Model training method and device, equipment and storage medium

Abstract

This application discloses a model training method, apparatus, device, and storage medium. The specific implementation scheme is as follows: The two parties involved in the exchange are determined based on the exchange data, and the feature datasets of both parties are obtained; each party is treated as a different node, and the feature dataset of the corresponding party is used as the node content of the corresponding node; the existing exchange relationships are drawn as edges connecting the corresponding nodes, resulting in a multilateral exchange relationship graph; the multilateral exchange relationship graph is used as input, and the annotation results of each node in the multilateral exchange relationship graph as an abnormal exchange node are used as output, to train an abnormal exchange node prediction model. This scheme implements a more accurate graph-based algorithm-based model training method and apparatus.

Description

Technical Field

[0001] The embodiments of this application relate to the field of big data technology, particularly to the field of data processing and deep learning technology, and especially to model training methods, apparatus, electronic devices and storage media. Background Art

[0002] Misconduct involving property online typically takes place via computer or mobile apps, is easy to operate, and spreads widely, making it a key target for crackdowns in internet governance. Combating online misconduct is a crucial part of payment platforms' risk control mechanisms, aimed at protecting the assets of those initiating such activities and maintaining the image and reputation of payment channels and businesses.

[0003] Existing identification schemes for parties exchanging property in online transactions mainly include rule-based identification methods based on statistical analysis and clustering methods based on machine learning algorithms. Rule-based identification methods based on statistical analysis have simple rule patterns and rely more on accumulated experience and risk mitigation after an event. Machine learning algorithm-based schemes typically use the characteristics of a class of objects to identify the parties involved in the exchange at the exchange level. Summary of the Invention

[0004] This application provides a model training method, apparatus, device, and storage medium, as well as a method, apparatus, device, and storage medium for generating information.

[0005] According to a first aspect of this application, a model training method is provided, comprising: determining the two parties involved in the exchange based on the exchange data, and obtaining the feature datasets of the two parties; treating each party involved in the exchange as a different node, using the feature dataset of the corresponding party as the node content of the corresponding node, and drawing the existing exchange relationships as edges connecting the corresponding nodes to obtain a multilateral exchange relationship graph; wherein, the same party in the multilateral exchange relationship graph appears only as a node; using the multilateral exchange relationship graph as input and the labeling results of each node in the multilateral exchange relationship graph being labeled as an abnormal exchange node as output, training an abnormal exchange node prediction model.

[0006] In some embodiments, the abnormal exchange node prediction model includes multiple node prediction sub-models; the abnormal exchange node prediction model is trained by taking a multilateral exchange relationship graph as input and the annotation results of each node in the multilateral exchange relationship graph being labeled as an abnormal exchange node as output, including: determining multiple multilateral exchange relationship subgraphs based on the multilateral exchange relationship graph; taking each multilateral exchange relationship subgraph as input and the annotation results of each node in the corresponding multilateral exchange relationship subgraph being labeled as an abnormal exchange node as output, and training each node prediction sub-model; and determining the abnormal exchange node prediction model based on each node prediction sub-model.

[0007] In some embodiments, determining multiple multilateral exchange relationship subgraphs based on a multilateral exchange relationship graph includes: sampling associated nodes for each node in the multilateral exchange relationship graph to generate a multilateral exchange relationship subgraph composed of each node in the multilateral exchange relationship graph and its corresponding associated nodes. The associated nodes include: each first associated node connected to the target node by an edge and each second associated node connected to each first associated node by an edge.

[0008] In some embodiments, the two parties to the exchange are the initiator and the provider of the exchange. Each party to the exchange is treated as a different node, the feature dataset of the corresponding party is treated as the node content of the corresponding node, and the existing exchange relationship is drawn as an edge connecting the corresponding nodes to obtain a multilateral exchange relationship graph. This includes: treating each initiator as a different initiator node, treating each provider as a different provider node, treating the feature dataset of the corresponding initiator and the feature dataset of the corresponding provider as the node content of the corresponding node, and drawing the existing exchange relationship as an edge connecting the corresponding initiator node and the provider node to obtain a multilateral exchange relationship graph.

[0009] In some embodiments, the abnormal exchange node prediction model is trained by taking a multilateral exchange relationship graph as input and the annotation results of each node in the multilateral exchange relationship graph being labeled as an abnormal exchange node as output.

[0010] In some embodiments, the abnormal exchange node prediction model is trained in a semi-supervised training manner. Before taking each of the two parties in the exchange as a different node, taking the feature dataset of the corresponding exchange party as the node content of the corresponding node, and drawing the existing exchange relationship as the edge connecting the corresponding node to obtain a multilateral exchange relationship graph, the method further includes: labeling the predetermined exchange party as a positive sample, and labeling other exchange parties besides the predetermined exchange party as negative samples to obtain each labeled exchange party.

[0011] In some embodiments, before obtaining the feature datasets of the two parties involved in the exchange, the method further includes: deleting the parties whose exchange value assignment information is below a threshold.

[0012] In some embodiments, the abnormal swap node prediction model and / or node prediction sub-model are constructed based on a deep heterogeneous graph convolutional network.

[0013] In some embodiments, the feature datasets of the exchanging parties are generated by processing the exchange data based on the historical information of each party. The historical information includes at least: historical exchange information, historical behavior information, and initiator profile.

[0014] According to a second aspect of this application, a method for generating information is provided, comprising: determining the two parties involved in the exchange based on exchange data, and obtaining feature datasets of the two parties; treating each party involved in the exchange as a different node, using the feature dataset of the corresponding party as the node content of the corresponding node, and drawing the existing exchange relationships as edges connecting the corresponding nodes to obtain a multilateral exchange relationship graph; wherein, in the multilateral exchange relationship graph, the same party involved in the exchange appears only as a node; inputting the multilateral exchange relationship graph into a pre-trained abnormal exchange node prediction model to generate a labeling result in which each node in the multilateral exchange relationship graph is labeled as an abnormal exchange node, wherein the abnormal exchange node prediction model is obtained through any of the implementation methods described in the first aspect.

[0015] In some embodiments, the method further includes: sorting the corresponding exchange parties according to the annotation results of each node in the multilateral exchange relationship graph to obtain the target exchange party; and outputting the first prediction information corresponding to the target exchange party.

[0016] In some embodiments, the method further includes: taking the target exchange party as one of the exchange parties, determining the other party among the exchange parties that are frequently exchanged in the multilateral exchange relationship graph based on the target exchange party; and outputting second prediction information corresponding to the determined other party.

[0017] According to a third aspect of this application, a model training apparatus is provided, comprising: an acquisition unit configured to determine the two parties involved in the exchange based on exchange data and acquire feature datasets of the two parties; a generation unit configured to treat each party involved in the exchange as a different node, use the feature dataset of the corresponding party as the node content of the corresponding node, and draw the existing exchange relationships as edges connecting the corresponding nodes to obtain a multilateral exchange relationship graph; wherein, the same party in the multilateral exchange relationship graph appears only as a node; and a training unit configured to take the multilateral exchange relationship graph as input and the labeling results of each node in the multilateral exchange relationship graph being labeled as an abnormal exchange node as output, and train an abnormal exchange node prediction model.

[0018] In some embodiments, the abnormal exchange node prediction model includes multiple node prediction sub-models; the training unit includes: a first determining module configured to determine multiple multilateral exchange relationship subgraphs based on the multilateral exchange relationship graph; a training module configured to take each multilateral exchange relationship subgraph as input and the labeling results of each node in the corresponding multilateral exchange relationship subgraph being labeled as an abnormal exchange node as output, and train to obtain each node prediction sub-model; and a second determining module configured to determine the abnormal exchange node prediction model based on each node prediction sub-model.

[0019] In some embodiments, the first determining module is further configured to sample associated nodes for each node in the multilateral exchange relationship graph to generate a multilateral exchange relationship subgraph composed of each node in the multilateral exchange relationship graph and its corresponding associated nodes. The associated nodes include: each first associated node connected to the target node and each second associated node connected to each first associated node.

[0020] In some embodiments, the two parties to the exchange are the initiator and the provider of the exchange behavior; the generation unit is further configured to treat each initiator as a different initiator node, treat each provider as a different provider node, treat the feature datasets of the corresponding initiator and the feature datasets of the corresponding provider as the node content of the corresponding node, and draw the existing exchange relationship as an edge connecting the corresponding initiator node and the provider node to obtain a multilateral exchange relationship graph.

[0021] In some embodiments, the training unit is further configured to take a multilateral exchange graph as input and the labeling results of each provider node in the multilateral exchange graph being labeled as an abnormal exchange node as output, and train an abnormal exchange node prediction model.

[0022] In some embodiments, the abnormal exchange node prediction model is trained in a semi-supervised training manner; the apparatus further includes: a labeling unit, configured to label predetermined exchange partners as positive samples and label other exchange partners besides the predetermined exchange partners as negative samples, thereby obtaining labeled exchange partners.

[0023] In some embodiments, the apparatus further includes a filtering unit configured to delete exchanges where the exchange value assignment information is below a threshold.

[0024] In some embodiments, the abnormal swap node prediction model and / or node prediction sub-model are constructed based on a deep heterogeneous graph convolutional network.

[0025] In some embodiments, the feature datasets of the exchanging parties are generated by processing the exchange data based on the historical information of each party. The historical information includes at least: historical exchange information, historical behavior information, and initiator profile.

[0026] According to a fourth aspect of this application, an apparatus for generating information is provided, comprising: a data acquisition unit configured to determine the parties involved in an exchange based on exchange data and acquire feature datasets of the parties involved in the exchange; a relationship graph generation unit configured to treat each party involved in the exchange as a different node, use the feature dataset of the corresponding party as the node content of the corresponding node, and draw the existing exchange relationships as edges connecting the corresponding nodes to obtain a multilateral exchange relationship graph; wherein, the same party involved in the multilateral exchange relationship graph appears only as a node; and an information generation unit configured to input the multilateral exchange relationship graph into a pre-trained abnormal exchange node prediction model to generate a labeling result in which each node in the multilateral exchange relationship graph is labeled as an abnormal exchange node, wherein the abnormal exchange node prediction model is obtained through any implementation of the first aspect.

[0027] In some embodiments, the apparatus further includes: a result sorting unit configured to sort the corresponding exchange parties according to the labeling results of each node in the multilateral exchange relationship graph to obtain the target exchange party; and a first output unit configured to output first prediction information corresponding to the target exchange party.

[0028] In some embodiments, the apparatus further includes: a swapping party determination unit configured to identify the target swapping party as one of the swapping parties, and determine, based on the target swapping party, the other party among the swapping parties identified as frequently exchanging in the multilateral swapping relationship graph; and a second output unit configured to output second prediction information corresponding to the determined other party.

[0029] According to a fifth aspect of this application, an electronic device is provided, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform a method as described in any implementation of the first or second aspect.

[0030] According to a sixth aspect of this application, a non-transitory computer-readable storage medium storing computer instructions is provided, characterized in that the computer instructions are used to cause a computer to perform a method as described in any implementation of the first or second aspect.

[0031] According to the technology of this application, the two parties involved in the exchange are determined based on the exchange data, and the feature datasets of the two parties are obtained. Each party is treated as a different node, and the feature dataset of the corresponding party is used as the node content of the corresponding node. The existing exchange relationships are drawn as edges connecting the corresponding nodes to obtain a multilateral exchange relationship graph. The multilateral exchange relationship graph is then used as input, and the annotation results of each node in the multilateral exchange relationship graph as an abnormal exchange node are used as output. An abnormal exchange node prediction model is trained to obtain the abnormal exchange node. Based on the features of the two parties and their exchange relationships, the abnormal situation of the exchange parties is predicted, realizing a more accurate graph algorithm-based model training method and device.

[0032] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of this application, nor is it intended to limit the scope of this application. Other features of this application will become readily apparent from the following description. Attached Figure Description

[0033] The accompanying drawings are provided to better understand this solution and do not constitute a limitation of this application.

[0034] Figure 1 This is a schematic diagram of a first embodiment of the model training method according to this application;

[0035] Figure 2 This is a scene diagram illustrating the model training method implemented in the embodiments of this application;

[0036] Figure 3 This is a schematic diagram of a second embodiment of the model training method according to this application;

[0037] Figure 4 This is a schematic diagram of a third embodiment of the model training method according to this application;

[0038] Figure 5 This is a schematic diagram of the multilateral exchange relationship diagram in the embodiments of this application;

[0039] Figure 6 This is a schematic diagram of an embodiment of the method for generating information according to this application;

[0040] Figure 7 This is a schematic diagram of the structure of one embodiment of the model training apparatus according to this application;

[0041] Figure 8 This is a schematic diagram of one embodiment of the apparatus for generating information according to this application;

[0042] Figure 9 This is a block diagram of an electronic device used to implement the model training method of the embodiments of this application. Detailed Implementation

[0043] The following description, in conjunction with the accompanying drawings, illustrates exemplary embodiments of this application, including various details to aid understanding. These should be considered merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of this application. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description.

[0044] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0045] Figure 1 A schematic diagram 100 of a first embodiment of the model training method according to this application is shown. The model training method includes the following steps:

[0046] Step 101: Determine the two parties involved in the exchange based on the exchanged data, and obtain the feature datasets of the two parties involved in the exchange.

[0047] In this embodiment, the executing entity (e.g., a server) can obtain exchange data from other electronic devices or locally via wired or wireless connections, determine the exchanging parties based on the exchange data, and then obtain the feature datasets of the exchanging parties again. The exchanging parties can be the initiator of the exchange and the provider of the exchange service, or two parties between providers, or two initiators, etc. Exchange data refers to the record of the exchange behavior between the exchanging parties, including exchange information such as the exchanging parties, exchange time, and exchange value assignment information. The feature dataset consists of multi-dimensional features of the exchanging parties, including various feature information of the exchanging parties, such as exchange-related information, behavioral information, and exchanging party registration information. It should be noted that the aforementioned wireless connection methods can include, but are not limited to, 3G, 4G, 5G, WiFi, Bluetooth, WiMAX, Zigbee, UWB (ultra-wideband), and other currently known or future wireless connection methods.

[0048] Step 102: Treat each party of the exchange as a different node, use the feature dataset of the corresponding exchange party as the node content of the corresponding node, and draw the existing exchange relationship as the edge connecting the corresponding node to obtain the multilateral exchange relationship graph.

[0049] In this embodiment, the executing entity can construct a graph structure using exchange relationships. Each party in the exchange is treated as a different node, the feature dataset of the corresponding party is used as the node content of the corresponding node, and the existing exchange relationships are drawn as edges connecting the corresponding nodes, resulting in a multilateral exchange relationship graph. In this graph, the same party in the same exchange relationship appears only as a single node. A multilateral exchange relationship graph is a complete graph composed of all nodes, all edges, and all node contents.

[0050] Step 103: Take the multilateral exchange relationship graph as input and the labeling results of each node in the multilateral exchange relationship graph as an abnormal exchange node as output to train an abnormal exchange node prediction model.

[0051] In this embodiment, the aforementioned execution entity can utilize graph algorithms and machine learning algorithms to train an abnormal exchange node prediction model by taking a multilateral exchange relationship graph as input and the annotation results of each node in the multilateral exchange relationship graph as an abnormal exchange node as output. An abnormal exchange node can indicate that the exchange party is a provider or initiator with malicious behavior, which may include abnormal assignment of exchange value information, etc. The annotation results can be node scores, node probability values, etc. The model training adopts a full graph training approach, that is, the complete graph structure is fed into the model for training at once. The training and prediction process of each node depends on the data of all nodes and edge relationships in the graph.

[0052] It should be noted that the above-mentioned abnormal exchange node prediction model can be, for example, a data table or a calculation formula. This embodiment does not limit the content in this regard. Those skilled in the art can set the model structure of the above-mentioned abnormal exchange node prediction model according to actual needs, such as logistic regression, decision tree, and other machine learning models. The embodiments disclosed herein do not limit this. The above-mentioned graph algorithms and machine learning algorithms are well-known technologies that are currently widely researched and applied, and will not be described in detail here.

[0053] For ease of understanding, a scenario that can implement the model training method of the embodiments of this application is provided; see [link to relevant documentation]. Figure 2In this embodiment, the model training method 200 runs on server 201, where the two parties involved in the exchange are the initiator and the provider of the exchange. Server 201 first determines the initiator and provider based on the exchange data and obtains the feature datasets of the initiator and the provider 202. Then, server 201 treats the initiator and provider as different nodes, uses the feature datasets of the corresponding initiator and the feature datasets of the corresponding provider as the node content of the corresponding node, and draws the existing exchange relationships as edges connecting the corresponding nodes, resulting in a multilateral exchange relationship graph 203. In the multilateral exchange relationship graph, the same exchanging party appears only as a node. Finally, server 201 uses the multilateral exchange relationship graph as input and the labeling results of each node in the multilateral exchange relationship graph as an abnormal exchange node as output to train an abnormal exchange node prediction model 204.

[0054] The model training method provided in the above embodiments of this application determines the two parties involved in the exchange based on the exchange data, obtains the feature datasets of the two parties, treats each party as a different node, uses the feature dataset of the corresponding party as the node content of the corresponding node, and draws the existing exchange relationships as edges connecting the corresponding nodes to obtain a multilateral exchange relationship graph. Then, the multilateral exchange relationship graph is used as input, and the labeling results of each node in the multilateral exchange relationship graph as an abnormal exchange node are used as output to train an abnormal exchange node prediction model. Based on the features of the two parties involved in the exchange and their exchange relationships, the abnormal situation of the exchanging parties is predicted, realizing a more accurate graph algorithm-based model training method. This method makes the maximum use of all available information of the two parties involved in the exchange as prediction samples, thereby improving the accuracy of model prediction.

[0055] Further reference Figure 3 The diagram 300 illustrates a second embodiment of the model training method. The method includes the following steps:

[0056] Step 301: Determine the two parties involved in the exchange based on the exchanged data, and obtain the feature datasets of the two parties involved in the exchange.

[0057] In some optional implementations of this embodiment, the feature datasets of the exchanging parties are generated by processing the exchange data based on the historical information of each party. The historical information includes at least: historical exchange information, historical behavior information, and initiator profiles. For example, enumerable and numerical features are processed for all involved providers based on recent exchanges, historical behavior, and registration information; enumerable and numerical features are processed for all involved initiators based on initiator profiles, historical behavior, registration information, recent exchanges, and frequently used devices. By processing the historical information of all exchanging parties, the feature dataset becomes accurate and comprehensive.

[0058] In some optional implementations of this embodiment, before obtaining the feature datasets of both parties in the exchange, the method further includes deleting parties whose exchange value is below a threshold. By filtering out providers of normal exchange information during data preprocessing, performance is ensured while computational efficiency is significantly improved.

[0059] Step 302: Treat each party of the exchange as a different node, use the feature dataset of the corresponding exchange party as the node content of the corresponding node, and draw the existing exchange relationship as the edge connecting the corresponding node to obtain the multilateral exchange relationship graph.

[0060] In some optional implementations of this embodiment, the training method for the abnormal exchange node prediction model can be semi-supervised training. Before treating each party in the exchange as a different node, using the feature dataset of the corresponding party as the node content, and drawing the existing exchange relationships as edges connecting the corresponding nodes to obtain a multilateral exchange relationship graph, the method further includes: labeling predetermined exchange parties as positive samples and labeling other exchange parties as negative samples, thus obtaining labeled exchange parties. Semi-supervised training refers to training where only some samples have explicit labels, unlike supervised training (where all samples have clear labels and are referenced by the model during training). During training, the model only refers to the recognition of these labeled samples to adjust its parameters; other samples are not referenced by the model, but they still affect the recognition of labeled samples. The predetermined exchange parties can be pre-identified exchange parties or a pre-determined specific exchange party or a certain type of exchange party. By using the feature similarity of exchange parties and the relationship between the exchanging parties for semi-supervised learning, more accurate predictions of abnormal exchange nodes can be achieved.

[0061] Step 303: Based on the multilateral exchange relationship graph, determine multiple multilateral exchange relationship subgraphs.

[0062] In this embodiment, the executing entity can group the multilateral exchange graph obtained in step 302 to obtain at least two grouped multilateral exchange subgraphs, so that the model training adopts the mini-batch gradient descent method, that is, grouping and training all nodes to be trained. Then, for each node in the multilateral exchange subgraph, associated nodes are sampled to generate a final multilateral exchange subgraph composed of each node in the multilateral exchange subgraph and its corresponding associated nodes. The associated nodes include: each first associated node connected to the target node and each second associated node connected to each first associated node.

[0063] In some optional implementations of this embodiment, multiple multilateral exchange relationship subgraphs are determined based on the multilateral exchange relationship graph. This includes: sampling associated nodes for each node in the multilateral exchange relationship graph to generate a multilateral exchange relationship subgraph composed of each node in the multilateral exchange relationship graph and its corresponding associated nodes. The associated nodes include: each first associated node connected to the target node and each second associated node connected to each first associated node. This solves the problem in the prior art that it is impossible to mine information about gangs engaging in misconduct, and that there may be some clustering characteristics at the exchange level among providers, which can only be utilized by graph relationship-based algorithms. By incorporating exchange relationship information during the training of each node (e.g., a provider node), the prediction is decoupled from the individual level. For second associated nodes of the same type (other associated provider nodes), it can identify whether the current node (current provider node) and other nodes of the same type (other provider nodes) are engaging in clustering activities, thus possessing the ability to identify gangs.

[0064] Step 304: Take each multilateral exchange relationship subgraph as input and the labeling results of each node in the corresponding multilateral exchange relationship subgraph as an abnormal exchange node as output, and train to obtain the prediction submodel for each node.

[0065] In this embodiment, the abnormal exchange node prediction model can include multiple node prediction sub-models. The executing entity can utilize machine learning algorithms and graph algorithms to train each node prediction sub-model by taking each multilateral exchange relation subgraph as input and the labeling results of each node in the corresponding multilateral exchange relation subgraph as an abnormal exchange node as output. Both the abnormal exchange node prediction model and the node prediction sub-models are built based on deep heterogeneous graph convolutional networks. Deep heterogeneous graph convolutional networks refer to deep graph convolutional networks, represented by the Graph Neural Network (GCN) algorithm, which evolved from convolutional computation in deep learning in the field of graph computing. Later, deep heterogeneous graph convolutional networks emerged to handle heterogeneous graphs (i.e., graphs with more than one type of node or edge). For example, the Modeling Relational Data with Graph Convolutional Networks (RGCN) algorithm is used. A multilateral exchange relation subgraph is a complete graph composed of all nodes, all edges, and all node content in the subgraph.

[0066] Step 305: Determine the prediction model for abnormal exchange nodes based on the prediction sub-models for each node.

[0067] In this embodiment, the executing entity can integrate the prediction sub-models of each node to obtain the abnormal exchange node prediction model of the entire multilateral exchange relationship graph.

[0068] In this embodiment, the specific operations of steps 301-302 are the same as... Figure 1 The operations of steps 101-102 in the illustrated embodiments are basically the same, and will not be repeated here.

[0069] from Figure 3 It can be seen from this that, with Figure 1 Compared to the corresponding embodiments, the schematic diagram 300 of the model training method in this embodiment uses a multilateral exchange relationship graph to determine multiple multilateral exchange relationship subgraphs. Each multilateral exchange relationship subgraph is used as input, and the labeling results of each node in the corresponding multilateral exchange relationship subgraph as an abnormal exchange node are used as output. The prediction submodel of each node is trained, and the prediction model of abnormal exchange nodes is determined based on each node prediction submodel. By changing the full-graph training method of the deep heterogeneous convolutional graph algorithm to sampling training (i.e., obtaining the small part of nodes with the closest association for each node), the training time of the model is greatly reduced while ensuring the model performance, improving the computational efficiency and reducing the cost of implementing the solution.

[0070] Further reference Figure 4 The diagram 400 illustrates a third embodiment of the model training method. The method includes the following steps:

[0071] Step 401: Determine the two parties involved in the exchange based on the exchanged data, and obtain the feature datasets of the two parties involved in the exchange.

[0072] In this embodiment, the executing entity can acquire exchange data, determine the two parties involved in the exchange based on the exchange data, and then acquire the feature datasets of the two parties involved in the exchange. The two parties involved in the exchange are the initiator of the exchange and the provider of the exchange.

[0073] Step 402: Treat each initiator as a different initiator node, treat each provider as a different provider node, treat the feature datasets of the corresponding initiator and the feature datasets of the corresponding provider as the node content of the corresponding node, and draw the existing exchange relationships as edges connecting the corresponding initiator node and the provider node to obtain a multilateral exchange relationship graph.

[0074] In this embodiment, the executing entity can construct a graph structure using exchange relationships. This involves treating providers and initiators as two types of nodes in the graph, with each initiator as a different initiator node and each provider as a different provider node. The feature datasets of the corresponding initiators and providers are used as the node content of their respective nodes. Existing exchange relationships are then drawn as edges connecting the corresponding initiator nodes and provider nodes, resulting in a multilateral exchange relationship graph, such as... Figure 5 As shown.

[0075] Step 403: Using the multilateral exchange graph as input and the annotation results of each provider node in the multilateral exchange graph being labeled as an abnormal exchange node as output, an abnormal exchange node prediction model is trained.

[0076] In this embodiment, the executing entity can use graph algorithms and machine learning algorithms to train an abnormal exchange node prediction model by taking a multilateral exchange relationship graph as input and the annotation results of each provider node in the multilateral exchange relationship graph as an abnormal exchange node as output. Here, an abnormal exchange node can represent a provider or initiator with misconduct. The abnormal exchange node prediction model is constructed based on a deep heterogeneous graph convolutional network.

[0077] from Figure 4 It can be seen from this that, with Figure 1 Compared to the corresponding embodiments, the schematic diagram 400 of the model training method in this embodiment uses each initiator as a different initiator node, each provider as a different provider node, and the feature datasets of the corresponding initiator and the corresponding provider as the node content of the corresponding node. The existing exchange relationships are drawn as edges connecting the corresponding initiator nodes and provider nodes to obtain a multilateral exchange relationship graph. The multilateral exchange relationship graph is used as input, and the annotation results of each provider node in the multilateral exchange relationship graph as an abnormal exchange node are used as output to train an abnormal exchange node prediction model. This makes up for the shortcomings of existing machine learning-based exchange party identification schemes that cannot simultaneously utilize the exchange relationship between the provider and the initiator as well as the features of the provider and the initiator. The identification scheme based on the deep heterogeneous graph convolution model can simultaneously process multiple types of objects and their features, which is more in line with real-world exchange scenarios.

[0078] Further reference Figure 6 The diagram 600 illustrates an embodiment of a method for generating information according to the present disclosure. The method for generating information includes the following steps:

[0079] Step 601: Determine the two parties involved in the exchange based on the exchanged data, and obtain the feature datasets of the two parties involved in the exchange.

[0080] In this embodiment, the executing entity (e.g., a server or terminal device) can obtain exchange data from other electronic devices or locally via wired or wireless connections. Based on the obtained exchange data, the two parties involved in the exchange are identified, and a feature dataset of both parties is obtained. The exchange data refers to the external order exchange records of providers whose providers have been identified and confirmed to be blacklisted by online business policies.

[0081] Step 602: Treat each party of the exchange as a different node, use the feature dataset of the corresponding exchange party as the node content of the corresponding node, and draw the existing exchange relationship as the edge connecting the corresponding node to obtain the multilateral exchange relationship graph.

[0082] In this embodiment, the executing entity can construct a graph structure using exchange relationships. This involves treating each party in the exchange as a different node, using the feature datasets of the corresponding parties as the node content of the corresponding nodes, and drawing the existing exchange relationships as edges connecting the corresponding nodes, thus obtaining a multilateral exchange relationship graph. In this multilateral exchange relationship graph, the same party in the exchange only appears as one node.

[0083] Step 603: Input the multilateral exchange relationship graph into the pre-trained abnormal exchange node prediction model to generate the annotation results of each node in the multilateral exchange relationship graph being marked as an abnormal exchange node.

[0084] In this embodiment, the executing entity can utilize a deep heterogeneous graph convolution algorithm and a machine learning algorithm to input the multilateral exchange graph obtained in step 602 into a pre-trained abnormal exchange node prediction model, generating a labeling result in which each node in the multilateral exchange graph is labeled as an abnormal exchange node. The abnormal exchange node prediction model is obtained through any of the implementation methods described in the first aspect.

[0085] In some optional implementations of this embodiment, the method further includes: sorting the corresponding exchange parties according to the annotation results of each node in the multilateral exchange relationship graph to obtain the target exchange party; and outputting the first prediction information corresponding to the target exchange party. This solves the problem in the prior art where some exchange parties may deliberately conceal their exchange characteristics to avoid being shut down by third-party payment platforms, making it difficult to monitor and detect the exchange party through effective channels, and thus achieves the monitoring of exchange parties.

[0086] In some optional implementations of this embodiment, the method further includes: identifying the target exchange party as one of the two exchange parties; determining the other party among the frequently exchanging parties in the multilateral exchange graph based on the target exchange party; and outputting the second prediction information corresponding to the determined other party. By employing a deep heterogeneous graph convolution algorithm, while identifying suspected providers, the initiating groups that frequently exchange with these suspected providers are extracted.

[0087] from Figure 6 It can be seen from this that, with Figure 1Compared to the corresponding embodiments, the process 600 of the method for generating information in this embodiment emphasizes the step of using a trained abnormal exchange node prediction model to generate the annotation results of each node in the multilateral exchange relationship graph. Therefore, the solution described in this embodiment can utilize a comprehensive and accurate prediction model to achieve targeted node prediction for different scenarios. In the foreign single exchange scenario, based on the identified black market providers, this embodiment uses a deep heterogeneous graph convolution algorithm to identify potential unidentified suspected providers that are closely related to and highly similar to the black market providers, supplementing existing online business rules and policy identification.

[0088] Further reference Figure 7 As a response to the above Figures 1-4 To implement the method shown, this application provides an embodiment of a model training device, which is similar to... Figure 1 Corresponding to the method embodiment shown, in addition to the features described below, the device embodiment may also include [features related to...]. Figure 1 The method embodiments shown have the same or corresponding features, and produce the same as Figure 1 The method embodiments shown have the same or corresponding effects, and the device can be specifically applied to various electronic devices.

[0089] like Figure 7 As shown, the model training device 700 of this embodiment includes: an acquisition unit 701, a generation unit 702, and a training unit 703. The acquisition unit is configured to determine the two parties involved in the exchange based on the exchange data and acquire the feature datasets of the two parties. The generation unit is configured to treat each party involved in the exchange as a different node, use the feature dataset of the corresponding party as the node content of the corresponding node, and draw the existing exchange relationships as edges connecting the corresponding nodes to obtain a multilateral exchange relationship graph. In the multilateral exchange relationship graph, the same party involved in the exchange only appears as a node. The training unit is configured to take the multilateral exchange relationship graph as input and the labeling results of each node in the multilateral exchange relationship graph being labeled as an abnormal exchange node as output, and train an abnormal exchange node prediction model.

[0090] In this embodiment, the specific processing of the acquisition unit 701, generation unit 702, and training unit 703 of the model training device 700 and the resulting technical effects can be referred to respectively. Figure 1 The relevant descriptions of steps 101 to 103 in the corresponding embodiments will not be repeated here.

[0091] In some optional implementations of this embodiment, the abnormal exchange node prediction model includes multiple node prediction sub-models; the training unit includes: a first determining module, configured to determine multiple multilateral exchange relationship sub-graphs based on the multilateral exchange relationship graph; a training module, configured to take each multilateral exchange relationship sub-graph as input and the labeling results of each node in the corresponding multilateral exchange relationship sub-graph being labeled as an abnormal exchange node as output, and train to obtain each node prediction sub-model; and a second determining module, configured to determine the abnormal exchange node prediction model based on each node prediction sub-model.

[0092] In some optional implementations of this embodiment, the first determining module is further configured to sample the associated nodes of each node in the multilateral exchange relationship graph to generate a multilateral exchange relationship subgraph composed of each node in the multilateral exchange relationship graph and the corresponding associated nodes. The associated nodes include: each first associated node connected to the target node and each second associated node connected to each first associated node.

[0093] In some optional implementations of this embodiment, the two parties to the exchange are the initiator and the provider of the exchange behavior; the generation unit is further configured to treat each initiator as a different initiator node, treat each provider as a different provider node, treat the feature datasets of the corresponding initiator and the feature datasets of the corresponding provider as the node content of the corresponding node, and draw the existing exchange relationship as an edge connecting the corresponding initiator node and the provider node to obtain a multilateral exchange relationship graph.

[0094] In some optional implementations of this embodiment, the training unit is further configured to take the multilateral exchange graph as input and the labeling results of each provider node in the multilateral exchange graph being labeled as an abnormal exchange node as output, and train an abnormal exchange node prediction model.

[0095] In some optional implementations of this embodiment, the abnormal exchange node prediction model is trained in a semi-supervised training manner; the device further includes: a labeling unit, configured to label a predetermined exchange party as a positive sample and label other exchange parties besides the predetermined exchange party as negative samples, thereby obtaining each labeled exchange party.

[0096] In some optional implementations of this embodiment, the apparatus further includes a filtering unit configured to delete exchanges whose exchange value assignment information is below a threshold.

[0097] In some optional implementations of this embodiment, the abnormal swap node prediction model and / or node prediction sub-model are constructed based on a deep heterogeneous graph convolutional network.

[0098] In some optional implementations of this embodiment, the feature datasets of the exchanging parties are generated by processing the exchange data based on the historical information of each party. The historical information includes at least: historical exchange information, historical behavior information, and the initiator's profile.

[0099] Continue to refer to Figure 8 As a response to the above Figure 6 The implementation of the method shown herein provides an embodiment of an apparatus for generating information, which is similar to... Figure 6 Corresponding to the method embodiment shown, in addition to the features described below, the device embodiment may also include [features related to...]. Figure 6 The method embodiments shown have the same or corresponding features, and produce the same as Figure 6 The method embodiments shown have the same or corresponding effects, and the device can be specifically applied to various electronic devices.

[0100] like Figure 8 As shown, the apparatus 800 for generating information in this embodiment includes: a data acquisition unit 801, a relationship graph generation unit 802, and an information generation unit 803. The data acquisition unit is configured to determine the exchanging parties based on the exchange data and acquire the feature datasets of the exchanging parties. The relationship graph generation unit is configured to treat each of the exchanging parties as a different node, use the feature dataset of the corresponding exchanging party as the node content of the corresponding node, and draw the existing exchange relationships as edges connecting the corresponding nodes to obtain a multilateral exchange relationship graph. In the multilateral exchange relationship graph, the same exchanging party appears only as a node. The information generation unit is configured to input the multilateral exchange relationship graph into a pre-trained abnormal exchange node prediction model to generate a labeling result in which each node in the multilateral exchange relationship graph is labeled as an abnormal exchange node. The abnormal exchange node prediction model is obtained through any of the implementation methods described in the first aspect.

[0101] In some optional implementations of this embodiment, the apparatus further includes: a result sorting unit configured to sort the corresponding exchange parties according to the labeling results of each node in the multilateral exchange relationship graph to obtain the target exchange party; and a first output unit configured to output the first prediction information corresponding to the target exchange party.

[0102] In some optional implementations of this embodiment, the apparatus further includes: a swapping party determination unit, configured to identify the target swapping party as one of the swapping parties, and determine the other party among the swapping parties that are frequently swapped in the multilateral swapping relationship graph based on the target swapping party; and a second output unit, configured to output second prediction information corresponding to the determined other party.

[0103] In this embodiment, the specific processing of the data acquisition unit 801, the relationship diagram generation unit 802, and the information generation unit 803 of the information generation device 800, and the resulting technical effects, can be referred to respectively. Figure 6 The relevant descriptions of steps 601 to 603 in the corresponding embodiments will not be repeated here.

[0104] It should be noted that if the technical solutions disclosed herein involve the acquisition, storage, and application of the initiator's personal information, they comply with the relevant laws and regulations and do not violate public order and good morals.

[0105] According to embodiments of this application, this application also provides an electronic device and a readable storage medium.

[0106] like Figure 9 The diagram shown is a block diagram of an electronic device for a model training method according to an embodiment of this application. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the present application described and / or claimed herein.

[0107] like Figure 9 As shown, the electronic device includes one or more processors 901, a memory 902, and interfaces for connecting the components, including high-speed interfaces and low-speed interfaces. The components are interconnected via different buses and can be mounted on a common motherboard or otherwise as required. The processors can process instructions executed within the electronic device, including instructions stored in or on memory to display graphical information of a GUI on an external input / output device (such as a display device coupled to the interface). In other embodiments, multiple processors and / or multiple buses can be used with multiple memories and multiple memory modules, if desired. Similarly, multiple electronic devices can be connected, each providing some of the necessary operations (e.g., as a server array, a group of blade servers, or a multiprocessor system). Figure 9 Take the 901 processor as an example.

[0108] The memory 902 is the non-transitory computer-readable storage medium provided in this application. The memory stores instructions executable by at least one processor to cause at least one processor to perform the model training method provided in this application. The non-transitory computer-readable storage medium of this application stores computer instructions for causing a computer to perform the model training method provided in this application.

[0109] Memory 902, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs, non-transitory computer-executable programs, and modules, such as the program instructions / modules corresponding to the model training method in the embodiments of this application (e.g., appendix). Figure 7 The acquisition unit 701, generation unit 702, and training unit 703 are shown. The processor 901 executes various functional applications of the server and model training by running non-transient software programs, instructions, and modules stored in the memory 902, thereby implementing the model training method in the above method embodiments.

[0110] The memory 902 may include a program storage area and a data storage area. The program storage area may store the operating system and applications required for at least one function; the data storage area may store data created based on the use of the model training electronic device. Furthermore, the memory 902 may include high-speed random access memory and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some embodiments, the memory 902 may optionally include memory remotely located relative to the processor 901, and these remote memories can be connected to the model training electronic device via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0111] The electronic device for the model training method may further include an input device 903 and an output device 904. The processor 901, memory 902, input device 903, and output device 904 can be connected via a bus or other means. Figure 9 Taking the example of a connection between China and Israel via a bus.

[0112] Input device 903 can receive input digital or character information, and generate key signal inputs related to the initiator settings and function control of the model training electronic device, such as a touch screen, keypad, mouse, trackpad, touchpad, joystick, one or more mouse buttons, trackball, joystick, etc. Output device 904 may include a display device, auxiliary lighting device (e.g., LED), and haptic feedback device (e.g., vibration motor). The display device may include, but is not limited to, a liquid crystal display (LCD), a light-emitting diode (LED) display, and a plasma display. In some embodiments, the display device may be a touch screen.

[0113] Various implementations of the systems and techniques described herein can be implemented in digital electronic circuit systems, integrated circuit systems, application-specific integrated circuits (ASICs), computer hardware, firmware, software, and / or combinations thereof. These various implementations may include: implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transferring data and instructions to the storage system, the at least one input device, and the at least one output device.

[0114] These computational programs (also referred to as programs, software, software applications, or code) include machine instructions for a programmable processor and can be implemented using high-level procedural and / or object-oriented programming languages, and / or assembly / machine languages. As used herein, the terms “machine-readable medium” and “computer-readable medium” refer to any computer program product, device, and / or apparatus (e.g., disk, optical disk, memory, programmable logic device (PLD)) used to provide machine instructions and / or data to a programmable processor, including machine-readable media that receive machine instructions as machine-readable signals. The term “machine-readable signal” refers to any signal used to provide machine instructions and / or data to a programmable processor.

[0115] To facilitate exchange with the initiator, the systems and techniques described herein can be implemented on a computer having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the initiator; and a keyboard and pointing device (e.g., a mouse or trackball) through which the initiator provides input to the computer. Other types of devices may also be used to facilitate exchange with the initiator; for example, feedback provided to the initiator may be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the initiator may be received in any form (including sound input, voice input, or tactile input).

[0116] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as a data server), or computing systems that include middleware components (e.g., an application server), or computing systems that include frontend components (e.g., an initiator computer with a graphical initiator interface or a web browser, through which the initiator can exchange with the implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication (e.g., a communication network) of any form or medium. Examples of communication networks include local area networks (LANs), wide area networks (WANs), and the Internet.

[0117] Computer systems can include clients and servers. Clients and servers are generally located far apart and typically exchange information via a communication network. The client-server relationship is created by computer programs running on the respective computers and having a client-server relationship with each other.

[0118] It should be understood that the various forms of processes shown above can be used to rearrange, add, or delete steps. For example, the steps described in this application can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution disclosed in this application can be achieved, and this is not limited herein.

[0119] The specific embodiments described above do not constitute a limitation on the scope of protection of this application. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A model training method, comprising: The parties involved in the exchange are identified based on the exchange data, and a feature dataset of the parties is obtained; wherein, the exchange data includes exchange behavior data of the parties involved. Each of the two parties in the exchange is treated as a different node, and the feature dataset of the corresponding party is used as the node content of the corresponding node. The existing exchange relationships are drawn as edges connecting the corresponding nodes to obtain a multilateral exchange relationship graph. In the multilateral exchange relationship graph, the same party in the exchange only appears as a node. The feature dataset of the two parties in the exchange is generated by processing the exchange data based on the historical information of the parties in the exchange. The historical information includes at least: historical exchange information, historical behavior information, and initiator profile. The abnormal exchange node prediction model is trained by taking the multilateral exchange relationship graph as input and the labeling results of each node in the multilateral exchange relationship graph as an abnormal exchange node as output, using all available information of the two parties in the exchange as prediction samples.

2. The method according to claim 1, wherein, The abnormal exchange node prediction model includes multiple node prediction sub-models; the abnormal exchange node prediction model is trained by taking the multilateral exchange relationship graph as input and the labeling results of each node in the multilateral exchange relationship graph as an abnormal exchange node as output, including: Based on the multilateral exchange relationship graph, multiple multilateral exchange relationship subgraphs are determined; Each of the multilateral exchange relationship subgraphs is taken as input, and the labeling results of each node in the corresponding multilateral exchange relationship subgraph being marked as an abnormal exchange node are taken as output, and a prediction submodel for each node is trained. The abnormal exchange node prediction model is determined based on each of the node prediction sub-models.

3. The method according to claim 2, wherein, The step of determining multiple multilateral exchange relationship subgraphs based on the multilateral exchange relationship graph includes: For each node in the multilateral exchange graph, associated node sampling is performed to generate a multilateral exchange subgraph composed of each node in the multilateral exchange graph and the corresponding associated nodes. The associated nodes include: each first associated node connected to the target node and each second associated node connected to each first associated node.

4. The method according to any one of claims 1-3, wherein, The two parties involved in the exchange are the initiator and the provider of the exchange. Each party is treated as a different node, and the feature dataset of the corresponding party is used as the node content of that node. The existing exchange relationships are then plotted as edges connecting the corresponding nodes, resulting in a multilateral exchange relationship graph, including: Each initiator is treated as a different initiator node, each provider as a different provider node, and the feature datasets of the corresponding initiator and the corresponding provider are treated as the node content of the corresponding node. The existing exchange relationships are drawn as edges connecting the corresponding initiator node and the provider node, thus obtaining a multilateral exchange relationship graph.

5. The method according to claim 4, wherein, The step of training an abnormal exchange node prediction model by taking the multilateral exchange graph as input and the annotation results of each node in the multilateral exchange graph as an abnormal exchange node as output includes: Using the multilateral exchange graph as input and the labeling results of each provider node in the multilateral exchange graph being marked as an abnormal exchange node as output, an abnormal exchange node prediction model is trained.

6. The method according to claim 1, wherein, The abnormal exchange node prediction model is trained using a semi-supervised training method. Before the step of treating each party in the exchange as a different node, using the feature dataset of the corresponding party as the node content of the corresponding node, and drawing the existing exchange relationships as edges connecting the corresponding nodes to obtain a multilateral exchange relationship graph, the method further includes: The predetermined exchange partners are labeled as positive samples, and other exchange partners besides the predetermined ones are labeled as negative samples, thus obtaining the labeled exchange partners.

7. The method according to any one of claims 1-3, wherein, The abnormal exchange node prediction model and / or the node prediction sub-model are constructed based on a deep heterogeneous graph convolutional network.

8. A method for generating information, comprising: The parties involved in the exchange are identified based on the exchange data, and a feature dataset of the parties is obtained; wherein, the exchange data includes exchange behavior data of the parties involved. Each of the two parties in the exchange is treated as a different node, the feature dataset of the corresponding party is used as the node content of the corresponding node, and the existing exchange relationships are drawn as edges connecting the corresponding nodes to obtain a multilateral exchange relationship graph; wherein, in the multilateral exchange relationship graph, the same party in the exchange only appears as a node; The multilateral exchange graph is input into a pre-trained abnormal exchange node prediction model to generate a labeling result in which each node in the multilateral exchange graph is labeled as an abnormal exchange node, wherein the abnormal exchange node prediction model is trained by the method described in any one of claims 1-7.

9. The method according to claim 8, further comprising: Based on the labeling results of each node in the multilateral exchange relationship graph, the corresponding exchange parties are sorted to obtain the target exchange parties; Output the first prediction information corresponding to the target exchange party.

10. The method of claim 9, further comprising: The target exchange party is taken as one of the two exchange parties. Based on the target exchange party, the other party in the multilateral exchange relationship graph that is identified as frequently exchanging is determined. Output the second prediction information corresponding to the determined other party.

11. A model training device, comprising: The acquisition unit is configured to determine the two parties involved in the exchange based on the exchange data, and acquire a feature dataset of the two parties involved in the exchange; wherein the exchange data includes exchange behavior data of the two parties involved in the exchange. The generation unit is configured to treat each of the two parties in the exchange as a different node, use the feature dataset of the corresponding party as the node content of the corresponding node, and draw the existing exchange relationships as edges connecting the corresponding nodes to obtain a multilateral exchange relationship graph; wherein, in the multilateral exchange relationship graph, the same party in the exchange appears only as a node, and the feature dataset of the two parties in the exchange is generated by processing the exchange data based on the historical information of the parties in the exchange, and the historical information includes at least: historical exchange information, historical behavior information, and initiator profile; The training unit is configured to take the multilateral exchange graph as input and the labeling results of each node in the multilateral exchange graph as an abnormal exchange node as output, so as to train an abnormal exchange node prediction model by using all available information of the two parties in the exchange as prediction samples.

12. The apparatus according to claim 11, wherein, The abnormal exchange node prediction model includes multiple node prediction sub-models; The training unit includes: The first determining module is configured to determine multiple multilateral exchange relationship subgraphs based on the multilateral exchange relationship graph; The training module is configured to take each of the multilateral exchange relationship subgraphs as input and the labeling results of each node in the corresponding multilateral exchange relationship subgraph being labeled as an abnormal exchange node as output, and train to obtain a prediction submodel for each node. The second determining module is configured to determine the abnormal exchange node prediction model based on each of the node prediction sub-models.

13. The apparatus according to claim 12, wherein, The first determining module is further configured to sample the associated nodes of each node in the multilateral exchange relationship graph to generate a multilateral exchange relationship subgraph composed of each node in the multilateral exchange relationship graph and the corresponding associated nodes. The associated nodes include: each first associated node connected to the target node and each second associated node connected to each first associated node.

14. The apparatus according to any one of claims 11-13, wherein, The two parties involved in the exchange are the initiator of the exchange and the provider of the exchange. The generation unit is further configured to treat each initiator as a different initiator node, each provider as a different provider node, the feature datasets of the corresponding initiator and the feature datasets of the corresponding provider as the node content of the corresponding node, and to draw the existing exchange relationship as an edge connecting the corresponding initiator node and the provider node to obtain a multilateral exchange relationship graph.

15. The apparatus according to claim 14, wherein, The training unit is further configured to take the multilateral exchange graph as input and the labeling results of each of the provider nodes in the multilateral exchange graph being labeled as abnormal exchange nodes as output, and train an abnormal exchange node prediction model.

16. An apparatus for generating information, comprising: The data acquisition unit is configured to determine the two parties involved in the exchange based on the exchange data, and to acquire the feature dataset of the two parties involved in the exchange; wherein the exchange data includes the exchange behavior data of the two parties involved in the exchange. The relationship graph generation unit is configured to treat each of the two parties in the exchange as a different node, use the feature dataset of the corresponding party as the node content of the corresponding node, and draw the existing exchange relationships as edges connecting the corresponding nodes to obtain a multilateral exchange relationship graph; wherein, in the multilateral exchange relationship graph, the same party in the exchange appears only as a node; The information generation unit is configured to input the multilateral exchange relationship graph into a pre-trained abnormal exchange node prediction model to generate a labeling result in which each node in the multilateral exchange relationship graph is labeled as an abnormal exchange node, wherein the abnormal exchange node prediction model is trained by the method described in any one of claims 1-7.

17. The apparatus of claim 16, further comprising: The sorting unit is configured to sort the corresponding exchange parties according to the labeling results of each node in the multilateral exchange relationship graph to obtain the target exchange party. The first output unit is configured to output the first prediction information corresponding to the target exchange party.

18. The apparatus of claim 17, further comprising: The exchange party determination unit is configured to identify the target exchange party as one of the exchange parties, and, based on the target exchange party, determine the other party among the frequently exchanging parties in the multilateral exchange relationship graph. The second output unit is configured to output the second prediction information corresponding to the determined other party.

19. An electronic device comprising: At least one processor; as well as A memory communicatively connected to the at least one processor; wherein, The memory stores instructions executable by the at least one processor, which, when executed by the at least one processor, enables the at least one processor to perform the method of any one of claims 1-10.

20. A non-transitory computer-readable storage medium storing computer instructions, wherein, The computer instructions are used to cause the computer to perform the method according to any one of claims 1-10.

Images