Shop violation registration identification method and device, equipment and medium

Abstract

The invention relates to a shop illegal registration identification method and device, equipment and a medium, and the method comprises the steps: obtaining a to-be-predicted registered shop sample set which comprises shop samples of a plurality of shops; constructing shop samples in the to-be-predicted registered shop sample set as newly added nodes in a graph object corresponding to the graph object constructed based on shop samples in a historical registered shop sample set so as to update the graph object; and inputting the updated graph object into a graph neural network model obtained by training shop samples of the historical registered shop sample set in advance, and predicting the risk probability that each shop in the to-be-predicted registered shop sample set belongs to an illegal registration event, and judging whether the corresponding shop belongs to an illegal registered shop according to the risk probability. According to the invention, the association information between shops can be effectively captured, so that the identification accuracy of illegal registration shops is improved.

Description

Technical Field

[0001] The present application relates to e-commerce information security technology, and in particular to a method for identifying illegal store registrations and its device, equipment, and medium. Background Art

[0002] With the rapid development of e-commerce, the number of stores on e-commerce platforms has exploded. While this phenomenon has led to economic prosperity, it has also been accompanied by the increasingly serious problem of illegally registered stores. Illegally registered stores usually refer to those that are registered for the purpose of malicious order brushing, wool-pulling, transaction fraud and other improper means. These behaviors not only harm the interests of consumers, but also seriously affect the reputation of the platform and the business environment for normal operation of stores.

[0003] The illegal registration identification technology currently used by e-commerce platforms is mainly based on traditional machine learning algorithms, such as logistic regression, GBDT, XGBoost, random forest, LightGBM, etc. These algorithms train machine learning classification models to predict the probability of illegal registration for each store, thereby determining whether the store is an illegal registration. However, this method fails to fully consider the network characteristics of illegal registration behavior, that is, stores with illegal registration tendencies tend to form clusters in the network. This clustering feature is a challenge for traditional machine learning models because they are usually unable to effectively identify and utilize the association information between stores. In addition, as illegal registration behaviors continue to evolve, the generalization ability and recognition accuracy of traditional machine learning models are also challenged when dealing with complex and changeable illegal registration behaviors.

[0004] Therefore, how to design a method that can effectively capture the correlation information between stores and improve the accuracy of identifying illegally registered stores has become a technical problem that e-commerce platforms urgently need to solve. Summary of the invention

[0005] The primary purpose of the present application is to solve at least one of the above problems and provide a method for identifying illegal store registrations and its device, equipment, and medium.

[0006] According to one aspect of the present application, a method for identifying illegal store registration is provided, comprising the following steps:

[0007] Obtaining a sample set of registered stores to be predicted, wherein the sample set of registered stores to be predicted includes store samples of multiple stores, and the store samples describe basic information and behavior characteristics of corresponding stores;

[0008] Corresponding to the graph object constructed based on the store samples in the historical registered store sample set, the store samples in the registered store sample set to be predicted are constructed as new nodes in the graph object to update the graph object;

[0009] The updated graph object is input into the graph neural network model pre-trained using the store samples of the historical registered store sample set, and the risk probability of each store in the registered store sample set to be predicted belonging to an illegal registration event is predicted, and whether the corresponding store is an illegally registered store is determined based on the risk probability.

[0010] According to another aspect of the present application, a store registration violation identification device is provided, comprising:

[0011] A sample set acquisition module to be predicted, configured to acquire a sample set of registered stores to be predicted, wherein the sample set of registered stores to be predicted includes store samples of multiple stores, and the store samples describe basic information and behavior characteristics of corresponding stores;

[0012] A graph object updating module is configured to correspond to a graph object constructed based on store samples in a historical registered store sample set, and to construct store samples in the registered store sample set to be predicted as newly added nodes in the graph object to update the graph object;

[0013] The risk probability prediction module is configured to input the updated graph object into the graph neural network model pre-trained using the store samples of the historical registered store sample set, predict the risk probability of each store in the registered store sample set to be predicted belonging to an illegal registration event, and determine whether the corresponding store is an illegally registered store based on the risk probability.

[0014] According to another aspect of the present application, a store registration violation identification device is provided, comprising a central processing unit and a memory, wherein the central processing unit is used to call and run a computer program stored in the memory to execute the steps of the store registration violation identification method described in the present application.

[0015] According to another aspect of the present application, a non-volatile readable storage medium is provided, which stores a computer program implemented according to the store illegal registration identification method in the form of computer-readable instructions, and when the computer program is called and executed by a computer, the steps included in the method are executed.

[0016] The technical solution of this application has many advantages, including but not limited to the following aspects:

[0017] The present application updates the graph object by constructing the store samples in the sample set of registered stores to be predicted as newly added nodes in the graph object based on the sample set of historical registered stores. The graph object can fully reflect the complex correlations between stores and the potential cluster characteristics, which enables the graph neural network model to accurately capture the mutual correlation patterns between stores by utilizing the information transfer between the store sample nodes in the graph object. The updated graph object is input into the graph neural network model that is pre-trained using the store samples of the historical registered store sample set, and the graph neural network model's ability to process graph structured data is utilized to learn the complex relationships between the store samples in the graph object, effectively identifying the potential patterns of illegal registration behaviors, thereby improving the accuracy and reliability of the prediction results. Specifically, the present application utilizes the graph neural network model to effectively capture the correlation information between stores, thereby improving the accuracy of identifying illegally registered stores. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 The corresponding network architecture is exemplarily deployed in the e-commerce platform for the store risk control system of this application;

[0019] Figure 2 A flowchart of an embodiment of a method for identifying illegal store registrations of the present application;

[0020] Figure 3 This is a functional block diagram of the store registration violation identification device of this application;

[0021] Figure 4 This is a structural diagram of a store registration violation identification device used in this application. DETAILED DESCRIPTION

[0022] See also Figure 1 The present application provides an exemplary network architecture suitable for e-commerce platform scenarios, including a terminal device 80, a security server 81 and a store server 82.

[0023] The terminal device 80 may be a smart phone, a personal computer, a laptop, a tablet computer, etc., which may be used to trigger a network request in order to use various application services in the store server 82, such as registering an online store. The terminal device 80 monitors the user's registration behavior when the user performs a store registration event, and sends the corresponding store sample generated by the user's registration behavior to the store server 82 for storage, wherein the store sample includes the store registration attribute information corresponding to the store registration event, such as registration time, registration method (mobile phone / email), mobile phone number, email, IP, device fingerprint, etc., and the store registration behavior feature data corresponding to the store registration event, and the store registration behavior feature data may be divided into two types: offline features and real-time features.

[0024] The generation and storage process of offline features includes analyzing the registration behavior of the same IP, mobile phone number, device fingerprint or user agent (UA) within a specific time range. These offline features include the number of registration applications, the number of registrations passed, the registration pass rate, and the number of registrations identified as illegal registrations for each IP, mobile phone number, device fingerprint or user agent within a specific period of time. These offline features are processed and output through the T+1 data processing process, which means that the data will be processed and summarized on the second day after the event. Specifically, these offline features will be stored in four tables with IP, mobile phone number, device fingerprint, and user agent as the primary keys. These tables correspond to the primary keys IP, mobile phone number, device fingerprint, and user agent, respectively, and are used to store the number of registration applications, the number of registrations passed, the registration pass rate, and the number of registrations identified as illegal registrations related to each primary key.

[0025] Real-time features are generated dynamically, focusing on the characteristics of the store registration behavior itself, and are closely related to the specific circumstances when the registration behavior occurs, including the total length of the registration process, the time required for users to complete the verification code, the time required to fill in the email or mobile phone number, the number of email or mobile phone number verifications, whether the IP address of the registered mobile phone number is consistent with the registered IP address, whether the geographic location information filled in during registration is consistent with the geographic location information of the IP address, and the number of registration applications, the number of registrations passed, the registration pass rate, and the number of registrations identified as illegal registrations with the same IP, mobile phone number, device fingerprint or user agent within a short time window before this registration event occurs. These real-time features are usually calculated immediately when the registration behavior occurs.

[0026] In addition, the store sample also includes a unique store identifier, such as a store ID. And illegal registration label data corresponding to whether the store corresponding to the store sample is an illegal registration store. The illegal registration label data can be manually annotated or obtained through the prediction of the graph neural network model trained in this application. The illegal registration label data corresponding to each store is encapsulated in the store sample and stored in the database of the store server 82. The above-mentioned store samples generated by the user performing the store registration behavior on the terminal device 80 are encapsulated into data packets and sent to the store server 82 through the security protocol.

[0027] The security server 81 can be used as the main executor of the store risk control system of the present application. The store risk control system adopts a graph neural network algorithm to effectively identify illegal registration behaviors with clustering characteristics by learning the association relationship information between registered stores. Specifically, before the implementation of the embodiment of the present application, the store risk control system accesses the database in the store server 82 to obtain the store unique identifier, store registration attribute information, store registration behavior feature data and illegal registration label data corresponding to each store sample in the historical registered store sample set, and then uses these store samples to construct nodes in the graph object, and establishes connections between the edges of these nodes based on the store registration attribute information and the preset edge connection rules to form a complete graph object. The graph object constructed based on the historical registered store sample set can be stored in the database of the security server 81 and used by the security server 81 to train the graph neural network model. In an embodiment of the present application, in response to a store registration violation identification event, the security server 81 receives a sample set of registered stores to be predicted, and extracts a graph object constructed based on a sample set of historical registered stores from a database, and constructs store samples in the sample set of registered stores to be predicted as newly added nodes in the graph object to update the graph object, and then inputs the updated graph object into a graph neural network model that has been pre-trained using store samples in the sample set of historical registered stores, and finally determines whether each store in the sample set of registered stores to be predicted is an illegally registered store.

[0028] According to the judgment result of whether each store in the sample set of registered stores to be predicted is an illegally registered store, the illegal registration label data of the corresponding store sample is marked with 0 (not an illegally registered store) and 1 (an illegally registered store), and the marked illegal registration label data is associated with each store sample in the sample set of registered stores to be predicted, and updated to the graph object constructed based on the historical registered store sample set in the database of the security server 81, and iterative training and optimization are performed to adapt to new changes in illegal registration behavior.

[0029] The store illegal registration identification method of the present application can be implemented as a computer program product through programming and implanted into the store risk control system. When the store risk control system predicts an online store as an illegally registered store, it can implement risk control and restriction processing on it. Of course, it can also be handed over to manual management users for further manual review and confirmation before implementing risk control and restriction processing.

[0030] See also Figure 2 According to a method for identifying illegal store registration provided by the present application, in one embodiment, the method comprises the following steps:

[0031] Step S5100: obtaining a sample set of registered stores to be predicted, wherein the sample set of registered stores to be predicted includes store samples of multiple stores;

[0032] In the daily operation of the e-commerce platform, a large number of new stores are usually registered every day. Whenever a user intends to open a new store on the platform, he will enter the e-commerce platform to register the store, triggering a request to submit a store registration to the store server. The store server can trigger the corresponding store registration event, and use the store unique identifier, store registration attribute information and store registration behavior feature data of the corresponding store generated by the registration behavior as the store sample of the store, and send the store samples of each store corresponding to the registration behavior to the store server for storage. In response to the store violation registration identification event of this application, the store risk control system will send a request to the store server to obtain a sample set of registered stores to be predicted, wherein the sample set of registered stores to be predicted includes store samples of multiple stores, and the number of store samples included in the sample set of registered stores to be predicted can be set as a fixed number by technical personnel in this field as needed, or it can be set to the number of store samples generated by registered stores within a specific time period.

[0033] It is further understood that the store sample includes the corresponding store registration attribute information when the store is registered, such as registration time, registration method (mobile phone / email), mobile phone number, email, IP, device fingerprint, etc. From the previously identified illegal registration behaviors, it can be known that illegal registration behaviors are often clustered. Usually, in order to register a large number of stores at one time, illegal registrants will use the same or similar store registration attribute information for rapid registration, that is, the store registration attribute information generated by these illegal stores when they are registered is often related. For example, if an IP address registers a large number of stores in a short period of time, or the device fingerprint matches the known illegal registration device fingerprint, or the illegal registrants often use similar email addresses and mobile phone numbers to register, then these registration behaviors should be considered high-risk. Using this feature, in-depth analysis of store registration attribute information is carried out to identify potential correlation patterns, provide rich contextual information for subsequent training of graph neural network models, and train more accurate graph neural network models, thereby effectively predicting and combating illegal registration behaviors.

[0034] The store sample also includes store registration behavior feature data as part of the training sample graph object of the training graph neural network model of this application, including offline features and real-time features. Offline features include the number of registration applications, the number of registration passes, the registration pass rate, and the number of registrations identified as illegal registrations for each IP, mobile phone number, device fingerprint or user agent within a specific period of time. By mining and analyzing offline features, hidden connections between different store registration attributes can be discovered, and these connections can often reveal the behavior patterns of illegal registrants. For example, when a mobile phone number is associated with multiple illegal registration stores, there may be a situation where the mobile phone number is used for batch registration, which is a common method used by illegal registrants. At this time, the number of registration applications, the number of registration passes, the registration pass rate, and the number of registrations identified as illegal registrations of the mobile phone number can be used to identify whether the store registered by the mobile phone number is a high-risk illegal registration. In addition, if an IP address is associated with a large number of registration behaviors in a short period of time, that is, a certain IP address has a large number of registration applications in a specific period of time, this is also a warning signal, because normal users are unlikely to register multiple stores from the same IP address in a short period of time.

[0035] Real-time features include the total length of the registration process, the time required for users to complete the verification code, the time required to fill in the email or mobile phone number, the number of email or mobile phone number verifications, whether the IP address of the registered mobile phone number is consistent with the registered IP address, whether the geographical location information filled in during registration is consistent with the geographical location of the IP address, and the number of registration applications, the number of registrations passed, the registration pass rate, and the number of registrations identified as illegal in a short time window before the current registration event. Among them, the total length of the registration process, the time required for users to complete the verification code, and the time required to fill in the email or mobile phone number are often considered illegal registrations. Because illegal registrants often pursue speed, they usually complete the registration process at a very fast speed, while normal users will be more cautious and need more time to think about filling in information. Similarly, registration behaviors that identify that the IP address of the registered mobile phone number is inconsistent with the registered IP address or the geographical location information filled in during registration is inconsistent with the geographical location of the IP address are often considered illegal registration behaviors. Under normal circumstances, the IP address of the user's registered mobile phone number should match its registered IP address. If there is a mismatch, it is a sign of illegal registration behavior. When training the graph neural network model, the illegal registration label data of each store sample will be used for training, and the characteristics of the above-mentioned real-time features will be learned to predict the risk probability of each store in the sample set of registered stores to be predicted belonging to the illegal registration event.

[0036] In addition, the number of registration applications, the number of registrations passed, the registration pass rate, and the number of registrations identified as illegal registrations with the same IP, mobile phone number, device fingerprint, or user agent within a short time window before the current registration event is counted. In one embodiment, the time slice is kept within 3 hours. The reason for selecting a time slice within 3 hours for analysis is that illegal registrants usually tend to concentrate on registration behavior in a short period of time, and such registration behavior can be quickly captured and analyzed. A shorter time window helps to focus on recent registration behaviors, and such data is more likely to show a direct connection between registration behaviors, because the behavior of registering with the same registration attributes in a short period of time is likely to be related. In addition, a shorter time window helps to reduce the consumption of computing resources and ensure stability and efficiency when processing large amounts of data. Finally, by limiting the time range, it is possible to avoid mistakenly regarding registration behaviors that are far apart in time and are actually unrelated as related, thereby improving the accuracy of identifying illegal registration behaviors. The store sample also includes a store unique identifier, such as a store ID. In summary, these store samples can provide rich contextual information for the graph neural network model, enabling it to understand registration behavior more comprehensively and predict illegal registration behavior more accurately.

[0037] Step S5200: corresponding to the graph object constructed based on the store samples in the historical registered store sample set, constructing the store samples in the registered store sample set to be predicted as new nodes in the graph object to update the graph object;

[0038] First, an initial graph object is constructed based on the store samples in the historical registered store sample set. The constructed graph object consists of multiple nodes and edges, where each node represents a store sample in the historical registered store sample set, and the edges are connected according to the association relationship between the store samples. Among them, the association relationship can be determined by IP address, device fingerprint, mobile phone number, email address, etc., and the specific definition can be determined based on experience and data exploration. For example, if two store samples have the same IP address, or similar attributes such as device fingerprint, mobile phone number, email address, etc., then an edge is added between the two nodes to indicate that there is some kind of association between them.

[0039] Add the store samples in the sample set of registered stores to be predicted one by one to the graph object built based on the sample set of historical registered stores as new nodes. Each new node also contains the store registration attribute information, store registration behavior feature data, and store unique identifier (such as store ID) of the store sample. In order to ensure the integrity and consistency of the graph object, the edge connection relationship between the new nodes and the historical nodes needs to be defined and constructed according to the same rules. For example, if the IP address of the new node is the same as the IP address of a historical node, or the attributes such as device fingerprint, mobile phone number, and email are similar to those of the historical node, then add an edge between the two nodes.

[0040] After the construction of the new node is completed, the entire graph object needs to be updated. This includes updating the store registration attribute information of the node, the store registration behavior feature data, the edge connection relationship, and the topological structure of the graph. The updated graph object not only contains the information of the historical registered store samples, but also incorporates the latest registered store samples to be predicted, thereby realizing the dynamic update of the graph object. This update mechanism enables the graph object to reflect the current registration behavior and association pattern in real time, providing a more comprehensive and accurate data basis for the subsequent identification of illegal registrations.

[0041] After the graph object is updated, the updated graph object can be used for further analysis and prediction. For example, the probability of illegal registration of newly added nodes can be predicted through the graph neural network model, so as to timely identify and handle potential illegal registration behaviors. The graph neural network model can effectively capture the complex relationship between nodes through the message passing mechanism, and predict the risk probability of illegal registration events for newly added nodes based on the relationship between the captured nodes. In this way, real-time monitoring and early warning of illegal registration behaviors can be achieved, improving the security and user experience of the e-commerce platform.

[0042] Step S5300: input the updated graph object into the graph neural network model pre-trained with the store samples of the historical registered store sample set, predict the risk probability of each store in the registered store sample set to be predicted belonging to an illegal registration event, and determine whether the corresponding store is an illegally registered store based on the risk probability.

[0043] First, the updated graph object is input into the graph neural network model that is pre-trained using the store samples of the historical registered store sample set. Among them, the graph neural network model is obtained by training the historical registered store sample set. For the specific training process, please refer to the subsequent specific implementation method, which will not be repeated here. After the updated graph object is input into the graph neural network model, the graph neural network model will predict the risk probability of illegal registration events for each newly added node in the graph object (that is, the registered store sample to be predicted). Specifically, the graph neural network model aggregates the feature information of each node with the feature information of its neighboring nodes through a message passing mechanism, thereby updating the representation of the node. This process will be repeated many times to ensure that the graph neural network model can fully capture the complex relationship between nodes.

[0044] After completing message delivery and feature aggregation, the graph neural network model will classify and predict each newly added node, and output the risk probability that the node belongs to an illegal registration event. In one embodiment, the risk probability obtained is a value between 0 and 1, indicating the possibility that the store corresponding to the node belongs to an illegal registration event. The higher the risk probability, the greater the possibility that the store belongs to an illegal registration event. Based on the predicted risk probability, it can be determined whether the corresponding store belongs to an illegal registration store. Usually, a risk probability threshold can be set. When the risk probability of a store exceeds this threshold, the store is considered to be an illegal registration store. This threshold can be adjusted according to business needs and actual conditions to balance the recognition accuracy and recall rate.

[0045] After determining that a store has been registered in violation of regulations, corresponding disposal measures can be taken, such as restricting certain functions of the store, requiring it to undergo further identity verification, or directly banning the store. By taking these measures, it is possible to effectively prevent stores that have been registered in violation of regulations from engaging in malicious order-brushing, wool-pulling, transaction fraud, and other behaviors, thereby maintaining the health of the e-commerce platform and the transaction satisfaction of buyers. In this way, the graph neural network model can not only monitor and warn of illegal registration behaviors in real time, but also dynamically update the graph neural network model based on the latest registration behaviors, thereby improving the accuracy and real-time performance of the model. This enables e-commerce platforms to respond more effectively to illegal registration behaviors, ensuring the normal operation of the platform and a good user experience.

[0046] In another embodiment, after determining whether each store in the sample set of stores to be predicted is an illegally registered store, the illegal registration label data of the corresponding store samples are marked with 0 (not an illegally registered store) and 1 (an illegally registered store) labels, and the store samples of the associated corresponding stores are stored in the store server. The marked store samples can be used to form a historical registered store sample set to iteratively train the graph neural network model of this application.

[0047] It can be known from the typical embodiments of the present application that the technical solution of the present application has many advantages, including but not limited to the following aspects:

[0048] This application updates the graph object by constructing the store samples in the sample set of registered stores to be predicted as newly added nodes in the graph object based on the sample set of historical registered stores. The graph object can fully reflect the complex correlations and potential cluster characteristics between stores. Since illegal registration behaviors often form clusters in the network, through the construction of the graph object, illegally registered stores can be more accurately identified. And the graph neural network model can use the information transfer between the store sample nodes in the graph object to accurately capture the mutual correlation pattern between stores. Specifically, the updated graph object is input into the graph neural network model that is pre-trained using the store samples of the historical registered store sample set, and the ability of the graph neural network model to process graph structure data is used to learn the complex relationship between each store sample in the graph object, effectively identify the potential pattern of illegal registration behavior, and thus improve the accuracy and reliability of the prediction results.

[0049] The graph neural network model in this application will show good generalization ability when dealing with complex and changeable illegal registration behaviors, and can adapt to the continuous evolution of illegal registration behaviors. That is, the graph neural network model can not only handle current illegal registration behaviors, but also adapt to new illegal registration behaviors that may appear in the future.

[0050] In general, this application uses a graph neural network model to effectively capture the correlation information between stores, thereby improving the accuracy of identifying illegally registered stores.

[0051] Based on any embodiment of the method of the present application, before obtaining the sample set of registered stores to be predicted, the method includes:

[0052] Step S6100: Acquire the historical registered store sample set, wherein the historical registered store sample set includes a plurality of store samples, each of which includes a store unique identifier of the corresponding store, store registration attribute information, store registration behavior feature data, and illegal registration label data corresponding to whether the store is illegally registered;

[0053] In this step, the store risk control system sends a request to the store server to obtain a sample set of historical registered stores, so as to obtain a sample set of historical registered stores. The sample set of historical registered stores is stored in the store server. A person skilled in the art can request multiple store samples as a sample set of historical registered stores in the store server by setting a time range or a quantity range. Each store sample includes a unique store identifier of the corresponding store, store registration attribute information, store registration behavior feature data, and illegal registration label data corresponding to whether it is an illegally registered store. Specifically, each store sample has a unique identifier, usually a store ID, which is used to distinguish different store samples to ensure that each store sample is unique in the data set. Each store sample also includes corresponding store registration attribute information to describe the context of store registration. In addition, each store sample also includes store registration behavior feature data, and illegal registration label data corresponding to whether it is an illegally registered store. The illegal registration label data is used to mark whether the store sample belongs to an illegally registered store. For example, a label data of 0 indicates that the store sample does not belong to an illegally registered store, and a label data of 1 indicates that the store sample belongs to an illegally registered store. By obtaining a detailed sample set of historical registered stores, we can provide a rich data foundation for the training of graph neural network models, which can help the model better understand store registration behavior and accurately predict illegal registration behavior.

[0054] Step S6200: constructing a graph object for training the graph neural network model according to the historical registered store sample set, wherein nodes of the graph object are constructed with the store samples;

[0055] A graph object for training the graph neural network model is constructed based on the historical registered store sample set. The graph object consists of multiple nodes and edges, where each node represents a historical registered store sample. Specifically, each store sample in the historical registered store sample set is used as a node of the graph object. Each node contains the store unique identifier, store registration attribute information, store registration behavior feature data, and illegal registration label data corresponding to whether it is an illegal registered store. In the process of constructing the graph object, it is necessary to define the edge connection relationship between nodes. Among them, the edge connection relationship is defined according to the correlation between store samples.

[0056] Specifically, the definition of an edge can be constructed in the following ways: using the device fingerprint of the terminal device used when registering a store as the store registration attribute information in the corresponding store sample, detecting whether the device fingerprints corresponding to any two nodes in the graph object are the same, and if they are the same, establishing an edge connection between the two nodes; using the mobile phone number registered with the store as the store registration attribute information in the corresponding store sample, deleting at least one digit of the mobile phone number, and comparing whether the remaining parts of the mobile phone numbers corresponding to any two nodes in the graph object are the same, and if they are the same, establishing an edge connection between the corresponding nodes; using the email information and registration time of store registration as the store registration attribute information in the corresponding store sample, Calculate the similarity of the corresponding email information between any two nodes in the graph object. If the similarity exceeds a preset first similarity threshold and the corresponding registration time interval between the two nodes does not exceed the preset first time threshold, establish an edge connection between the corresponding nodes; use the IP address, email information and registration time when the store is registered as the store registration attribute information in the corresponding store sample, compare whether the corresponding IP addresses between any two nodes in the graph object are the same. If they are the same and the corresponding email information between the two nodes exceeds a preset second similarity threshold and the registration time interval between the two nodes does not exceed the preset second time threshold, establish an edge connection between the corresponding nodes.

[0057] Through the definition of these association relationships for edge connections, a graph object containing rich association information can be constructed. This graph object not only contains the store unique identifier, store registration attribute information, store registration behavior feature data, and illegal registration label data corresponding to whether it is an illegal registration store, but also captures the association pattern between store samples through edge connection relationships. This graph object can provide a comprehensive data foundation for the training of graph neural network models, help the model better understand store registration behavior, and accurately predict illegal registration behavior. In this way, the constructed graph object can better reflect the complex relationship between store samples and provide more accurate and comprehensive data support for the training of graph neural network models.

[0058] Step S6300: input the constructed graph object into the graph neural network model, and use the illegal registration label data of each store sample for supervised training to train the graph neural network model to a convergence state.

[0059] During the training process, the constructed graph object is input into the graph neural network model of the store risk control system, and the illegal registration label data of each store sample is used as a supervisory signal to guide the training of the graph neural network model. Specifically, the graph neural network model aggregates the feature information of each node with the feature information of its neighboring nodes through a message passing mechanism to update the representation of the node. This process is repeated many times to ensure that the graph neural network model can fully capture the complex relationship between the nodes. In each iteration, the graph neural network model will make classification predictions based on the current node representation and calculate the loss between the predicted result and the true label. The loss function usually uses cross entropy loss to measure the difference between the predicted result of the graph neural network model and the true label.

[0060] After calculating the loss, the graph neural network model will perform back propagation and update the parameters of the graph neural network model to minimize the loss function. This process will be repeated many times until the loss of the graph neural network model no longer decreases significantly, that is, the graph neural network model reaches a convergence state. The convergence state means that the graph neural network model has fully learned the characteristics and association patterns of historical registered store samples and can accurately predict illegal registration behaviors.

[0061] During the training process, some optimization strategies can also be used, such as learning rate adjustment, regularization, batch normalization, etc., to improve the training effect and generalization ability of the graph neural network model. Through these optimization strategies, the graph neural network model can learn the characteristics and association patterns of historical registered store samples more stably and efficiently, thereby improving the prediction accuracy of the graph neural network model.

[0062] In this way, the constructed graph object is input into the graph neural network model, and the illegal registration label data of each store sample is used for supervised training, so that the graph neural network model can be trained to a convergence state. This trained graph neural network model can effectively capture the complex relationship between store samples and accurately predict illegal registration behavior, thereby improving the security and health of the e-commerce platform.

[0063] Based on any embodiment of the method of the present application, a graph object for training the graph neural network model is constructed according to the historical registered store sample set, including:

[0064] Step S6210: construct corresponding nodes in the graph object using the unique store identifiers of the store samples in the historical registered store sample set;

[0065] The corresponding nodes in the graph object are constructed with the store unique identifiers of the store samples in the historical registered store sample set. Specifically, the store unique identifier of each store sample is first extracted from the historical registered store sample set, usually the store ID. Then, these store unique identifiers are used as nodes in the graph object. Each node represents a store sample, and the node stores the store unique identifier of the store sample, which is used to distinguish different store samples and ensure that each store sample is unique in the graph object. In this way, the initial node set of the graph object is constructed, laying the foundation for subsequent edge connections and feature associations.

[0066] Step S6220: Based on the store registration attribute information in the store sample and a preset edge connection rule, determine whether the store registration attribute information of any two nodes of the graph object satisfies the edge connection rule; if so, establish an edge connection between the two nodes;

[0067] First, the store registration attribute information of each store sample is extracted from the historical registered store sample set. These store registration attribute information usually include but are not limited to registration time, registration method (mobile phone / email), mobile phone number, email, IP, device fingerprint, etc. These store registration attribute information is an important basis for judging the correlation between store samples. Secondly, the preset edge connection rules are based on the analysis and experience summary of historical illegal registration behaviors. For example, the rules may include: if the IP addresses of two store samples are the same, or the attributes such as device fingerprints, mobile phone numbers, and emails are similar, then an edge connection is established between the two nodes. In addition, the registration time interval, geographic location information, etc. can also be considered to more comprehensively capture the correlation between store samples. Compare any two nodes in the graph object to determine whether their store registration attribute information meets the preset edge connection rules. If the edge connection rules are met, an edge is established between the two nodes, indicating that there is a certain correlation between them. When the edge connection rules are met, an edge connection is established between the two nodes. The establishment of the edge is not only based on a single attribute, but also multiple attributes can be considered comprehensively to more accurately capture the correlation pattern between store samples. For example, if two nodes have the same IP address and a short registration time interval, an edge connection is established between the two nodes, indicating that there is a strong correlation between them. Through the above steps, the edge connection relationship of the graph object is constructed. These edge connection relationships capture the association patterns between store samples and provide rich contextual information for the subsequent graph neural network model training.

[0068] This step ensures that the edge connection relationship of the graph object can accurately capture the correlation between store samples, providing a solid foundation for the subsequent identification of illegal registrations. The edge connection rules can effectively capture the correlation between store samples by defining the connection conditions between nodes. Through these rules, the graph object can capture the complex associations between store samples and provide rich contextual information for subsequent analysis and prediction. The edge connection rules determine the topological structure of the graph object, that is, the connection mode between nodes. A good topological structure can better reflect the true association relationship between store samples, thereby improving the training effect of the graph neural network model. For example, illegal registration behaviors usually have local clustering, that is, a large number of registrations are made using the same store registration attribute information such as IP addresses and device fingerprints in a short period of time. Through the edge connection rules, a graph structure with local clustering can be constructed, so that the graph neural network model can better capture this clustering. The edge connection rules directly affect the training effect of the graph neural network model. By defining reasonable edge connection rules, a graph object with rich association information can be constructed, thereby improving the training effect of the graph neural network model. Specifically, the graph neural network model aggregates the behavioral characteristics of each node with the behavioral characteristics of its neighboring nodes through a message passing mechanism. The edge connection rule determines which nodes are neighbor nodes, thus affecting the effect of feature aggregation. Reasonable edge connection rules can ensure the effectiveness of feature aggregation and improve the prediction accuracy of the graph neural network model.

[0069] Edge connection rules also support dynamic updating and adaptability of graph objects. As new store samples are continuously added, graph objects need to be continuously updated to reflect the latest registration behaviors and association patterns. Edge connection rules can ensure that the update process of graph objects is reasonable and effective, thereby maintaining the real-time and accuracy of the graph neural network model. Through these operations, edge connection rules play a vital role in the entire solution. They not only define the connection relationship between nodes in the graph object, but also directly affect the training effect of the graph neural network model and the final accuracy of illegal registration identification. By capturing the association pattern between store samples, constructing a reasonable graph object topology, improving the training effect of the graph neural network model, and enhancing the accuracy of illegal registration identification, edge connection rules provide a solid foundation for the successful implementation of the entire solution.

[0070] Step S6230: import the store registration behavior feature data and the illegal registration label data corresponding to each store sample into the graph object, associate them with the corresponding nodes in the graph object, and form a complete graph object.

[0071] Extract the store registration behavior feature data of each store sample from the historical registered store sample set. These feature data include offline features and real-time features. These feature data can fully reflect the registration behavior characteristics of the store sample and provide rich contextual information for subsequent analysis and prediction. Extract the illegal registration label data of each store sample. These label data are used to mark whether the store sample belongs to an illegally registered store. For example, a label data of 0 indicates that the store sample does not belong to an illegally registered store, and a label data of 1 indicates that the store sample belongs to an illegally registered store. Illegal registration label data is an important supervisory signal for the training of graph neural network models, which can guide the model to learn the characteristics and association patterns of store samples, thereby improving the accuracy of illegal registration identification.

[0072] The store registration behavior feature data and illegal registration label data of each store sample are imported into the graph object and associated with the corresponding nodes in the graph object. Specifically, the store registration behavior feature data of each store sample is used as the feature vector of the node and associated with the node. At the same time, the illegal registration label data of each store sample is used as the label of the node and associated with the node. In this way, each node in the graph object not only contains the store unique identification and store registration attribute information of the store sample, but also contains the store registration behavior feature data and illegal registration label data, thus forming a complete graph object.

[0073] Through the above steps, a graph object containing rich information is constructed. This graph object not only contains the unique store identifier, store registration attribute information, store registration behavior feature data of the historical registered store samples, and the corresponding illegal registration label data that indicates whether it is an illegal registration store, but also captures the association pattern between store samples through edge connection relationships. This complete graph object can provide a comprehensive data foundation for the training of the graph neural network model, helping the model to better understand the store registration behavior and accurately predict the illegal registration behavior. In this way, the store registration behavior feature data and illegal registration label data corresponding to each store sample are imported into the graph object, associated with the corresponding nodes in the graph object, and form a complete graph object, which provides a solid foundation for the subsequent illegal registration identification.

[0074] On the basis of any embodiment of the method of the present application, based on the store registration attribute information in the store sample and the preset edge connection rule, it is determined whether the store registration attribute information of any two nodes of the graph object satisfies the edge connection rule, and if so, an edge connection is established between the two nodes, including any one or more of the following steps:

[0075] Step S6221: using the device fingerprint of the terminal device used when registering a store as the store registration attribute information in the corresponding store sample, detecting whether the device fingerprints corresponding to any two nodes in the graph object are the same; if they are the same, establishing an edge connection between the two nodes;

[0076] In this step, the device fingerprint of the terminal device of each store sample is extracted from the historical registered store sample set. The device fingerprint of the terminal device is a unique identifier generated by collecting and encoding various features of the device, which can accurately identify the uniqueness of the device. These device fingerprints usually include the hardware configuration, operating system, browser type and version, plug-in information, etc. of the device. With this information, a unique device fingerprint can be generated to distinguish different devices. The device fingerprint of the terminal device of each store sample is imported into the graph object as part of the store registration attribute information. Each node represents a store sample, and the node stores the store unique identifier, store registration attribute information, store registration behavior feature data, and illegal registration label data of the store sample. The device fingerprint of the terminal device is associated with the node as part of the store registration attribute information.

[0077] Compare any two nodes in the graph object to detect whether the device fingerprints of their terminal devices are the same. Specifically, by comparing the device fingerprints of the terminal devices of the two nodes, determine whether they are completely consistent. If the device fingerprints of the terminal devices of the two nodes are the same, it is considered that there is some association between the two nodes, that is, they may be registered by the same device. In this case, an edge is established between the two nodes, indicating that there is some association between them.

[0078] In this way, the edge connection relationship of the graph object can accurately capture the association pattern between store samples, providing a solid foundation for the subsequent identification of illegal registrations. The same device fingerprint of the terminal device usually reflects the characteristics of illegal registration behavior, because illegal registrants often use the same device for batch registration. By capturing this association pattern, the graph object can better reflect the complex relationship between store samples and provide more accurate and comprehensive data support for the training of the graph neural network model.

[0079] Step S6222: using the contact information and registration time of the store registration as the store registration attribute information in the corresponding store sample, calculating the similarity between the contact information corresponding to any two nodes in the graph object; if the similarity exceeds a preset first similarity threshold, and the corresponding registration time interval between the two nodes does not exceed a preset first time threshold, establishing an edge connection between the two nodes;

[0080] The contact information and registration time of each store sample are extracted from the historical registered store sample set. Contact information usually includes mobile phone number, email address, etc. Contact information is an important basis for judging the correlation between store samples. Registration time is the registration timestamp of the store sample, which is used to judge the registration time interval between store samples. The contact information and registration time of each store sample are imported into the graph object as part of the store registration attribute information. Each node represents a store sample, and the node stores the store unique identifier, store registration attribute information, store registration behavior feature data, and illegal registration label data of the store sample. Contact information and registration time are associated with the node as part of the store registration attribute information.

[0081] Compare any two nodes in the graph object and detect the similarity between their contact information. Specifically, by calculating the similarity between the contact information of the two nodes, determine whether they are similar. Similarity calculation can be performed using a variety of methods, such as string similarity calculation, cosine similarity calculation, etc. If the similarity between the contact information of the two nodes exceeds the preset first similarity threshold, it is considered that there is a certain association between the two nodes, that is, they may be registered by the same illegal registrant or group. Further compare the registration time intervals of the two nodes. Specifically, calculate the difference between the registration times of the two nodes to determine whether they are within the preset first time threshold. If the registration time interval of the two nodes does not exceed the preset first time threshold, it is considered that there is a strong association between the two nodes, that is, they may be registered by the same illegal registrant or group in a short period of time. Among them, the first similarity threshold and the first time threshold can be set by technicians in this field based on experience or data exploration.

[0082] When the above two conditions are met, that is, when the similarity between the contact information exceeds the preset first similarity threshold and the registration time interval does not exceed the preset first time threshold, an edge is established between the two nodes, indicating that there is a certain association between them. In this way, it is ensured that the edge connection relationship of the graph object can accurately capture the association pattern between store samples, providing a solid foundation for the subsequent identification of illegal registrations. The similarity and time interval between contact information and registration time usually reflect the characteristics of illegal registration behavior, because illegal registrants tend to use similar contact information for batch registration and complete the registration in a short time. By capturing this association pattern, the graph object can better reflect the complex relationship between store samples and provide more accurate and comprehensive data support for the training of the graph neural network model.

[0083] Step S6223: Use the IP address, contact information and registration time when the store is registered as the store registration attribute information in the corresponding store sample, and compare whether the corresponding IP addresses between any two nodes in the graph object are the same; if they are the same and the similarity of the corresponding contact information between the two nodes exceeds a preset second similarity threshold and the registration time interval between the two nodes does not exceed a preset second time threshold, establish an edge connection between the two nodes.

[0084] Extract the IP address, contact information and registration time of each store sample from the historical registered store sample set. The IP address is the network address used when the store sample is registered. The IP address, contact information and registration time of each store sample are imported into the graph object as part of the store registration attribute information. Compare any two nodes in the graph object to detect whether their IP addresses are the same. Specifically, by comparing the IP addresses of the two nodes, determine whether they are completely consistent. If the IP addresses of the two nodes are the same, it is considered that there is a certain association between the two nodes, that is, they may be registered by the same device or the same network environment. Further compare the similarity between the contact information of the two nodes. Specifically, by calculating the similarity between the contact information of the two nodes, determine whether they are similar. If the similarity between the contact information of the two nodes exceeds the preset second similarity threshold, it is considered that there is a certain association between the two nodes, that is, they may be registered by the same illegal registrant or group. Finally, further compare the registration time intervals of the two nodes. If the registration time interval of the two nodes does not exceed the preset second time threshold, it is considered that there is a strong association between the two nodes, that is, they may be registered by the same illegal registrant or group in a short time.

[0085] When the above three conditions are met, that is, the IP addresses are the same, the similarity between the contact information exceeds the preset second similarity threshold, and the registration time interval does not exceed the preset second time threshold, an edge is established between the two nodes, indicating that there is a certain association between them. The similarity and time interval of IP addresses, contact information and registration time usually reflect the characteristics of illegal registration behavior, because illegal registrants often use the same IP address and similar contact information for batch registration and complete the registration in a short time. By capturing this association pattern, the graph object can better reflect the complex relationship between store samples and provide more accurate and comprehensive data support for the training of graph neural network models.

[0086] On the basis of any embodiment of the method of the present application, corresponding to the graph object constructed based on the store samples in the historical registered store sample set, the store samples in the registered store sample set to be predicted are constructed as new nodes in the graph object to update the graph object, including:

[0087] Step S5210: acquiring store samples in the sample set of registered stores to be predicted and a graph object based on the sample set of historical registered stores, wherein each store sample includes a store unique identifier of the corresponding store, store registration attribute information, and store registration behavior feature data;

[0088] In the process of obtaining store samples in the sample set of registered stores to be predicted and the graph objects based on the sample set of historical registered stores, it is first necessary to clarify that the store samples in the sample set of registered stores to be predicted are newly registered store samples in the daily operation of the e-commerce platform. These newly registered store samples need to be tested for illegal registration behaviors to ensure the security and health of the platform.

[0089] First, the store risk control system sends a request to the store server to obtain the sample set of registered stores to be predicted. The store server responds to this request and sends the sample data of the newly registered store to the store risk control system. Store samples are the basis for identifying illegal registrations because they contain key information about store registration behavior. At the same time, it is also crucial to obtain a graph object based on the historical registered store sample set. The historical registered store sample set is a collection of all registered store samples in the past period of time. These samples have been marked by manual annotation or other means to indicate whether they are illegal registered stores.

[0090] The purpose of obtaining the store samples in the sample set of registered stores to be predicted and the graph object based on the sample set of historical registered stores is to associate the newly registered store samples with the historical registered store samples, thereby constructing a graph object containing the latest registration behavior. This updated graph object not only contains the information of the historical registered store samples, but also incorporates the latest registered store samples to be predicted, thereby realizing the dynamic update of the graph object. This update mechanism enables the graph object to reflect the current registration behavior and association pattern in real time, providing a more comprehensive and accurate data foundation for subsequent illegal registration identification. In this way, the integrity and consistency of the graph object can be ensured, so that the graph neural network model can better capture the complex relationship between store samples, and predict the risk probability of illegal registration events for the newly added nodes based on the relationship between the captured nodes. This helps to achieve real-time monitoring and early warning of illegal registration behaviors and improve the security and user experience of e-commerce platforms.

[0091] Step S5220: using the unique identifier of the store in the store sample as a newly added node of the graph object, judging whether the connection rules are satisfied between two nodes of the graph object according to the store registration attribute information corresponding to each node in the graph object and the preset edge connection rule; if satisfied, establishing edge connections between the corresponding nodes to integrate the newly added node into the graph object;

[0092] By integrating newly registered store samples into the graph object as new nodes, the dynamic update of the graph object is realized. This dynamic update mechanism enables the graph object to reflect the current registration behavior and association pattern in real time, providing a more comprehensive and accurate data basis for subsequent illegal registration identification. With the continuous addition of new store samples, the graph object can be continuously updated to maintain its timeliness and accuracy, thereby improving the real-time and effectiveness of illegal registration identification. The preset edge connection rules are based on the analysis and experience summary of historical illegal registration behaviors, and can effectively capture the complex association patterns between store samples.

[0093] After the newly added nodes are integrated into the graph object, the updated graph object can be used for further analysis and prediction. For example, the probability of illegal registration of the newly added nodes is predicted through the graph neural network model, so as to timely identify and handle potential illegal registration behaviors. This real-time monitoring and early warning mechanism can effectively prevent the illegally registered stores from maliciously brushing orders, plucking wool, transaction fraud, etc., thereby maintaining the health of the e-commerce platform and the transaction satisfaction of buyers. In this way, real-time monitoring and early warning of illegal registration behaviors can be achieved, and the security and user experience of the e-commerce platform can be improved. In one embodiment, after determining whether each store in the sample set of stores to be predicted is an illegally registered store, the illegal registration label data of the corresponding store sample can be marked as 0 (not an illegally registered store) and 1 (an illegally registered store), and the store samples of the corresponding stores are associated and stored in the store server. The marked store samples can be used to form a historical registered store sample set, and iteratively train the graph neural network model of this application. In this way, the performance of the graph neural network model can be continuously optimized and improved, so that it can better adapt to the changing illegal registration behavior pattern.

[0094] Step S5230: Import the store registration behavior feature data of the sample set of registered stores to be predicted into the graph object, associate them with corresponding nodes in the graph object, and update the graph object.

[0095] The store registration behavior feature data of each store sample is extracted from the sample set of registered stores to be predicted, and the store registration behavior feature data of each store sample is imported into the graph object and associated with the corresponding node in the graph object. In this way, each node in the graph object not only contains the store unique identifier and store registration attribute information of the store sample, but also contains the store registration behavior feature data, thus forming a complete graph object. The import and association of this store registration behavior feature data enables the graph object to fully reflect the registration behavior characteristics of the store sample, providing a more comprehensive and accurate data basis for the subsequent identification of illegal registrations.

[0096] Based on any embodiment of the method of the present application, after determining whether the corresponding store is an illegally registered store according to the risk probability, the following steps are included:

[0097] Step S5310: When the store is confirmed to be an illegally registered store, the large language model is called to generate an illegal registration report corresponding to the store according to the store registration attribute information and store registration behavior feature data of the store, and the illegal registration report is sent to the user corresponding to the store;

[0098] After confirming that a store is an illegally registered store, the store risk control system will call the big language model. The big language model is a natural language processing model based on deep learning that can understand and generate natural language text. By calling the big language model, the store risk control system can automatically generate a detailed illegal registration report.

[0099] Specifically, the large language model will generate an illegal registration report based on the store registration attribute information and store registration behavior feature data of the store. The illegal registration report usually includes the store's unique identifier, store registration attribute information, store registration behavior feature data, risk assessment results, and corresponding disposal suggestions. For example, the illegal registration report points out that the store's IP address has registered a large number of stores in a short period of time, or the device fingerprint matches the known illegal registration device fingerprint, or the IP address of the registered mobile phone number is inconsistent with the registered IP address. The report will also provide corresponding disposal suggestions, such as restricting certain functions of the store, requiring it to conduct further identity authentication, or directly banning the store. Furthermore, the store risk control system will send the generated illegal registration report to the user corresponding to the store. The sending method can be through email, SMS, platform message, etc. By sending an illegal registration report, users can be notified in time that their store has been identified as an illegal registration store, and corresponding disposal suggestions can be provided, thereby helping users take timely measures to prevent further malicious behavior.

[0100] This step ensures the timely identification and handling of illegally registered stores, improving the security and user experience of the e-commerce platform. By generating a detailed illegal registration report, it can provide users with clear guidance to help them understand and deal with illegal registration behaviors, thereby maintaining the normal operation of the platform and a good user experience.

[0101] Step S5320, iterating the illegal registration report generation event until the number of generated illegal registration reports exceeds a preset report number threshold, calling the large language model to generate corresponding illegal registration regulations according to the generated illegal registration reports, the illegal registration regulations including multiple illegal registration details;

[0102] The store risk control system will record each illegal registration report generated and count the number of reports generated. When the number of illegal registration reports generated exceeds the preset report number threshold, the store risk control system will trigger further processing. The preset report number threshold can be adjusted according to business needs and actual conditions to ensure that timely measures can be taken when illegal registration behaviors are concentrated. Then the store risk control system will call the large language model to generate corresponding illegal registration regulations based on the generated illegal registration reports. Illegal registration regulations are a summary and summary of illegal registration behaviors, usually including multiple illegal registration details. Each illegal registration detail corresponds to a specific illegal registration behavior pattern, such as an IP address registering a large number of stores in a short period of time, the device fingerprint matches the known illegal registration device fingerprint, and the IP address of the registered mobile phone number is inconsistent with the registered IP address.

[0103] It can be further understood that the large language model will analyze the generated illegal registration reports, extract key information and behavior patterns, and generate corresponding illegal registration details. For example, if multiple illegal registration reports mention that a certain IP address has registered a large number of stores in a short period of time, the large language model will extract this behavior pattern as an illegal registration detail. In this way, illegal registration behaviors can be systematically summarized and summarized, providing more comprehensive and accurate guidance for subsequent illegal registration identification.

[0104] Finally, the generated illegal registration regulations will be stored in the store risk control system for subsequent illegal registration identification and disposal. Illegal registration regulations can be used as training data for graph neural network models to help the model better understand illegal registration behaviors and improve the model's prediction accuracy. In addition, illegal registration regulations can also be used to generate illegal registration prevention guidelines for the platform to help users understand and deal with illegal registration behaviors, thereby improving the platform's security and user experience.

[0105] Step S5330: After determining that the illegal registration details exceed the preset threshold number of details, after obtaining the sample set of registered stores to be predicted, call the large language model to determine whether each store in the sample set of registered stores to be predicted is an illegal registration store based on the illegal registration details and the sample set of registered stores to be predicted.

[0106] The store risk control system will first check whether the number of illegal registration details exceeds the preset threshold of the number of details. If it exceeds, the store risk control system can obtain the latest sample set of registered stores to be predicted, call the large language model, and use the illegal registration details and the sample set of registered stores to be predicted as input for comprehensive analysis. The large language model will analyze each store sample in the sample set of registered stores to be predicted one by one according to the illegal registration behavior patterns summarized in the illegal registration details. Specifically, the large language model will extract the store registration attribute information and store registration behavior feature data of each store sample, and match them with the behavior patterns in the illegal registration details. If the registration behavior characteristics of a store sample are highly matched with a detail in the illegal registration details, the large language model will determine that the store sample is an illegal registration store and generate the corresponding risk assessment results. In this way, the large language model can use the behavior patterns summarized in the illegal registration details to quickly and accurately determine whether each store in the sample set of registered stores to be predicted is an illegal registration store, thereby improving the efficiency and accuracy of illegal registration identification. Ultimately, the store risk control system will take corresponding measures based on the judgment results of the large language model, such as restricting certain functions of the store, requiring further identity verification, or directly banning the store to maintain the security of the e-commerce platform and user experience.

[0107] Based on any embodiment of the method of the present application, after determining whether each store in the sample set of registered stores to be predicted is an illegally registered store, the method includes:

[0108] Step S5331, obtaining a sample set of illegally registered stores corresponding to the illegally registered stores judged by the large language model in the sample set of registered stores to be predicted;

[0109] After determining whether each store in the sample set of registered stores to be predicted is an illegally registered store, it is first necessary to obtain the sample set of illegally registered stores that corresponds to the illegally registered stores in the sample set of registered stores to be predicted that are determined to be illegally registered stores by the large language model. Specifically, the store risk control system will identify which stores in the sample set of registered stores to be predicted are determined to be illegally registered stores based on the judgment results of the large language model. Then all the store samples that are determined to be illegally registered stores are collected together to form a sample set of illegally registered stores. This sample set of illegally registered stores contains store samples that are determined to be illegally registered stores by the large language model.

[0110] Step S5332: Obtain a graph object based on a sample set of historically registered stores, and update the graph object in combination with the sample set of illegally registered stores;

[0111] The sample set of illegally registered stores is a set of store samples that are judged as illegally registered stores by the large language model in the previous step. The store risk control system will first obtain a graph object based on the sample set of historical registered stores, and add the store samples in the sample set of illegally registered stores one by one to the graph object constructed based on the sample set of historical registered stores, and update the graph object as new nodes. The updated graph object not only contains the information of historical registered store samples, but also includes the latest samples of illegally registered stores. These samples of illegally registered stores are stores that are judged as illegally registered by the large language model. The subsequent store risk control system uses the trained graph neural network model to predict the risk probability of each store in the sample of illegally registered stores, so as to determine the accuracy of the large language model in predicting illegal registrations. Please refer to the subsequent specific implementation method for this step, which will not be repeated here.

[0112] Step S5333: input the updated graph object into the trained graph neural network model, determine the number of illegally registered stores in the sample set of illegally registered stores, divide the number by the total number of stores in the sample set of illegally registered stores, and obtain the accuracy rate of the large language model in determining illegally registered stores;

[0113] Input the updated graph object in the previous step into the graph neural network model that was pre-trained using the store samples of the historical registered store sample set. After the updated graph object is input into the graph neural network model, the graph neural network model will predict the risk probability of illegal registration events for each newly added node in the graph object (i.e., the sample of illegally registered stores). Then, count the number of illegally registered stores in the sample set of illegally registered stores that are judged as illegally registered by the graph neural network model, and divide this number by the total number of stores in the sample set of illegally registered stores to obtain the accuracy of the large language model in judging illegally registered stores. This accuracy reflects the performance of the large language model in identifying illegally registered stores. In this way, the store risk control system can evaluate the accuracy of the large language model in identifying illegally registered stores, thereby providing a basis for subsequent model optimization and decision-making.

[0114] Step S5334: When the accuracy reaches a preset accuracy threshold, the large language model is enabled to identify illegal stores.

[0115] The store risk control system sets a preset accuracy threshold to determine whether the performance of the large language model in identifying illegally registered stores meets the expected standards. This accuracy threshold can be adjusted according to business needs and actual conditions to ensure that the performance of the large language model in identifying illegally registered stores can meet the security and user experience requirements of the platform. When the accuracy of the large language model in determining illegally registered stores reaches or exceeds the preset accuracy threshold, the store risk control system can choose to enable the large language model to identify illegal stores. After enabling the large language model, the store risk control system will identify and handle potential illegally registered stores in real time based on the judgment results of the large language model, thereby improving the security and user experience of the e-commerce platform. In this way, the store risk control system can ensure that the performance of the large language model in identifying illegally registered stores meets the expected standards, thereby providing guarantees for the normal operation of the platform and a good user experience.

[0116] In one embodiment, the trained graph neural network model and the large language model can be used simultaneously for prediction, and then the illegal registration judgment condition can be set to determine whether it is an illegal registration. For example, a comprehensive risk probability threshold can be set. When the comprehensive risk probability of the prediction results of the graph neural network model and the large language model exceeds the threshold, the store sample is determined to be an illegally registered store. It can also be set that when at least one of the graph neural network model and the large language model determines that a store sample is an illegally registered store, the store sample is determined to be an illegally registered store.

[0117] See also Figure 3 According to one aspect of the present application, a store illegal registration identification device is provided, including a sample set acquisition module 5100 to be predicted, a graph object update module 5200, and a risk probability prediction module 5300, wherein the sample set acquisition module 5100 to be predicted is configured to obtain a sample set of registered stores to be predicted, the sample set of registered stores to be predicted includes store samples of multiple stores, and the store samples describe the basic information and behavior characteristics of the corresponding stores; the graph object update module 5200 is configured to correspond to a graph object constructed based on store samples in a historical registered store sample set, and construct the store samples in the sample set of registered stores to be predicted as new nodes in the graph object to update the graph object; the risk probability prediction module 5300 is configured to input the updated graph object into a graph neural network model pre-trained using store samples of the historical registered store sample set, predict the risk probability of each store in the sample set of registered stores to be predicted belonging to an illegal registration event, and determine whether the corresponding store belongs to an illegal registration store based on the risk probability.

[0118] On the basis of any embodiment of the device of the present application, the sample set to be predicted acquisition module 5100 includes: a historical sample set acquisition submodule, which is configured to acquire the historical registered store sample set, wherein the historical registered store sample set includes multiple store samples, and each store sample includes the store unique identifier of the corresponding store, store registration attribute information, store registration behavior feature data, and illegal registration label data corresponding to whether it is an illegally registered store; a graph object construction submodule, which is configured to construct a graph object for training the graph neural network model according to the historical registered store sample set, wherein the nodes of the graph object are constructed with the store samples; a graph neural network model training submodule, which is configured to input the constructed graph object into the graph neural network model, and use the illegal registration label data of each store sample for supervised training to train the graph neural network model to a convergence state.

[0119] Based on any embodiment of the device of the present application, the graph object construction submodule includes: a node construction submodule, which is configured to construct the corresponding node in the graph object with the store unique identifier of the store sample in the historical registered store sample set; an edge connection submodule, which is configured to determine whether the store registration attribute information of any two nodes of the graph object meets the edge connection rule based on the store registration attribute information in the store sample and the preset edge connection rule, and if so, establish an edge connection between the two nodes; a graph object feature import submodule, which is configured to import the store registration behavior feature data and the illegal registration label data corresponding to each store sample into the graph object, and associate them with the corresponding nodes in the graph object to form a complete graph object.

[0120] On the basis of any embodiment of the device of the present application, the edge connection submodule includes any one or more of the following modules: a device fingerprint judgment submodule, configured to use the device fingerprint of the terminal device used when registering a store as the store registration attribute information in the corresponding store sample, detect whether the device fingerprints corresponding to any two nodes in the graph object are the same, and if they are the same, establish an edge connection between the two nodes; a contact information judgment submodule, configured to use the contact information and registration time when registering a store as the store registration attribute information in the corresponding store sample, calculate the similarity between the contact information corresponding to any two nodes in the graph object, and if the similarity exceeds a preset first similarity threshold, and the corresponding registration time interval between the two nodes does not exceed a preset first time threshold, establish an edge connection between the two nodes; an IP address judgment submodule, configured to use the IP address, contact information and registration time when registering a store as the store registration attribute information in the corresponding store sample, compare whether the corresponding IP addresses between any two nodes in the graph object are the same, and if they are the same and the similarity of the corresponding contact information between the two nodes exceeds a preset second similarity threshold and the registration time interval between the two nodes does not exceed a preset second time threshold, establish an edge connection between the two nodes.

[0121] Based on any embodiment of the device of the present application, the graph object updating module 5200 includes: a data acquisition submodule, configured to acquire store samples in the sample set of registered stores to be predicted and a graph object constructed based on a historical registered store sample set, wherein the store samples in the sample set of registered stores to be predicted include a store unique identifier of the corresponding store, store registration attribute information, and store registration behavior feature data; a new node construction submodule, configured to use the store unique identifier in the store sample as a new node of the graph object, and determine whether the connection rules are met between each node of the graph object according to the corresponding store registration attribute information of each node in the graph object and the preset edge connection rule. If so, an edge connection is established between the corresponding nodes to integrate the new node into the graph object; a feature data importing submodule, configured to import the store registration behavior feature data of the sample set of registered stores to be predicted into the graph object, associate it with the corresponding node in the graph object, and update the graph object.

[0122] On the basis of any embodiment of the device of the present application, the risk probability prediction module 5300 includes: an illegal registration report generation submodule, which is configured to call the large language model to generate a corresponding illegal registration report for the store according to the store registration attribute information and store registration behavior characteristic data of the store when confirming that the store is an illegally registered store, and send the illegal registration report to the corresponding user of the store; an illegal registration regulations generation submodule, which is configured to iterate the above-mentioned illegal registration report generation event until the number of generated illegal registration reports exceeds a preset report number threshold, and then call the large language model to generate corresponding illegal registration regulations according to the generated illegal registration report, and the illegal registration regulations include multiple illegal registration details; an illegal registration store judgment submodule, which is configured to call the large language model to determine whether each store in the sample set of registered stores to be predicted is an illegally registered store after determining that the illegal registration details exceed the preset detail number threshold, after obtaining the sample set of registered stores to be predicted. According to the illegal registration details and the sample set of registered stores to be predicted, each store in the sample set of registered stores to be predicted is an illegally registered store.

[0123] On the basis of any embodiment of the device of the present application, after the illegal registration store judgment submodule, it includes: an illegal registration sample set acquisition submodule, which is configured to obtain the illegal registration store sample set corresponding to the illegal registration store judged as the illegal registration store by the large language model in the registered store sample set to be predicted; a graph object update submodule one, which is configured to obtain a graph object composed of a historical registered store sample set, and update the graph object in combination with the illegal registration store sample set; a judgment accuracy calculation submodule, which is configured to input the updated graph object into a trained graph neural network model, determine the number of illegal registration stores in the illegal registration store sample set, divide the number by the total number of stores in the illegal registration store sample set, and obtain the accuracy of the large language model in judging the illegal registration stores; a large language model activation submodule, which is configured to enable the large language model to identify illegal stores when the accuracy reaches a preset accuracy threshold.

[0124] In order to solve the above technical problems, the present application also provides a computer device. Figure 4As shown, a schematic diagram of the internal structure of a computer device. The computer device includes a processor, a computer-readable storage medium, a memory, and a network interface connected via a system bus. Among them, the computer-readable storage medium of the computer device stores an operating system, a database, and computer-readable instructions. The database may store a control information sequence. When the computer-readable instructions are executed by the processor, the processor can implement a method for identifying illegal store registrations. The processor of the computer device is used to provide computing and control capabilities to support the operation of the entire computer device. The memory of the computer device may store computer-readable instructions. When the computer-readable instructions are executed by the processor, the processor can execute the method for identifying illegal store registrations of the present application. The network interface of the computer device is used to connect and communicate with a terminal. Those skilled in the art can understand that Figure 4 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0125] In this embodiment, the processor is used to execute Figure 3 The memory stores the program code and data required to execute the above modules or submodules. The network interface is used to transmit data between user terminals or servers. The memory in this embodiment stores the program code and data required to execute all modules / submodules in the store registration violation identification device of this application, and the server can call the program code and data of the server to execute the functions of all submodules.

[0126] The present application also provides a storage medium storing computer-readable instructions. When the computer-readable instructions are executed by one or more processors, the one or more processors execute the steps of the store registration violation identification method of any embodiment of the present application.

[0127] A person skilled in the art can understand that all or part of the processes in the above-mentioned embodiments of the present application can be implemented by instructing the relevant hardware through a computer program, and the computer program can be stored in a computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, the aforementioned storage medium can be a computer-readable storage medium such as a disk, an optical disk, a read-only memory (ROM), or a random access memory (RAM).

[0128] It will be understood by those skilled in the art that the various operations, methods, steps, measures, and schemes in the processes discussed in this application may be alternated, changed, combined, or deleted. Furthermore, other steps, measures, and schemes in the various operations, methods, and processes discussed in this application may also be alternated, changed, rearranged, decomposed, combined, or deleted. Furthermore, the steps, measures, and schemes in the various operations, methods, and processes in the prior art that are open source and in this application may also be alternated, changed, rearranged, decomposed, combined, or deleted.

[0129] The above description is only a partial implementation method of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.

Claims

1. A method for identifying illegal store registration, characterized in that: include: Obtaining a sample set of registered stores to be predicted, wherein the sample set of registered stores to be predicted includes store samples of multiple stores, and the store samples describe basic information and behavior characteristics of corresponding stores; Corresponding to the graph object constructed based on the store samples in the historical registered store sample set, the store samples in the registered store sample set to be predicted are constructed as new nodes in the graph object to update the graph object; The updated graph object is input into the graph neural network model pre-trained using the store samples of the historical registered store sample set, and the risk probability of each store in the registered store sample set to be predicted belonging to an illegal registration event is predicted, and whether the corresponding store is an illegally registered store is determined based on the risk probability.

2. The method for identifying illegal store registration according to claim 1, characterized in that: Before obtaining the sample set of registered stores to be predicted, include: Acquire the historical registered store sample set, the historical registered store sample set comprising a plurality of store samples, each store sample comprising a store unique identifier of the corresponding store, store registration attribute information, store registration behavior feature data, and illegal registration label data corresponding to whether the store is illegally registered; Constructing a graph object for training the graph neural network model according to the historical registered store sample set, wherein nodes of the graph object are constructed with the store samples; The constructed graph object is input into the graph neural network model, and supervised training is performed using the illegal registration label data of each store sample to train the graph neural network model to a convergence state.

3. The method for identifying illegal store registration according to claim 2, characterized in that: A graph object for training the graph neural network model is constructed according to the historical registered store sample set, including: Constructing corresponding nodes in the graph object with store unique identifiers of store samples in the historical registered store sample set; Based on the store registration attribute information in the store sample and a preset edge connection rule, determine whether the store registration attribute information of any two nodes of the graph object satisfies the edge connection rule, and if so, establish an edge connection between the two nodes; The store registration behavior feature data and the illegal registration label data corresponding to each store sample are imported into the graph object, associated with corresponding nodes in the graph object, and form a complete graph object.

4. The method for identifying illegal store registration according to claim 3, characterized in that: Based on the store registration attribute information in the store sample and the preset edge connection rule, it is determined whether the store registration attribute information of any two nodes of the graph object satisfies the edge connection rule, and if so, an edge connection is established between the two nodes, including any one or more of the following steps: Using the device fingerprint of the terminal device used when registering a store as the store registration attribute information in the corresponding store sample, detecting whether the device fingerprints corresponding to any two nodes in the graph object are the same, and if they are the same, establishing an edge connection between the two nodes; Using the contact information and registration time when the store is registered as the store registration attribute information in the corresponding store sample, calculate the similarity between the contact information corresponding to any two nodes in the graph object, and if the similarity exceeds a preset first similarity threshold, and the corresponding registration time interval between the two nodes does not exceed the preset first time threshold, establish an edge connection between the two nodes; The IP address, contact information and registration time when the store is registered are used as the store registration attribute information in the corresponding store sample, and the corresponding IP addresses between any two nodes in the graph object are compared to see whether they are the same; if they are the same and the similarity of the corresponding contact information between the two nodes exceeds a preset second similarity threshold and the registration time interval between the two nodes does not exceed a preset second time threshold, an edge connection is established between the two nodes.

5. The method for identifying illegal store registration according to any one of claims 1 to 4, characterized in that: Corresponding to the graph object constructed based on the store samples in the historical registered store sample set, constructing the store samples in the registered store sample set to be predicted as new nodes in the graph object to update the graph object, including: Acquire store samples in the sample set of registered stores to be predicted and a graph object formed based on the sample set of historical registered stores, wherein the store samples in the sample set of registered stores to be predicted include store unique identifiers, store registration attribute information, and store registration behavior feature data of corresponding stores; The unique identification of the store in the store sample is used as a newly added node of the graph object, and according to the store registration attribute information corresponding to each node in the graph object and the preset edge connection rule, it is determined whether the connection rule is satisfied between two nodes of the graph object, and if so, an edge connection is established between the corresponding nodes to integrate the newly added node into the graph object; The store registration behavior feature data of the sample set of registered stores to be predicted is imported into the graph object, associated with corresponding nodes in the graph object, and the graph object is updated.

6. The method for identifying illegal store registration according to any one of claims 1 to 4, characterized in that: After judging whether the corresponding store is a store registered in violation of regulations based on the risk probability, the following steps are performed: When a store is confirmed to be an illegally registered store, the large language model is called to generate an illegal registration report corresponding to the store based on the store registration attribute information and store registration behavior feature data of the store, and the illegal registration report is sent to the user corresponding to the store; Iterate the illegal registration report generation event until the number of generated illegal registration reports exceeds a preset report number threshold, and call the large language model to generate corresponding illegal registration regulations according to the generated illegal registration reports, wherein the illegal registration regulations include multiple illegal registration details; After determining that the illegal registration details exceed the preset detail number threshold, after obtaining the sample set of registered stores to be predicted, the large language model is called to judge whether each store in the sample set of registered stores to be predicted is an illegal registration store based on the illegal registration details and the sample set of registered stores to be predicted.

7. The method for identifying illegal store registration according to claim 6, characterized in that: After determining whether each store in the sample set of registered stores to be predicted is an illegally registered store, the following steps are performed: Obtain a sample set of illegally registered stores corresponding to the illegally registered stores judged as illegally registered stores by the large language model in the sample set of registered stores to be predicted; Obtain a graph object based on a sample set of historically registered stores, and update the graph object in combination with the sample set of illegally registered stores; Input the updated graph object into the trained graph neural network model, determine the number of illegally registered stores in the sample set of illegally registered stores, divide the number by the total number of stores in the sample set of illegally registered stores, and obtain the accuracy rate of the large language model in determining illegally registered stores; When the accuracy reaches a preset accuracy threshold, the large language model is enabled to identify illegal stores.

8. A device for identifying illegal store registration, characterized in that: include: A sample set acquisition module to be predicted, configured to acquire a sample set of registered stores to be predicted, wherein the sample set of registered stores to be predicted includes store samples of multiple stores, and the store samples describe basic information and behavior characteristics of corresponding stores; A graph object updating module is configured to correspond to a graph object constructed based on store samples in a historical registered store sample set, and to construct store samples in the registered store sample set to be predicted as newly added nodes in the graph object to update the graph object; The risk probability prediction module is configured to input the updated graph object into the graph neural network model pre-trained using the store samples of the historical registered store sample set, predict the risk probability of each store in the registered store sample set to be predicted belonging to an illegal registration event, and determine whether the corresponding store is an illegally registered store based on the risk probability.

9. A store registration violation identification device, comprising a central processor and a memory, characterized in that: The central processing unit is used to call and run the computer program stored in the memory to execute the steps of the method according to any one of claims 1 to 7.

10. A non-volatile readable storage medium, characterized in that: It stores a computer program implemented according to the method described in any one of claims 1 to 7 in the form of computer-readable instructions, and when the computer program is called and executed by a computer, the steps included in the corresponding method are executed.

Images