Abnormal application user determination method and device, equipment, medium and program product

Abstract

The invention discloses a method, a device and equipment for determining an abnormal application user, a medium and a program product, and relates to the technical field of artificial intelligence and financial science and technology. The method comprises the following steps: in response to a first application instruction of a first user, determining each second user associated with the first user; respectively determining first structured data of the first user and second structured data of each second user; determining first feature data of the first user based on the first structured data and the second structured data, and generating first label data of the first user according to the first feature data; and inputting the first structured data and the first tag data into a target large model obtained by fine tuning in advance to obtain an output result, and determining whether the first user is an abnormal application user or not according to the output result. According to the scheme, the abnormal application user can be quickly and accurately determined, and help is provided for preventing abnormal application in the financial field.

Description

Technical Field

[0001] The present invention relates to the fields of artificial intelligence and financial technology, and in particular to a method, apparatus, device, medium and program product for determining abnormal applicants. Background Art

[0002] In recent years, in the financial field, abnormal applications such as false loan applications, credit card fraud applications, and applications for opening accounts under false names have occurred frequently; these abnormal applications are not only a threat to the interests of financial institutions themselves, but also pose a challenge to the healthy and stable development of the entire financial market.

[0003] How to quickly and accurately identify abnormal application users and prevent abnormal applications from occurring is a key research issue in the industry. Summary of the Invention

[0004] The present invention provides a method, apparatus, device, medium and program product for determining abnormal applicants, so as to quickly and accurately determine abnormal applicants and provide assistance in preventing the occurrence of abnormal applications in the financial field.

[0005] According to one aspect of the present invention, a method for determining abnormal applicants is provided, the method comprising:

[0006] In response to a first application instruction of a first user, determining second users associated with the first user;

[0007] determining first structured data of the first user and second structured data of each of the second users respectively;

[0008] determining first feature data of the first user based on the first structured data and each of the second structured data, and generating first tag data of the first user according to the first feature data;

[0009] The first structured data and the first label data are input into a pre-fine-tuned target large model to obtain an output result, and whether the first user is an abnormal application user is determined based on the output result.

[0010] According to another aspect of the present invention, a device for determining an abnormal applicant user is provided, the device comprising:

[0011] a first determining module, configured to determine second users associated with a first user in response to a first application instruction of the first user;

[0012] a second determining module, configured to respectively determine the first structured data of the first user and the second structured data of each of the second users;

[0013] a third determining module, configured to determine first feature data of the first user based on the first structured data and each of the second structured data, and generate first tag data of the first user according to the first feature data;

[0014] The fourth determination module is used to input the first structured data and the first label data into the first large model obtained by pre-fine-tuning to obtain an output result, and determine whether the first user is an abnormal application user based on the output result.

[0015] According to another aspect of the present invention, an electronic device is provided, comprising:

[0016] at least one processor; and

[0017] a memory communicatively connected to the at least one processor; wherein,

[0018] The memory stores a computer program that can be executed by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute the method for determining abnormal applicant users described in any embodiment of the present invention.

[0019] According to another aspect of the present invention, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a processor to implement the method for determining abnormal applicant users described in any embodiment of the present invention when executed.

[0020] According to another aspect of the present invention, a computer program product is provided, comprising a computer program, wherein when the computer program is executed by a processor, the computer program implements the method for determining an abnormal applicant user according to any embodiment of the present invention.

[0021] The technical solution of an embodiment of the present invention determines each second user associated with the first user in response to a first application instruction of the first user; determines the first structured data of the first user and the second structured data of each second user respectively; determines the first feature data of the first user based on the first structured data and the second structured data, and generates the first label data of the first user according to the first feature data; inputs the first structured data and the first label data into a pre-fine-tuned target large model to obtain an output result, and determines whether the first user is an abnormal application user based on the output result. Abnormal application users can be quickly and accurately identified, which helps to prevent the occurrence of abnormal applications in the financial field.

[0022] It should be understood that the content described in this section is not intended to identify the key or important features of the embodiments of the present invention, nor is it intended to limit the scope of the present invention. Other features of the present invention will become readily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0024] Figure 1 This is a flowchart of a method for determining abnormal applicants provided in accordance with the first embodiment of the present invention;

[0025] Figure 2 This is a flow chart of a fine-tuning process of a target large model according to the second embodiment of the present invention;

[0026] Figure 3 This is a flowchart of a method for determining abnormal applicants provided in accordance with the third embodiment of the present invention;

[0027] Figure 4 This is a schematic diagram of the structure of a device for determining abnormal applicants provided in accordance with a fourth embodiment of the present invention;

[0028] Figure 5 It is a structural diagram of an electronic device for implementing the method for determining abnormal applicants according to an embodiment of the present invention. DETAILED DESCRIPTION

[0029] In order to enable those skilled in the art to better understand the solutions of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of the present invention.

[0030] It should be noted that the terms "first", "second", etc. in the description and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that the numbers used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0031] Example 1

[0032] Figure 1 This is a flow chart of a method for determining abnormal applicants provided in accordance with the first embodiment of the present invention. This embodiment is applicable to quickly determining abnormal applicants. The method can be executed by an abnormal applicant determination device. The abnormal applicant determination device can be implemented in the form of hardware and / or software. The abnormal applicant determination device can be configured in a computer, server, or tablet electronic device. Figure 1 As shown, the method includes:

[0033] Step 110: In response to a first application instruction from a first user, determine each second user associated with the first user.

[0034] Among them, the first user can be an applicant for any financial business, such as a loan applicant, a debit card applicant, a credit card applicant, or a financial product purchase applicant, etc., which is not limited in this embodiment; correspondingly, in this embodiment, the first application instruction of the first user can be a loan application, debit card application, or credit card application issued by the first user, etc., which is not limited in this embodiment.

[0035] Optionally, in this embodiment, after receiving the first application instruction of the first user, the user information of the first user can be further obtained, such as the first user's user name, address, identification code or mobile phone number, etc.; the first user's historical application information can also be obtained, such as other applications related to the first application submitted by the first user in the past five years, and relevant information of these applications can be obtained, such as approval materials, agents or third-party personnel, etc.; the first user's historical transaction information can also be obtained, such as the payee of historical payment information, payment time or specific amount, etc., which are not limited in this embodiment.

[0036] Furthermore, the second user associated with the first user can be determined based on the above-mentioned information obtained; illustratively, the user associated with the first user (for example, a relative of the first user) can be determined based on the user name of the first user and determined as the second user; the agent or third-party personnel in the price history application information can also be determined as the second user; the payee of the payment information or the payee of the transfer information can also be determined as the second user, which is not limited in this embodiment.

[0037] It should be noted that, in this embodiment, the user information is obtained only after authorization by the user, and the method of obtaining it is reasonable and legal.

[0038] Step 120: Determine the first structured data of the first user and the second structured data of each of the second users.

[0039] Optional. In this embodiment, after obtaining the relevant information of the first user (e.g., the user information, historical application information, or historical transaction information involved in the above steps), the relevant information can be further preprocessed. For example, the above information can be cleaned, missing data can be filled, duplicate data can be deleted, or outliers can be processed, thereby obtaining processed relevant data. The processed relevant data can then be converted into structured data, i.e., obtaining first structured data of the first user.

[0040] Optionally, in this embodiment, after determining the second user associated with the first user, relevant information of each second user can be further obtained. For example, user information, historical application information or historical transaction information of each second user can be obtained, and the relevant information of each second user can be preprocessed to obtain the preprocessed relevant data, and the preprocessed relevant data can be structured to obtain the second structured data of each second user.

[0041] Step 130: Determine first feature data of the first user based on the first structured data and each piece of the second structured data, and generate first tag data of the first user according to the first feature data.

[0042] Optionally, in this embodiment, after obtaining the first structured data of the first user and the second structured data of each second user, the first feature data of the first user can be further determined based on the first structured data of the first user and the second structured data of each second user. Furthermore, the first label data of the first user can be generated based on the first feature data.

[0043] In an optional implementation of this embodiment, after obtaining the first structured data and each second structured data, the first structured data and each second structured data can be converted into feature data respectively, for example, the first structured data can be converted into a first feature matrix, and each second structured data can be converted into an adjacent-order feature matrix; further, the target feature can be obtained from each adjacent feature matrix, and each target feature can be added to the first feature matrix to obtain the first feature data of the first user; further, the first label data of the first user can be generated based on the first feature data.

[0044] In this embodiment, obtaining target features from each adjacent feature matrix may include obtaining features related to the first application from each adjacent feature, and determining the features as target features; after adding each target feature to the first feature matrix to obtain first feature data, the first feature data may be further identified or parsed to obtain first label data; it can be understood that the first label data is the annotation data of the target data after the first structured data is superimposed with important information of the second structured data, which may have one or more label data, and the specific number of label data is not limited in this embodiment.

[0045] In another optional implementation of this embodiment, after obtaining the first structured data and each second structured data, the first structured data and each second structured data can be respectively input into the graph neural network. Further, the first feature matrix and each adjacent-order feature matrix can be determined through the graph neural network; further, the first feature data of the first user can be determined based on the first feature matrix and each adjacent-order feature matrix, and first label data matching the first feature data can be generated.

[0046] Step 140: Input the first structured data and the first label data into the pre-fine-tuned target large model to obtain an output result, and determine whether the first user is an abnormal application user based on the output result.

[0047] Among them, the target large model can be a large model obtained by fine-tuning based on a large model obtained by pre-training such as a natural language processing large model or a multimodal model.

[0048] Optionally, after generating the first label data of the first user, the first label data and the first structured data of the first user can be further input into the pre-fine-tuned target large model. The large model processes the first label data and the first structured data to obtain a processing result, and outputs the processing result. Furthermore, it can be determined whether the first user is an abnormal application user based on the output result of the target large model.

[0049] The technical solution of this embodiment determines each second user associated with the first user in response to the first application instruction of the first user; determines the first structured data of the first user and the second structured data of each second user respectively; determines the first feature data of the first user based on the first structured data and the second structured data, and generates the first label data of the first user according to the first feature data; inputs the first structured data and the first label data into the target large model obtained by pre-fine-tuning to obtain an output result, and determines whether the first user is an abnormal application user based on the output result. Abnormal application users can be quickly and accurately identified, which helps to prevent the occurrence of abnormal applications in the financial field.

[0050] Example 2

[0051] Figure 2 This is a flow chart of the fine-tuning process of the target large model provided in the second embodiment of the present invention. This embodiment is a further refinement of the above technical solution. The technical solution in this embodiment can be combined with various optional solutions in one or more of the above embodiments. Figure 2 As shown in Figure 2, the fine-tuning process of the target large model includes:

[0052] Step 210: Obtain the target structured dataset.

[0053] The target structured data set includes: target users, historical transaction information of each target user, and relationship information of each user; the target users include at least one of the following: abnormal application users, normal application users, third-party users, and agent users.

[0054] Optionally, in this embodiment, user information of abnormal applicants, normal applicants, third-party users, and proxy users can be obtained, and historical application information and historical transaction information of each of the above users within a preset time period (for example, the past six months, the past year, or the past five years, etc., which are not limited in this embodiment) can also be obtained; further, the relationship information between each target user can be determined based on the historical transaction information of each target user; exemplarily, the relationship information may include: proxy information, entrustment information, guarantee information, transfer information, or social relationship information, etc., which are not limited in this embodiment.

[0055] Furthermore, the acquired user information, the historical transaction information of each target user, and the relationship information of each user can be preprocessed. For example, duplicate data or invalid data can be deleted, missing data can be supplemented, etc., so as to obtain the preprocessed data. Furthermore, the preprocessed data can be structured to obtain various structured data, and the various structured data are combined to obtain the target structured data set.

[0056] Step 220: Establish a target graph based on the target structured data set, and determine node label data of each node according to the target graph.

[0057] Among them, the number of target graphs is 1, the target graph can contain multiple nodes, and one node represents one user.

[0058] Optionally, in this embodiment, after obtaining the target structured data set, a target graph can be further established based on the structured data in the target structured data set. For example, one user can represent a vertex (node), and the vertices can be connected according to the relationship information between the users. For example, user A and user B are in an agent relationship, then the line between vertex A (user A) and vertex B (user B) can be connected. When the relationship information between all users is connected, the target graph can be obtained; further, the node label data of each node can be determined according to the target graph.

[0059] Optionally, in this embodiment, a target graph is established based on the target structured data set, and node label data of each node is determined according to the target graph, which may include: determining each target user as each node of the target graph, and determining the user information of each target user as the attribute of each node; connecting the edges of each node according to the historical transaction information of each target user and the relationship information of each target user; obtaining the target graph based on each node, the attributes of each node and each edge; inputting the target graph into a pre-trained target graph neural network to obtain the node features of the target node and the adjacency matrix of each reference node associated with the target node; generating the target node label data of the target node according to the node features of the target node and each adjacency matrix.

[0060] In an optional implementation of this embodiment, after obtaining the target structured data set, each target user in the target structured data set can be determined as a node of the target graph, that is, each target user can be determined as a vertex of the target graph, and the user information of each target user can also be determined as an attribute of each node; further, each node can be connected according to the historical transaction information of each target user and the relationship information of each target user, thereby forming each edge of the target graph, and finally obtaining a target graph that matches the target structured data set.

[0061] Furthermore, the obtained target graph can be input into a pre-trained target graph neural network, and the node features of the target node and the adjacency matrix of each reference node associated with the target node can be learned through the target graph neural network; wherein, the target node can be any node in the target graph, and the reference nodes associated with the target node can be nodes directly adjacent to the target node, or can be multi-hop nodes of the target node, that is, as long as the target node is connected by lines, other nodes can all be reference nodes associated with the target node.

[0062] Furthermore, the target graph neural network can generate target node label data of the target node based on the node features of the target node and the adjacency matrix of each reference node associated with the target node.

[0063] In this embodiment, by converting the acquired target structured data set into a target graph, other customers associated with the target user can be quickly and accurately determined; at the same time, the node features of each node are extracted through the graph neural network to generate node label data for each node. The labeled data of each data can be obtained without labeling the data of each user, which saves a lot of manpower costs, improves the efficiency of data labeling, and provides assistance for the fine-tuning of subsequent large models.

[0064] Step 230: Input the node label data and the target structured data set into the target large model to fine-tune the target large model.

[0065] Optionally, in this embodiment, after obtaining the label data of each node in the target graph, the obtained node label data of each node and each target structured data set can be further input into the target large model for iterative training. When the iteration stopping condition is met, the fine-tuned target large model is obtained.

[0066] It should be noted that, in this embodiment, the target large model can use the cross entropy loss function in combination with a regularization strategy to prevent model overfitting.

[0067] The solution of this embodiment is to obtain a target structured data set; establish a target graph based on the target structured data set, and determine the node label data of each node according to the target graph; input each of the node label data and the target structured data set into a target large model to fine-tune the target large model. The known large model can be fine-tuned based on the obtained target structured data set, thereby obtaining a target large model for identifying abnormal applicants in the financial field, which helps to improve the accuracy and efficiency of identifying abnormal applicants.

[0068] Example 3

[0069] Figure 3 This is a flow chart of a method for determining abnormal application users provided in Example 3 of the present invention. This embodiment is a further refinement of the above technical solution. The technical solution in this embodiment can be combined with various optional solutions in one or more of the above embodiments. Figure 3 As shown, the method includes:

[0070] Step 310: In response to a first application instruction from a first user, determine each second user associated with the first user.

[0071] Optionally, in this embodiment, determining each second user associated with the first user may include: obtaining historical application information and / or historical transaction information of the first user; determining a third-party user or an agent user based on the historical application information, and determining the third-party user or the agent user as the second user; and / or determining a transaction applicant or a transaction recipient based on the historical transaction information, and determining the transaction applicant or the transaction recipient as the second user.

[0072] In an optional implementation of this embodiment, after receiving the first application instruction of the first user, the historical application information and historical transaction information of the first user within a preset time period can be further obtained; further, the third-party user or agent user can be determined based on the historical application information, and further, the third-party user and the agent user can be determined as the second user; exemplarily, if the first user applied for a housing loan in the previous year, and the loan matters were all completed by agent user A, then agent user A is the second user associated with the first user; if the first user applied for a vehicle loan in the previous year, and the loan matter was guaranteed by third-party user B, then third-party user B is the second user associated with the first user.

[0073] In another optional implementation of this embodiment, after obtaining the historical transaction information of the first user, the historical transaction information can be further parsed. For example, the transaction applicant or the transaction recipient of the historical transaction information can be obtained. For example, if the historical transaction information is transfer information, then the transaction applicant can be the transfer initiator and the transaction recipient can be the payee; further, the transaction applicant or the transaction recipient can be determined as the second user.

[0074] In another optional implementation of this embodiment, after obtaining the historical transaction information of the transaction applicant or the transaction recipient, the historical transaction information of the transaction applicant or the transaction recipient can be further obtained, and the second user can be determined based on the historical transaction information of the transaction applicant or the transaction recipient. In this way, not only can the user directly associated with the first user be determined as the second user, but the user indirectly associated with the first user can also be determined as the second user, enriching the second users associated with the first user and providing assistance for the subsequent identification of abnormal applicant users.

[0075] The solution of this embodiment determines the second users associated with the first user based on the first user's historical application information and historical transaction information, and can quickly and accurately determine each second user, thereby helping to subsequently determine whether the first user is an abnormal application customer.

[0076] Step 320: Determine the user information, historical application information, and historical transaction information of the first user and each of the second users, respectively, and structure the user information, historical application information, and historical transaction information to obtain the first structured data and each of the second structured data.

[0077] Among them, the application information includes at least one of the following: timestamp, amount and frequency; exemplarily, if the historical application is a loan application, then the application information may include the time of submission of the loan application, the loan amount, and the number of loan applications; the historical transaction information includes at least one of the following: timestamp, transaction applicant, transaction recipient, amount and frequency; exemplarily, if the historical transaction information is remittance information, then the historical transaction information may include remittance time, remitter, payee, remittance finance and number of remittances, etc.

[0078] Optionally, in this embodiment, after determining each second user associated with the first user, user information of the first user and each second user may be further obtained. For example, the user name, residential address, contact information, or account information of the first user and each second user may be obtained, which is not limited in this embodiment. Historical application information and historical transaction information of the first user and each second user may also be obtained. Furthermore, the obtained information may be structured to obtain first structured data of the first user and second structured data of each second user.

[0079] The advantage of such a setting is that the structured data of the first user and each second user can be quickly obtained, providing a basis for the subsequent generation of the tag data of the first user.

[0080] Step 330: Input the first structured data and each of the second structured data into the target graph neural network to obtain the first user features of the first user and the second user features of each of the second users; generate the first label data of the first user based on the first user features and each of the second user features.

[0081] Optionally, in this embodiment, after obtaining the first structured data of the first user and the second structured data of each second user, the first structured data and each second structured data can be further input into the target graph neural network, so as to extract the first user features of the first user and the second user features of each second user based on the first structured data and each second structured data; further, the target graph neural network can generate the first label data of the first user based on the first user features and the second user features.

[0082] The advantage of this setting is that the first user's labeled data can be quickly obtained based on the graph neural network, which helps to determine whether the application submitted by the first user is an abnormal application.

[0083] Step 340: Input the first structured data and the first label data into the pre-fine-tuned target large model to obtain an output result, and determine whether the first user is an abnormal application user based on the output result.

[0084] The output result of the target large model is a probability value, for example, 0.6, 0.8 or 0.96, etc., which is not limited in this embodiment.

[0085] In an optional implementation of this embodiment, determining whether the first user is an abnormal applicant user according to the output result may include: when the probability value is greater than or equal to a set threshold, determining that the first user is an abnormal applicant user.

[0086] The threshold value may be set to a value such as 0.8 or 0.85, which is not limited in this embodiment.

[0087] Optionally, in this embodiment, after the first structured data and the first label data are input into the target large model, the target large model outputs a probability value that the application submitted by the first user is an abnormal application based on the input data; when the output probability value (0.9) is greater than the set threshold (0.8), it can be determined that the first user is an abnormal application customer; otherwise, the first user is a normal application customer.

[0088] The solution of this embodiment can quickly determine whether the application submitted by the first user is an abnormal application through the probability value output by the target large model, thereby improving the efficiency of identifying users with abnormal applications and effectively preventing the occurrence of abnormal applications in the financial field.

[0089] In order to better understand the method for determining abnormal applicants involved in this embodiment, a specific example is used below to explain it, which mainly includes the following:

[0090] 1. Data preparation

[0091] In this embodiment, known entity nodes such as loan fraud clients, personal loan clients, third-party counterparties, and transaction agents can be collected, as well as direct or indirect transfers, loan issuance records, repayment records, and transaction agent relationships between these entities. Furthermore, the acquired data can be cleaned to address missing, duplicate, and outlier values ​​to ensure data integrity and consistency.

[0092] 2. Graph Model Construction

[0093] In this embodiment, loan fraud customers, personal loan customers, third-party counterparties, transaction agents, etc. are regarded as entity nodes; the relationships between entity nodes (such as capital flow, transaction agents) are regarded as edges; the characteristics of entity nodes and edges; illustratively, they can include loan customer personal portrait labels, timestamps, amounts or frequencies, etc.; further, the above-mentioned entity nodes, edges, and attributes can be used to determine the graph model structure.

[0094] 3. Graph Construction

[0095] In this embodiment, based on the graph model determined in the above steps, knowledge extraction and point-edge relationship mapping can be performed on the cleaned data to complete the construction of the loan customer association relationship graph.

[0096] 4. Graph Neural Networks

[0097] In this embodiment, a graph neural network can be used to capture the structured information of entity nodes in the graph. First, the initial features of each graph model node and the adjacency matrix of the graph model are input. Then, messages are passed through a multi-layer graph neural network, and each layer includes two steps: information aggregation and feature conversion: collecting information from neighboring nodes and updating the representation of the current node. During the training process, node classification tasks are used for optimization so that the representation of the node can well reflect its category information and local structural features, and finally high-quality vector representations and category predictions for each node are obtained. These representations contain both the node's own features and the structural information of the graph.

[0098] 5. Labeling

[0099] Among them, labeling processing mainly converts the node representation learned by the graph neural network into a structured text description.

[0100] In this embodiment, a basic label containing the node's primary type information can be generated based on the entity node's category prediction results. Next, the node's local structural features are analyzed, and the relationship information of neighboring nodes is integrated into the label. A multi-level label template is designed, and the level of description detail is determined based on the node's importance and prediction confidence. In this embodiment, more attributes and contextual information can be added to key nodes, while a concise basic description is maintained for secondary nodes.

[0101] 6. Feature Fusion

[0102] In this embodiment, text input and graph structure labels can be organized into a suitable input format by using pre-prompts or interleaved insertion. Then, the enhanced information is processed by a specially designed graph neural network layer, which effectively integrates the structured knowledge contained in the labels. The specific steps are:

[0103] Input preparation phase: Design an appropriate input template and use pre-prompts or interleaved insertion to organize text input and graph structure labels into a suitable input format;

[0104] Feature processing stage: Information is processed through a specially designed graph neural network layer to achieve information interaction at the node level and structure level, ensuring that the structural knowledge of the graph is effectively preserved;

[0105] Fusion strategy stage: Design a feature fusion gating mechanism to balance the weights of original text and graph structure information, and use the attention mechanism through graph neural networks to dynamically adjust feature weights to adapt to different tasks;

[0106] Output processing stage: Generate the final fused representation and ensure that the output format meets the requirements of downstream tasks.

[0107] It should be noted that, in this embodiment, a cross entropy loss function can be used in combination with a regularization strategy to prevent overfitting.

[0108] The solution of this embodiment identifies known loan fraud customers, personal loan customers, third-party counterparties, transaction agents, etc. as entity nodes, and constructs a loan customer association graph through related elements such as direct or indirect transfers, loan records, repayment records, and transaction agency relationships. It also uses a graph neural network to learn node representations and perform labeling processing, effectively integrating labeled graph structure information and text information, enhancing the knowledge acquisition ability of the large model, and using the fact-based information of the graph to improve the output illusion of the large model. The solution of the embodiment of the present invention not only retains the structural advantages of the graph, but also fully utilizes the semantic understanding ability of the large model, improves the performance of the model on structured tasks, and can more comprehensively and reliably mine loan fraud customers.

[0109] Example 4

[0110] Figure 4 Schematic diagram of a device for determining abnormal applicants according to the fourth embodiment of the present invention. Figure 4 As shown, the apparatus includes: a first determination module 410 , a second determination module 420 , a third determination module 430 and a fourth determination module 440 .

[0111] The first determining module 410 is configured to determine each second user associated with the first user in response to the first application instruction of the first user;

[0112] A second determining module 420, configured to respectively determine the first structured data of the first user and the second structured data of each of the second users;

[0113] a third determining module 430, configured to determine first feature data of the first user based on the first structured data and each of the second structured data, and generate first tag data of the first user according to the first feature data;

[0114] The fourth determination module 440 is used to input the first structured data and the first label data into the first large model obtained by pre-fine-tuning to obtain an output result, and determine whether the first user is an abnormal application user based on the output result.

[0115] The solution of this embodiment is to determine each second user associated with the first user in response to the first application instruction of the first user through a first determination module; determine the first structured data of the first user and the second structured data of each second user respectively through a second determination module; determine the first feature data of the first user based on the first structured data and each second structured data through a third determination module, and generate the first label data of the first user based on the first feature data; input the first structured data and the first label data into the first large model obtained by pre-fine-tuning through a fourth determination module to obtain an output result, and determine whether the first user is an abnormal application user based on the output result. Abnormal application users can be quickly and accurately determined, which helps to prevent the occurrence of abnormal applications in the financial field.

[0116] In an optional implementation of this embodiment, the apparatus for determining abnormal applicants further includes a large model fine-tuning module configured to obtain a target structured data set; the target structured data set includes target users, historical transaction information of each target user, and relationship information of each user; the target users include at least one of the following: abnormal applicants, normal applicants, third-party users, and proxy users;

[0117] Establishing a target graph based on the target structured data set, and determining node label data of each node according to the target graph;

[0118] Input each of the node label data and the target structured data set into a target large model to fine-tune the target large model.

[0119] In an optional implementation of this embodiment, the large model fine-tuning module is further configured to determine each of the target users as a node of the target graph, and to determine the user information of each of the target users as an attribute of each of the nodes;

[0120] Connecting edges of the nodes according to the historical transaction information of each target user and the relationship information of each target user;

[0121] Obtaining the target graph based on each of the nodes, the attributes of each of the nodes, and each of the edges;

[0122] Inputting the target graph into a pre-trained target graph neural network to obtain node features of the target node and an adjacency matrix of each reference node associated with the target node;

[0123] Target node label data of the target node is generated according to the node characteristics of the target node and each of the adjacency matrices.

[0124] In an optional implementation of this embodiment, the first determining module 410 is specifically configured to obtain the historical application information and / or historical transaction information of the first user;

[0125] Determine a third-party user or a proxy user based on the historical application information, and determine the third-party user or the proxy user as the second user;

[0126] And / or, a transaction applicant or a transaction recipient is determined based on the historical transaction information, and the transaction applicant or the transaction recipient is determined as the second user.

[0127] In an optional implementation of this embodiment, the second determining module 420 is specifically configured to determine user information, historical application information, and historical transaction information of the first user and each of the second users, respectively, and perform structured representation of the user information, historical application information, and historical transaction information to obtain the first structured data and each of the second structured data;

[0128] The application information includes at least one of the following: a timestamp, an amount, and a frequency;

[0129] The historical transaction information includes at least one of the following: timestamp, transaction applicant, transaction recipient, amount, and frequency.

[0130] In an optional implementation of this embodiment, the third determining module 430 is specifically configured to input the first structured data and each of the second structured data into a target graph neural network to obtain a first user feature of the first user and a second user feature of each of the second users;

[0131] First label data of the first user is generated according to the first user feature and each of the second user features.

[0132] In an optional implementation of this embodiment, the output result is a probability value;

[0133] Correspondingly, the fourth determining module 440 is specifically configured to determine that the first user is an abnormal application user when the probability value is greater than or equal to a set threshold.

[0134] The device for determining abnormal applicants provided in the embodiment of the present invention can execute the method for determining abnormal applicants provided in any embodiment of the present invention, and has functional modules and beneficial effects corresponding to the execution method.

[0135] In the technical solutions of the embodiments of the present invention, the collection, storage, use, processing, transmission, provision and disclosure of user information (such as user name, residential address, contact information or account information, etc.) involved comply with the provisions of relevant laws and regulations and do not violate public order and good morals.

[0136] Example 5

[0137] Figure 5 A schematic diagram of the structure of an electronic device 10 that can be used to implement an embodiment of the present invention is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices (such as helmets, glasses, watches, etc.) and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present invention described and / or claimed herein.

[0138] like Figure 5As shown, the electronic device 10 includes at least one processor 11 and a memory, such as a read-only memory (ROM) 12, a random access memory (RAM) 13, etc., which is communicatively connected to the at least one processor 11. The memory stores a computer program that can be executed by the at least one processor. The processor 11 can perform various appropriate actions and processes according to the computer program stored in the read-only memory (ROM) 12 or the computer program loaded from the storage unit 18 into the random access memory (RAM) 13. Various programs and data required for the operation of the electronic device 10 can also be stored in the RAM 13. The processor 11, ROM 12, and RAM 13 are connected to each other via a bus 14. An input / output (I / O) interface 15 is also connected to the bus 14.

[0139] Multiple components in the electronic device 10 are connected to the I / O interface 15, including an input unit 16, such as a keyboard, a mouse, etc.; an output unit 17, such as various types of displays, speakers, etc.; a storage unit 18, such as a magnetic disk, an optical disk, etc.; and a communication unit 19, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 19 allows the electronic device 10 to exchange information / data with other devices via a computer network such as the Internet and / or various telecommunication networks.

[0140] The processor 11 can be various general and / or special processing components with processing and computing capabilities. Some examples of the processor 11 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various dedicated artificial intelligence (AI) computing chips, various processors that run machine learning model algorithms, digital signal processors (DSPs), and any appropriate processors, controllers, microcontrollers, etc. The processor 11 executes the various methods and processes described above, such as a method for determining abnormal application users, which includes: responding to a first application instruction of a first user, determining each second user associated with the first user; determining the first structured data of the first user and the second structured data of each second user respectively; determining the first feature data of the first user based on the first structured data and each second structured data, and generating the first label data of the first user based on the first feature data; inputting the first structured data and the first label data into the target large model obtained by pre-fine-tuning to obtain an output result, and determining whether the first user is an abnormal application user based on the output result.

[0141] In some embodiments, the method for determining abnormal applicant users can be implemented as a computer program, which is tangibly contained in a computer-readable storage medium, such as a storage unit 18. In some embodiments, part or all of the computer program can be loaded and / or installed on the electronic device 10 via the ROM 12 and / or the communication unit 19. When the computer program is loaded into the RAM 13 and executed by the processor 11, one or more steps of the method for determining abnormal applicant users described above can be performed. Alternatively, in other embodiments, the processor 11 can be configured to execute the method for determining abnormal applicant users in any other appropriate manner (for example, by means of firmware).

[0142] Various embodiments of the systems and techniques described herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), system-on-chip systems (SOCs), programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments can include being implemented in one or more computer programs that are executable and / or interpreted on a programmable system that includes at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0143] Computer programs for implementing the methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when the computer program is executed by the processor, the functions / operations specified in the flowcharts and / or block diagrams are implemented. The computer program may be executed entirely on the machine, partially on the machine, as a stand-alone software package, partially on the machine and partially on a remote machine, or entirely on a remote machine or server.

[0144] In the context of the present invention, computer-readable storage media can be tangible media that can contain or store a computer program for use with an instruction execution system, device or equipment or used in combination with an instruction execution system, device or equipment. Computer-readable storage media can include but are not limited to electronic, magnetic, optical, electromagnetic, infrared or semiconductor systems, devices or equipment, or any suitable combination of the foregoing. Alternatively, computer-readable storage media can be machine-readable signal media. More specific examples of machine-readable storage media can include electrical connections based on one or more lines, portable computer disks, hard disks, random access memories (RAM), read-only memories (ROM), erasable programmable read-only memories (EPROM or flash memory), optical fibers, portable compact disk read-only memories (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0145] To provide interaction with a user, the systems and techniques described herein can be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user can provide input to the electronic device. Other types of devices can also be used to provide interaction with the user; for example, the feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including acoustic input, voice input, or tactile input).

[0146] The systems and techniques described herein can be implemented in a computing system that includes back-end components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes front-end components (e.g., a user computer with a graphical user interface or web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such back-end components, middleware components, or front-end components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: a local area network (LAN), a wide area network (WAN), a blockchain network, and the Internet.

[0147] A computing system may include a client and a server. The client and server are typically remote from each other and typically interact via a communication network. This client-server relationship arises through computer programs running on the respective computers, creating a client-server relationship. The server may be a cloud server, also known as a cloud computing server or cloud host. This server is a hosting product within a cloud computing service ecosystem that addresses the management difficulties and limited scalability of traditional physical hosting and VPS services.

[0148] It should be understood that the various forms of the processes shown above can be used to reorder, add, or delete steps. For example, the steps described in the present invention can be performed in parallel, sequentially, or in a different order, as long as the desired results of the technical solution of the present invention can be achieved. This is not limited herein.

[0149] The above specific embodiments do not limit the scope of protection of the present invention. Those skilled in the art will appreciate that various modifications, combinations, sub-combinations, and substitutions may be made based on design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention are intended to be included within the scope of protection of the present invention.

[0150] Example 6

[0151] An embodiment of the present invention further provides a computer program product, including a computer program, which, when executed by a processor, implements the database detection method provided in any embodiment of the present application.

[0152] The computer program product may be implemented by writing computer program code for performing the operations of the present invention in one or more programming languages, or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, C++, and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a stand-alone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving a remote computer, the remote computer may be connected to the user's computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0153] It should be noted that in the embodiments of the present application, certain software, components, models and other existing solutions in the industry may be mentioned. They should be regarded as exemplary. Their purpose is only to illustrate the feasibility of implementing the technical solution of the present application, but it does not mean that the applicant has or will necessarily use the solution.

[0154] Note that the above are only preferred embodiments of the present invention and the technical principles employed. Those skilled in the art will appreciate that the present invention is not limited to the specific embodiments herein, and that various obvious changes, readjustments, and substitutions are possible for those skilled in the art without departing from the scope of protection of the present invention. Therefore, although the present invention has been described in detail through the above embodiments, the present invention is not limited to the above embodiments and may include many other equivalent embodiments without departing from the scope of the present invention. The scope of the present invention is determined by the scope of the appended claims.

Claims

1. A method for determining abnormal application users, characterized in that: include: In response to a first application instruction of a first user, determining second users associated with the first user; respectively determining first structured data of the first user and second structured data of each of the second users; Determine first feature data of the first user based on the first structured data and each of the second structured data, and generate first tag data of the first user according to the first feature data; The first structured data and the first label data are input into a pre-fine-tuned target large model to obtain an output result, and whether the first user is an abnormal application user is determined according to the output result.

2. The method for determining abnormal application users according to claim 1, characterized in that: The target large model is obtained by fine-tuning the following steps: Acquire a target structured data set; the target structured data set includes: target users, historical transaction information of each target user, and relationship information of each user; the target user includes at least one of the following: abnormal application users, normal application users, third-party users, and proxy users; Establishing a target graph based on the target structured data set, and determining node label data of each node according to the target graph; Each of the node label data and the target structured data set is input into the target large model to fine-tune the target large model.

3. The method for determining abnormal application users according to claim 2, characterized in that: The step of establishing a target graph based on the target structured data set and determining node label data of each node according to the target graph includes: Determine each of the target users as each node of the target graph, and determine the user information of each of the target users as the attribute of each of the nodes; Connecting edges of each node according to the historical transaction information of each target user and the relationship information of each target user; Obtaining the target graph based on each of the nodes, the attributes of each of the nodes, and each of the edges; Inputting the target graph into a pre-trained target graph neural network to obtain node features of the target node and an adjacency matrix of each reference node associated with the target node; Generate target node label data of the target node according to the node characteristics of the target node and each of the adjacency matrices.

4. The method for determining abnormal application users according to claim 1, characterized in that: The determining of each second user associated with the first user includes: Acquiring historical application information and / or historical transaction information of the first user; Determine a third-party user or a proxy user according to the historical application information, and determine the third-party user or the proxy user as the second user; And / or, determining a transaction applicant or a transaction recipient according to the historical transaction information, and determining the transaction applicant or the transaction recipient as the second user.

5. The method for determining abnormal application users according to claim 1, characterized in that: The respectively determining the first structured data of the first user and the second structured data of each of the second users includes: respectively determining user information, historical application information, and historical transaction information of the first user and each of the second users, and performing structured representation on the user information, historical application information, and historical transaction information to obtain the first structured data and each of the second structured data; The application information includes at least one of the following: a timestamp, an amount, and a frequency; The historical transaction information includes at least one of the following: timestamp, transaction applicant, transaction recipient, amount, and frequency.

6. The method for determining abnormal application users according to claim 1, characterized in that: The determining the first feature data of the first user based on the first structured data and each of the second structured data, and generating the first tag data of the first user according to the first feature data includes: Inputting the first structured data and each of the second structured data into a target graph neural network to obtain a first user feature of the first user and a second user feature of each of the second users; First label data of the first user is generated according to the first user feature and each of the second user features.

7. The method for determining abnormal application users according to claim 1, characterized in that: The output result is a probability value; Correspondingly, determining whether the first user is an abnormal application user according to the output result includes: When the probability value is greater than or equal to a set threshold, it is determined that the first user is an abnormal application user.

8. A device for determining abnormal application users, characterized in that: include: A first determining module, configured to determine second users associated with a first user in response to a first application instruction of the first user; A second determining module, configured to respectively determine the first structured data of the first user and the second structured data of each of the second users; a third determining module, configured to determine first feature data of the first user based on the first structured data and each of the second structured data, and generate first tag data of the first user according to the first feature data; The fourth determination module is used to input the first structured data and the first label data into the first large model obtained by pre-fine-tuning to obtain an output result, and determine whether the first user is an abnormal application user based on the output result.

9. An electronic device, characterized in that: The electronic device comprises: at least one processor; and a memory communicatively connected to the at least one processor; wherein, The memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute the method for determining an abnormal applicant user according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a processor to implement the method for determining an abnormal applicant user according to any one of claims 1 to 7 when executed.

11. A computer program product, comprising a computer program, wherein when the computer program is executed by a processor, the computer program implements the method for determining an abnormal applicant according to any one of claims 1 to 7.

Images