Registered garbage shop troubleshooting method and device, equipment and medium

Abstract

The invention relates to a junk registered shop troubleshooting method and device, equipment and medium, and the method comprises the steps: responding to a shop registration event, and recognizing whether a to-be-detected shop belongs to a junk registered shop or not based on the registration feature data of the to-be-detected shop; if the to-be-detected shop does not belong to the junk registered shop, monitoring and acquiring feature data after registration of the to-be-detected shop, inputting the feature data of the to-be-detected shop during registration and the feature data after registration into a junk registration detection model, and judging whether the to-be-detected shop belongs to the junk registered shop or not; and marking whether the registered feature data and the registered feature data of the to-be-detected shop belong to a junk registration shop or not, storing the junk registration shop in a training data set, and periodically iteratively updating the junk registration detection model by using the training data set. According to the invention, the accuracy of identifying the junk registered shop can be obviously improved.

Description

Technical Field

[0001] The present application relates to e-commerce information security technology, and in particular to a method for checking spam-registered stores 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 junk-registered stores. Junk-registered stores usually refer to stores 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] At present, the spam registration detection methods commonly used by e-commerce platforms mainly include the authenticity verification of registration information, such as SMS and email verification, and various verification code technologies to prevent automated spam registration. In addition, it also includes related searches such as IP, mobile phone number, and email address based on historical reserve blacklists. However, most of these methods focus on detection when registration occurs, and lack continuous monitoring and analysis after the registration behavior is completed, and cannot fully identify and prevent spam registration behavior.

[0004] In addition, existing detection methods are also limited in accuracy and timeliness. Although some methods can accurately identify spam registrations, they often require complex calculations and analysis, resulting in slow detection; while other methods can respond quickly, but their accuracy is not high and they are prone to missed detections or false detections. Therefore, how to design a fast and accurate method to detect spam registration stores has become a technical problem that e-commerce platforms urgently need to solve. Summary of the invention

[0005] The purpose of this application is to solve the above-mentioned problem and provide a method for detecting junk registered stores and its corresponding devices, equipment, non-volatile readable storage medium, and computer program product.

[0006] According to one aspect of the present application, a method for checking junk registered stores is provided, comprising the following steps:

[0007] Respond to store registration events and identify whether the store to be detected is a spam registration store based on the registration feature data of the store to be detected;

[0008] If the store to be detected is not a junk registration store, monitor and obtain its post-registration feature data, input the registration feature data and post-registration feature data of the store to be detected into the junk registration detection model, and determine whether the store to be detected is a junk registration store;

[0009] The registration feature data and post-registration feature data of the store to be detected are marked as whether they belong to junk registration stores and stored in a training data set, and the junk registration detection model is regularly iterated and updated with the training data set.

[0010] According to another aspect of the present application, a device for checking junk registered stores is provided, comprising:

[0011] The in-process detection module is configured to respond to store registration events and identify whether the store to be detected is a spam registration store based on the characteristic data of the store to be detected during registration;

[0012] The post-detection module is configured to monitor and obtain post-registration feature data of the store to be detected if the store to be detected is not a junk registration store, input the registration feature data and post-registration feature data of the store to be detected into a junk registration detection model, and determine whether the store to be detected is a junk registration store;

[0013] The iterative update module is configured to mark the registration feature data and post-registration feature data of the store to be detected as whether it is a junk registration store and store it in a training data set, and regularly iterate and update the junk registration detection model with the training data set.

[0014] According to another aspect of the present application, a device for checking junk registered stores 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 junk registered store checking 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 junk registered store screening 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] According to another aspect of the present application, a computer program product is provided, including a computer program / instruction, wherein when the computer program / instruction is executed by a processor, the steps of the method described in any one of the embodiments of the present application are implemented.

[0017] Compared with the prior art, this application has the following significant technical advantages and beneficial effects:

[0018] First, by responding to store detection events, the store's registration feature data is detected during the process, and by monitoring the store's post-registration feature data for post-detection, this application can more comprehensively identify spam registration behavior by combining the in-process detection and post-detection. In-process detection uses the registration feature data to quickly screen suspicious stores, while post-detection further confirms whether it is a spam registration by performing subsequent detection on the post-registration feature data of the stores that were missed during the in-process detection, thereby significantly improving the accuracy and coverage of identification.

[0019] Secondly, this application realizes dynamic assessment of store risk status by continuously monitoring the post-registration feature data of the store and combining it with the feature data at the time of registration. This dynamic assessment mechanism enables the screening method to adapt to the changes in spam registration behavior and promptly discover new suspicious patterns, thereby providing continuous risk management capabilities for e-commerce platforms.

[0020] In addition, this application uses an automated data annotation and storage mechanism to continuously optimize the spam registration detection model. Through regular iterative updates, the model learns the latest registration behavior patterns, automatically adapts and improves its ability to predict future spam registration behaviors. This approach not only improves the efficiency of investigation, but also ensures the long-term effectiveness and adaptability of the model, which helps to maintain the operating order of the e-commerce platform in the long term. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 The corresponding network architecture is exemplarily deployed in the e-commerce platform for the store risk control system of this application;

[0022] Figure 2 A flowchart of an embodiment of a method for checking junk registered stores of the present application;

[0023] Figure 3 This is a functional block diagram of the garbage registration store screening device of this application;

[0024] Figure 4 This is a structural diagram of a junk registered store screening device used in this application. DETAILED DESCRIPTION

[0025] 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.

[0026] The security server 81 can be used as the main execution body of the store risk control system of the present application. The store risk control system predicts whether the online store is a junk registered store based on the characteristic data of the online store at the time of registration and the characteristic data after registration in the e-commerce platform. The junk registered store screening 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 a junk registered store, it can implement risk control and restriction processing on it. Of course, it can also be handed over to the manual management user for further manual review and confirmation before implementing risk control and restriction processing.

[0027] The store server 82 can be used to deploy one or more online stores on the Internet platform, and various network requests for accessing online stores can be responded to by the corresponding application services provided by the store server. Each online store will generate various corresponding feature data during registration and configuration operations after registration, such as the user name, mobile phone number, email address submitted during registration, the IP address and feature code of the registration device used, etc., which can be directly used as the feature data during registration, or further determined based on other derived feature data during registration or feature data after registration, etc.; for example, after registration, the online store needs to perform corresponding payment method configuration, product shelf configuration, store logistics configuration, etc., and various post-registration feature data will also be generated accordingly. Of course, the feature data during registration and feature data after registration of online stores are not limited to the examples here.

[0028] The terminal device 80 may be used to trigger a network request so as to use various application services in the store server 82, such as browsing the online store, executing an ordering process for merchandise items in the online store, etc.

[0029] The security server 81 can read various characteristic data of the online store in the store server 82, which is used by the store risk control system to determine whether the online store is a junk registered store, and to conduct further manual investigation when necessary.

[0030] The store risk control system running in the security server 81 can, with the help of the method of the present application, use a spam registration analysis model to identify whether the online store that needs to be detected, that is, the store to be detected, is a spam registration store. This registration analysis model can also use a self-learning mechanism, with the help of various feature data continuously obtained from each online store and the factual data corresponding to whether these online stores are spam registration stores as a training data set. Through this training data set, the spam registration analysis model is iteratively trained and updated regularly or irregularly, thereby continuously improving the spam registration analysis model's ability to identify the risk level of whether each online store is a spam registration store, and continuously improving the e-commerce platform's risk control ability for spam registration stores.

[0031] See also Figure 2 According to a method for checking junk registered stores provided by the present application, in one embodiment, the method comprises the following steps:

[0032] Step S1100, responding to a store registration event, identifying whether the store to be detected is a junk registration store based on the registration feature data of the store to be detected;

[0033] In the daily operation of e-commerce platforms, the registration of new stores is a standard and frequent event. Whenever a merchant intends to open a new store on the platform, he enters the e-commerce platform to register a new store, triggering the submission of a store registration request to the store server, which can trigger the corresponding store registration event. When the security server receives the store registration event, it can use the corresponding new store as a store to be detected, and perform security detection based on the registration feature data formed during the registration process of the store to be detected, so as to identify whether the store to be detected is a spam registration store.

[0034] The characteristic data at registration time is a set of key information collected during the store registration process of the store to be tested, which constitutes the basis for evaluating the registration behavior. These characteristic data can not only include the basic information submitted by the registrant, such as contact data, identity information data, etc., but also cover the behavior patterns in the registration process, the equipment and network environment information used, and the specific time point when the registration behavior occurs, etc. For example, the IP address, device type, operating system information, page jump frequency, speed of filling in information, etc. used by the registrant can all constitute part of the characteristic data at registration time. The nature of these data is diverse, including but not limited to text information, timestamps, device identifiers, etc., and in specific applications, they can be flexibly implemented by those skilled in the art according to the inspiration of the embodiments of this application.

[0035] Based on these registration feature data, the process of identifying whether the store to be detected is a junk registered store may include single or multiple levels of analysis. For example, any one of the following embodiments can be used to determine whether the store to be detected is a junk registered store.

[0036] In one embodiment, the store risk control system adapts to the situation where the characteristic information at the time of registration uses contact information data. Based on the characteristic information at the time of registration, the store risk control system checks whether the contact information used for the registration of the store to be tested appears in a known blacklist. Once a match is successful, the store is identified as a spam registration store and becomes a high-risk store.

[0037] In another embodiment, to adapt to the situation where the characteristic information at the time of registration uses contact data, the store risk control system verifies the authenticity of the registration information by sending a verification code. If the registrant fails to respond correctly within the specified time, his registration behavior may be deemed suspicious, and thus it will be identified as a spam registration store.

[0038] In another embodiment, behavior analysis during the registration process is also key, such as abnormally fast filling or frequent page refreshes, which may be signs of automated script operations. Therefore, the time information corresponding to the filling of various registration information during the registration process can also be used as registration feature data to identify whether the store to be detected is a spam registration store.

[0039] In another embodiment, analysis of the device and network information used by the registrant is equally important. For example, an IP address associated with multiple store registrations may indicate organized spam registration behavior. Therefore, the UUID and network information of the user's device can also be used as feature data during registration to identify whether the store to be tested is a spam registration store.

[0040] In another embodiment, the time series corresponding to the same device executing multiple new store registrations can also reveal a large number of registration requests initiated from the same IP address in a short period of time, which is usually related to batch registration behavior. Therefore, the registration behavior data initiated by the store to be detected on the same device in the same period of time can be used as the registration feature data of the store to be detected, to determine whether the current store to be detected is a spam registration store.

[0041] In another embodiment, rule extraction can also be performed in advance on the feature data corresponding to regular registered shops and junk registered shops formed in the e-commerce platform to generate one or more real-time policy rules corresponding to identifying junk registered shops and their required registration feature data. Then, when it is necessary to determine whether the shop to be detected is a junk registered shop, the registration feature data corresponding to the real-time policy rules are first extracted for the shop to be detected, and then the registration feature data is judged according to the real-time policy rules. When the real-time policy rules are hit, the shop to be detected can be determined as a junk registered shop.

[0042] It is not difficult to understand that any one or more of the various embodiments listed above can determine in advance the characteristic data at the time of registration corresponding to the store to be detected, and then use the corresponding method of the embodiment to detect the characteristic data at the time of registration corresponding to the store to be detected, thereby effectively determining whether the store to be detected is a junk registered store. It can be further understood that by responding to the store registration event triggered by each newly registered store based on the event response mechanism through the store risk control system, potential junk registered stores can be identified from a large number of newly registered stores, providing solid technical support for the risk management and credit system of the e-commerce platform.

[0043] Step S1200: If the store to be detected is not a junk registration store, monitor and obtain its post-registration feature data, input the registration feature data and post-registration feature data of the store to be detected into a junk registration detection model, and determine whether the store to be detected is a junk registration store;

[0044] In the operation of the e-commerce platform, once the registration request of a new store is submitted and the store registration event is triggered, step S1100 will be automatically executed to analyze the store's registration feature data to identify possible junk registration behavior. However, not all stores to be detected will be clearly identified as junk registration stores when they are registered. For those stores to be detected that are not determined to be junk registration stores in the preliminary detection of step S1100, step S1200 provides further monitoring and analysis means to significantly improve the accuracy of identifying junk registration stores.

[0045] The core of step S1200 is to monitor and obtain the behavioral data of the store after registration is completed, and construct the post-registration feature data based on this. The post-registration feature data is mainly used to reflect the behavior pattern and operation characteristics of the store to be detected after successful registration. The post-registration feature data is extracted by reflecting the basic behavior data corresponding to these behavior patterns and operation characteristics, which can be used to evaluate the true intention and operation status of the store to be detected. The post-registration feature data may include but is not limited to any number of the key items such as the store's product listing frequency, payment method settings, customer service interaction, order processing speed, relationship network characteristics, etc., which can be flexibly set.

[0046] After the store completes registration, the store risk control system can automatically start a monitoring mechanism for the store, or only automatically start a monitoring mechanism for the stores to be detected that are not identified as junk registered stores in step S1100, to collect and analyze these post-registration feature data. For example, if a store quickly puts a large number of goods on the shelves after registration, and the prices of the goods are abnormally low or the descriptions are unclear, this may be a sign of junk registration behavior. Similarly, if the store has a single payment method that only supports cash on delivery and lacks online payment options, this may also indicate that the store is risky. Based on this, feature extraction can be performed based on the behavioral data corresponding to these operations to obtain the corresponding post-registration feature data.

[0047] In a further optimized embodiment, the store risk control system can also analyze the store's operational configuration behavior, such as the complexity of the store's decoration, the level of detail in the product description, and the response speed to customer feedback, etc., and extract features based on the behavioral data corresponding to these behaviors to obtain corresponding post-registration feature data. Since these behavioral features can provide additional information about the intentions and capabilities of the store operator, they can be used to identify whether the store to be tested is a junk registration store. For example, a normally operating store will usually invest resources to optimize the store page and product descriptions, while a junk registration store may invest less in this regard, or not optimize at all. These two types of stores will inevitably reflect identifiable differences in the post-registration feature data.

[0048] The store risk control system introduces a spam registration detection model that has been pre-trained to convergence to identify whether the store to be detected is a spam registration store. The spam registration detection model comprehensively analyzes the store's registration feature data and post-registration feature data, and uses corresponding data analysis technology and machine learning algorithms to conduct risk assessment on the store's registration behavior. The design of this model aims to improve the accuracy and efficiency of identifying spam registration behavior through automated means.

[0049] The training process of the spam registration detection model begins with the analysis of a large amount of historical store registration data, which includes samples of normal registrations and spam registrations. By learning and extracting features from these samples, the model can identify the key features that distinguish normal from spam registrations. In the feature engineering stage, the model relies on both the registration feature data used in step S1100 and the post-registration feature data extracted in step S1200, as well as whether the corresponding store is marked as a spam registration store as a supervisory label. Based on these feature data and supervisory labels, the spam registration detection model is subjected to classification training, thereby enabling the spam registration detection model to have the ability to identify whether the store to be detected is a spam registration store based on the corresponding registration feature data and post-registration feature data of the store to be detected.

[0050] It is not difficult to understand that by comprehensively considering the feature data at the time of registration and the feature data after registration, the store risk control system will input these data into the spam registration detection model for in-depth analysis, thereby realizing the learning and identification of the store's behavior patterns based on the in-process data and post-data corresponding to the store registration, thereby determining whether the store to be detected is a spam registration store.

[0051] In some embodiments, during the training phase of the spam registration detection model, algorithms such as decision trees, support vector machines, random forests or neural networks can be used to perform pattern recognition training on feature data. By continuously adjusting model parameters and optimizing algorithms, the model can achieve a high-accuracy classification effect. In addition, the performance of the model is evaluated through indicators such as accuracy and recall rate to ensure its effectiveness in practical applications.

[0052] The trained spam registration detection model can analyze the real-time collected feature data of the stores to be detected during and after registration, and generate a probability value for each store to determine whether it is a spam registration store. This probability value is based on the similarity between the store's behavior and the spam registration behavior pattern, thus helping e-commerce platforms quickly identify potential spam registration stores.

[0053] It is not difficult to understand that the spam registration detection model provides e-commerce platforms with a powerful tool to maintain a healthy operating environment for the platform, protect the interests of consumers and honest merchants, and greatly improve the identification and response speed to spam registration behaviors.

[0054] Step S1300: annotate the registration feature data and post-registration feature data of the store to be detected as junk registration stores and store them in a training data set, and regularly iterate and update the junk registration detection model with the training data set.

[0055] The training data set is the basic data used in this application to dynamically update the spam registration detection model. It consists of the marked historical store registration feature data and post-registration feature data, which reflect the store's behavior patterns during and after the registration process. For example, the training data set can not only include the store's basic information, such as contact information, identity information, equipment and network usage, through the store's registration feature data, but also include behavioral analysis data, such as page jump frequency, information filling speed, etc., through the registration feature data. In addition, the training data set also incorporates post-registration feature data that reflects the store's operating status, such as corresponding feature data such as product listing behavior and payment method settings.

[0056] These feature data are the specific data determined to be used in step S1100 and step S1200. After careful screening and labeling in advance, normal stores and junk registered stores are clearly distinguished, providing rich learning and verification materials for the junk registration detection model. When a store to be detected is identified as a junk registered store through step S1100 and step S1200, or is manually confirmed as a junk registered store or a normal store, that is, a regular registered store, the corresponding feature data at the time of registration and the feature data after registration in the training data set can be labeled accordingly, and the store to be detected also becomes a historical store in the training data set in the sense of time scale.

[0057] Through continuous data accumulation and updating, the training data set can reflect the latest risk trends and behavior patterns, ensuring that the spam registration detection model can adapt to the ever-changing risk environment and improve the accuracy and efficiency of identifying spam registration behaviors. In this step, by annotating the store's registration feature data and post-registration feature data, we can clearly distinguish between spam registration stores and normal stores, and systematically store these annotated data in the training data set. This process not only provides rich learning materials for the model, but also enables the training data set to reflect the latest risk trends and pattern changes through continuous data accumulation.

[0058] In the daily operation of e-commerce platforms, new junk registration methods and behavior patterns will continue to emerge as time goes by and the transaction volume increases. Step S1300 ensures that the junk registration detection model can learn the latest data features by regularly updating the training data set, thereby improving the model's ability to identify emerging risks. This iterative update mechanism is the key to maintaining high accuracy and adaptability of the model.

[0059] The updated training data set will be used to retrain the spam registration detection model. In different embodiments, this process can be performed by full retraining or incremental learning. Full retraining refers to using the updated training data set to fully retrain the model, while incremental learning is based on the existing model, only learning the newly added data in the training data set to update the model's judgment criteria.

[0060] Iterative training of the junk registration detection model can be performed regularly or irregularly. Considering that the daily store registration events of the e-commerce platform are massive, in one embodiment, the junk registration detection model is restarted at a certain time point every day with the updated training data set in units of natural days. After the training is completed, the latest version of the junk registration detection model is obtained, and the latest version is used to replace the original version in use, and the update of the updated junk registration detection model is completed. After the update, the latest junk registration detection model can be used to identify newly registered stores to determine whether the newly registered stores are junk registration stores. In another embodiment, the cumulative number of new store registrations in the e-commerce platform can be accumulated and counted since the last time the junk registration detection model was updated. At a specific time of the day when the cumulative number of new store registrations reaches a preset threshold, such as 12 o'clock midnight, the retraining and updating of the junk registration detection model is started. In this way, frequent training of the junk registration detection model can be avoided, saving resource overhead.

[0061] In one embodiment, the training samples and supervision labels in the training data set can also be used to verify the spam registration detection model that has just completed training, and the recall accuracy of the spam registration detection model can be statistically calculated based on the verification results. When the recall accuracy is lower than the corresponding preset threshold, the spam registration detection model trained this time will replace the original version. Otherwise, the original version will be maintained and used to avoid weakening the ability of the store risk control system to identify spam registration stores.

[0062] It can be seen that the implementation of step S1300 enables this application to provide a dynamic and self-improving risk management mechanism. Through continuous data updates and model optimization, the e-commerce platform can more accurately identify and manage spam registration behaviors, effectively maintain a healthy operating environment for the platform, and protect the interests of consumers and honest merchants.

[0063] According to the above embodiments, it can be seen that the present application has the following significant technical advantages and beneficial effects:

[0064] First, by responding to store detection events, the store's registration feature data is detected during the process, and by monitoring the store's post-registration feature data for post-detection, this application can more comprehensively identify spam registration behavior by combining the in-process detection and post-detection. In-process detection uses the registration feature data to quickly screen suspicious stores, while post-detection further confirms whether it is a spam registration by performing subsequent detection on the post-registration feature data of the stores that were missed during the in-process detection, thereby significantly improving the accuracy and coverage of identification.

[0065] Secondly, this application realizes dynamic assessment of store risk status by continuously monitoring the post-registration feature data of the store and combining it with the feature data at the time of registration. This dynamic assessment mechanism enables the screening method to adapt to the changes in spam registration behavior and promptly discover new suspicious patterns, thereby providing continuous risk management capabilities for e-commerce platforms.

[0066] In addition, this application uses an automated data annotation and storage mechanism to continuously optimize the spam registration detection model. Through regular iterative updates, the model learns the latest registration behavior patterns, automatically adapts and improves its ability to predict future spam registration behaviors. This approach not only improves the efficiency of investigation, but also ensures the long-term effectiveness and adaptability of the model, which helps to maintain the operating order of the e-commerce platform in the long term.

[0067] On the basis of any embodiment of the method of the present application, based on the registration feature data of the store to be detected, identifying whether the store to be detected is a junk registration store includes any one or more of the following steps:

[0068] Step S1110: using the contact information data of the store to be detected as the characteristic data at the time of registration, detecting whether the characteristic data at the time of registration is stored in a preset blacklist, and if it exists in the blacklist, identifying the store to be detected as a junk registration store;

[0069] Step S1110 can be used alone or selectively in combination with step S1120 and step S1130 to form a specific embodiment of step S1100, which focuses on using the contact data of the store to be detected as part of the characteristic data during registration to identify potential junk registered stores. On an e-commerce platform, when registering a new store, the merchant must provide valid contact information, which usually includes a mobile phone number, an email address, or both. These contact information is not only used to verify the identity of the merchant, but also an important channel for the store to communicate with consumers.

[0070] In step S1110, the store risk control system compares the collected contact data with a pre-set blacklist. The blacklist is a list of contact information known to be associated with spam registration behavior, which can come from past spam registration events, user complaints, or other risk monitoring activities. When the contact information provided by the store to be tested matches an entry in the blacklist, the system marks the store as high risk and identifies it as a spam registration store.

[0071] The implementation of this step involves real-time monitoring and rapid response to registration information. The system needs to be able to quickly check the contact information submitted by each newly registered store and compare it with the blacklist. This process can be achieved through automated tools to reduce manual intervention and improve detection efficiency. At the same time, in order to ensure accuracy and reduce false positives, the system can also combine other characteristic data at the time of registration for comprehensive evaluation.

[0072] For example, if a newly registered store uses the same email address as an entry in the blacklist, and the email address has been recorded as being associated with multiple spam registration events, the system can identify the store as a spam registration store based on this information. Such a mechanism helps e-commerce platforms to promptly identify and deal with stores that may pose a threat to the health of the platform and the interests of consumers.

[0073] Through the implementation of step S1110, the e-commerce platform can more effectively screen and identify spam registration behaviors, thereby protecting the trading environment of the platform and safeguarding the legitimate rights and interests of consumers and other merchants.

[0074] Step S1120: using the contact information data of the store to be detected as the characteristic data at the time of registration, and performing authenticity verification based on the characteristic data at the time of registration. If the authenticity verification fails, the store to be detected is identified as a junk registration store.

[0075] Similarly, step S1120 can be used alone or selectively in combination with step S1110 and step S1130 to form a specific embodiment of step S1100, which focuses on identifying junk registered stores by verifying the authenticity of the registration information. In this step, the store risk control system will use the contact data collected during registration as the feature data during registration to perform authenticity verification to ensure that the contact information provided by the registrant is authentic and valid.

[0076] Specifically, the system will require registrants to provide a contact method that can receive a verification code, such as a mobile phone number or email address, when submitting registration information. Subsequently, the system will automatically send a verification code to the contact method provided by the registrant. The registrant needs to enter the received verification code within the specified time to complete the verification process. This mechanism is designed to ensure that the entity behind the registration behavior is real and can respond in real time, thereby reducing the possibility of registration by automated scripts or robots.

[0077] In addition to the optional embodiments corresponding to the verification code verification above, in other embodiments, additional methods such as sliding verification code verification and robot phone voice verification can be provided to enhance the credibility of verifying the characteristic data during registration. If the registrant fails to provide the correct verification code within the specified time, or the system detects an abnormal pattern of verification code request, such as too high a request frequency or too many incorrect verification code inputs, the system will regard these behaviors as suspicious signs. In this case, the store risk control system can identify the store to be detected as a junk registration store and take corresponding risk control measures.

[0078] It can be seen that the implementation of step S1120 improves the defense capability of the e-commerce platform against spam registration behavior. Through real-time interactive verification, the system can more accurately distinguish between normal registration and spam registration behavior, thereby protecting the platform from malicious registration behavior.

[0079] Step S1130: Use the offline feature data and real-time feature data of the store to be detected as feature data during registration to match the real-time policy rules. When the real-time policy rules are hit, the store to be detected is identified as a junk registration store, wherein the offline feature data includes historical registration event statistics determined based on the contact data and / or identity feature data of the store to be detected, and the real-time feature data includes the registration event statistics of the day corresponding to the contact data and / or identity feature data of the store to be detected on the day of registration.

[0080] Similarly, step S1130 can be used alone or selectively in combination with step S1110 and step S1120 to form a specific embodiment of step S1100. Step S1130 uses real-time policy rules to conduct an in-depth analysis of store registration behavior to identify possible spam registration stores. The implementation of this step is based on a detailed analysis and modeling of historical registration data, from which a series of policy rules are extracted, which can capture the key features of spam registration behavior.

[0081] In step S1130, the store risk control system uses these real-time policy rules to perform real-time matching of the characteristic data of the store to be detected during registration. These characteristic data include but are not limited to the contact information, device information, IP address, etc. submitted by the registrant, as well as the behavior patterns during the registration process, such as the speed of filling in information, the frequency of page jumps, etc. The system compares these characteristic data with the real-time policy rules to identify any signs that may indicate spam registration behavior.

[0082] In one embodiment, the registration feature data used to apply the real-time policy rules include offline feature data and real-time feature data. Offline feature data, offline feature data includes historical registration event statistics determined based on the contact data and / or identity feature data of the store to be detected. As a collection of historical statistical information, offline feature data provides the system with a means to evaluate registration behavior from a macro perspective. For example, the system can analyze the number of registration requests initiated by a certain IP address over a period of time in the past, or check the frequency of registration events associated with a certain contact method. These statistics help the system identify patterns that are significantly different from normal registration behavior.

[0083] At the same time, the real-time feature data includes the contact data and / or identity feature data of the store to be detected, which corresponds to the registration event statistics on the day of registration. The real-time feature data provides insights into the registration behavior at the micro level. For example, consider a scenario where a store to be detected is registered on an e-commerce platform. During the registration process, the real-time feature data can capture the specific behavior patterns of the contact data used by the store, such as email address or mobile phone number, on the day of registration. Specifically, if the store initiates multiple registration attempts from the same IP address or device within a very short period of time after submitting the registration information, this abnormal registration frequency may indicate the use of automated scripts or batch registration tools. In addition, if the information filled in during the registration process (such as verification code input) is abnormally fast and inconsistent with normal human behavior, this may also trigger real-time policy rules to identify the store as a potential spam registration store. The collection and analysis of this type of real-time feature data provides e-commerce platforms with the ability to instantly identify and respond to spam registration behaviors, enhancing the platform's risk prevention and control mechanism.

[0084] Combining offline feature data and real-time feature data, the real-time policy rules in step S1130 can comprehensively evaluate the registration behavior. For example, if the IP address used by a newly registered store is associated with the registration behaviors of multiple other stores in a short period of time, or the registration information is filled in abnormally quickly, these signs may trigger the real-time policy rules, causing the store to be identified as a potential spam registration store.

[0085] Real-time policy rules can be set manually or constructed through statistical methods or machine learning algorithms. Distinctive registration feature data can be extracted from the registration-related data corresponding to the historical stores, and preliminary policy rules can be established based on these feature data. For example, if a certain IP address is found to have initiated a large number of registration requests in a very short period of time, this can be used as a registration feature data to form a policy rule, namely "too many registration requests from the same IP address in a short period of time." By continuously testing and verifying the effectiveness of these preliminary rules, the parameters of the rules can be further adjusted and optimized to improve their accuracy and coverage. In actual applications, these real-time policy rules can be converted into executable conditional statements and embedded in the store risk control system to monitor and evaluate the behavior of newly registered stores in real time.

[0086] Through this comprehensive analysis, step S1130 not only improves the ability to identify spam registration behaviors, but also provides a more dynamic and flexible risk management method for e-commerce platforms. The implementation of this method ensures that e-commerce platforms can respond to and handle potential spam registration behaviors in a timely manner, thereby maintaining a healthy operating environment for the platform and the interests of consumers.

[0087] It should be emphasized that steps S1110 to S1130 describe multiple specific embodiments of step S1100. A flexible combination of any one or any multiple steps among them can constitute a specific embodiment of step S1100. As long as one of the steps identifies the store to be detected as a junk registered store, it can be confirmed that the store to be detected is a junk registered store. Otherwise, the store to be detected still needs to be further confirmed through step S1200 to determine whether it is a junk registered store.

[0088] It is not difficult to find that the embodiments of step S1110 to step S1130 demonstrate a multi-dimensional and multi-level mechanism for identifying junk registration stores. These steps are applied independently or complement each other to jointly build a powerful risk prevention and control system. By comparing contact data with blacklists, performing authenticity verification, and applying real-time policy rule matching, these steps can evaluate the store's registration behavior from different angles, significantly improving the accuracy and efficiency of identifying junk registrations.

[0089] In particular, step S1110 uses the blacklist mechanism to quickly screen out high-risk stores, providing an efficient means for preliminary identification. Step S1120 further ensures the reliability of registration information through authenticity verification, effectively blocking automated or false registration behaviors. Step S1130 uses real-time strategy rules to conduct in-depth behavioral pattern recognition by comprehensively analyzing offline and real-time feature data, which not only enhances the ability to identify complex spam registration behaviors, but also provides good adaptability and flexibility for the system.

[0090] Overall, the technical effect of these steps is that they can comprehensively cover multiple risk indicators from single features to complex behaviors, forming a comprehensive identification system. This system can not only quickly respond to emerging junk registration methods, but also improve its identification capabilities through continuous learning and optimization, thereby providing an e-commerce platform with a stable and reliable risk management solution. The combined use of these embodiments greatly enhances the e-commerce platform's defense capabilities against junk registration behaviors, protects the platform's trading environment, and safeguards the legitimate rights and interests of consumers and other merchants.

[0091] On the basis of any embodiment of the method of the present application, the offline feature data and the real-time feature data of the store to be detected are used as the feature data at the time of registration to match the real-time policy rules. When the real-time policy rules are hit, before identifying the store to be detected as a junk registration store, the steps include:

[0092] Step S2100: Obtain offline feature data and real-time feature data of all historical stores within a specific period to form corresponding registration feature data as training samples, and obtain sample labels corresponding to whether the training samples are junk registered stores determined according to the corresponding registration feature data, and store each training sample and its sample label in a training data set;

[0093] The purpose of step S2100 is to provide rich and representative sample data for the training of the spam registration detection model. In this step, the system collects offline feature data and real-time feature data of all historical stores on the e-commerce platform within a specific period of time. These data together constitute the store's registration feature data, which are the basis of the training samples.

[0094] As mentioned above, offline feature data mainly comes from the information provided by the store during the registration process, such as contact information, identity information, and statistics of historical registration events. This data helps analyze whether the store's registration behavior has abnormal patterns or similarities with known spam registration behaviors. For example, if a contact method is associated with multiple stores that have been previously identified as spam registrations, then a new store using that contact method may also have the risk of spam registrations.

[0095] Similarly, as mentioned above, real-time feature data mainly focuses on the store's behavior on the day of registration, including the registration event statistics of the day of contact information and identity feature data. These data can reflect the immediate behavior patterns when the store registers, such as registration speed, consistency of information filling, etc. By analyzing these real-time data, the system can capture signs that may indicate automated scripts or batch registration behavior.

[0096] After collecting these feature data, the system can further obtain sample labels corresponding to these training samples, that is, determine whether each sample belongs to a spam registration store. This label is usually based on confirmed spam registration events in historical data or the results obtained through manual review. Each training sample and its corresponding sample label will be stored in the training data set, providing necessary information for subsequent spam registration detection model training and real-time policy rule generation.

[0097] By implementing step S2100, a comprehensive and highly representative training data set can be constructed, which not only contains rich historical data but also reflects the latest registration behavior trends. Such a data set provides a solid foundation for the training of the spam registration detection model and the generation of real-time policy rules, ensuring that the model can accurately learn and identify spam registration behaviors.

[0098] Step S2200: input each training sample and its sample label in the training data set into a classification model fitted according to a decision tree algorithm for training, and obtain each feature rule set based on each leaf node of the classification model;

[0099] This step uses the decision tree algorithm to train the classification model on the training data set. The decision tree algorithm is a supervised learning algorithm that classifies new data instances by learning simple decision rules from data features.

[0100] The construction of a decision tree begins with the analysis of the training data set. The algorithm will identify the attributes that best represent the data characteristics from the specific feature data of the training samples in the training data set based on the correspondence between the sample labels in the training data set and the training samples, and then build the root node of the decision tree based on these attributes. Subsequently, the algorithm will recursively repeat this process on each subset of the data set, creating branches of the decision tree for each subset until a certain termination condition is reached, such as all samples belonging to the same category, or the number of samples is lower than a preset threshold.

[0101] At each branch node of the decision tree, the algorithm will make judgments based on the attribute values ​​of the samples, divide the samples into different subsets, and generate corresponding feature rules for each subset. These feature rule sets will eventually form the leaf nodes of the decision tree, and each leaf node represents a final decision result, that is, the judgment of whether the sample belongs to a junk registered store or a normal store.

[0102] Through the above training process, the system can build a model that can classify and predict newly registered stores based on the feature rule set formed by the training data set, namely the classification model. This classification model can capture the key features of the registration behavior and conduct risk assessment on the store's registration behavior based on these features.

[0103] Decision tree algorithms are a series of algorithms used in machine learning for classification and regression tasks. They classify or predict from data features by learning simple decision rules. These algorithms recursively select the best attributes for node splitting and build easy-to-understand tree models. Optional decision tree algorithms include the classic ID3 algorithm, which makes decisions based on information gain; the C4.5 algorithm, which improves ID3 and uses information gain rate to reduce the risk of overfitting; the CART algorithm, which can handle classification and regression problems and uses G ini impurity as the splitting criterion; and ensemble learning methods such as random forest and gradient boosted decision tree (GBDT). In addition, there are algorithms optimized for large data sets, such as XGBoost and LightGBM, as well as CatBoost, which is particularly suitable for handling classification problems, and Hoeffding Tree for online learning scenarios. The M5P algorithm focuses on regression problems and uses an instance-based prediction method. These algorithms have their own advantages and are suitable for different application scenarios and data types. Given the characteristics of massive registration behaviors on e-commerce platforms, it is recommended to use LightGBM or XGBoost algorithms. Choosing this type of decision tree algorithm can significantly improve the prediction ability and accuracy of the model.

[0104] The advantage of the decision tree algorithm lies in the interpretability of its results. It can clearly show the logical path from registration features to risk determination, which is of great significance for subsequent feature rule optimization and model iteration. By continuously optimizing the construction process of the decision tree, the system can improve the accuracy and efficiency of identifying spam registration stores and provide strong technical support for the risk management of e-commerce platforms.

[0105] Step S2300: Calculate the recall accuracy corresponding to each feature rule set based on the training samples and sample labels in the training data set, and store the feature rule set with a recall accuracy exceeding a preset threshold as a real-time strategy rule for identifying junk registered stores.

[0106] This step can perform a detailed statistical analysis on each feature rule set generated during the training classification model based on the training samples and their sample labels in the training data set to determine its effectiveness in identifying spam registered stores.

[0107] Specifically, the system can calculate the recall accuracy of each feature rule set, which is a measure of the classification model's ability to identify spam registration behavior. Recall accuracy reflects the proportion of stores that the model correctly identifies among all stores that are actually spam registrations. For example, if a feature rule set identifies 90 out of 100 known spam registrations, then its recall accuracy is 90%.

[0108] In one embodiment, a recall accuracy threshold can be pre-set, and only those feature rule sets whose recall accuracy exceeds this threshold will be considered valid and stored as real-time policy rules. These rules will be used in the subsequent junk registration store identification process. For example, if the threshold is set to 90%, only those rule sets that can identify at least 90% of the known junk registration stores will be adopted.

[0109] In another embodiment, the classification model can also be tuned. Specifically, the system can combine or adjust parameters of different feature rule sets to achieve a higher recall accuracy, or improve the accuracy of the model while maintaining a high recall accuracy. For example, if it is found that some feature rule sets have a high recall accuracy but a relatively high false alarm rate, the parameters of these rules can be adjusted or combined with other rule sets to reduce the false alarm rate.

[0110] Through the above processing, the most effective and accurate feature rule set can be screened out to form a powerful real-time strategy rule base. These rules will be used in the real-time monitoring system of the e-commerce platform to quickly and accurately assess the risks of newly registered stores, thereby effectively identifying and preventing spam registration behaviors and protecting the healthy operating environment of the e-commerce platform.

[0111] Through the implementation of steps S2100 to S2300, further technical advantages can be obtained for this application. First, by comprehensively using offline and real-time feature data, the present invention can construct a comprehensive and representative training data set, which provides a solid foundation for training accurate spam registration detection models and generating real-time policy rules. Secondly, the decision tree algorithm and its derivative algorithms, such as LightGBM or XGBoost, adopted in this application not only improve the efficiency of model training and the accuracy of prediction, but also provide a clear direction for further optimization and iteration of the model through its interpretability. In addition, this application ensures the effectiveness of the selected feature rule set by setting a recall accuracy threshold, and further balances the recall rate and false alarm rate through model tuning, making the real-time policy rule base more accurate and efficient. Finally, through the automated feature rule set screening and real-time policy rule generation, a dynamic and adaptive risk management solution is provided for the e-commerce platform, which significantly improves the recognition ability and response speed of spam registration behavior, and effectively maintains the healthy operating environment and consumer interests of the e-commerce platform.

[0112] Based on any embodiment of the method of the present application, if the store to be detected is not a junk registration store, monitoring and obtaining its post-registration feature data, inputting the registration feature data and post-registration feature data of the store to be detected into the junk registration detection model, and determining whether the store to be detected is a junk registration store, including:

[0113] Step S3100: constructing registration relationship feature data of each historical store based on the contact data of all historical stores registered within a specific period, and constructing a store registration relationship network based on the registration relationship feature data of all historical stores;

[0114] In order to deeply analyze store registration behavior and build a relationship network, the system first determines a specific period of time. The selection of this period depends on the representativeness of the data and the accuracy of the analysis. The specific period can be the past week, month, quarter, or any time range set according to the business characteristics and risk management requirements of the e-commerce platform. For example, selecting the last month as a specific period can ensure that the collected registration data has a high timeliness and can also reflect the recent registration behavior trends.

[0115] After determining a specific period of time, the system will collect contact data for all historical stores registered during that period. This data includes, but is not limited to, email addresses, phone numbers, and other contact information that may be related to identity verification provided when registering a store. These contact data are the basis for building store registration relationship feature data because they can establish potential connections between different stores.

[0116] Using these contact data, the system will construct registration relationship feature data for each historical store. The registration relationship feature data includes not only the contact information of a single store, but also the contact information of other stores with the same or similar contact information as the store. For example, if two stores use the same email address to register, this will be recorded as a registration relationship feature between them.

[0117] Subsequently, the system will construct a store registration relationship network based on the registration relationship feature data of all historical stores. In this network, each store is a node, and the connections between stores (such as shared contact information) constitute the edges of the network. This network can reveal the potential connections between stores and help identify possible group registration behaviors or organized spam registration activities.

[0118] The construction of the store registration relationship network enables the system to examine store registration behavior from a macro perspective and identify groups of stores with similar registration characteristics. These groups may appear as highly interconnected community structures, thus providing an important data foundation for further community discovery and identification of spam registration stores.

[0119] Step S3200: Apply a preset community discovery algorithm to perform community grouping on the store registration relationship network, and calibrate whether the store in the training data set is a junk registration store based on the community grouping result.

[0120] After building the store registration relationship network, the community discovery algorithm can be applied to identify the community structure in the network to further analyze and identify spam registration stores. The community discovery algorithm is a graph theory algorithm used to identify tightly connected groups of nodes in the network that are less connected to other parts of the network.

[0121] In this step, the system will use a preset community discovery algorithm to analyze the store registration relationship network. Specific implementations may include but are not limited to any of the following algorithms:

[0122] Louvain algorithm: This is a greedy algorithm based on modularity optimization, which is used to discover community structures in networks. The Louvain algorithm divides communities by maximizing modularity, which is an indicator of the tightness of connections within a community.

[0123] Girvan-Newman algorithm: This is a method based on network modularity, which gradually reveals the community structure of the network by gradually removing the edges (or nodes) with the highest connectivity in the network.

[0124] Label Propagation Algorithm: This is a simple and effective method to identify communities by propagating information between nodes. Each node starts with a unique label, and then the node propagates the label to neighboring nodes, and eventually similar labels will gather together to form communities.

[0125] The results of community clustering will obtain specific data, such as the number of stores contained in each community, the connection density within the community, and the connection between communities. The relevant data constitutes the community clustering results. Based on the community clustering results, the system can identify communities that may be associated with spam registration behavior. For example, if a community has very close internal connections and fewer connections with other parts of the network, this may indicate that the stores in the community belong to the same operating entity and may be involved in spam registration behavior. In addition, the system can also analyze the post-registration feature data of stores in the community, such as product listing behavior, payment method settings, etc., to further verify the suspicion of spam registration.

[0126] Taking the Louvain algorithm as a specific embodiment of the community discovery algorithm, the Louvain algorithm is applied to the community discovery process of the store registration relationship network, aiming to optimize the annotation of the training data set. The Louvain algorithm divides the nodes in the network into communities by calculating the modularity, which reflects the tightness of the internal connection of the community and the sparseness of the external connection of the community.

[0127] First, using the contact data of all historical stores registered within a specific period, a store registration relationship network was constructed, in which each store was used as a node in the network, and the similarity or commonality of contact information between stores was used as the edge between nodes. Then, the Louvain algorithm was applied to this network to identify closely connected groups of nodes, namely communities.

[0128] In the initialization phase of the algorithm, each node is assigned a separate community label. Subsequently, the algorithm enters an iterative process to adjust the node's community affiliation by evaluating whether redistributing the node to another community can improve modularity. This iterative process continues until the community affiliation of all nodes in the network reaches a stable state, that is, the modularity cannot be further improved by redistributing the node.

[0129] The community clustering results obtained through the Louvain algorithm reveal the closely connected groups in the network. On this basis, for the stores identified as high-risk communities, their labels in the training dataset will be corrected and may be marked as candidates for spam registration stores. In particular, the labels of stores in high-modularity communities will be updated to reflect their risk status because they are closely connected internally and have few external connections.

[0130] Ultimately, these corrected annotation information will be fed back into the training process of the spam registration detection model, allowing the model to more accurately learn and predict whether a store's registration behavior is spam. Through this annotation correction method based on community discovery, e-commerce platforms can more effectively manage and control the quality of store registrations and improve the efficiency and effectiveness of risk management.

[0131] In other embodiments, a graph clustering algorithm may be used to cluster the store registration relationship network to achieve the purpose of community discovery, which may also have the same effect as the community discovery algorithm.

[0132] The above embodiment builds a store registration relationship network and applies a community discovery algorithm for analysis, so that the system can identify communities that may be related to spam registration behavior and correct the store labels in the training data set, thereby improving the accuracy and efficiency of spam registration detection. The outstanding effect of this technical solution is that it can not only fully capture potential spam registration behaviors, but also dynamically update and optimize the detection model, so that the e-commerce platform can respond and adapt to the ever-changing risk environment in a timely manner, and effectively maintain the healthy operation of the platform and the interests of consumers.

[0133] Based on any embodiment of the method of the present application, a preset community discovery algorithm is applied to perform community grouping on the store registration relationship network, and after correcting whether the store in the training data set is a junk registration store according to the community grouping result, the method includes:

[0134] Step S3300: construct post-registration feature data corresponding to all historical stores within a specific period and store them in the training data set, wherein the post-registration feature data includes relationship network features extracted from the store registration relationship network, operation configuration behavior features on the day of store registration, and operation time-consuming features corresponding to store configuration;

[0135] The construction and update of post-registration feature data of all historical stores within a specific period can be stored in the training data set for further analysis. In this step, the system first determines a specific period with representativeness and accuracy requirements, such as the most recent month or quarter, in order to collect store registration data that meets the requirements of timeliness and trend.

[0136] In order to construct the post-registration feature data required for training the spam registration detection model, the relationship network features, such as the degree centrality and PageRank value of the nodes, are extracted from the store registration relationship network. These features reveal the importance and influence of the store in the network. At the same time, the system also collects the operational configuration behavior characteristics of the store on the day of registration, including the frequency of product listing, payment method settings, and store decoration operations. These characteristics reflect the store's operating model and activity. In addition, the system also measures and records the time required for the store to perform specific configuration operations, such as the time required to set up payment methods and decorate the store. These operation time characteristics help identify possible abnormal behaviors.

[0137] The corresponding data collected from these aspects constitute the post-registration feature data, which not only enriches the content of the training data set, but also provides a multi-dimensional perspective for the model to evaluate the store's registration behavior. The updated training data set contains comprehensive feature data from the time of registration to after registration, providing a solid foundation for subsequent model training and performance verification. The implementation of this step ensures that the spam registration detection model can learn and predict based on the latest and most comprehensive data, thereby improving the accuracy and adaptability of the model.

[0138] Step S3400: input the registration feature data and post-registration feature data of the store in the training data set into the junk registration detection model, perform supervised training according to whether the store is a junk registration store, and train the junk registration detection model to a convergence state;

[0139] First, we need to integrate all relevant feature data collected by the store during and after registration. As shown in the previous example, feature data during registration may include basic information, contact information, identity verification information, etc. of the store, while feature data after registration covers the store's operational behavior, configuration behavior, and operation time. The collection of these feature data provides a comprehensive perspective for the model to evaluate whether the store's registration behavior is normal.

[0140] Subsequently, these comprehensive feature data in the training data set are fed into the spam registration detection model one by one for training, and the model will begin to learn the patterns that distinguish spam registration stores from normal stores. Supervised training means that the model will adjust its parameters based on known labels (i.e. whether the store is a spam registration) to better classify. During the training process, the model may use various machine learning algorithms, such as support vector machines, decision trees, random forests, or deep neural networks, to identify patterns and associations in the data.

[0141] Once the model training is completed and reaches convergence, it will be able to predict and classify new or unknown store registration behaviors. Convergence means that the model has stabilized and further training will not significantly improve its performance. At this point, the model can be used in actual spam registration detection tasks, providing real-time protection for e-commerce platforms to identify and prevent potential spam registration behaviors.

[0142] Step S3500: Use the training data set to verify whether the recall accuracy of the trained spam registration detection model exceeds a preset threshold. If it exceeds, stop using the spam registration detection model in use and publish the trained spam registration detection model as the used version.

[0143] In order to ensure that the model can achieve the expected performance standards in practical applications, in this step, the recall accuracy of the model is first evaluated using the training data set. Recall accuracy is an important indicator that measures the model's ability to correctly identify spam-registered stores, that is, the proportion of stores that the model can accurately identify from all stores that are actually registered as spam.

[0144] During the validation process, the system can use a series of evaluation methods to test the performance of the model. This includes comparing the model's prediction results with the known labels in the training data set to calculate the model's recall accuracy. If the model's recall accuracy exceeds the preset threshold, it indicates that the model has shown sufficient accuracy and reliability in identifying spam registration stores. At this point, the system will stop using the current model version and release the trained and verified new model version as the official version in use to replace the old model.

[0145] This verification mechanism not only ensures that the model has met certain performance requirements before deployment, but also provides feedback for the continuous optimization and iteration of the model. If the recall accuracy of the model fails to exceed the preset threshold, the system will return to step S3400 to further train and adjust the model until it meets the performance standard.

[0146] In addition, step S3500 can also evaluate other performance indicators of the model, such as accuracy, F1 score, etc., and test the generalization ability of the model to ensure that the model can maintain stable performance when processing new and unseen data. Through this rigorous verification process, it can ensure that the e-commerce platform can deploy an efficient and accurate spam registration detection model, effectively improve the risk management capabilities of the platform, and protect the interests of consumers and honest merchants.

[0147] After the whole process from step S3100 to step S3500 is executed, the junk registration detection model can be used to comprehensively and accurately detect the data during and after the registration. The system not only captures key feature data during the store registration, but also continues to collect and analyze the store's behavior data in the post-registration stage after the registration is completed, thus realizing the monitoring of the entire life cycle of the store before and after registration.

[0148] In the mid-stage, the system responds to store registration events, quickly collects registration feature data, and performs preliminary screening in real time. This rapid response mechanism effectively intercepts obvious abnormalities or known spam registration behaviors. In the post-event stage, the system further deepens its understanding of store behavior patterns by monitoring post-registration feature data, such as operational configuration behavior and operation time, so that the model can identify potential spam registration stores that were not directly identified at the time of registration.

[0149] Through the implementation of steps S3300 and S3400, the training data set is continuously updated and optimized, ensuring that the spam registration detection model can learn the latest behavior patterns and improve the model's adaptability to emerging spam registration methods. The model verification process in step S3500 ensures that the model will only be updated to the official version when the recall accuracy exceeds the preset threshold. This strict verification mechanism ensures the efficiency and accuracy of the model in practical applications.

[0150] The spam registration detection model itself can also conduct comprehensive detection based on in-process and post-process data, so that the technical solution of this application can not only quickly respond to known risks, but also deeply analyze and identify new or more hidden spam registration behaviors. This comprehensive advantage provides e-commerce platforms with a more reliable and forward-looking risk management tool, significantly improving the platform's overall defense capabilities against spam registration behaviors, ensuring a clean trading environment and enhancing user trust.

[0151] Based on any embodiment of the method of the present application, post-registration feature data corresponding to all historical stores within a specific period of time is constructed and stored in the training data set, including:

[0152] Step S3310: based on the store registration relationship network, extract the relationship network characteristics of each store node in the network within a specific period of time, including node degree, local clustering coefficient, betweenness centrality, closeness centrality, eigenvector centrality and any number of PageRank indicators;

[0153] Step S3320: collect the operation configuration behavior characteristics of different periods within the specific period of store registration, including the occurrence frequency and configuration mode corresponding to any multiple items of product listing behavior, payment method setting, store decoration operation, and store subscription package;

[0154] Step S3330: extract the operation time-consuming features corresponding to the store configuration within a specific period from the store buried point data, and determine the time required to complete the specific operation configuration behavior as the operation time-consuming features, including any multiple of the time for setting the payment method, the time for decorating the store, the time for viewing a single page, the time for configuring the store payment, and the time for configuring the store logistics data;

[0155] Step S3340: construct the relationship network characteristics, operation configuration behavior characteristics and operation time-consuming characteristics of each store into the post-registration feature data corresponding to the store and store them in the training data set.

[0156] Steps S3310 to S3340 constitute a specific embodiment for constructing post-registration feature data, which is used to collect and construct post-registration feature data of historical stores within a specific period of time and store them in a training data set. The collection of these feature data provides a rich source of information for the subsequent spam registration detection model, thereby improving the prediction ability and accuracy of the model.

[0157] Step S3310 involves extracting relationship network features from the store registration relationship network. This step is based on the principles of graph theory and identifies potential connections between stores by analyzing the network relationships of each store node in the network. For example, node degree can reflect the number of connections between a store and other stores; local clustering coefficient and betweenness centrality can reveal the importance of the store's position in the network; and closeness centrality and eigenvector centrality can measure the influence of the store in the network. The PageRank algorithm can evaluate the importance of nodes and is usually used to identify key stores or influential nodes in the network.

[0158] Step S3320 focuses on collecting the store's operational configuration behavior characteristics during a specific period of time. This includes recording the frequency and pattern of key operational activities such as product listing behavior, payment method settings, store decoration operations, and store subscription packages. These behavioral characteristics can reveal the store's operating model and activity, and provide a basis for identifying abnormal or suspicious operational behaviors.

[0159] Step S3330 focuses on extracting operation time characteristics from the store's buried data. The buried data records the detailed process of the user's interaction with the website. By analyzing this data, the time required to complete a specific operation configuration behavior can be determined. For example, the time to set up the payment method, the time to decorate the store, the time to view a single page, etc. These time-consuming characteristics help identify abnormal fast registration or configuration behavior, which may be a sign of spam registration behavior.

[0160] Finally, step S3340 combines the relationship network features, operation configuration behavior features, and operation time-consuming features extracted in the above steps to construct post-registration feature data for each store. These feature data are organized and stored in the training data set, providing a comprehensive perspective for the spam registration detection model to evaluate the store's registration behavior.

[0161] Through this series of steps, this application can comprehensively collect and analyze the post-registration behavior of stores, thereby providing an effective tool for e-commerce platforms to identify and manage spam registration behavior. This method not only improves the accuracy of detection, but also provides solid technical support for the risk management and credit system of e-commerce platforms.

[0162] Based on any embodiment of the method of the present application, monitoring and obtaining post-registration feature data includes:

[0163] Step S1210: automatically deploy a monitoring mechanism for the store to be detected to collect post-registration behavior data of the store to be detected within a specific monitoring period;

[0164] Automatically deploy monitoring mechanisms to collect store behavior data within a specific period after registration. The implementation of this process involves the following aspects:

[0165] First, the system will set a monitoring cycle, which is determined according to the operating characteristics and risk control requirements of the e-commerce platform. For example, the monitoring cycle can be set to 1 hour, 3 hours, 6 hours, etc. after the store is registered. This period of time is usually enough to observe the initial operation of the store.

[0166] Next, the system will automatically start monitoring tools for newly registered stores. These tools can be monitoring software integrated into the backend of the e-commerce platform, or independent applications deployed on cloud servers. The monitoring tools are responsible for tracking and recording various behaviors of the store, such as product listing, payment method configuration, store page editing, etc.

[0167] To ensure the accuracy and completeness of monitoring data, the system can use efficient data collection technology. This may include using API interfaces to capture store operation logs, or using front-end embedding technology to collect user interaction data. The collected data will be transmitted to the data processing system in real time for storage and preliminary analysis.

[0168] In addition, the system will perform quality control on the collected data. This includes data cleaning, removing invalid or redundant information, and data verification to ensure that each record accurately reflects the actual behavior of the store. The data cleaning and verification process can be implemented through preset rules or automated scripts.

[0169] Through the above implementation methods, the automatically deployed monitoring mechanism can ensure that the e-commerce platform conducts comprehensive and detailed monitoring of the behavior of newly registered stores, providing a solid foundation for subsequent feature data analysis and identification of spam registration behavior.

[0170] Step S1220: Analyze the collected post-registration behavior data to identify abnormal features that are inconsistent with the normal store operation mode;

[0171] After the automatically deployed monitoring mechanism collects the store's post-registration behavior data, the next key step is to conduct in-depth analysis of this data to identify abnormal characteristics that are inconsistent with the normal store operation mode. The analysis process involves the following aspects:

[0172] First, the system can use data analysis techniques, such as statistical analysis, pattern recognition, and machine learning algorithms, to process and analyze the collected behavioral data. These techniques can help the system extract meaningful information from large amounts of data and discover potential abnormal patterns.

[0173] Secondly, the system predefines a set of standards or baselines to measure the normality of store behavior. These standards can be based on the behavioral characteristics of normal stores in historical data, including the frequency of product listings, the diversity of payment method settings, the complexity of store decoration, etc. By comparing the behavioral data of the store to be tested with these standards, the system can identify behaviors that deviate greatly.

[0174] For example, if a newly registered store puts a large number of products on the shelves in a very short period of time, or its payment method settings are single and unconventional, these behaviors will be considered abnormal characteristics. Similarly, if the store decoration operation is too simple or no updates are made for a long time after registration, it may also indicate that the store is risky.

[0175] In addition, the system can also take into account time factors and evaluate whether the speed and order of store actions are normal. For example, under normal circumstances, the store will not put the goods on the shelves until the decoration is completed. If this order is reversed or completed too quickly, it may indicate the use of automation tools.

[0176] The system can also analyze user interaction data, such as page browsing time, click-through rate, etc., to identify abnormal user behavior patterns. Normal stores usually have users browsing and interacting for a certain period of time, while spam registered stores may lack such user engagement.

[0177] Step S1230: determine the risk score of the store to be detected based on the abnormal features obtained through analysis; when the risk score exceeds a preset safety range, obtain post-registration feature data of the store to be detected to further determine whether the store to be detected is a junk registered store.

[0178] In the risk management process of e-commerce platforms, the risk score of a store determined based on the abnormal features obtained through analysis can be used to generate post-registration feature data, thereby facilitating subsequent comprehensive evaluations to determine whether the corresponding store has abnormal behavior, thereby identifying spam registered stores.

[0179] It should be noted that post-registration feature data and post-registration behavior data can be distinguished. Post-registration feature data refers to a series of characteristics exhibited by a store after completing registration. These characteristics can reflect the store's operating model and activity. In contrast, post-registration behavior data focuses more on describing the specific behavior of a store within a specific time. The selection of the specific content of post-registration behavior data and post-registration feature data can be set as needed by those skilled in the art in accordance with the spirit of this application.

[0180] The process of determining the risk score first requires collecting and organizing the store's post-registration behavior data. This data will be input into a risk assessment model, which uses preset algorithms and parameters to quantitatively analyze the store's operating model. For example, if a store puts a large number of products on the shelves in a short period of time, and the quality of these product information is low, this will be captured as an abnormal feature by the risk assessment model and may cause the store's risk score to increase.

[0181] The risk assessment model assigns a risk score to each store based on the degree of abnormality of the feature data. This score is a quantitative indicator that reflects the degree to which the store deviates from the normal operating model. The higher the risk score, the more obvious the abnormal behavior of the store is, and the greater the possibility of spam registration.

[0182] Once the risk score is determined, the system will judge the score based on the preset safety range. If the store's risk score exceeds this safety range, the system will determine that the store has a high risk of spam registration, and thus start collecting post-registration feature data for the store, thereby improving the accuracy of the collected post-registration feature data.

[0183] The innovative measures of the automatic monitoring mechanism enhance the technical advantages of this application. Through this mechanism, the e-commerce platform can capture and analyze the behavior patterns of newly registered stores in real time, thereby identifying potential spam registration behaviors at the earliest stage. The implementation of this method not only improves the accurate correspondence between spam registered stores in the training data set and their post-registration feature data, which can save the cost of annotation and review, but also provides an objective and standardized assessment tool for risk management by quantifying risk scores. In addition, the automated nature of the mechanism reduces the need for manual intervention, reduces operating costs, and increases processing speed.

[0184] See also Figure 3 According to one aspect of the present application, a device for checking junk-registered stores is provided, comprising an in-process detection module 1100, a post-detection module 1200, and an iterative update module 1300, wherein the in-process detection module 1100 is configured to respond to store registration events, and identify whether the store to be detected is a junk-registered store based on the feature data of the store to be detected at the time of registration; the post-detection module 1200 is configured to monitor and obtain its post-registration feature data if the store to be detected is not a junk-registered store, input the feature data of the store to be detected at the time of registration and the feature data after registration into a junk registration detection model, and determine whether the store to be detected is a junk-registered store; the iterative update module 1300 is configured to mark whether the feature data of the store to be detected at the time of registration and the feature data after registration are junk-registered stores and store them in a training data set, and regularly iterate and update the junk registration detection model with the training data set.

[0185] On the basis of any embodiment of the device of the present application, based on the characteristic data of the store to be detected at the time of registration, it is identified whether the store to be detected is a junk registration store, including any one or more of the following modules: a filtering and identification module, which is configured to use the contact data of the store to be detected as the characteristic data at the time of registration, detect whether the characteristic data at the time of registration is stored in a preset blacklist, and when it exists in the blacklist, identify the store to be detected as a junk registration store; a verification and identification module, which is configured to use the contact data of the store to be detected as the characteristic data at the time of registration, and perform authenticity verification based on the characteristic data at the time of registration, and when the authenticity verification fails, identify the store to be detected as a junk registration store; a rule identification module, which is configured to use the offline characteristic data and the real-time characteristic data of the store to be detected as the characteristic data at the time of registration to perform real-time policy rule matching, and when the real-time policy rule is hit, identify the store to be detected as a junk registration store, wherein the offline characteristic data includes historical registration event statistics determined according to the contact data and / or identity characteristic data of the store to be detected, and the real-time characteristic data includes the registration event statistics of the day corresponding to the contact data and / or identity characteristic data of the store to be detected on the day of registration.

[0186] On the basis of any embodiment of the device of the present application, prior to the rule identification module, it includes: a data preparation module, which is configured to obtain offline feature data and real-time feature data of all historical stores within a specific period of time to form corresponding registration feature data as training samples, and obtain sample labels corresponding to whether the training samples determined according to the corresponding registration feature data belong to junk registered stores, and store each training sample and its sample label in a training data set; a training execution module, which is configured to input each training sample and its sample label in the training data set into a classification model fitted according to a decision tree algorithm for training, and obtain each feature rule set based on each leaf node of the classification model; a rule optimization module, which is configured to calculate the recall accuracy corresponding to each feature rule set according to the training samples and sample labels in the training data set, and store the feature rule set with a recall accuracy exceeding a preset threshold as a real-time strategy rule for identifying junk registered stores.

[0187] On the basis of any embodiment of the device of the present application, prior to the post-detection module 1200, it includes: a relationship construction module, which is configured to construct the registration relationship feature data of each historical store based on the contact data of all historical stores registered within a specific period of time, and construct a store registration relationship network according to the registration relationship feature data of all historical stores; a clustering processing module, which is configured to apply a preset community discovery algorithm to perform community clustering on the store registration relationship network, and correct whether the store in the training data set is a junk registration store according to the community clustering result.

[0188] On the basis of any embodiment of the device of the present application, after the clustering processing module, it includes: a feature construction module, which is configured to construct post-registration feature data corresponding to all historical stores within a specific period of time and store them in the training data set, and the post-registration feature data includes relationship network features extracted from the store registration relationship network, operation configuration behavior features on the day of store registration, and operation time-consuming features corresponding to the store configuration; an import training module, which is configured to input the registration feature data and post-registration feature data of the store in the training data set into the junk registration detection model, implement supervised training according to whether the store is a junk registration store, and train the junk registration detection model to a convergence state; a version update module, which is configured to use the training data set to verify whether the recall accuracy of the trained junk registration detection model exceeds a preset threshold. When it exceeds, the junk registration detection model is stopped from being used in the version, and the trained junk registration detection model is published as the version in use.

[0189] Based on any embodiment of the device of the present application, the feature construction module includes: a relationship feature extraction module, which is configured to extract the relationship network features of each store node in the network within a specific period of time based on the store registration relationship network, including any number of node degree, local clustering coefficient, betweenness centrality, closeness centrality, feature vector centrality and PageRank indicators; a configuration feature collection module, which is configured to collect the operation configuration behavior features of different periods within a specific period of store registration, including the occurrence frequency and configuration mode corresponding to any number of product listing behavior, payment method setting, store decoration operation, and store subscription package; a time-consuming feature extraction module, which is configured to extract the operation time-consuming features corresponding to the store configuration within a specific period of time from the store buried point data, and determine the time required to complete a specific operation configuration behavior as the operation time-consuming feature, including the time for setting the payment method, the time for decorating the store, the single-page viewing time, the store payment configuration time, and the store logistics data configuration time; a feature construction storage module, which is configured to construct the relationship network features, operation configuration behavior features and operation time-consuming features of each store into the corresponding post-registration feature data of the store and store them in the training data set.

[0190] Based on any embodiment of the device of the present application, the post-detection module 1200 includes: a monitoring startup module, which is configured to automatically deploy a monitoring mechanism for the store to be detected to collect post-registration behavior data of the store to be detected within a specific monitoring period; a behavior analysis module, which is configured to perform analysis based on the collected post-registration behavior data to identify abnormal features that are inconsistent with the normal store operation mode; an abnormality handling module, which is configured to determine the risk score of the store to be detected based on the abnormal features obtained by analysis, and when the risk score exceeds a preset safety range, obtain the post-registration feature data of the store to be detected to further determine whether the store to be detected is a junk registration store.

[0191] Another embodiment of the present application also provides a device for checking junk registered stores. Figure 4 As shown, a schematic diagram of the internal structure of a junk-registered store screening device is shown. The junk-registered store screening device includes a processor, a computer-readable storage medium, a memory, and a network interface connected through a system bus. Among them, the computer-readable non-volatile storage medium of the junk-registered store screening device stores an operating system, a database, and computer-readable instructions. The database may store an information sequence. When the computer-readable instructions are executed by the processor, the processor may implement a junk-registered store screening method.

[0192] The processor of the garbage registration store screening device is used to provide computing and control capabilities to support the operation of the entire garbage registration store screening device. The memory of the garbage registration store screening device may store computer-readable instructions, and when the computer-readable instructions are executed by the processor, the processor may execute the garbage registration store screening method of the present application. The network interface of the garbage registration store screening device is used to connect and communicate with a terminal.

[0193] Those skilled in the art will understand that Figure 4 The structure shown in the figure is merely a block diagram of a portion of the structure related to the present application scheme, and does not constitute a limitation on the garbage registration store inspection device to which the present application scheme is applied. The specific garbage registration store inspection device may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.

[0194] In this embodiment, the processor is used to execute Figure 3 The memory stores the program code and various data required to execute the above modules or submodules. The network interface is used to realize data transmission between user terminals or servers. The non-volatile readable storage medium in this embodiment stores the program code and data required to execute all modules in the junk registered store screening device of this application, and the server can call the program code and data of the server to execute the functions of all modules.

[0195] The present application also provides a non-volatile readable 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 method for detecting junk registered stores in any embodiment of the present application.

[0196] The present application also provides a computer program product, including a computer program / instruction, which implements the steps of the method described in any embodiment of the present application when executed by one or more processors.

[0197] Those 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 non-volatile 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).

[0198] In summary, this application achieves comprehensive identification and effective prevention of spam registration behaviors on e-commerce platforms by combining real-time identification during the process and in-depth analysis after the process. It not only improves the accuracy and response speed of identifying spam registrations, but also enhances the ability to adapt to emerging risk behaviors through automated data updates and model iteration mechanisms, thus providing e-commerce platforms with a continuous and dynamic risk management solution, ensuring a healthy trading environment and enhanced user trust.

Claims

1. A method for checking junk registered stores, characterized in that: include: Respond to store registration events and identify whether the store to be detected is a spam registration store based on the registration feature data of the store to be detected; If the store to be detected is not a junk registration store, monitor and obtain its post-registration feature data, input the registration feature data and post-registration feature data of the store to be detected into the junk registration detection model, and determine whether the store to be detected is a junk registration store; The registration feature data and post-registration feature data of the store to be detected are marked as whether they belong to junk registration stores and stored in a training data set, and the junk registration detection model is regularly iterated and updated with the training data set.

2. The method for checking junk registered stores according to claim 1, characterized in that: Based on the registration feature data of the store to be detected, identifying whether the store to be detected is a junk registration store includes any one or more of the following steps: The contact data of the store to be detected is used as the characteristic data at the time of registration, and the characteristic data at the time of registration is detected to see whether it is stored in a preset blacklist. If it exists in the blacklist, the store to be detected is identified as a junk registration store; The contact information data of the store to be tested is used as the characteristic data at the time of registration, and authenticity verification is performed based on the characteristic data at the time of registration. When the authenticity verification fails, the store to be tested is identified as a junk registration store; The offline feature data and real-time feature data of the store to be detected are used as feature data at the time of registration to match the real-time policy rules. When the real-time policy rules are hit, the store to be detected is identified as a junk registration store, wherein the offline feature data includes historical registration event statistics determined based on the contact data and / or identity feature data of the store to be detected, and the real-time feature data includes the registration event statistics of the day corresponding to the contact data and / or identity feature data of the store to be detected on the day of registration.

3. The method for checking junk registered stores according to claim 2, characterized in that: The offline feature data and real-time feature data of the store to be detected are used as the feature data at the time of registration to match the real-time policy rules. When the real-time policy rules are hit, before the store to be detected is identified as a junk registration store, the following steps are included: Obtain offline feature data and real-time feature data of all historical stores within a specific period to form corresponding registration feature data as training samples, and obtain sample labels corresponding to whether the training samples are junk registration stores determined according to the corresponding registration feature data, and store each training sample and its sample label in a training data set; Input each training sample and its sample label in the training data set into a classification model fitted according to a decision tree algorithm for training, and obtain each feature rule set based on each leaf node of the classification model; The recall accuracy corresponding to each feature rule set is calculated based on the training samples and sample labels in the training data set, and the feature rule set with a recall accuracy exceeding a preset threshold is stored as a real-time strategy rule for identifying junk registered stores.

4. The method for checking junk registered stores according to any one of claims 1 to 3, characterized in that: If the store to be detected is not a junk registration store, the post-registration feature data of the store to be detected is obtained by monitoring, and the registration feature data and post-registration feature data of the store to be detected are input into the junk registration detection model to determine whether the store to be detected is a junk registration store, including: Based on the contact information data of all historical stores registered within a specific period of time, the registration relationship feature data of each historical store is constructed, and the store registration relationship network is constructed according to the registration relationship feature data of all historical stores; A preset community discovery algorithm is applied to perform community grouping on the store registration relationship network, and the stores in the training data set are corrected to determine whether they are junk registration stores based on the community grouping results.

5. The method for checking junk registered stores according to claim 4, characterized in that: Applying a preset community discovery algorithm to perform community grouping on the store registration relationship network, and correcting whether the store in the training data set is a junk registration store according to the community grouping result, including: Constructing post-registration feature data corresponding to all historical stores within a specific period and storing them in the training data set, the post-registration feature data includes relationship network features extracted from the store registration relationship network, operation configuration behavior features on the day of store registration, and operation time-consuming features corresponding to store configuration; Inputting the registration feature data and post-registration feature data of the store in the training data set into the spam registration detection model, performing supervised training according to whether the store is a spam registration store, and training the spam registration detection model to a convergence state; The training data set is used to verify whether the recall accuracy of the trained spam registration detection model exceeds a preset threshold. If it exceeds, the in-use version of the spam registration detection model is stopped, and the trained spam registration detection model is released as the in-use version.

6. The method for checking junk registered stores according to claim 5, characterized in that: Constructing the post-registration feature data corresponding to all historical stores within a specific period and storing them in the training data set, including: Based on the store registration relationship network, extract the relationship network characteristics of each store node in the network within a specific period of time, including node degree, local clustering coefficient, betweenness centrality, closeness centrality, eigenvector centrality and any number of PageRank indicators; Collect operational configuration behavior characteristics at different times during a specific period of store registration, including the corresponding occurrence frequency and configuration mode of any number of items in product listing behavior, payment method settings, store decoration operations, and store subscription packages; Extract the operation time characteristics corresponding to the store configuration in a specific period of time from the store tracking data, and determine the time required to complete a specific operation configuration behavior as the operation time characteristics, including any number of the time for setting the payment method, the time for decorating the store, the time for viewing a single page, the time for store payment configuration, and the time for store logistics data configuration; The relationship network features, operation configuration behavior features, and operation time-consuming features of each store are constructed as post-registration feature data corresponding to the store and stored in the training data set.

7. The method for checking junk registered stores according to any one of claims 1 to 3, characterized in that: Monitor and obtain post-registration feature data, including: Automatically deploy monitoring mechanisms for stores to be tested to collect post-registration behavior data of stores to be tested within a specific monitoring period; Analyze the collected post-registration behavior data to identify abnormal features that are inconsistent with normal store operation patterns; The risk score of the store to be tested is determined based on the abnormal features obtained through analysis. When the risk score exceeds the preset safety range, the post-registration feature data of the store to be tested is obtained to further determine whether the store to be tested is a spam registration store.

8. A device for checking junk registered stores, characterized in that: include: The in-process detection module is configured to respond to store registration events and identify whether the store to be detected is a spam registration store based on the characteristic data of the store to be detected during registration; The post-detection module is configured to monitor and obtain post-registration feature data of the store to be detected if the store to be detected is not a spam registration store, input the registration feature data and post-registration feature data of the store to be detected into a spam registration detection model, and determine whether the store to be detected is a spam registration store; The iterative update module is configured to mark the registration feature data and post-registration feature data of the store to be detected as whether it is a junk registration store and store it in a training data set, and regularly iterate and update the junk registration detection model with the training data set.

9. A device for checking junk registered stores, comprising a central processing unit 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