Resource interaction risk prediction processing method, device and equipment and storage medium

Abstract

The application relates to a resource interaction risk prediction processing method, device and equipment. The method comprises the following steps: determining a plurality of resource interaction object groups and resource interaction amounts associated with each resource interaction object group according to resource interaction data in a preset time period; constructing a resource interaction network according to each resource interaction object group and the resource interaction amounts associated with each resource interaction object group; determining a resource transfer probability matrix corresponding to the resource interaction network; obtaining an object node sequence by sampling nodes based on the resource interaction network and the resource transfer probability matrix; constructing a node coding vector pair according to the node coding vectors corresponding to each object node in the object node sequence; obtaining risk labels of each node coding vector pair; training an initial interaction risk prediction model according to each node coding vector pair and the risk labels of each node coding vector pair; and obtaining a trained interaction risk prediction model. The method can improve the resource interaction risk prediction accuracy.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and in particular to a resource interaction risk prediction and processing method, apparatus, computer equipment, computer-readable storage medium, and computer program product. Background Technology

[0002] With the development of artificial intelligence technology and the promotion and application of various resource interaction services, the security performance requirements for interactive resources and resource information involved in the resource interaction service process are increasing. In order to ensure resource security in the resource interaction service process and avoid malicious resource interaction or resource loss, the approach of predicting the interaction risks in the resource interaction service process has emerged, so as to deploy corresponding risk handling measures in advance.

[0003] In traditional technologies, when predicting the interaction risks of resource interaction services, a combination of rule-based strategies and logistic regression-based scoring card models are typically used. This approach assesses the risk of resource interaction and blocks resource interaction services based on the characteristics of both parties involved in the resource interaction (including the resource sender and the resource receiver).

[0004] However, under the current methods of resource interaction risk assessment and interaction business blocking, the rule combination strategies adopted are relatively simple in terms of their rule combination learning based on decision trees, and the scorecard model based on logistic regression is essentially a single linear model. Therefore, neither can fully simulate the complex resource interaction behavior under different resource interaction scenarios. When using rule combination strategies and scorecard models based on logistic regression for resource interaction risk assessment and prediction, there are still problems such as low risk prediction accuracy and inability to block risky interaction business in a timely and accurate manner. Summary of the Invention

[0005] Therefore, it is necessary to provide a resource interaction risk prediction and processing method, apparatus, computer equipment, computer-readable storage medium, and computer program product that can improve the accuracy of interaction risk prediction in the resource interaction business processing process, in response to the above-mentioned technical problems.

[0006] In a first aspect, this application provides a method for predicting and processing resource interaction risks, comprising: acquiring resource interaction data within a preset time period; determining multiple resource interaction object groups and resource interaction quantities associated with the resource interaction object groups based on the resource interaction data; constructing a resource interaction network based on each resource interaction object group and the resource interaction quantities associated with the resource interaction object groups, and determining a resource transfer probability matrix corresponding to the resource interaction network; performing object node sampling processing based on the resource interaction network and the resource transfer probability matrix to obtain an object node sequence; acquiring the node encoding vector corresponding to each object node in the object node sequence, constructing node encoding vector pairs based on each node encoding vector, and acquiring risk labels for each node encoding vector pair; and training an initial interaction risk prediction model based on each node encoding vector pair and its respective risk labels to obtain a trained interaction risk prediction model.

[0007] Secondly, this application also provides a resource interaction risk prediction and processing device, comprising: a resource interaction data acquisition module, used to acquire resource interaction data within a preset time period, and determine multiple resource interaction object groups and resource interaction amounts associated with the resource interaction object groups based on the resource interaction data; a resource transfer probability matrix determination module, used to construct a resource interaction network based on each of the resource interaction object groups and the resource interaction amounts associated with the resource interaction object groups, and determine a resource transfer probability matrix corresponding to the resource interaction network; an object node sequence acquisition module, used to perform object node sampling processing based on the resource interaction network and the resource transfer probability matrix to obtain an object node sequence; a risk label acquisition module, used to acquire the node encoding vector corresponding to each object node in the object node sequence, construct node encoding vector pairs based on each of the node encoding vectors, and acquire risk labels for each of the node encoding vector pairs; and an interaction risk prediction model acquisition module, used to train an initial interaction risk prediction model based on each of the node encoding vector pairs and the risk labels for each of the node encoding vector pairs to obtain a trained interaction risk prediction model.

[0008] Thirdly, this application also provides a computer device, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to perform the following steps: acquiring resource interaction data within a preset time period; determining multiple resource interaction object groups and resource interaction quantities associated with the resource interaction object groups based on the resource interaction data; constructing a resource interaction network based on each resource interaction object group and the resource interaction quantities associated with the resource interaction object groups, and determining a resource transfer probability matrix corresponding to the resource interaction network; performing object node sampling processing based on the resource interaction network and the resource transfer probability matrix to obtain an object node sequence; acquiring the node encoding vector corresponding to each object node in the object node sequence; constructing node encoding vector pairs based on each node encoding vector, and acquiring risk labels for each node encoding vector pair; training an initial interaction risk prediction model based on each node encoding vector pair and its respective risk labels to obtain a trained interaction risk prediction model.

[0009] Fourthly, this application also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, performs the following steps: acquiring resource interaction data within a preset time period; determining multiple resource interaction object groups and resource interaction quantities associated with the resource interaction object groups based on the resource interaction data; constructing a resource interaction network based on each resource interaction object group and the resource interaction quantities associated with the resource interaction object groups, and determining a resource transfer probability matrix corresponding to the resource interaction network; performing object node sampling processing based on the resource interaction network and the resource transfer probability matrix to obtain an object node sequence; acquiring the node encoding vector corresponding to each object node in the object node sequence, constructing node encoding vector pairs based on each node encoding vector, and acquiring risk labels for each node encoding vector pair; training an initial interaction risk prediction model based on each node encoding vector pair and its respective risk label to obtain a trained interaction risk prediction model.

[0010] Fifthly, this application also provides a computer program product, including a computer program that, when executed by a processor, performs the following steps: acquiring resource interaction data within a preset time period; determining multiple resource interaction object groups and resource interaction quantities associated with the resource interaction object groups based on the resource interaction data; constructing a resource interaction network based on each resource interaction object group and the resource interaction quantities associated with the resource interaction object groups, and determining a resource transfer probability matrix corresponding to the resource interaction network; performing object node sampling processing based on the resource interaction network and the resource transfer probability matrix to obtain an object node sequence; acquiring the node encoding vector corresponding to each object node in the object node sequence; constructing node encoding vector pairs based on each node encoding vector, and acquiring risk labels for each node encoding vector pair; training an initial interaction risk prediction model based on each node encoding vector pair and its respective risk labels to obtain a trained interaction risk prediction model.

[0011] In the aforementioned resource interaction risk prediction and processing method, apparatus, computer equipment, computer-readable storage medium, and computer program product, resource interaction data within a preset time period is acquired. Based on the resource interaction data, multiple resource interaction object groups and the resource interaction quantities associated with the resource interaction object groups are determined. A resource interaction network is constructed based on each resource interaction object group and the resource interaction quantities associated with the resource interaction object groups. By determining the resource transfer probability matrix corresponding to the resource interaction network, object node sampling processing is performed based on the resource interaction network and the resource transfer probability matrix. This allows the acquisition of an object node sequence containing the specific location information of the resource interaction objects in the resource interaction network, while simultaneously determining the structural characteristics and resource interaction records of the resource interaction objects in the resource interaction network. Furthermore, by obtaining the node encoding vector corresponding to each object node in the object node sequence, constructing node encoding vector pairs based on each node encoding vector, and obtaining the risk label of each node encoding vector pair, the initial interaction risk prediction model is trained based on each node encoding vector and its respective risk label to obtain a trained interaction risk prediction model. This model can then be used to predict the interaction risks of different resource interaction objects, accurately and quickly identify the interaction risks of the corresponding resource interaction objects, and promptly implement risk control and resource interaction blocking. This improves the accuracy of interaction risk prediction during resource interaction business processing and effectively reduces resource losses for the resource interaction platform and other resource interaction objects. Attached Figure Description

[0012] To more clearly illustrate the technical solutions in the embodiments of this application or related technologies, the drawings used in the description of the embodiments of this application or related technologies will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0013] Figure 1 This is an application environment diagram of the resource interaction risk prediction and processing method in one embodiment;

[0014] Figure 2 This is a flowchart illustrating a resource interaction risk prediction and processing method in one embodiment;

[0015] Figure 3 This is a schematic diagram of the word vector network model in one embodiment;

[0016] Figure 4 This is a schematic diagram illustrating the process of constructing a resource interaction network in one embodiment;

[0017] Figure 5 This is a schematic diagram of a resource interaction network constructed for resource interaction data in one embodiment;

[0018] Figure 6 This is a flowchart illustrating the process of obtaining a sequence of object nodes in one embodiment;

[0019] Figure 7 This is a flowchart illustrating the process of obtaining a trained interactive risk prediction model in one embodiment.

[0020] Figure 8 This is a schematic diagram of the structure of the initial interaction risk prediction model in one embodiment;

[0021] Figure 9 This is a flowchart illustrating the resource interaction risk prediction and processing method in another embodiment;

[0022] Figure 10 This is a schematic diagram illustrating the concatenation of the target node's encoding vector in one embodiment;

[0023] Figure 11 This is a schematic diagram illustrating the risk prediction process based on an interactive risk prediction model in one embodiment.

[0024] Figure 12 This is a flowchart illustrating the resource interaction risk prediction and processing method in another embodiment;

[0025] Figure 13 This is a schematic diagram of the overall processing procedure of the resource interaction risk prediction and processing method in one embodiment;

[0026] Figure 14 This is a structural block diagram of a resource interaction risk prediction and processing device in one embodiment;

[0027] Figure 15 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0028] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0029] The resource interaction risk prediction and processing method provided in this application embodiment involves artificial intelligence technology and can be applied to various scenarios such as online media, instant messaging, and online financial transactions. Specifically, it can be applied to, for example... Figure 1 In the application environment shown, terminal 102 communicates with server 104 via a network. A data storage system can store the data that server 104 needs to process. The data storage system can be integrated onto server 104 or placed on the cloud or other network servers. Terminal 102 can be, but is not limited to, various personal computers, laptops, smartphones, tablets, IoT devices, portable wearable devices, and aircraft. IoT devices can include smart speakers, smart TVs, smart air conditioners, smart in-vehicle devices, and projection devices. Portable wearable devices can include smartwatches, smart bracelets, and head-mounted devices. Head-mounted devices can be virtual reality (VR) devices, augmented reality (AR) devices, smart glasses, etc. Server 104 can be a standalone physical server, a server cluster consisting of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms. The terminal 102 and the server 104 can be connected directly or indirectly through wired or wireless communication, and this embodiment does not impose any restrictions on this.

[0030] Both terminal 102 and server 104 can be used independently to execute the resource interaction risk prediction processing method provided in this embodiment, or they can work together to execute the same method. For example, taking the collaborative execution of the resource interaction risk prediction processing method by terminal 102 and server 104, server 104 acquires resource interaction data within a preset time period and determines multiple resource interaction object groups and the resource interaction amounts associated with those groups based on the data. Further, based on each resource interaction object group and the associated resource interaction amounts, server 104 constructs a resource interaction network and determines the resource transfer probability matrix corresponding to the resource interaction network. Furthermore, server 104 performs object node sampling processing based on the resource interaction network and resource transfer probability matrix to obtain an object node sequence. It then obtains the node encoding vector corresponding to each object node in the object node sequence, constructs node encoding vector pairs based on each node encoding vector, and obtains the risk label of each node encoding vector pair. Thus, based on each node encoding vector pair and its respective risk label, the initial interaction risk prediction model can be trained to obtain a trained interaction risk prediction model. When a resource interaction risk prediction request triggered by terminal 102 is received, the trained interaction risk prediction model can be used to perform risk prediction on the second resource receiving object and the second resource transferring object corresponding to the resource interaction risk prediction request, obtain the corresponding risk prediction results, and feed back the obtained risk prediction results to terminal 102.

[0031] In one exemplary embodiment, such as Figure 2 As shown, a resource interaction risk prediction and processing method is provided, which is applied to... Figure 1 Taking server 104 as an example, the explanation includes the following steps S202 to S210. Wherein:

[0032] Step S202: Obtain resource interaction data within a preset time period, and determine multiple resource interaction object groups and the resource interaction volume associated with the resource interaction object groups based on the resource interaction data.

[0033] The resource interaction data within the preset time period can be understood as transaction data between resource interaction objects over a period of time. This can include multiple resource transaction records. When parsing the multiple resource transaction records included in the resource interaction data, each resource transaction record can be obtained, including the resource transferor, the resource receiver, and the initial resource interaction amount. A resource transferor and a resource receiver with a resource interaction relationship can be considered as a resource interaction object group. To simplify the subsequent processing of resource interaction data and reduce cumbersome operations, at least two resource transaction records belonging to the same resource interaction object group need to be aggregated. Specifically, the initial interaction amounts of at least two resource interaction pairs belonging to the same resource interaction pair are superimposed to obtain the resource interaction amount associated with the corresponding resource interaction object group.

[0034] Specifically, the server parses resource interaction data within a preset time period to obtain multiple resource transaction records included in the resource interaction data, and determines the resource interaction object group and initial interaction amount associated with each resource transaction record. The resource interaction object group associated with a resource transaction record refers to a resource transferor and a resource receiver that have engaged in resource interaction. The initial interaction amount refers to the amount of resources transferred between the resource transferor and the resource receiver within a single resource transaction record.

[0035] Furthermore, if at least two resource transaction records belonging to the same resource interaction object group are detected, the at least two resource transaction records belonging to the same resource interaction object group are aggregated, that is, the at least two initial interaction quantities belonging to the same resource interaction object group are superimposed to obtain the resource interaction quantity associated with the resource interaction object group.

[0036] For example, as shown in Table 1 below (a summary table of resource transaction records within a preset time period), multiple resource transaction records included in the resource interaction data within the preset time period are provided:

[0037] Table 1. Summary of resource transaction records within the preset time period

[0038]

[0039] As shown in Table 1, by summarizing at least two resource transaction records belonging to the same resource interaction object group, Table 1 can only display the summarized resource interaction volume (e.g., the resource interaction volume between resource transfer object A and resource receiving object B) for the same resource interaction object group (e.g., the resource interaction volume between resource transfer object A and resource receiving object B is 100). It does not display multiple resource transaction records corresponding to the same resource interaction object group separately, thereby reducing the number of resource transaction records that need to be monitored and calculated, simplifying the subsequent processing flow of resource interaction data, and reducing cumbersome operations.

[0040] Step S204: Construct a resource interaction network based on each resource interaction object group and the resource interaction amount associated with the resource interaction object group, and determine the resource transfer probability matrix corresponding to the resource interaction network.

[0041] Specifically, for each resource interaction object group, the server needs to determine the first resource receiving object and the first resource transferring object in each resource interaction object group, and further determine the resource transfer direction between the first resource receiving object and the first resource transferring object.

[0042] It is understandable that the same resource interaction object can be both a resource sender and a resource receiver in different resource interaction object groups. For example, consider resource interaction objects A through E. In one resource interaction object group, resource interaction object B is a resource receiver, while in another group, resource interaction object B is a resource sender. Specifically, in one resource interaction object group, there are first resource sender object B and first resource receiver object A, and the resource transfer direction between them is B→A. In the other resource interaction object group, there are first resource sender object A and first resource receiver object B, and the resource transfer direction between them is A→B.

[0043] Furthermore, after determining the first resource receiving object and the first resource sending object included in each resource interaction object group, as well as the resource transfer direction between the first resource receiving object and the first resource sending object, the server treats each first resource receiving object and each first resource sending object as network nodes in the resource interaction network. The resource transfer direction between the first resource receiving object and the first resource sending object is used as the data flow between network nodes. For example, if the resource transfer direction between the first resource sending object A and the first resource receiving object B is A→B, then A→B can be used as the data flow between network node A and network node B. A directed connection is established between network node A and network node B, resulting in a directed edge between network node A and network node B. The resource interaction amount between the first resource sending object A and the first resource receiving object B is used as the weight of the directed edge between network node A and network node B. By constructing directed connections and setting the weights of directed edges for each resource interaction object group in sequence, a resource interaction network corresponding to the resource interaction data can be constructed.

[0044] In an exemplary embodiment, after the server constructs the resource interaction network, it needs to further determine the resource transfer probability matrix corresponding to the resource interaction network, so that object node sampling processing can be performed based on the resource interaction network and the resource transfer probability matrix.

[0045] For a resource interaction network consisting of N network nodes, the resource transfer probability of each of the N network nodes can be determined. By summing up the resource transfer probabilities of the N network nodes, an N-order matrix M can be constructed as the resource transfer probability matrix corresponding to the resource interaction network.

[0046] Specifically, for each of the N network nodes in the resource interaction network, the resource transfer probability corresponding to the network node needs to be determined based on at least one directed edge corresponding to the network node and the weight corresponding to the at least one directed edge. Thus, an N-order resource transfer probability matrix corresponding to the resource interaction network can be constructed based on the resource transfer probabilities corresponding to the N network nodes.

[0047] Step S206: Based on the resource interaction network and resource transfer probability matrix, perform object node sampling processing to obtain the object node sequence.

[0048] Specifically, when sampling object nodes for the resource interaction network and the resource transfer probability matrix, a network node should be randomly sampled with the same probability as the starting object node for the resource interaction network. For example, if the resource interaction network includes network nodes A to E, and network node A is randomly sampled with the same probability, then network node A will be used as the starting object node.

[0049] After determining the initial object node, the resource transfer vector corresponding to the initial object node is determined from the resource transfer probability matrix. For example, a row of resource transfer probabilities corresponding to network node A is determined from the resource transfer probability matrix as the resource transfer vector corresponding to network node A. Further, based on the multiple resource transfer probabilities corresponding to network node A, for example, when the resource interaction network includes network nodes A to E, the multiple resource transfer probabilities corresponding to network node A refer to the 5 resource transfer probabilities corresponding to network nodes A to E.

[0050] Furthermore, after determining the resource transfer probability vector corresponding to the initial object node from the resource transfer probability matrix, such as the resource transfer probability vector corresponding to network node A, the resource interaction network is sampled based on the resource transfer probability vector corresponding to network node A, i.e., the five resource transfer probabilities corresponding to network nodes A to E, to obtain the next object node. For example, when the resource interaction network is sampled based on the five resource transfer probabilities corresponding to network nodes A to E, the next object node obtained is network node D.

[0051] Since the length of the object node sequence is preset, i.e. the preset number threshold of object nodes included in the object node sequence, the object node sampling process needs to be repeated until the number of object nodes obtained by sampling reaches the preset number threshold, at which point the object node sampling process is stopped and the object node sequence is obtained.

[0052] Furthermore, by repeatedly executing the step of sampling object nodes based on the resource interaction network and resource transition probability matrix to obtain object node sequences, the process continues until the number of obtained object node sequences reaches a preset sequence number threshold. At this point, the process stops, for example, obtaining T object node sequences. The preset sequence number threshold can be set and adjusted according to actual needs and specific application scenarios, and is not limited to a single or certain values.

[0053] Specifically, this involves treating a sequence of T object nodes as a whole, that is, summarizing all the object nodes in the T object node sequences to obtain a single overall object node sequence.

[0054] Step S208: Obtain the node encoding vector corresponding to each object node in the object node sequence, construct node encoding vector pairs based on each node encoding vector, and obtain the risk label of each node encoding vector pair.

[0055] Specifically, the server obtains the initial node vectors corresponding to each object node in the object node sequence, trains the initial word vector model based on each initial node vector, obtains the trained word vector model, and obtains the model parameter matrix corresponding to the trained word vector model, so as to determine the node encoding vector corresponding to each object node based on the model parameter matrix.

[0056] The process of training the initial word vector model based on each initial node vector can be understood as the initial word vector model learning the embedding vector representation (i.e., the embedding representation, which refers to representing high-dimensional data (such as text, images, audio, etc.) with low-dimensional vectors) corresponding to each initial node vector. The training of the initial word vector model specifically refers to adjusting the model parameters of the initial word vector model. After obtaining the trained word vector model, it is necessary to obtain the model parameter matrix corresponding to the trained word vector model and use the model parameter matrix corresponding to the trained word vector model as the node encoding vector matrix corresponding to the object node sequence.

[0057] Specifically, in the model parameter matrix corresponding to the trained word vector model, each column of the N-dimensional vector corresponds to a reduced-dimensional representation of an initial node vector. In other words, each column of the node encoding vector matrix corresponding to the object node sequence corresponds to a node encoding vector for an object node. Therefore, by using the model parameter matrix corresponding to the trained word vector model as the node encoding vector matrix corresponding to the object node sequence, the node encoding vector for each object node can be determined based on the node encoding vector matrix.

[0058] For example, such as Figure 3 As shown, a word vector network model is provided, referring to... Figure 3As can be seen, the word vector network model can specifically be a word2vec model (i.e., a deep learning model used to generate word vectors), which includes an Input Layer -> a Hidden Layer -> an Output Layer. Each initial node vector serves as the input data to the Input Layer of the word2vec model. The output data of the Input Layer is then input into the Hidden Layer of the word2vec model to adjust its parameters, resulting in the adjusted model parameter matrix W. The Output Layer of the word2vec model outputs the node encoding vectors after encoding the initial node vectors; specifically, the adjusted model parameter matrix W is output as the node encoding vector matrix.

[0059] Furthermore, the server constructs node encoding vector pairs based on the encoding vectors of each node. That is, the server determines the resource transfer relationship between the encoding vectors of each node. For example, if the object node A corresponding to node encoding vector A transfers resources to the object node B corresponding to node encoding vector B, then the object node A and the node encoding vector B corresponding to node encoding vector A with resource transfer relationship can be regarded as a node encoding vector pair. Multiple node encoding vector pairs are obtained in sequence, and the risk label corresponding to each node encoding vector pair is obtained. For example, a risk label of 0 indicates no resource interaction risk, while a risk label of 1 indicates resource interaction risk.

[0060] For example, for node encoding vectors A to E, there exist the following: object node A transfers resources to object node B, object node A transfers resources to object node C, object node A transfers resources to object node D, object node B transfers resources to object node A, object node B transfers resources to object node C, object node D transfers resources to object node E, object node C transfers resources to object node A, object node C transfers resources to object node E, object node D transfers resources to object node B, object node E transfers resources to object node A, and object node E transfers resources to object node D. The resulting node encoding vector pairs specifically include: A→B, A→C, A→D, B→A, B→C, D→E, C→A, C→E, D→B, E→A, and E→D.

[0061] Each node encoding vector pair has a corresponding risk label, specifically including: a risk label of 0, indicating no resource interaction risk, and a risk label of 1, indicating resource interaction risk. For example, if the risk label for node encoding vector pair A→B is 0, it means there is no resource interaction risk for node encoding vector pair A→B; and if the risk label for node encoding vector pair A→C is 1, it means there is resource interaction risk for node encoding vector pair A→B.

[0062] Step S210: Train the initial interactive risk prediction model based on the encoding vector pairs of each node and their respective risk labels to obtain the trained interactive risk prediction model.

[0063] Specifically, the server can train the initial interaction risk prediction model based on the encoding vector pairs of each node and the risk labels corresponding to each node encoding vector. When the model training termination condition is met, the trained interaction risk prediction model is obtained.

[0064] In the above-mentioned resource interaction risk prediction and processing method, resource interaction data within a preset time period is acquired. Based on the resource interaction data, multiple resource interaction object groups and the resource interaction quantities associated with the resource interaction object groups are determined. Based on each resource interaction object group and the resource interaction quantities associated with the resource interaction object groups, a resource interaction network is constructed. By determining the resource transition probability matrix corresponding to the resource interaction network, object node sampling processing is performed based on the resource interaction network and the resource transition probability matrix. This allows the acquisition of an object node sequence containing the specific location information of the resource interaction objects in the resource interaction network, while simultaneously determining the structural characteristics and resource interaction records of the resource interaction objects in the resource interaction network. Furthermore, by obtaining the node encoding vector corresponding to each object node in the object node sequence, constructing node encoding vector pairs based on each node encoding vector, and obtaining the risk label of each node encoding vector pair, the initial interaction risk prediction model is trained based on each node encoding vector and its respective risk label to obtain a trained interaction risk prediction model. This model can then be used to predict the interaction risks of different resource interaction objects, accurately and quickly identify the interaction risks of the corresponding resource interaction objects, and promptly implement risk control and resource interaction blocking. This improves the accuracy of interaction risk prediction during resource interaction business processing and effectively reduces resource losses for the resource interaction platform and other resource interaction objects.

[0065] In one exemplary embodiment, such as Figure 4 As shown, the steps for constructing a resource interaction network, namely, constructing the resource interaction network based on each resource interaction object group and the resource interaction volume associated with the resource interaction object group, specifically include the following steps S402 to S406. Wherein:

[0066] Step S402: Based on each resource interaction object group, determine the first resource receiving object, the first resource transferring object, and the resource transfer direction in each resource interaction object group.

[0067] Specifically, for each resource interaction object group, the server determines the first resource receiving object, the first resource sending object, and the resource transfer direction between the first resource receiving object and the first resource sending object within each resource interaction object group. The same resource interaction object can be either a resource sending object or a resource receiving object in different resource interaction object groups.

[0068] For example, specifically including resource interaction objects A to B, in one group of resource interaction objects, resource interaction object A is the resource sender, while in another group, resource interaction object A is the resource receiver. For instance, if one group of resource interaction objects includes a first resource sender A and a first resource receiver B, the resource transfer direction between them is A→B. Similarly, if another group of resource interaction objects includes a first resource sender B and a first resource receiver A, the resource transfer direction between them is B→A.

[0069] Step S404: The first resource receiving object and the first resource sending object in each resource interaction object group are determined as network nodes of the resource interaction network.

[0070] In constructing a resource interaction network, it is necessary to determine the network nodes included in the resource interaction network, the connection relationships and directions between the network nodes, and the weights of the directed edges formed by two network nodes that are connected. Thus, the resource interaction network can be constructed based on the network nodes included in the resource interaction network, the connection relationships and directions between the network nodes, and the weights of the directed edges formed by two network nodes that are connected.

[0071] Specifically, after determining the first resource receiving object and the first resource sending object in each resource interaction object group, the server identifies each first resource receiving object and each first resource sending object as a network node of the resource interaction network.

[0072] For example, referring to Table 1 (a summary of resource transaction records within a preset time period), it can be seen that there are 11 resource interaction object groups and the resource transfer directions within each resource interaction object group, including: First resource transferor A → First resource receiver B, First resource transferor A → First resource receiver C, First resource transferor A → First resource receiver D, First resource transferor B → First resource receiver A, First resource transferor B → First resource receiver C, First resource transferor D → First resource receiver E, First resource transferor C → First resource receiver A, First resource transferor C → First resource receiver E, First resource transferor D → First resource receiver B, First resource transferor E → First resource receiver A, and First resource transferor E → First resource receiver D.

[0073] Therefore, for the above 11 resource interaction object groups, it can be determined that there are first resource transfer objects A to E, and first resource receiving objects A to E. Since the same resource interaction object can be both a resource transfer object and a resource receiving object in different resource interaction object groups, it can be determined that the network nodes included in the resource interaction network are specifically resource interaction objects A to E. Since resource interaction objects A to E can each serve as the first resource transfer object and the first resource receiving object, it is also necessary to determine the resource transfer direction between any two of resource interaction objects A to E.

[0074] Step S406: According to the resource transfer direction in each resource interaction object group, establish directed connections between each network node to obtain the directed edges of the resource interaction network, determine the resource interaction quantity as the weight of the directed edges, and construct the resource interaction network.

[0075] Specifically, after determining the network nodes included in the resource interaction network, the server needs to establish directed connections between the network nodes and directed edges of the resource interaction network according to the resource transfer direction in each resource interaction object group.

[0076] For example, if the network nodes in the resource interaction network are specifically resource interaction objects A to E, a directed connection needs to be established between any two resource interaction objects A to E based on the resource transfer direction between any two of them. This will yield the directed edges of the resource interaction network, including A→B, A→C, A→D, B→A, B→C, D→E, C→A, C→E, D→B, E→A, and E→D.

[0077] Furthermore, the amount of resource interaction between the first resource transferor and the first resource receiver in each resource interaction object group is determined as the weight of the corresponding directed edge to construct a resource interaction network. For example, referring to Table 1 (a summary of resource transaction records within a preset time period), the resource interaction amount (i.e., weight) between A and B can be determined to be 100, the resource interaction amount (i.e., weight) between A and C to be 200, the resource interaction amount (i.e., weight) between A and D to be 300, the resource interaction amount (i.e., weight) between B and A to be 400, the resource interaction amount (i.e., weight) between B and C to be 500, the resource interaction amount (i.e., weight) between D and E to be 600, the resource interaction amount (i.e., weight) between C and A to be 700, the resource interaction amount (i.e., weight) between C and E to be 800, the resource interaction amount (i.e., weight) between D and B to be 900, the resource interaction amount (i.e., weight) between E and A to be 1000, and the resource interaction amount (i.e., weight) between E and D to be 1000.

[0078] In one exemplary embodiment, such as Figure 5 As shown, a resource interaction network constructed for resource interaction data is provided, referring to... Figure 5 It can be seen that the network nodes included in the resource interaction network are specifically network nodes A to E. Based on the resource transfer direction between any two network nodes A to E, a directed connection is established between any two network nodes A to E, resulting in the directed edges of the resource interaction network, including A→B, A→C, A→D, B→A, B→C, D→E, C→A, C→E, D→B, E→A, and E→D.

[0079] Furthermore, referring to Figure 5 It can be seen that the weight between A and B is 100, the weight between A and C is 200, the weight between A and D is 300, the weight between B and A is 400, the weight between B and C is 500, the weight between D and E is 600, the weight between C and A is 700, the weight between C and E is 800, the weight between D and B is 900, the weight between E and A is 1000, and the weight between E and D is 1000.

[0080] In an exemplary embodiment, after constructing the resource interaction network, the resource transfer probability matrix corresponding to the resource interaction network is further determined, including:

[0081] For each network node in the resource interaction network, the resource transfer probability corresponding to the network node is determined based on at least one directed edge corresponding to the network node and the weight corresponding to the at least one directed edge. Based on the resource transfer probabilities corresponding to each network node, a resource transfer probability matrix corresponding to the resource interaction network is constructed. Specifically, for each network node in the resource interaction network, including network nodes A to E, the server needs to determine the resource transfer probability corresponding to the network node based on at least one directed edge corresponding to the network node and the weight corresponding to the at least one directed edge.

[0082] For example, for network node A, in the resource interaction network, the directed edges corresponding to network node A include A→B, A→C, A→D, B→A, C→A, and E→A. The weight between A→B is 100, the weight between A→C is 200, the weight between A→D is 300, the weight between B→A is 400, the weight between C→A is 700, and the weight between E→A is 1000. Specifically, the resource transfer probability corresponding to network node A is determined based on the directed edges corresponding to network node A, including A→B, A→C, A→D, B→A, C→A, and E→A, and the weights of each directed edge.

[0083] In an exemplary embodiment, if the resource interaction network includes N network nodes, then the resource transfer probability matrix corresponding to the resource interaction network is specifically an N-order matrix M, which is represented by the following formula (1):

[0084] ;Formula (1)

[0085] Where, m ij Let represent the element in the i-th row and j-th column of an N-order matrix M, which is the resource transfer probability corresponding to a network node in the resource interaction network. N indicates that the resource interaction network includes N network nodes.

[0086] Furthermore, the element m in the i-th row and j-th column of the N-order matrix M ij (That is, the resource transfer probability corresponding to the network node in the resource interaction network), which is specifically determined by the following formula (2):

[0087] ;Formula (2)

[0088] Where, m ij This represents the percentage of resource interactions where resource interaction object i interacts with resource interaction object j. In a resource interaction network, it can be understood as the probability of a transition from network node i to network node j, where m is the number of resource interactions. ijThe larger the value of , the greater the proportion of resource interaction volume of network node i in the resource interaction network, and the more active the resource interaction operations related to it.

[0089] Furthermore, after determining the resource transfer probability corresponding to each network node in the resource interaction network, the server can construct a resource transfer probability matrix corresponding to the resource interaction network based on the resource transfer probability of each network node. For example, for... Figure 5 The resource transfer probability matrix M1 obtained from the resource interaction network shown is represented by the following formula (3):

[0090] ;Formula (3)

[0091] Referring to formula (3), it can be seen that in formula (3) This represents the resource transfer probability vector corresponding to network node A. Similarly, each row in the resource transfer probability matrix M1 corresponds to the resource transfer probability vector from network node A to network node E.

[0092] In this embodiment, by determining the first resource receiving object, the first resource sending object, and the resource transfer direction in each resource interaction object group, the first resource receiving object and the first resource sending object in each resource interaction object group are identified as network nodes of the resource interaction network. Directed connections are established between the network nodes according to the resource transfer direction in each resource interaction object group, resulting in directed edges of the resource interaction network. The resource interaction quantity is determined as the weight of the directed edges, thus constructing the resource interaction network. This allows for a comprehensive display of the resource interaction relationships and quantities between different resource interaction objects in the resource interaction data, avoiding data omissions. Furthermore, after further determining the resource transfer probability matrix between different resource interaction objects based on the resource interaction network, the resource transfer probability between resource interaction objects can be more accurately determined. Resource interaction objects with abnormal resource transfer probabilities can be monitored and their resource interactions blocked in advance, effectively reducing resource losses for the resource interaction platform and other resource interaction objects.

[0093] In one exemplary embodiment, such as Figure 6 As shown, the steps for obtaining the object node sequence, namely, the steps of sampling object nodes based on the resource interaction network and the resource transition probability matrix to obtain the object node sequence, specifically include steps S602 to S608. Wherein:

[0094] Step S602: Determine the object node based on the resource interaction network.

[0095] Specifically, when the server performs object node sampling, it needs to randomly sample a network node with the same probability as the starting object node for the resource interaction network. For example, if the resource interaction network includes network nodes A to E, and network node A is randomly sampled with the same probability, then network node A will be used as the starting object node.

[0096] Step S604: Determine the probability transfer vector corresponding to the object node from the resource transfer probability matrix, and perform object node sampling processing on the resource interaction network according to the resource transfer probabilities in the probability transfer vector to obtain the next object node.

[0097] Specifically, after determining the initial object node, the server determines the resource transfer vector corresponding to the initial object node from the resource transfer probability matrix. For example, it determines the row of resource transfer probabilities corresponding to network node A from the resource transfer probability matrix M1. , which serves as the resource transfer vector corresponding to network node A.

[0098] Among them, for each resource transfer probability in the resource transfer vector corresponding to network node A, including This allows us to determine the probability that, after collecting data from network node A, the next network node collected will be one of the following: network node A through network node E. For example, the probability of collecting data from network node B is... The probability of collecting network node C is The probability of collecting network node D is The probability of collecting network node A and network node F is 0.

[0099] Furthermore, after determining the resource transfer vector corresponding to the initial object node, the resource interaction network is sampled according to the resource transfer probabilities in the probability transfer vector to obtain the next object node, for example, according to the resource transfer vector corresponding to network node A. The next object node obtained is network node D.

[0100] Step S606: Repeat step S604 to determine whether the number of obtained object nodes has reached the preset number threshold.

[0101] Specifically, the process involves repeatedly executing the steps of determining the probability transfer vector corresponding to the object node from the resource transfer probability matrix, sampling the object node in the resource interaction network according to the resource transfer probability in the probability transfer vector, obtaining the next object node, counting the number of object nodes obtained, and determining whether the number of object nodes obtained reaches a preset threshold.

[0102] The preset quantity threshold refers to the threshold number of object nodes included in the object node sequence. It is related to the pre-set length of the object node sequence, meaning that the preset quantity threshold is determined based on the pre-set length of the object node sequence. The length of the object node sequence can be set and adjusted according to actual needs or specific application scenarios, and is not limited to one or more specific values.

[0103] Step S608: If the number of obtained object nodes reaches a preset threshold, then an object node sequence is constructed based on each object node.

[0104] Specifically, by obtaining the length of a pre-set object node sequence and determining a preset quantity threshold based on the length of the object node sequence, the obtained number of object nodes is compared with the preset quantity threshold. If it is determined that the obtained number of object nodes reaches the preset quantity threshold, the object node processing is stopped, and an object node sequence is constructed based on each object node.

[0105] For example, if the preset quantity threshold is set to 10, then based on the resource interaction network and the resource transfer probability matrix, object node sampling processing is performed, and the resulting object node sequence can specifically be: [A, D, B, C, E, A, C, A, B, C].

[0106] In an exemplary embodiment, the process of obtaining object node sequences by repeatedly performing object node sampling processing based on the resource interaction network and resource transfer probability matrix is ​​stopped until the number of obtained object node sequences reaches a preset sequence number threshold. For example, if T object node sequences are obtained, the subsequent process is to perform node encoding vector conversion and risk label determination on the overall object node sequence composed of the T object node sequences.

[0107] In this embodiment, object nodes are determined based on the resource interaction network. From the resource transfer probability matrix, the probability transfer vector corresponding to the object node is determined. According to the resource transfer probabilities in the probability transfer vector, object node sampling processing is performed on the resource interaction network to obtain the next object node. This process is repeated until the number of obtained object nodes reaches a preset threshold. Then, based on each object node, an object node sequence is constructed, realizing resource-based... By sampling object nodes using the source interaction network and resource transfer probability matrix, a sequence of object nodes containing the specific location information of resource interaction objects in the resource interaction network can be obtained. At the same time, the structural characteristics and resource interaction records of resource interaction objects in the resource interaction network can be determined, and comprehensive and rich association information with resource interaction objects can be obtained. This allows for the further acquisition of node encoding vectors that include resource interaction objects and their association information, thereby improving the prediction accuracy of the interaction risk prediction model trained using these node encoding vectors. This enables accurate and rapid identification of the interaction risks existing in the corresponding resource interaction objects, allowing for timely risk management and resource interaction blocking.

[0108] In one exemplary embodiment, such as Figure 7 As shown, the steps for training the interactive risk prediction model, namely, training the initial interactive risk prediction model based on the encoding vector pairs of each node and their respective risk labels, to obtain the trained interactive risk prediction model, specifically include the following steps S702 to S704. Wherein:

[0109] Step S702: Determine the model training loss value during the model training process based on the encoding vector pairs of each node and the risk labels of each encoding vector pair.

[0110] Specifically, the server trains the initial interactive risk prediction model based on the encoding vector pairs of each node and the risk labels of each encoding vector pair, and determines the model training loss value during the model training process.

[0111] For example, the model loss function value L of the initial interactive risk prediction model is represented by the following formula (4):

[0112] ;Formula (4)

[0113] Where L represents the model training loss value of the initial interaction risk prediction model during the training process, S represents the number of all node encoding vector pairs, and p iThe initial interaction risk prediction model predicts the risk of node encoding vector pair i, y. i It is the risk label (which can be 0 or 1) of the sample node encoding vector pair i.

[0114] Step S704: Adjust the parameters of the initial interaction risk prediction model based on the model training loss value to obtain the trained interaction risk prediction model.

[0115] Specifically, the server adjusts the parameters of the initial interaction risk prediction model based on the determined model training loss value until the model parameters meet the model training termination condition, thus obtaining a trained interaction risk prediction model. The training process for the model parameters employs gradient descent.

[0116] Once the trained interactive risk prediction model is obtained, its performance needs to be evaluated to determine if it meets expectations. Specifically, evaluation metrics include IV (Information Value, used for feature selection and variable screening), AUC (Area Under Curve, used to measure the model's ability to distinguish between positive and negative samples; AUC values ​​range from 0.5 to 1, with values ​​closer to 1 indicating better model performance), and KS (Kolmogorov-Smirnov, used to evaluate the performance of binary classification models, primarily assessing the model's ability to distinguish between positive and negative examples). If the model performs as expected, it can be deployed for actual prediction tasks. If the model does not perform as expected, it needs to be retrained.

[0117] In one exemplary embodiment, such as Figure 8 As shown, an initial interaction risk prediction model is provided, referring to... Figure 8 It can be seen that the initial interaction risk prediction model can be a fully connected neural network model composed of multiple fully connected layers (FC). Its input data are node encoding vector pairs and risk labels for the node encoding vector pairs. A node encoding vector pair usually includes the node encoding vector of the resource receiving object and the node encoding vector of the resource transferring object. The output data of the initial interaction risk prediction model represents the risk prediction result for the node encoding vector pair (including the node encoding vector of the resource receiving object and the node encoding vector of the resource transferring object).

[0118] The risk prediction result can be a specific value in (0,1), such as 0.2, 0.5, 0.8, etc., to indicate the degree of resource interaction risk. For example, a risk prediction result of 0.8 means that when the node's encoding vector interacts with the corresponding resource interaction object group, there is an 80% chance of resource interaction risk.

[0119] In this embodiment, by utilizing the node encoding vector pairs and their risk labels, the model training loss value is determined during the model training process. Based on this loss value, the parameters of the initial interaction risk prediction model can be adjusted to obtain a trained interaction risk prediction model. This allows the initial interaction risk prediction model to learn the resource interaction records, resource interaction amounts, and the existence of interaction risks carried by each node encoding vector pair, thereby improving the model accuracy of the obtained interaction risk prediction model. This further enhances the accuracy of interaction risk prediction during resource interaction business processing using the interaction risk prediction model.

[0120] In one exemplary embodiment, such as Figure 9 As shown, a resource interaction risk prediction and processing method is provided, which is applied to... Figure 1 Taking server 104 as an example, the explanation includes the following steps S902 to S906. Wherein:

[0121] Step S902: Receive a resource interaction risk prediction request and obtain the second resource receiving object and the second resource transferring object corresponding to the resource interaction risk prediction request.

[0122] Specifically, the resource interaction object can trigger a resource interaction risk prediction request based on the terminal. When the terminal detects the resource interaction risk prediction request, it sends the resource interaction risk prediction request to the server. After receiving the resource interaction risk prediction request, the server can obtain the second resource receiving object and the second resource transferring object corresponding to the resource interaction risk prediction request by parsing the resource interaction risk prediction request.

[0123] Step S904: Determine the node encoding vector corresponding to the second resource receiving object and the node encoding vector corresponding to the second resource sending object, and concatenate the node encoding vector corresponding to the second resource receiving object and the node encoding vector corresponding to the second resource sending object to obtain the target node encoding vector.

[0124] Specifically, the server determines the initial encoding vector corresponding to the second resource receiving object and the initial encoding vector corresponding to the second resource sending object, and uses the model parameter matrix corresponding to the trained word vector model to determine the node encoding vectors corresponding to the initial encoding vectors corresponding to the second resource receiving object and the initial encoding vectors corresponding to the second resource sending object, thus obtaining the node encoding vectors corresponding to the second resource receiving object and the node encoding vectors corresponding to the second resource sending object.

[0125] Furthermore, by concatenating the node encoding vector corresponding to the second resource receiving object and the node encoding vector corresponding to the second resource sending object, the target node encoding vector for inputting into the trained interactive risk prediction model can be obtained.

[0126] For example, such as Figure 10 As shown, this provides a schematic diagram for concatenating the target node's encoding vector, according to... Figure 10 It can be seen that by concatenating the node encoding vector corresponding to the second resource receiving object and the node encoding vector corresponding to the second resource sending object, the target node encoding vector of the input trained interactive risk prediction model can be obtained.

[0127] Step S906: Based on the trained interactive risk prediction model, perform risk prediction processing on the target node encoding vector to obtain the risk prediction result corresponding to the target node encoding vector.

[0128] Specifically, the server inputs the spliced ​​target node encoding into the trained interactive risk prediction model, and then uses the trained interactive risk prediction model to perform risk prediction processing on the target node encoding vector to obtain the risk prediction result corresponding to the target node encoding vector.

[0129] The risk prediction result can be a specific value in (0,1), such as 0.2, 0.5, 0.8, etc., to indicate the degree of resource interaction risk. For example, a risk prediction result of 0.8 means that when the node's encoding vector interacts with the corresponding resource interaction object group, there is an 80% chance of resource interaction risk.

[0130] In one exemplary embodiment, such as Figure 11 As shown, a process for risk prediction based on an interactive risk prediction model is provided, referring to... Figure 11 It can be seen that the process of risk prediction and processing based on the interactive risk prediction model specifically includes the deployment stage, the prediction and inference stage, and the disposal stage, among which:

[0131] A. Deployment phase: (1) Save the mapping relationship between resource interaction objects and node encoding vectors: The node encoding vectors corresponding to the resource receiving object and the resource sending object, determined by the trained word vector model, are saved in the form of (resource interaction object identifier, node encoding vector) data pairs. This will solidify the mapping relationship between resource interaction objects and node encoding vectors in the system. When risk prediction processing is required later, the node encoding vector corresponding to the resource interaction object identifier of the resource interaction object to be processed for risk prediction can be obtained for subsequent risk prediction processing. (2) Model deployment: The interaction risk prediction model is trained and used to perform inference prediction when the actual resource interaction operation occurs, and the risk prediction result is obtained.

[0132] B. Prediction and Inference Stage: When the actual resource interaction operation occurs: (1) Obtain the resource interaction object identifiers of the resource interaction objects involved in the resource interaction operation, including the resource transfer object identifier corresponding to the resource transfer object and the resource receiving object identifier corresponding to the resource receiving object. (2) Determine the node encoding vector according to the resource interaction object identifier: Based on the stored data pairs of (resource interaction object identifier, node encoding vector), according to the resource interaction object identifier, including the resource transfer object identifier and the resource receiving object identifier, obtain the node encoding vector corresponding to the resource transfer object identifier and the node encoding vector corresponding to the resource receiving object identifier. (3) Model input data splicing: Splice the node encoding vector corresponding to the resource transfer object identifier and the node encoding vector corresponding to the resource receiving object identifier to obtain the target node encoding vector of the input interaction risk prediction model. (4) Model inference prediction: Input the spliced ​​target node encoding vector into the trained interaction risk prediction model, perform inference prediction calculation, and obtain the risk prediction result. The output of the interaction risk prediction model, i.e. the risk prediction result, can be any value in (0,1). This value represents the degree of resource interaction risk of the current resource interaction operation. The larger the value of the risk prediction result, the higher the degree of resource interaction risk of the current resource interaction operation.

[0133] C. Handling Phase: Based on the risk prediction results, the current resource interaction operation is blocked or handled. Specifically, if the risk prediction result is greater than a preset threshold p (e.g., 0.7), the current resource interaction operation is determined to be high-risk, and the risk control system blocks the current resource interaction operation.

[0134] Specifically, the system can provide reminders and intercept resource interactions for resource transfer recipients, indicating that there is a risk of resource interaction and preventing resource loss for the transfer recipients. Alternatively, it can restrict resource reception for resource receiving recipients to reduce resource losses for the resource interaction platform and other resource interaction recipients.

[0135] In the aforementioned resource interaction risk prediction and processing method, a resource interaction risk prediction request is received, and the second resource receiving object and the second resource transferring object corresponding to the request are obtained. The node encoding vectors corresponding to the second resource receiving object and the second resource transferring object are determined. These vectors are then concatenated to obtain the target node encoding vector. A trained interaction risk prediction model is then used to perform risk prediction processing on the target node encoding vector, obtaining the corresponding risk prediction result. This method enables the use of the interaction risk prediction model to predict the interaction risks of different resource interaction objects, accurately and quickly identifying the interaction risks present in the corresponding resource interaction objects, and promptly implementing risk control and resource interaction blocking. This improves the accuracy of interaction risk prediction during resource interaction business processing and effectively reduces resource losses for the resource interaction platform and other resource interaction objects.

[0136] In one exemplary embodiment, such as Figure 12 As shown, a resource interaction risk prediction and processing method is provided, which is applied to... Figure 1 Taking server 104 as an example, the explanation includes the following steps S1202 to S1215. Wherein:

[0137] Step S1201: Obtain resource interaction data within a preset time period, and determine the resource interaction object group and the resource interaction volume associated with the resource interaction object group based on the resource interaction data.

[0138] Specifically, the resource interaction data within the preset time period includes multiple resource transaction records. The server parses these records to obtain the resource interaction object group and initial interaction volume associated with each transaction. The resource interaction object group associated with a transaction record refers to a resource transferor and a resource receiver that engage in resource interaction. The initial interaction volume refers to the amount of resources transferred between the resource transferor and receiver within a single transaction record.

[0139] Furthermore, if at least two resource transaction records belonging to the same resource interaction object group are detected, the at least two resource transaction records belonging to the same resource interaction object group are aggregated, that is, the at least two initial interaction quantities belonging to the same resource interaction object group are superimposed to obtain the resource interaction quantity associated with the resource interaction object group.

[0140] Step S1202: Based on each resource interaction object group, determine the first resource receiving object, the first resource transferring object, and the resource transfer direction in each resource interaction object group.

[0141] Specifically, the same resource interaction object can be either a resource transferor or a resource receiver in different resource interaction object groups. Therefore, for each resource interaction object group, it is necessary to determine the first resource receiver, the first resource transferor, and the resource transfer direction in each resource interaction object group. For example, resource interaction object B may be a resource receiver in one resource interaction object group, while it may be a resource transferor in another resource interaction object group.

[0142] Step S1203: The first resource receiving object and the first resource sending object in each resource interaction object group are determined as network nodes of the resource interaction network.

[0143] Specifically, the server determines the first resource receiving object and the first resource sending object included in each resource interaction object group, and treats each first resource receiving object and each first resource sending object as network nodes in the resource interaction network.

[0144] Step S1204: According to the resource transfer direction in each resource interaction object group, establish directed connections between each network node to obtain the directed edges of the resource interaction network, determine the resource interaction quantity as the weight of the directed edges, and construct the resource interaction network.

[0145] Specifically, the server treats each first resource receiving object and each first resource sending object as network nodes in the resource interaction network, and the resource transfer direction between the first resource receiving object and the first resource sending object as the data flow between network nodes. Thus, directed connections can be established between network nodes according to the resource transfer direction in each resource interaction object group, resulting in directed edges of the resource interaction network. The resource interaction amount is determined as the weight of the directed edge. By constructing directed connections and setting the weights of the directed edges for each resource interaction object group in sequence, a resource interaction network corresponding to the resource interaction data can be constructed.

[0146] Step S1205: For each network node in the resource interaction network, determine the resource transfer probability corresponding to the network node based on at least one directed edge corresponding to the network node and the weight corresponding to the at least one directed edge.

[0147] Specifically, for each network node in the resource interaction network, including network node A to network node E, the server needs to determine the resource transfer probability corresponding to the network node based on at least one directed edge corresponding to the network node and the weight corresponding to at least one directed edge.

[0148] Step S1206: Based on the resource transfer probability corresponding to each network node, construct the resource transfer probability matrix corresponding to the resource interaction network.

[0149] Specifically, after determining the resource transfer probability of each network node in the resource interaction network, the server can construct a resource transfer probability matrix corresponding to the resource interaction network based on the resource transfer probability of each network node.

[0150] Step S1207: Determine the object node based on the resource interaction network.

[0151] Specifically, when the server performs object node sampling, it needs to randomly sample a network node with the same probability as the starting object node for the resource interaction network. For example, if the resource interaction network includes network nodes A to E, and network node A is randomly sampled with the same probability, then network node A will be used as the starting object node.

[0152] Step S1208: Determine the probability transfer vector corresponding to the object node from the resource transfer probability matrix, and perform object node sampling processing on the resource interaction network according to the resource transfer probabilities in the probability transfer vector to obtain the next object node.

[0153] Specifically, after determining the initial object node, the server determines the resource transfer vector corresponding to the initial object node from the resource transfer probability matrix. For example, it determines the row of resource transfer probabilities corresponding to network node A from the resource transfer probability matrix M1, and uses it as the resource transfer vector corresponding to network node A.

[0154] Step S1209: Repeat step S1208 and determine whether the number of obtained object nodes reaches the preset number threshold.

[0155] Specifically, the process involves repeatedly executing the steps of determining the probability transfer vector corresponding to the object node from the resource transfer probability matrix, sampling the object node in the resource interaction network according to the resource transfer probability in the probability transfer vector, obtaining the next object node, counting the number of object nodes obtained, and determining whether the number of object nodes obtained reaches a preset threshold.

[0156] Step S1210: If it is determined that the number of obtained object nodes reaches a preset number threshold, then an object node sequence is constructed based on each object node.

[0157] Specifically, by obtaining the length of a pre-set object node sequence and determining a preset quantity threshold based on the length of the object node sequence, the obtained number of object nodes is compared with the preset quantity threshold. If it is determined that the obtained number of object nodes reaches the preset quantity threshold, the object node processing is stopped, and an object node sequence is constructed based on each object node.

[0158] Step S1211: Obtain the node encoding vector corresponding to each object node in the object node sequence, construct node encoding vector pairs based on each node encoding vector, and obtain the risk label of each node encoding vector pair.

[0159] Specifically, the server obtains the initial node vectors corresponding to each object node in the object node sequence, trains the initial word vector model based on each initial node vector, obtains the trained word vector model, and obtains the model parameter matrix corresponding to the trained word vector model, so as to determine the node encoding vector corresponding to each object node based on the model parameter matrix.

[0160] Furthermore, the server constructs node encoding vector pairs based on the encoding vectors of each node. That is, the server determines the resource transfer relationship between the encoding vectors of each node. For example, if the object node A corresponding to node encoding vector A transfers resources to the object node B corresponding to node encoding vector B, then the object node A and the node encoding vector B corresponding to node encoding vector A with resource transfer relationship can be regarded as a node encoding vector pair. Multiple node encoding vector pairs are obtained in sequence, and the risk label corresponding to each node encoding vector pair is obtained. For example, a risk label of 0 indicates no resource interaction risk, while a risk label of 1 indicates resource interaction risk.

[0161] Step S1212: Based on the encoding vector pairs of each node and the risk labels of each encoding vector pair, determine the model training loss value during the model training process, and adjust the parameters of the initial interaction risk prediction model according to the model training loss value to obtain the trained interaction risk prediction model.

[0162] Specifically, the server can train the initial interaction risk prediction model based on the encoding vector pairs of each node and the risk labels corresponding to each node encoding vector. When the model training termination condition is met, the trained interaction risk prediction model is obtained.

[0163] Step S1213: Receive a resource interaction risk prediction request and obtain the second resource receiving object and the second resource transferring object corresponding to the resource interaction risk prediction request.

[0164] Specifically, after receiving a resource interaction risk prediction request, the server can obtain the second resource receiving object and the second resource transferring object corresponding to the resource interaction risk prediction request by parsing the resource interaction risk prediction request.

[0165] Step S1214: Determine the node encoding vector corresponding to the second resource receiving object and the node encoding vector corresponding to the second resource sending object, and concatenate the node encoding vector corresponding to the second resource receiving object and the node encoding vector corresponding to the second resource sending object to obtain the target node encoding vector.

[0166] Specifically, the server determines the initial encoding vector corresponding to the second resource receiving object and the second resource sending object, and uses the model parameter matrix corresponding to the trained word vector model to determine the node encoding vectors corresponding to the initial encoding vectors corresponding to the second resource receiving object and the second resource sending object, respectively. In other words, it obtains the node encoding vectors corresponding to the second resource receiving object and the second resource sending object. By concatenating the node encoding vectors corresponding to the second resource receiving object and the second resource sending object, the target node encoding vector used as input to the trained interactive risk prediction model can be obtained.

[0167] Step S1215: Based on the trained interactive risk prediction model, perform risk prediction processing on the target node encoding vector to obtain the risk prediction result corresponding to the target node encoding vector.

[0168] Specifically, the server inputs the spliced ​​target node encoding into the trained interactive risk prediction model, and then uses the trained interactive risk prediction model to perform risk prediction processing on the target node encoding vector to obtain the risk prediction result corresponding to the target node encoding vector.

[0169] In one exemplary embodiment, such as Figure 13 As shown, this paper presents the overall processing procedure of a resource interaction risk prediction and processing method, referring to... Figure 13 The overall processing procedure of the resource interaction risk prediction method includes: P1 determining the resource interaction object group and the resource interaction quantity associated with the resource interaction object group; P2 constructing the resource interaction network; P3 determining the resource transfer probability matrix; P4 processing the object nodes to obtain the object node sequence; P5 constructing node encoding vector pairs based on the object node sequence; P6 training the initial interaction risk prediction model based on the node encoding vector pairs; P7 evaluating the model training effect; and P8 performing risk prediction processing based on the trained interaction risk prediction model. Among these:

[0170] P1 determines the resource interaction object group and resource interaction volume: The server acquires and parses resource interaction data within a preset time period to obtain multiple resource transaction records included in the resource interaction data, and determines the resource interaction object group and initial interaction volume associated with each resource transaction record. When at least two resource transaction records belonging to the same resource interaction object group are detected, the at least two resource transaction records belonging to the same resource interaction object group are aggregated, that is, the at least two initial interaction volumes belonging to the same resource interaction object group are superimposed to obtain the resource interaction volume associated with that resource interaction object group.

[0171] P2 Constructing a Resource Interaction Network: For each resource interaction object group, it is necessary to determine the first resource receiving object and the first resource sending object in each resource interaction object group, and further determine the resource transfer direction between the first resource receiving object and the first resource sending object. Each first resource receiving object and each first resource sending object is regarded as a network node in the resource interaction network, and the resource transfer direction between the first resource receiving object and the first resource sending object is regarded as the data flow between network nodes. According to the resource transfer direction in each resource interaction object group, directed connections are established between each network node to obtain the directed edges of the resource interaction network, and the resource interaction amount is determined as the weight of the directed edge. By constructing directed connections and setting the weight of the directed edges for each resource interaction object group in sequence, the resource interaction network corresponding to the resource interaction data can be constructed.

[0172] P3 determines the resource transfer probability matrix: For each network node in the resource interaction network, the resource transfer probability corresponding to the network node is determined based on at least one directed edge corresponding to the network node and the weight corresponding to at least one directed edge. Based on the resource transfer probability corresponding to each network node, the resource transfer probability matrix corresponding to the resource interaction network is constructed.

[0173] P4 Object Node Sampling to Obtain the Object Node Sequence: For the resource interaction network, a network node is randomly sampled with the same probability as the starting object node. From the resource transition probability matrix, the resource transition vector corresponding to the initial object node is determined. According to the resource transition probabilities in the probability transition vector, the resource interaction network is sampled to obtain the next object node. By repeatedly executing the steps of determining the probability transition vector corresponding to the object node from the resource transition probability matrix, and sampling the object node according to the resource transition probabilities in the probability transition vector to obtain the next object node, the number of obtained object nodes is counted. When the number of obtained object nodes reaches a preset threshold, an object node sequence is constructed based on each object node.

[0174] P5 constructs node encoding vector pairs: by obtaining the initial node vectors corresponding to each object node in the object node sequence, and training the initial word vector model based on each initial node vector, a trained word vector model is obtained, and the model parameter matrix corresponding to the trained word vector model is obtained. Based on the model parameter matrix, the node encoding vectors corresponding to each object node are determined, and node encoding vector pairs are constructed based on each node encoding vector. The risk label corresponding to each node encoding vector pair is also obtained. For example, a risk label of 0 indicates no risk of resource interaction, while a risk label of 1 indicates a risk of resource interaction.

[0175] P6 trains the initial interaction risk prediction model based on the node encoding vector pairs: Based on each node encoding vector pair and its risk label, the model training loss value is determined during the model training process, and the parameters of the initial interaction risk prediction model are adjusted based on the model training loss value to obtain the trained interaction risk prediction model.

[0176] P7 Model Training Performance Evaluation: After obtaining the trained interactive risk prediction model, evaluate whether the model's performance meets expectations based on the evaluation metrics used to assess the model. If the model performs as expected, it can be deployed and used for actual prediction tasks. If the model does not perform as expected, it needs to be retrained.

[0177] P8 performs risk prediction based on the interaction risk prediction model: After receiving a resource interaction risk prediction request, the system parses the request to obtain the second resource receiving object and the second resource sending object corresponding to the request. It then determines the node encoding vectors corresponding to the second resource receiving object and the second resource sending object, concatenates these vectors, and obtains the target node encoding vector. Based on the trained interaction risk prediction model, the system performs risk prediction processing on the target node encoding vector to obtain the corresponding risk prediction result.

[0178] In the aforementioned resource interaction risk prediction and processing method, when sampling object nodes based on the resource interaction network and resource transition probability matrix, a sequence of object nodes containing the specific location information of resource interaction objects in the resource interaction network can be obtained. Simultaneously, the structural characteristics and resource interaction records of the resource interaction objects in the resource interaction network are determined. Furthermore, by obtaining the node encoding vector corresponding to each object node in the object node sequence, node encoding vector pairs are constructed based on each node encoding vector, and risk labels for each node encoding vector pair are obtained. Based on each node encoding vector and its respective risk label, the initial interaction risk prediction model is trained to obtain a trained interaction risk prediction model, which can then be used to predict resource interaction risks. When predicting a request, the target node encoding vector is obtained by concatenating the node encoding vector of the second resource receiving object and the node encoding vector of the second resource sending object in the resource interaction risk prediction request. This allows the trained interaction risk prediction model to perform risk prediction processing on the target node encoding vector, obtaining the corresponding risk prediction result. This enables the use of the interaction risk prediction model to predict the interaction risks of different resource interaction objects, accurately and quickly identifying the interaction risks present in the corresponding resource interaction objects, and promptly implementing risk control and resource interaction blocking. This improves the accuracy of interaction risk prediction during resource interaction business processing and effectively reduces resource losses for the resource interaction platform and other resource interaction objects.

[0179] It should be understood that although the steps in the flowcharts of the above embodiments are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the above embodiments may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0180] Based on the same inventive concept, this application also provides a resource interaction risk prediction and processing apparatus for implementing the resource interaction risk prediction and processing method described above. The solution provided by this apparatus is similar to the implementation scheme described in the above method; therefore, the specific limitations in one or more resource interaction risk prediction and processing apparatus embodiments provided below can be found in the limitations of the resource interaction risk prediction and processing method described above, and will not be repeated here.

[0181] In one exemplary embodiment, such as Figure 14As shown, a resource interaction risk prediction and processing device is provided, including: a resource interaction data acquisition module 1402, a resource transfer probability matrix determination module 1404, an object node sequence acquisition module 1406, a risk label acquisition module 1408, and an interaction risk prediction model acquisition module 1410, wherein:

[0182] The resource interaction data acquisition module 1402 is used to acquire resource interaction data within a preset time period, and determine multiple resource interaction object groups and the resource interaction volume associated with the resource interaction object groups based on the resource interaction data; the resource transfer probability matrix determination module 1404 is used to construct a resource interaction network based on each resource interaction object group and the resource interaction volume associated with the resource interaction object group, and determine the resource transfer probability matrix corresponding to the resource interaction network; the object node sequence acquisition module 1406 is used to perform object node sampling processing based on the resource interaction network and the resource transfer probability matrix to obtain an object node sequence; the risk label acquisition module 1408 is used to acquire the node encoding vector corresponding to each object node in the object node sequence, construct node encoding vector pairs based on each node encoding vector, and obtain the risk label of each node encoding vector pair; the interaction risk prediction model acquisition module 1410 is used to train the initial interaction risk prediction model based on each node encoding vector pair and the risk label of each node encoding vector pair to obtain the trained interaction risk prediction model.

[0183] In the aforementioned resource interaction risk prediction and processing device, resource interaction data within a preset time period is acquired. Based on the resource interaction data, multiple resource interaction object groups and the resource interaction quantities associated with the resource interaction object groups are determined. A resource interaction network is constructed based on each resource interaction object group and the resource interaction quantities associated with the resource interaction object groups. By determining the resource transfer probability matrix corresponding to the resource interaction network, object node sampling processing is performed based on the resource interaction network and the resource transfer probability matrix. This allows the acquisition of an object node sequence containing the specific location information of the resource interaction objects in the resource interaction network, while simultaneously determining the structural characteristics and resource interaction records of the resource interaction objects in the resource interaction network. Furthermore, by obtaining the node encoding vector corresponding to each object node in the object node sequence, constructing node encoding vector pairs based on each node encoding vector, and obtaining the risk label of each node encoding vector pair, the initial interaction risk prediction model is trained based on each node encoding vector and its respective risk label to obtain a trained interaction risk prediction model. This model can then be used to predict the interaction risks of different resource interaction objects, accurately and quickly identify the interaction risks of the corresponding resource interaction objects, and promptly implement risk control and resource interaction blocking. This improves the accuracy of interaction risk prediction during resource interaction business processing and effectively reduces resource losses for the resource interaction platform and other resource interaction objects.

[0184] In an exemplary embodiment, the resource transfer probability matrix determination module is further configured to: determine the first resource receiving object, the first resource transferring object, and the resource transfer direction in each resource interaction object group according to each resource interaction object group; determine the first resource receiving object and the first resource transferring object in each resource interaction object group as network nodes of the resource interaction network; establish directed connections between each network node according to the resource transfer direction in each resource interaction object group to obtain directed edges of the resource interaction network; determine the resource interaction amount as the weight of the directed edges; and construct the resource interaction network.

[0185] In an exemplary embodiment, the resource transfer probability matrix determination module is further configured to: for each network node in the resource interaction network, determine the resource transfer probability corresponding to the network node based on at least one directed edge corresponding to the network node and the weight corresponding to the at least one directed edge; and construct a resource transfer probability matrix corresponding to the resource interaction network based on the resource transfer probability corresponding to each network node. In an exemplary embodiment, the object node sequence acquisition module is further configured to: determine object nodes based on the resource interaction network; determine the probability transfer vector corresponding to the object node from the resource transfer probability matrix; perform object node sampling processing on the resource interaction network according to each resource transfer probability in the probability transfer vector to obtain the next object node; repeatedly execute the steps of determining the probability transfer vector corresponding to the object node from the resource transfer probability matrix and performing object node sampling processing on the resource interaction network according to each resource transfer probability in the probability transfer vector to obtain the next object node until the number of obtained object nodes reaches a preset number threshold; and construct an object node sequence based on each object node.

[0186] In an exemplary embodiment, the interactive risk prediction model acquisition module is further configured to: determine the model training loss value during the model training process based on the node encoding vector pairs and the risk labels of the node encoding vector pairs; and adjust the parameters of the initial interactive risk prediction model based on the model training loss value to obtain the trained interactive risk prediction model.

[0187] In an exemplary embodiment, a resource interaction risk prediction processing apparatus is provided, further comprising a risk prediction module, configured to: receive a resource interaction risk prediction request; obtain a second resource receiving object and a second resource transferring object corresponding to the resource interaction risk prediction request; determine a node encoding vector corresponding to the second resource receiving object and a node encoding vector corresponding to the second resource transferring object; concatenate the node encoding vector corresponding to the second resource receiving object and the node encoding vector corresponding to the second resource transferring object to obtain a target node encoding vector; and perform risk prediction processing on the target node encoding vector according to a trained interaction risk prediction model to obtain a risk prediction result corresponding to the target node encoding vector.

[0188] Each module in the aforementioned resource interaction risk prediction and processing device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the corresponding operations of each module.

[0189] In one exemplary embodiment, a computer device is provided, which may be a server or a terminal. Taking the computer device as a server as an example, its internal structure diagram can be as follows: Figure 15 As shown, this computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operating system and computer programs stored in the non-volatile storage media. The database stores resource interaction data for a preset time period, resource interaction object groups, resource interaction quantities associated with resource interaction object groups, resource interaction networks, resource transition probability matrices corresponding to the resource interaction networks, object node sequences, node encoding vectors, risk labels for node encoding vector pairs formed by node encoding vectors, initial interaction risk prediction models, and trained interaction risk prediction models. The I / O interfaces are used for exchanging information between the processor and external devices. The communication interface is used for communicating with external terminals via a network connection. When the computer program is executed by the processor, it implements a resource interaction risk prediction and processing method.

[0190] Those skilled in the art will understand that Figure 15 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0191] In one embodiment, a computer device is also provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above method embodiments.

[0192] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon that, when executed by a processor, implements the steps in the above method embodiments.

[0193] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above method embodiments.

[0194] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0195] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile memory and volatile memory. Non-volatile memory may include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory may include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, artificial intelligence (AI) processors, etc., and are not limited to these.

[0196] The technical features of the above embodiments can be combined arbitrarily. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as the combination of these technical features does not contradict each other, it should be considered within the scope of this application. The above embodiments only illustrate several implementation methods of this application, and their descriptions are relatively specific and detailed, but they should not be construed as limiting the scope of this application. It should be noted that for those skilled in the art, several modifications and improvements can be made without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A method for predicting and processing resource interaction risks, characterized in that, The method includes: Obtain resource interaction data within a preset time period, and determine multiple resource interaction object groups and the resource interaction volume associated with the resource interaction object groups based on the resource interaction data; Based on each of the resource interaction object groups and the amount of resource interaction associated with the resource interaction object groups, a resource interaction network is constructed, and a resource transfer probability matrix corresponding to the resource interaction network is determined. Based on the resource interaction network and the resource transfer probability matrix, object node sampling processing is performed to obtain an object node sequence; Obtain the node encoding vector corresponding to each object node in the object node sequence, construct node encoding vector pairs based on each node encoding vector, and obtain the risk label of each node encoding vector pair; The initial interaction risk prediction model is trained based on the node encoding vector pairs and their respective risk labels to obtain a trained interaction risk prediction model.

2. The method according to claim 1, characterized in that, The step of constructing a resource interaction network based on each of the resource interaction object groups and the resource interaction volume associated with the resource interaction object groups includes: Based on each of the resource interaction object groups, determine the first resource receiving object, the first resource transferring object, and the resource transfer direction in each of the resource interaction object groups; The first resource receiving object and the first resource sending object in each of the resource interaction object groups are determined as network nodes of the resource interaction network. According to the resource transfer direction in each of the resource interaction object groups, a directed connection is established between each of the network nodes to obtain the directed edges of the resource interaction network. The resource interaction amount is determined as the weight of the directed edge to construct the resource interaction network.

3. The method according to claim 2, characterized in that, Determining the resource transfer probability matrix corresponding to the resource interaction network includes: For each network node in the resource interaction network, the resource transfer probability corresponding to the network node is determined based on at least one directed edge corresponding to the network node and the weight corresponding to the at least one directed edge. Based on the resource transfer probability corresponding to each network node, a resource transfer probability matrix corresponding to the resource interaction network is constructed.

4. The method according to any one of claims 1 to 3, characterized in that, The step of sampling object nodes based on the resource interaction network and the resource transfer probability matrix to obtain an object node sequence includes: The object node is determined based on the resource interaction network; From the resource transfer probability matrix, determine the probability transfer vector corresponding to the object node, and perform object node sampling processing on the resource interaction network according to each resource transfer probability in the probability transfer vector to obtain the next object node; Repeat the steps of determining the probability transfer vector corresponding to the object node from the resource transfer probability matrix, sampling the object node of the resource interaction network according to each resource transfer probability in the probability transfer vector, and obtaining the next object node until the number of obtained object nodes reaches a preset number threshold. Based on each of the object nodes, a sequence of object nodes is constructed.

5. The method according to any one of claims 1 to 3, characterized in that, The step of training the initial interaction risk prediction model based on each node encoding vector pair and its respective risk label to obtain a trained interaction risk prediction model includes: Based on each node encoding vector pair and the risk label of each node encoding vector pair, the model training loss value is determined during the model training process. Based on the training loss value of the model, the parameters of the initial interaction risk prediction model are adjusted to obtain a trained interaction risk prediction model.

6. The method according to any one of claims 1 to 3, characterized in that, The method further includes: Receive a resource interaction risk prediction request and obtain a second resource receiving object and a second resource transferring object corresponding to the resource interaction risk prediction request; Determine the node encoding vector corresponding to the second resource receiving object and the node encoding vector corresponding to the second resource sending object, and concatenate the node encoding vector corresponding to the second resource receiving object and the node encoding vector corresponding to the second resource sending object to obtain the target node encoding vector; Based on the trained interactive risk prediction model, risk prediction processing is performed on the target node encoding vector to obtain the risk prediction result corresponding to the target node encoding vector.

7. A resource interaction risk prediction and processing device, characterized in that, The device includes: The resource interaction data acquisition module is used to acquire resource interaction data within a preset time period, and determine multiple resource interaction object groups and the resource interaction volume associated with the resource interaction object groups based on the resource interaction data. The resource transfer probability matrix determination module is used to construct a resource interaction network based on each resource interaction object group and the resource interaction amount associated with the resource interaction object group, and to determine the resource transfer probability matrix corresponding to the resource interaction network. The object node sequence acquisition module is used to perform object node sampling processing based on the resource interaction network and the resource transfer probability matrix to obtain the object node sequence; The risk label acquisition module is used to acquire the node encoding vector corresponding to each object node in the object node sequence, construct node encoding vector pairs based on each node encoding vector, and acquire the risk label of each node encoding vector pair; The interactive risk prediction model acquisition module is used to train the initial interactive risk prediction model based on the node encoding vector pairs and their respective risk labels, so as to obtain the trained interactive risk prediction model.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.

Images