous graph neural network, only the features of first-order related object nodes affect the update of the features of the object node to be evaluated.

[0117] S207: Conduct a credibility assessment of the object to be evaluated based on the aggregated features.

[0118] By the end of S206, the aggregated features of the nodes to be evaluated are a fixed-dimensional feature vector. In one optional implementation, a multi-layer fully connected network can be used to obtain the probability value of a node being a risky object (i.e., an object with low credibility) based on its aggregated features. This probability value characterizes the risk level of the object to be evaluated; a higher probability indicates a higher risk level and lower credibility, while a lower probability indicates a lower risk level and higher credibility.

[0119] The preceding section introduced how a credibility assessment model can be used to assess the credibility of an object. The following section, with accompanying diagrams, describes the training process of this credibility assessment model. Figure 10 This is a schematic diagram for training a credibility assessment model. (For example...) Figure 10 As shown, the first step is to obtain the heterosubgraphs corresponding to the sample objects. Sample objects refer to the objects whose reliability needs to be evaluated during training. The data for the sample objects and their heterosubgraphs can be obtained from historical data.

[0120] In this embodiment, the platform where the sample objects are hosted constructs a heterogeneous graph of the hosted objects. Similar to the heterogeneous graph of hosted objects described above, it lists the features of the objects hosted on the platform as nodes in the heterogeneous graph. In the heterogeneous graph of hosted objects, nodes represent objects, nodes record the features of the objects, and edges represent the relationships between nodes, with the edge type corresponding to the type of relationship.

[0121] To train the model, it is necessary to extract the heterogeneous subgraphs corresponding to the sample objects from the heterogeneous graph of the inbound object samples. The heterogeneous subgraph corresponding to the sample object includes the sample object node, the associated object nodes of the sample object node, and the edges between the nodes in the heterogeneous subgraph corresponding to the sample object; the inbound time of the associated objects of the sample object is earlier than the inbound time of the sample object.

[0122] The training model aggregates features of the sample object nodes and their associated object nodes in the heterogeneous subgraph corresponding to the sample object based on the edge type of the heterogeneous subgraph, obtaining aggregated features of the sample object nodes. Specifically, the number of layers in the heterogeneous graph neural network in the training model matches the maximum order of the associated object nodes in the heterogeneous subgraph corresponding to the sample object.

[0123] Next, the credibility of the sample object is evaluated based on the aggregated features of the sample object nodes, and the credibility evaluation result is obtained. For example, the probability of the sample object being an untrustworthy object (risk object) is obtained.

[0124] Each sample object has a credibility tag. Taking a merchant as an example, this credibility tag indicates the likelihood that the sample merchant will become a blacklisted merchant (a merchant with very high risk, hindering normal transactions on the platform, such as having serious credit or service problems). For example, a credibility tag of 1 indicates that it is a blacklisted merchant, while a credibility tag of 0 indicates that it is a whitelisted merchant (without any characteristics indicating untrustworthiness).

[0125] Since the model has already obtained the credibility evaluation results during the training process, the coefficients of the heterogeneous graph neural network in the model to be trained can be adjusted according to the gap between the credibility evaluation results of the sample objects and the credibility labels of the sample objects, until the training cutoff condition is met and the credibility evaluation model is obtained.

[0126] The training cutoff condition here can be either reaching a preset threshold for the number of model training iterations, or the loss value being less than a preset value. Specifically, the loss function can be a function of the difference between the credibility assessment result of the sample object and the credibility label of the sample object. A higher loss value indicates a larger difference.

[0127] By applying the object credibility assessment method provided in this application, when a new object joins and submits basic information, the associated heterogeneous subgraphs and feature vectors of associated objects in the subgraphs can be quickly obtained from the graph based on this basic information. This information is then used to perform predictions through a trained model to obtain the credibility assessment result. In one optional implementation, objects with low credibility in the credibility assessment result can be rejected from successfully joining the platform.

[0128] During the onboarding phase, the available information is less compared to the transaction phase, making it difficult to develop effective strategies based on information beyond the penalty records of associated entities. Furthermore, relying solely on penalty records makes it difficult to cover the large number of untrusted entities associated with already onboarded entities at the time of onboarding, thus hindering timely and effective control over the onboarding of untrusted entities.

[0129] This solution constructs a heterogeneous graph of new entrants, extracts heterogeneous subgraphs of these new entrants from the graph, trains a heterogeneous graph neural network model, and extracts features associated with objects through different paths. It then uses transaction information, penalty information, and company information from multi-level associated objects to comprehensively assess the credibility of new entrants, ultimately aiding in decision-making regarding penalties or investigations. This solution effectively aggregates information from heterogeneous subgraphs using multi-stage association information, distinguishes different association relationships, and automatically assigns weights to different relationships. Compared to methods that typically convert data into structured data, this solution improves the accuracy of credibility assessment, ultimately enabling more precise detection of malicious registrations, saving on manual review costs during entrant registration, and providing early prevention of credibility issues.

[0130] Based on the object credibility assessment method provided in the foregoing embodiments, this application also provides an object credibility assessment device. The implementation of the object credibility assessment device will be described below with reference to embodiments and accompanying drawings.

[0131] Figure 11 This is a schematic diagram of the structure of an object credibility assessment device provided in the application embodiment. Figure 11 The object credibility assessment device 110 shown includes:

[0132] The feature acquisition unit 111 is used to obtain the features of the related objects of the object to be evaluated; the features include historical investigation features and / or transaction features; the related objects and the object to be evaluated are registered on the target platform;

[0133] Feature aggregation unit 112 is used to perform feature aggregation on the object to be evaluated and the associated objects based on the type of the association relationship between the object to be evaluated and the associated objects, so as to obtain the aggregated features of the object to be evaluated.

[0134] The object credibility assessment unit 113 is used to assess the credibility of the object to be assessed based on the aggregated features.

[0135] Optionally, the feature acquisition unit 111 includes:

[0136] The unit for determining the object to be evaluated is used to determine the object to be evaluated; the related objects and the object to be evaluated are registered on the target platform, and the registration time of the related objects is earlier than that of the object to be evaluated;

[0137] The heterogeneous subgraph extraction unit is used to extract the heterogeneous subgraph corresponding to the object to be evaluated from the heterogeneous graph of the inbound objects of the target platform. In the heterogeneous graph of the inbound objects, objects are represented by nodes, which record the characteristics of the objects, and the relationships between nodes are represented by edges, with the type of the edge corresponding to the type of the relationship. The heterogeneous subgraph includes the node of the object to be evaluated, the nodes of the associated objects, and the edges between the nodes in the heterogeneous subgraph.

[0138] The feature backtracking unit is used to backtrack the features of nodes in the heterogeneous subgraph to the entry time of the object to be evaluated.

[0139] The feature aggregation unit 112 is specifically used 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 using the credibility evaluation model, so as to obtain the aggregated features of the object node to be evaluated.

[0140] Optionally, the feature aggregation unit 112 includes:

[0141] The meta-path identification unit is used to identify multiple meta-paths in a heterogeneous subgraph based on the different types of edges; a meta-path includes nodes with edges of the same type.

[0142] A node-level feature aggregation unit is used to perform feature aggregation between nodes in the metapath to obtain the first aggregated features of the nodes in the metapath.

[0143] The semantic-level feature aggregation unit is used to perform feature aggregation between meta-paths on the same node based on the first aggregated features aggregated from different meta-paths, and obtain the second aggregated features of the same node.

[0144] Optionally, the node-level feature aggregation unit includes:

[0145] The first normalization subunit is used to obtain the first aggregation weight of the target neighbor node on the target node based on the characteristics of the target node, the characteristics of the target neighbor node, and the characteristics of the non-target neighbor node in the metapath; the target node is any node in the metapath, the target neighbor node is any neighbor node in the metapath connected to the target node through an edge, and the non-target neighbor node is any other neighbor node of the target node in the metapath other than the target neighbor node.

[0146] The first transformation subunit is used to perform a linear transformation on the features of the target node using the normalized first aggregation weights of the target neighbor nodes in the metapath, so as to obtain the first linear transformation result of the target neighbor nodes on the features of the target node.

[0147] The first accumulation unit is used to accumulate the first linear transformation results of each neighbor node of the target node in the metapath to the target node, and obtain the accumulation result.

[0148] The second transformation subunit is used to perform a nonlinear transformation on the accumulated result through a nonlinear activation function to obtain the first aggregated feature of the target node in the metapath.

[0149] Optionally, the semantic-level feature aggregation unit includes:

[0150] The third transformation subunit is used to transform the first aggregated feature of the node in the metapath into a weight scalar.

[0151] The average calculation subunit is used to obtain the average of the weight scalars of all nodes in the metapath as the second aggregation weight;

[0152] The second normalization subunit is used to obtain the normalized second aggregation weight of the target meta-path based on the second aggregation weight of the target meta-path and the second aggregation weight of other meta-paths besides the target meta-path; the target meta-path is any meta-path among multiple meta-paths of the heterogeneous subgraph.

[0153] The fourth transformation subunit is used to perform a linear transformation on the first aggregated features of the nodes in the target meta-path using the normalized second aggregation weights of the target meta-path, so as to obtain the second linear transformation result of the node features of the target meta-path.

[0154] The second accumulation subunit is used to accumulate the second linear transformation results of the features of the same node from different meta-paths to obtain the second aggregated features of the same node.

[0155] Optionally, the object credibility assessment device 110 may further include:

[0156] The order determination unit is used to determine the target order N, where the target order N is a positive integer greater than 1;

[0157] The heteroprotic subgraph extraction unit is specifically used to extract the N-order heteroprotic subgraph corresponding to the object to be evaluated from the heteroprotic subgraph of the objects on the target platform, based on the target order N; the associated object nodes in the N-order heteroprotic subgraph include associated object nodes with an order less than or equal to N.

[0158] Optionally, the credibility assessment model includes an N-layer heterogeneous graph neural network, which is computed layer by layer from the first layer to the Nth layer.

[0159] Feature aggregation unit 112 is used for:

[0160] The m-th layer heterogeneous graph neural network performs feature aggregation based on the type of edges between the object node to be evaluated and the associated object nodes of order 1 to N-m+1 in the N-order heterogeneous subgraph, so as to update the N-order heterogeneous subgraph; m is an integer from 1 to N-1.

[0161] The Nth layer heterogeneous graph neural network is used to aggregate the features of the object node to be evaluated and the first-order associated object node in the Nth-order heterogeneous subgraph updated by the (N-1)th layer heterogeneous graph neural network. This process is based on the type of edge between the object node to be evaluated and the first-order associated object node in the Nth-order heterogeneous subgraph. The aggregated features of the object node to be evaluated are obtained by the Nth layer heterogeneous graph neural network.

[0162] Optionally, the object credibility assessment device 110 further includes:

[0163] The heterogeneous subgraph processing unit is used to add self-loop edges to the nodes in the heterogeneous subgraph before performing feature aggregation on the nodes to be evaluated and the associated nodes in the heterogeneous subgraph based on the edge type in the heterogeneous subgraph using the credibility evaluation model. The self-loop edges are assigned the type of each edge in the heterogeneous subgraph. The edges in the heterogeneous subgraph are converted into two directed edges to obtain the processed heterogeneous subgraph.

[0164] Optionally, the heteroprotic processing unit is also used to delete directed edges in the heteroprotic graph that originate from the node to be evaluated.

[0165] Optionally, the credibility assessment model is trained using a model training unit in the object credibility assessment device 110. The model training unit includes:

[0166] The training graph acquisition unit is used to extract the heterogeneous subgraph corresponding to the sample object from the heterogeneous graph of the inbound object samples. In the heterogeneous graph of the inbound object samples, objects are represented by nodes, each node records the characteristics of the object, and edges represent the relationships between nodes, with the type of the edge corresponding to the type of the relationship. The heterogeneous subgraph corresponding to the sample object includes the sample object node, the associated object nodes of the sample object node, and the edges between the nodes in the heterogeneous subgraph corresponding to the sample object. The inbound time of the associated object of the sample object is earlier than the inbound time of the sample object.

[0167] The model processing unit is used to perform feature aggregation on the sample object nodes and their associated object nodes in the heterogeneous subgraph corresponding to the sample object based on the edge type in the heterogeneous subgraph corresponding to the sample object, and obtain the aggregated features of the sample object nodes.

[0168] The evaluation unit is used to evaluate the credibility of a sample object based on the aggregated features of the sample object nodes and obtain the credibility evaluation result.

[0169] The coefficient adjustment unit is used to adjust the coefficients of the heterogeneous graph neural network in the model to be trained based on the gap between the credibility assessment result of the sample object and the credibility label of the sample object, until the training cutoff condition is met and the credibility assessment model is obtained.

[0170] Figure 12 This is a schematic diagram of a server structure provided in an embodiment of this application. The server 900 can vary significantly due to different configurations or performance. It may include one or more central processing units (CPUs) 922 (e.g., one or more processors) and memory 932, and one or more storage media 930 (e.g., one or more mass storage devices) for storing application programs 942 or data 944. The memory 932 and storage media 930 can be temporary or persistent storage. The program stored in the storage media 930 may include one or more modules (not shown in the diagram), each module may include a series of instruction operations on the server. Furthermore, the CPU 922 may be configured to communicate with the storage media 930 and execute the series of instruction operations in the storage media 930 on the server 900.

[0171] 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 X™, Unix™, Linux™, FreeBSD™, etc.

[0172] The steps performed by the server in the above embodiments can be based on this Figure 12 The server structure shown.

[0173] CPU 922 is used to perform the following steps:

[0174] Obtain the characteristics of the related objects of the object to be evaluated; the characteristics include historical investigation characteristics and / or transaction characteristics; the related objects and the object to be evaluated are registered on the target platform;

[0175] Based on the type of association between the object to be evaluated and the associated objects, feature aggregation is performed on the object to be evaluated and the associated objects to obtain the aggregated features of the object to be evaluated.

[0176] The credibility of the object to be evaluated is assessed based on the aggregated features.

[0177] This application also provides another object credibility assessment device, such as... Figure 13 As shown, for ease of explanation, only the parts related to the embodiments of this application are shown. For specific technical details not disclosed, please refer to the method section of the embodiments of this application. The terminal can be any terminal device including mobile phones, tablets, personal digital assistants (PDAs), point-of-sale (POS) terminals, in-vehicle computers, etc. Taking a mobile phone as an example:

[0178] Figure 13 This is a block diagram illustrating a portion of the structure of a mobile phone related to the terminal provided in the embodiments of this application. (Reference) Figure 13 The mobile phone includes: a radio frequency (RF) circuit 1010, a memory 1020, an input unit 1030, a display unit 1040, a sensor 1050, an audio circuit 1060, a wireless fidelity (WiFi) module 1070, a processor 1080, and a power supply 1090, etc. Those skilled in the art will understand that... Figure 13The mobile phone structure shown does not constitute a limitation on the mobile phone and may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0179] The following is combined Figure 13 A detailed introduction to each component of a mobile phone:

[0180] The RF circuit 1010 can be used for receiving and transmitting signals during information transmission or calls. Specifically, it receives downlink information from the base station and processes it with the processor 1080; additionally, it transmits uplink data to the base station. Typically, the RF circuit 1010 includes, but is not limited to, an antenna, at least one amplifier, a transceiver, a coupler, a low-noise amplifier (LNA), a duplexer, etc. Furthermore, the RF circuit 1010 can also communicate wirelessly with networks and other devices. The aforementioned wireless communications may use any communication standard or protocol, including but not limited to Global System for Mobile Communications (GSM), General Packet Radio Service (GPRS), Code Division Multiple Access (CDMA), Wideband Code Division Multiple Access (WCDMA), Long Term Evolution (LTE), email, Short Messaging Service (SMS), etc.

[0181] The memory 1020 can be used to store software programs and modules. The processor 1080 executes various mobile phone functions and data processing by running 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. The program storage area may store the operating system, applications required for at least one function (such as sound playback function, image playback function, etc.), etc.; the data storage area may store data created according to the use of the mobile phone (such as audio data, phonebook, etc.). In addition, the memory 1020 may include high-speed random access memory, and may also include non-volatile memory, such as at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0182] The input unit 1030 can be used to receive input numerical or character information, and to generate key signal inputs related to user settings and function control of the mobile phone. Specifically, the input unit 1030 may include a touch panel 1031 and other input devices 1032. The touch panel 1031, also known as a touch screen, can collect touch operations performed by the user on or near it (such as operations performed by the user using a finger, stylus, or any suitable object or accessory on or near the touch panel 1031), and drive the corresponding connection devices according to a pre-set program. Optionally, the touch panel 1031 may include two parts: a touch detection device and a touch controller. The touch detection device detects the user's touch position and the signal generated by the touch operation, and transmits the signal to the touch controller; the touch controller receives touch information from the touch detection device, converts it into touch point coordinates, and sends it to the processor 1080, and can also receive and execute commands sent by the processor 1080. In addition, the touch panel 1031 can be implemented using various types such as resistive, capacitive, infrared, and surface acoustic wave. In addition to the touch panel 1031, the input unit 1030 may also include other input devices 1032. Specifically, other input devices 1032 may include, but are not limited to, one or more of the following: physical keyboard, function keys (such as volume control buttons, power buttons, etc.), trackball, mouse, joystick, etc.

[0183] The display unit 1040 can be used to display information input by the user or information provided to the user, as well as various menus of the mobile phone. The display unit 1040 may include a display panel 1041, which may optionally be configured as a Liquid Crystal Display (LCD), Organic Light-Emitting Diode (OLED), or similar display panel 1041. Furthermore, a touch panel 1031 may cover the display panel 1041. When the touch panel 1031 detects a touch operation on or near it, it transmits the information to the processor 1080 to determine the type of touch event. Subsequently, the processor 1080 provides corresponding visual output on the display panel 1041 based on the type of touch event. Although in Figure 13 In this embodiment, the touch panel 1031 and the display panel 1041 are two separate components to realize the input and output functions of the mobile phone. However, in some embodiments, the touch panel 1031 and the display panel 1041 can be integrated to realize the input and output functions of the mobile phone.

[0184] The mobile phone may also include at least one sensor 1050, such as a light sensor, a motion sensor, and other sensors. Specifically, the light sensor may include an ambient light sensor and a proximity sensor. The ambient light sensor can adjust the brightness of the display panel 1041 according to the ambient light level, and the proximity sensor can turn off the display panel 1041 and / or backlight when the phone is moved to the ear. As a type of motion sensor, an accelerometer sensor can detect the magnitude of acceleration in various directions (generally three axes). When stationary, it can detect the magnitude and direction of gravity and can be used for applications that recognize the phone's posture (such as landscape / portrait switching, related games, magnetometer posture calibration), vibration recognition-related functions (such as pedometer, taps), etc. Other sensors that may be configured in the mobile phone, such as gyroscopes, barometers, hygrometers, thermometers, and infrared sensors, will not be described in detail here.

[0185] The audio circuit 1060, speaker 1061, and microphone 1062 provide an audio interface between the user and the mobile phone. The audio circuit 1060 converts the received audio data into electrical signals and transmits them to the speaker 1061, where the speaker 1061 converts them into sound signals for output. On the other hand, the microphone 1062 converts the collected sound signals into electrical signals, which are then received by the audio circuit 1060, converted into audio data, and then processed by the processor 1080 before being transmitted via the RF circuit 1010 to, for example, another mobile phone, or the audio data can be output to the memory 1020 for further processing.

[0186] WiFi is a short-range wireless transmission technology. Through the WiFi module 1070, mobile phones can help users send and receive emails, browse web pages, and access streaming media, providing users with wireless broadband internet access. Although Figure 13 The WiFi module 1070 is shown, but it is understood that it is not an essential component of a mobile phone and can be omitted as needed without changing the essence of the invention.

[0187] The processor 1080 is the control center of the mobile phone, connecting various parts of the phone through various interfaces and lines. It executes various functions and processes data by running or executing software programs and / or modules stored in the memory 1020 and calling data stored in the memory 1020. Optionally, the processor 1080 may include one or more processing units; preferably, the processor 1080 may integrate an application processor and a modem processor, wherein the application processor mainly handles the operating system, user interface, and applications, and the modem processor mainly handles wireless communication. It is understood that the modem processor may not be integrated into the processor 1080.

[0188] The mobile phone also includes a power supply 1090 (such as a battery) that supplies power to various components. Preferably, the power supply can be logically connected to the processor 1080 through a power management system, thereby enabling functions such as charging, discharging, and power consumption management through the power management system.

[0189] Although not shown, mobile phones may also include a camera, Bluetooth module, etc., which will not be described in detail here.

[0190] In this embodiment of the application, the processor 1080 included in the terminal also has the following functions:

[0191] Obtain the characteristics of the related objects of the object to be evaluated; the characteristics include historical investigation characteristics and / or transaction characteristics; the related objects and the object to be evaluated are registered on the target platform;

[0192] Based on the type of association between the object to be evaluated and the associated objects, feature aggregation is performed on the object to be evaluated and the associated objects to obtain the aggregated features of the object to be evaluated.

[0193] The credibility of the object to be evaluated is assessed based on the aggregated features.

[0194] This application also provides a computer-readable storage medium for storing program code that executes any one of the implementation methods of the object credibility assessment method described in the foregoing embodiments.

[0195] This application also provides a computer program product including instructions that, when run on a computer, cause the computer to execute any one of the implementation methods of the object credibility assessment method described in the foregoing embodiments.

[0196] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0197] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be an indirect coupling or communication connection between apparatuses or units through some interfaces, and may be electrical, mechanical, or other forms.

[0198] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0199] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0200] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes: USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, optical disks, and other media capable of storing program code.

[0201] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A method for evaluating the credibility of an object, characterized in that, include: Identify the object to be evaluated; the associated objects of the object to be evaluated and the object to be evaluated are registered on the same target platform, and the registration time of the associated objects is earlier than the registration time of the object to be evaluated. The association between the associated object and the object to be evaluated is that they have the same type of registration information on the target platform; The type of the association relationship is the type of the registration information; the object to be evaluated is a new entrant to the target platform. The heterogeneous subgraph corresponding to the object to be evaluated is extracted from the heterogeneous graph of the objects on the target platform. In the heterogeneous graph of the objects on the platform, objects are represented by nodes, each node records the characteristics of the object, and edges represent the relationships between nodes, with the edge type corresponding to the type of relationship. The heterogeneous subgraph includes the object to be evaluated node, associated object nodes, and the edges between the nodes in the heterogeneous subgraph. The object to be evaluated node in the heterogeneous subgraph is connected by edges of different types. The characteristics of the associated objects include historical investigation characteristics and / or transaction characteristics. The features of the nodes in the heterogeneous subgraph are traced back to the entry time of the object to be evaluated, and features generated after the entry time of the object to be evaluated are removed. Add self-loop edges to the nodes in the heterogeneous subgraph, and assign the self-loop edges the type of each edge in the heterogeneous subgraph; The edges in the heterosubgraph are converted into two directed edges to obtain the processed heterosubgraph; Delete the directed edges in the heterogeneous subgraph that originate from the loop edges and start from the node to be evaluated. Based on the type of edges in the heterogeneous subgraph, the credibility assessment model performs feature aggregation on the node to be evaluated and the associated node in the heterogeneous subgraph to obtain the aggregated features of the node to be evaluated. The credibility of the object to be evaluated is assessed based on the aggregated features.

2. The method according to claim 1, characterized in that, The step of using a credibility assessment model to perform feature aggregation on the object node to be evaluated and the associated object node in the heterogeneous subgraph based on the edge type in the heterogeneous subgraph to obtain the aggregated features of the object node to be evaluated includes: The heterogeneous subgraph is identified as multiple meta-paths based on the different types of edges; each meta-path includes nodes with edges of the same type. In the metapath, feature aggregation is performed between nodes to obtain the first aggregated features of the nodes in the metapath; For the same node, feature aggregation between meta-paths is performed based on the first aggregated features aggregated from different meta-paths to obtain the second aggregated features of the same node.

3. The method according to claim 2, characterized in that, The step of performing feature aggregation between nodes in the metapath to obtain the first aggregated features of the nodes in the metapath includes: In the metapath, based on the features of the target node, the features of the target neighbor nodes, and the features of the non-target neighbor nodes, a normalized first aggregation weight of the target neighbor nodes on the target node is obtained; the target node is any node in the metapath, the target neighbor node is any neighbor node in the metapath connected to the target node through an edge, and the non-target neighbor node is any other neighbor node of the target node in the metapath other than the target neighbor node. The features of the target node are linearly transformed using the normalized first aggregation weights of the target neighbor nodes in the metapath to obtain the first linear transformation result of the features of the target node by the target neighbor nodes. The first linear transformation results of each neighbor node of the target node in the metapath to the target node are summed to obtain the summed result; The accumulated result is nonlinearly transformed by a nonlinear activation function to obtain the first aggregated feature of the target node in the metapath.

4. The method according to claim 2, characterized in that, The step of performing feature aggregation between meta-paths based on the first aggregated features aggregated from different meta-paths for the same node, to obtain the second aggregated features for the same node, includes: The first aggregated feature of the node in the meta-path is transformed into a weight scalar; The average of the weight scalars of all nodes in the meta-path is obtained as the second aggregation weight; The normalized second aggregation weight of the target meta-path is obtained based on the second aggregation weight of the target meta-path and the second aggregation weights of other meta-paths besides the target meta-path; the target meta-path is any meta-path among the multiple meta-paths of the heterogeneous subgraph. The first aggregated features of the nodes in the target meta-path are linearly transformed using the normalized second aggregation weight of the target meta-path to obtain the second linear transformation result of the node features of the target meta-path. The second linear transformation results of the features of the same node from different meta-paths are accumulated to obtain the second aggregated features of the same node.

5. The method according to claim 1, characterized in that, Also includes: Determine the target order N, where the target order N is a positive integer greater than 1; The step of extracting the heterogeneous subgraph corresponding to the object to be evaluated from the heterogeneous graph of the inbound objects of the target platform includes: Based on the target order N, extract the N-order heterogeneous subgraph corresponding to the object to be evaluated from the heterogeneous graph of the inbound objects of the target platform; The associated object nodes in the N-order heteroprotic subgraph include associated object nodes with an order less than or equal to N.

6. The method according to claim 5, characterized in that, The credibility assessment model includes an N-layer heterogeneous graph neural network, which is operated layer by layer from the 1st layer to the Nth layer. The step of using a credibility assessment model to perform feature aggregation on the object node to be evaluated and the associated object node in the heterogeneous subgraph based on the edge type in the heterogeneous subgraph to obtain the aggregated features of the object node to be evaluated includes: The Nth-order heterogeneous subgraph is updated by performing feature aggregation based on the type of edges between the object node to be evaluated and the associated object nodes of order 1 to N-m+1 in the Nth-order heterogeneous subgraph using the m-th layer heterogeneous graph neural network; m is an integer from 1 to N-1. The Nth layer heterogeneous graph neural network aggregates the features of the object node to be evaluated by using the type of edge between the object node to be evaluated and the first-order associated object node in the Nth-order heterogeneous subgraph updated by the (N-1)th layer heterogeneous graph neural network. This process yields the aggregated features of the object node to be evaluated after passing through the Nth layer heterogeneous graph neural network.

7. The method according to any one of claims 1-6, characterized in that, The credibility assessment model is trained in the following way: The heterogeneous subgraph corresponding to the sample object is extracted from the heterogeneous graph of the inbound object samples. In the heterogeneous graph of the inbound object samples, objects are represented by nodes, each node records the characteristics of the object, and edges represent the relationships between nodes, with the type of the edge corresponding to the type of the relationship. The heterogeneous subgraph corresponding to the sample object includes the sample object node, the associated object nodes of the sample object node, and the edges between the nodes in the heterogeneous subgraph corresponding to the sample object. The inbound time of the associated object of the sample object is earlier than the inbound time of the sample object. Based on the type of edges in the heterogeneous subgraph corresponding to the sample object, the training model performs feature aggregation on the sample object node and its associated object node in the heterogeneous subgraph corresponding to the sample object to obtain the aggregated features of the sample object node. The credibility of the sample object is evaluated based on the aggregated features of the sample object nodes to obtain the credibility evaluation result. Based on the gap between the credibility assessment result of the sample object and the credibility label of the sample object, the coefficients of the heterogeneous graph neural network in the model to be trained are adjusted until the training cutoff condition is met, and the credibility assessment model is obtained.

8. The method according to any one of claims 1-6, characterized in that, Both the object to be evaluated and the associated object are merchants, and the registration information includes any of the following types: Contact person's mobile phone number, legal representative's ID card, bank card number, corporate credit code, shareholder representative, and full name of the merchant or contact email address.

9. The method according to any one of claims 1-6, characterized in that, The historical screening characteristics include positive screening result characteristics or negative screening result characteristics, and the transaction characteristics include positive transaction result characteristics or negative transaction result characteristics; Both the negative investigation results and the negative transaction results characteristics point to unreliability.

10. An object credibility assessment device, characterized in that, include: The object to be evaluated is determined by the object to be evaluated; the associated object of the object to be evaluated and the object to be evaluated are registered on the same target platform, and the registration time of the associated object is earlier than the registration time of the object to be evaluated. The association between the associated object and the object to be evaluated is that they have the same type of registration information on the target platform; The type of the association relationship is the type of the registration information; the object to be evaluated is a new entrant to the target platform. The heterogeneous subgraph extraction unit is used to extract the heterogeneous subgraph corresponding to the object to be evaluated from the heterogeneous graph of the objects on the target platform. In the heterogeneous graph of the objects on the target platform, objects are represented by nodes, each node records the characteristics of the object, and edges represent the relationships between nodes, with the edge type corresponding to the type of relationship. The heterogeneous subgraph includes the object to be evaluated node, associated object nodes, and the edges between the nodes in the heterogeneous subgraph. The object to be evaluated node in the heterogeneous subgraph is connected by edges of different types. The characteristics of the associated objects include historical investigation characteristics and / or transaction characteristics. The feature backtracking unit is used to backtrack the features of the nodes in the heterogeneous subgraph to the entry time of the object to be evaluated, and to remove features generated after the entry time of the object to be evaluated. A heterogeneous subgraph processing unit is used to add self-loop edges to nodes in the heterogeneous subgraph, assign each type of edge in the heterogeneous subgraph to the self-loop edges; convert the edges in the heterogeneous subgraph into two directed edges to obtain a processed heterogeneous subgraph; and delete the directed edges in the heterogeneous subgraph that start from the node to be evaluated, excluding the self-loop edges. The feature aggregation unit is used 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 using a credibility evaluation model, so as to obtain the aggregated features of the object node to be evaluated. An object credibility assessment unit is used to assess the credibility of the object to be assessed based on the aggregated features.

11. The apparatus according to claim 10, characterized in that, The feature aggregation unit includes: A meta-path identification unit is used to identify the heterogeneous subgraph as multiple meta-paths based on the different types of edges; the meta-paths include nodes with edges of the same type; A node-level feature aggregation unit is used to perform feature aggregation between nodes in the metapath to obtain the first aggregated features of the nodes in the metapath. A semantic-level feature aggregation unit is used to perform feature aggregation between meta-paths on the same node based on the first aggregated features aggregated from different meta-paths, to obtain the second aggregated features of the same node.

12. The apparatus according to claim 11, characterized in that, The node-level feature aggregation unit includes: The first normalization subunit is used to obtain a normalized first aggregation weight of the target neighbor node to the target node based on the characteristics of the target node, the characteristics of the target neighbor node, and the characteristics of the non-target neighbor node in the metapath; the target node is any node in the metapath, the target neighbor node is any neighbor node in the metapath connected to the target node by an edge, and the non-target neighbor node is any other neighbor node of the target node in the metapath other than the target neighbor node. The first transformation subunit is used to perform a linear transformation on the features of the target node using the normalized first aggregation weight of the target neighbor node in the metapath, so as to obtain the first linear transformation result of the target neighbor node on the features of the target node. The first accumulation unit is used to accumulate the first linear transformation results of each neighbor node of the target node in the metapath to the target node, and obtain the accumulation result. The second transformation subunit is used to perform a nonlinear transformation on the accumulated result through a nonlinear activation function to obtain the first aggregated feature of the target node in the metapath.

13. The apparatus according to claim 11, characterized in that, The semantic-level feature aggregation unit includes: The third transformation subunit is used to transform the first aggregated feature of the node in the meta-path into a weight scalar. The average calculation subunit is used to obtain the average of the weight scalars of all nodes in the metapath as the second aggregation weight; The second normalization subunit is used to obtain the normalized second aggregation weight of the target meta-path based on the second aggregation weight of the target meta-path and the second aggregation weights of other meta-paths besides the target meta-path; the target meta-path is any meta-path among the multiple meta-paths of the heterogeneous subgraph. The fourth transformation subunit is used to perform a linear transformation on the first aggregated features of the nodes in the target meta-path using the normalized second aggregation weight of the target meta-path, so as to obtain the second linear transformation result of the node features of the target meta-path. The second accumulation subunit is used to accumulate the second linear transformation results of the features of the same node from different meta-paths to obtain the second aggregated features of the same node.

14. The apparatus according to claim 10, characterized in that, Also includes: An order determination unit is used to determine a target order N, wherein the target order N is a positive integer greater than 1; The heterogeneous subgraph extraction unit is specifically used to extract the N-order heterogeneous subgraph corresponding to the object to be evaluated from the heterogeneous subgraph of the inbound objects of the target platform according to the target order N; the associated object nodes in the N-order heterogeneous subgraph include associated object nodes with an order less than or equal to N.

15. The apparatus according to claim 14, characterized in that, The credibility assessment model includes an N-layer heterogeneous graph neural network, which is operated layer by layer from the 1st layer to the Nth layer. The feature aggregation unit is used for: The Nth-order heterogeneous subgraph is updated by performing feature aggregation based on the type of edges between the object node to be evaluated and the associated object nodes of order 1 to N-m+1 in the Nth-order heterogeneous subgraph using the m-th layer heterogeneous graph neural network; m is an integer from 1 to N-1. The Nth layer heterogeneous graph neural network aggregates the features of the object node to be evaluated by using the type of edge between the object node to be evaluated and the first-order associated object node in the Nth-order heterogeneous subgraph updated by the (N-1)th layer heterogeneous graph neural network. This process yields the aggregated features of the object node to be evaluated after passing through the Nth layer heterogeneous graph neural network.

16. The apparatus according to any one of claims 10-15, characterized in that, The credibility assessment model is trained through a model training unit; the model training unit includes... The training graph acquisition unit is used to extract the heterogeneous subgraph corresponding to the sample object from the heterogeneous graph of the inbound object samples. In the heterogeneous graph of the inbound object samples, objects are represented by nodes, each node records the characteristics of the object, and edges represent the relationships between nodes, with the edge type corresponding to the type of the relationship. The heterogeneous subgraph corresponding to the sample object includes the sample object node, the associated object nodes of the sample object node, and the edges between the nodes in the heterogeneous subgraph corresponding to the sample object. The inbound time of the associated object of the sample object is earlier than the inbound time of the sample object. The model processing unit is used to perform feature aggregation on the sample object node and its associated object node in the heterogeneous subgraph corresponding to the sample object based on the type of the edge in the heterogeneous subgraph corresponding to the sample object by the model to be trained, so as to obtain the aggregated features of the sample object node. An evaluation unit is used to evaluate the credibility of the sample object based on the aggregated features of the sample object nodes, and obtain a credibility evaluation result. The coefficient adjustment unit is used to adjust the coefficients of the heterogeneous graph neural network in the model to be trained based on the gap between the credibility assessment result of the sample object and the credibility label of the sample object, until the training cutoff condition is met, and the credibility assessment model is obtained.

17. The apparatus according to any one of claims 10-15, characterized in that, Both the object to be evaluated and the associated object are merchants, and the registration information includes any of the following types: Contact person's mobile phone number, legal representative's ID card, bank card number, corporate credit code, shareholder representative, and full name of the merchant or contact email address.

18. The apparatus according to any one of claims 10-15, characterized in that, The historical screening characteristics include positive screening result characteristics or negative screening result characteristics, and the transaction characteristics include positive transaction result characteristics or negative transaction result characteristics; Both the negative investigation results and the negative transaction results characteristics point to unreliability.

19. A computer device, characterized in that, The device includes a processor and a memory: The memory is used to store program code and transmit the program code to the processor; The processor is configured to execute the object credibility assessment method according to any one of claims 1-9 according to the instructions in the program code.

20. A computer-readable storage medium, characterized in that, The computer-readable storage medium is used to store program code for executing the object credibility assessment method according to any one of claims 1-9.

21. A computer program product comprising instructions, characterized in that, When it is run on a computer, it causes the computer to perform the object credibility assessment method according to any one of claims 1-9.

Images