Methods, devices, equipment, media, and products for identifying physical stores

Abstract

This application relates to a method, apparatus, device, medium, and product for identifying stores with the same physical entity. The method includes: acquiring store profiles of online stores in all store groups, where each store group includes a store profile corresponding to at least one online store, and multiple online stores in the same store group have registered the same communication characteristics; determining the semantic similarity between the store profiles of a first store group and any other store group, and identifying any other store group containing the online store with the highest semantic similarity to each online store in the first store group as a similar store group; determining the voting rate of each similar store group based on the number of similar store groups, and determining that the similar store group is a store with the same physical entity when the voting rate of a similar store group exceeds a preset threshold. This application can accurately identify online stores with the same physical entity on e-commerce platforms, effectively maintaining the operational order of e-commerce platforms.

Description

Technical Field

[0001] This application relates to e-commerce information processing technology, and more particularly to a method, apparatus, device, medium, and product for identifying physical stores. Background Technology

[0002] E-commerce, or simply e-commerce, can be categorized into platform-based and independent website-based models. Platform-based e-commerce centrally integrates a vast array of branded products, allowing users to focus on the products themselves when making purchases. Unlike platform-based e-commerce, independent website-based e-commerce attracts users primarily through the brand itself, relying on brand traffic. In independent website-based e-commerce, merchants need to choose an e-commerce platform to build their own online store. Furthermore, merchants can use the same platform to create multiple similar online stores. However, some merchants may use the creation of multiple similar online stores to sell counterfeit or fake goods for illegal purposes. Therefore, identifying similar merchants on an e-commerce platform—that is, how to uncover all merchants belonging to the same business entity—is crucial to the e-commerce platform's risk management capabilities.

[0003] Traditional methods for identifying similar businesses often involve linking them by business account name, phone number, etc. This method is easily circumvented; once a business changes its registered account information, the method becomes ineffective. Another method is to calculate the similarity of email addresses to find similar businesses. While this adds a dimension to the identification process, judging similarity based solely on email addresses is prone to false positives.

[0004] Another shortcoming of traditional technologies for identifying businesses with the same physical location is that they are limited to identifying static information about the business's store. This static information refers to information that does not change easily within a certain time frame, such as mobile phone numbers and email addresses. While this excessive focus on static information takes advantage of its unchanging nature, it also has its drawbacks. It reduces the channels for capturing the characteristics of businesses with the same physical location, making it impossible to effectively identify such businesses.

[0005] In conclusion, the identification technology for physical stores needs further improvement in order to enhance e-commerce platforms' ability to identify store risks. Summary of the Invention

[0006] The purpose of this application is to solve the above-mentioned problems by providing a method for identifying physical stores and corresponding devices, equipment, non-volatile readable storage media, and computer program products.

[0007] According to one aspect of this application, a method for identifying physical stores is provided, comprising the following steps:

[0008] Obtain the store profiles of online stores in all store groups. Each store group includes a store profile corresponding to at least one online store. Multiple online stores in the same store group are registered with the same communication characteristics. The store profile of each online store includes multiple specific sampling features.

[0009] Determine the semantic similarity between the store profiles of the first store group and any other store group in all store groups, and identify any other store group containing the online store with the highest semantic similarity in the first store group as a similar store group.

[0010] The voting rate of each similar store group is determined based on the number of similar store groups. When the voting rate of a similar store group exceeds a preset threshold, the similar store group is determined to be the same entity store group of the first store group.

[0011] Optionally, retrieve the store profiles of all online stores in the store group, including:

[0012] Based on the communication characteristics registered by all online stores, online stores that have registered the same communication characteristics are grouped into the same store group. The communication characteristics include any one of telephone number, email address, and mailing address.

[0013] For each online store, multiple specific sampling features are obtained to form a store profile. These multiple specific sampling features include dynamic product information features and static merchant information features.

[0014] Optionally, determine the semantic similarity between the store profiles of each pair of online stores in the first store group and any other store group, and identify the other arbitrary store groups containing the online store with the highest semantic similarity in the first store group as similar store groups, including:

[0015] Based on the Cartesian product, the pairs of online stores between the first store group and any other store group are determined, forming the store pairs to be compared.

[0016] Convert the store profile of each online store in the store pair to a profile feature representation.

[0017] The profile feature representations of the stores to be compared are mapped to the classification space to determine the semantic similarity between two online stores.

[0018] Iterate through all the online stores in the first store group. For each online store, determine the store group containing the online store with the highest semantic similarity to it as the similar store group.

[0019] Optionally, the store profile of each online store in the comparison store pair is converted into a profile feature representation, including:

[0020] The product dynamic information features in the store profiles of the two online stores in the store pair to be compared are compiled into one-hot encoded vectors and constructed as a joint representation of dynamic features;

[0021] The static information features of merchants in the store profiles of the two online stores in the store pair to be compared are compiled into text feature vectors and constructed as a joint representation of static features;

[0022] The joint representation of dynamic features and the joint representation of static features are used to construct a portrait feature representation.

[0023] Optionally, the product dynamic information features include category coverage features and / or operation coverage features, and the merchant static information features include email address and / or network address.

[0024] Optionally, the voting rate of each similar store group is determined based on the number of similar store groups. When the voting rate of a similar store group exceeds a preset threshold, the similar store group is determined to be the same entity store group of the first store group, including:

[0025] Construct a store group and message, which includes the store profiles of the first store group and its physical store groups;

[0026] Send the combined store information to the review interface;

[0027] In response to the confirmation instruction returned by the audit interface, the same entity store group and its first store group are merged into the same store group.

[0028] Optionally, the voting rate of each similar store group is determined based on the number of similar store groups. When the voting rate of a similar store group exceeds a preset threshold, the similar store group is determined to be the same entity store group of the first store group, including:

[0029] Exclude the first store group and its corresponding physical store group from all store groups;

[0030] Continue to use any store group that has not been excluded from all store groups as the first store group, and iterate the above process to determine the same entity store group for it.

[0031] According to another aspect of this application, a physical store identification device is provided, comprising:

[0032] The profile acquisition module is set to acquire the profiles of online stores in all store groups. Each store group includes at least one online store with a corresponding profile. Multiple online stores in the same store group are registered with the same communication characteristics. Each online store's profile includes multiple specific sampling characteristics.

[0033] The similarity recognition module is set to determine the semantic similarity between the store profiles of the first store group and any other store group in all store groups, and to determine the other store group containing the online store with the highest semantic similarity to each online store in the first store group as the similar store group.

[0034] The similarity and difference identification module is set to determine the vote rate of each similar store group based on the number of similar store groups. When the vote rate of a similar store group exceeds a preset threshold, the similar store group is identified as the same entity store group of the first store group.

[0035] According to another aspect of this application, a physical store identification device is provided, including a central processing unit and a memory, wherein the central processing unit is used to invoke and run a computer program stored in the memory to perform the steps of the physical store identification method described in this application.

[0036] According to another aspect of this application, a non-volatile readable storage medium is provided, which stores a computer program implemented according to the aforementioned physical store identification method in the form of computer-readable instructions, wherein the computer program, when invoked by a computer, executes the steps included in the method.

[0037] According to another aspect of this application, a computer program product is provided, including a computer program / instructions that, when executed by a processor, implement the steps of the method described in any embodiment of this application.

[0038] Compared with existing technologies, this application has numerous technical advantages, including but not limited to:

[0039] First, this application divides different store groups based on the communication characteristics of merchant registration. Then, it determines the semantic similarity between the store profiles of each pair of online stores in each store group. Next, taking each online store in the first store group as the standard, it determines the store group containing the online store with the highest semantic similarity to each online store as the similar store group. Finally, it determines some similar store groups as the same entity store groups of the first store group based on the vote rate of the similar store groups, thus realizing the batch identification of same entity store groups across the entire platform, which is very efficient.

[0040] Secondly, in the process of identifying similar store groups of the first store, this application has already divided the online stores into groups in advance through communication features. On this basis, it mainly uses multiple specific sampling features in the store profile of the online stores to perform semantic matching to determine the similarity relationship between the online stores. This method can intelligently determine the similarity between any two online stores based on the semantic understanding of specific sampling features, thereby improving the accurate identification ability of whether any two online stores are similar. On this basis, it is more accurate to determine whether two store groups belong to the same entity.

[0041] Furthermore, this application is economical, practical, and highly efficient in improving the ability to identify similar stores on e-commerce platforms and enhancing their risk control capabilities, thereby effectively maintaining the operational order of e-commerce platforms. Attached Figure Description

[0042] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0043] Figure 1 This provides an example of deploying the corresponding network architecture for the store risk control system of this application on an e-commerce platform;

[0044] Figure 2 This is a flowchart illustrating one embodiment of the physical store identification method of this application;

[0045] Figure 3 This is a schematic diagram of the process for determining the similar store group of the first store group in an embodiment of this application;

[0046] Figure 4 This is a schematic diagram illustrating the process of obtaining the profile feature representation of the store pair to be compared in an embodiment of this application;

[0047] Figure 5 This is a flowchart illustrating the review process for physical store groups in an embodiment of this application;

[0048] Figure 6 This is a schematic diagram illustrating the process of iteratively identifying all physical stores on an e-commerce platform in an embodiment of this application;

[0049] Figure 7 This is a schematic diagram of the physical store identification device of this application;

[0050] Figure 8 This is a schematic diagram of the structure of a physical store identification device used in this application. Detailed Implementation

[0051] Please see Figure 1 This application provides an exemplary network architecture applicable to e-commerce platform scenarios, including a terminal device 80, a security server 81, and a store server 82.

[0052] The security server 81 can serve as the main execution entity of the store risk control system of this application. The store risk control system is a computer program product programmed according to the same entity store identification method of this application. When it runs, it executes each step of the method, thereby identifying the same entity stores (groups) of a given first store (group) in the e-commerce platform.

[0053] The store server 82 can be used to deploy one or more online stores on an internet platform. Various network requests to access the online stores can be responded to by the corresponding application services provided by the store server. Each online store has various basic feature data suitable for creating its store profile, such as the store's category coverage feature data, operation coverage feature data, email address, phone number, etc. Any number of these feature data can be used to construct the store profile.

[0054] The terminal device 80 can be used to trigger the network request in order to use various application services in the store server 82, such as browsing the online store and performing the order process for product items in the online store.

[0055] The security server 81 can read the relevant basic feature data of the online stores in the store server 82 to update the store profile of the online stores. Then, the store risk control system can determine whether two online stores or two groups of online stores belong to the same entity store (group) based on the various feature data in the store profile. This enables the identification of the similarities and differences between the back-end entities of online stores, thereby improving the monitoring level of the e-commerce platform in identifying risky stores.

[0056] Please see Figure 2 According to a method for identifying physical stores provided in this application, in one embodiment, the method includes the following steps:

[0057] Step S1100: Obtain the store profiles of online stores in all store groups. Each store group includes at least one store profile corresponding to an online store. Multiple online stores in the same store group are registered with the same communication characteristics. The store profile of each online store includes multiple specific sampling features.

[0058] In e-commerce platforms, different merchants register their own online stores, resulting in a massive number of online stores. When some merchants register multiple online stores using the same communication characteristics, they are considered to naturally belong to the same business entity. These online stores can be grouped into a single store group to facilitate entity-based identification across all online stores on the platform. It's easy to understand that each resulting store group includes at least one online store.

[0059] The aforementioned communication feature, in one embodiment, could be a merchant's phone number; in another embodiment, it could be a merchant's email address. It's easy to understand that merchants using the same phone number or email address are usually owned by the same merchant, and therefore should be grouped together. When a merchant operates multiple online stores on the same platform, even if the products sold in each online store are not exactly the same, some operational characteristics exhibited in maintaining each online store may be highly similar. Alternatively, even if the operational characteristics differ, there may often be overlap in product items or product categories across different online stores. In short, among different online stores registered with the same communication feature, there are certain semantic relationships hidden between the characteristics exhibited in the online stores.

[0060] This association can actually be extended to different online stores that haven't registered with the same communication characteristics, and this application utilizes this feature for extraction. Therefore, the online store profiles are further designed so that each store profile contains multiple specific sampling features. These specific sampling features are determined based on their suitability for capturing the semantic association described above. For example:

[0061] In one embodiment, the specific sampling features include the category coverage features of online stores. The category coverage features are used to describe the distribution of the product items listed by the corresponding online stores in each category of a preset product classification system. Thus, the similarity of the product items operated by different online stores in terms of category distribution can be mined through the category coverage features.

[0062] In another embodiment, the specific sampling features include the operation coverage features of online stores. The operation coverage features are used to describe the distribution of the corresponding online stores' operation behaviors such as listing and delisting products, filling in product sales information, and filling in logistics orders. Thus, the similarity of different operation behaviors of merchants between different online stores can be mined through the operation coverage features.

[0063] In another embodiment, the specific sampling features include the email address and / or network address registered by the online store. Considering that the same merchant may use the same or similar email address to keep in touch with customers, or use the same device to log in to the back-end of different online stores, using this information as specific sampling features can, to some extent, help identify the semantic similarity between different online stores.

[0064] In practice, those skilled in the art can also add other specific sampling features suitable for mining semantic associations between different online stores as needed, so as to sample various basic feature data of the corresponding online stores based on these specific sampling features, thereby constructing a store profile of the corresponding online store. Each online store profile can implicitly contain the semantics corresponding to the business activities of the underlying merchants, so as to match the similarity of business activities between different online stores by mining these semantics, and determine whether there is the same business entity behind different online stores (groups) through the matching results.

[0065] Step S1200: Determine the semantic similarity between the store profiles of the first store group and any other store group in the entire store group, and determine the other store group containing the online store with the highest semantic similarity in the first store group as a similar store group.

[0066] When determining whether the operating entities of different store groups on an e-commerce platform are the same, a store group can be used as a unit, which is called the first store group. The first store group is matched one by one with any other store group (which can be called the second store group) to determine the similarity of the online stores in the first store group and the other second store groups in terms of business activities. Then, it is determined which second store groups constitute the same entity store group of the first store group.

[0067] To this end, we first determine the semantic similarity between the store profiles of each online store in the first store group and each online store in a selected second store group. This way, for each pair of stores in the first and second store groups, we obtain the semantic similarity between each pair of stores. By extension, we can obtain the corresponding semantic similarity between any online store in the first store group and any pair of stores in any other second store group.

[0068] Considering that the purpose of this application is to achieve similarity and difference matching between store groups, for each online store in the first store group, after determining the semantic similarity of all online stores in its second store group, only the store pair corresponding to the highest semantic similarity is used, and other store pairs can be ignored. In this way, for each online store in the first store group, there is a second store group to which the online store with the highest semantic similarity belongs. Therefore, in fact, each specified online store in the first store group not only has the highest semantic similarity mapped to a certain second store group, but also obtains the label of this second store group. This label can be used to identify this second store group, and the second store group pointed to by this label becomes a similar store group that is similar to the first store group to which the specified online store belongs.

[0069] In one embodiment, similar store groups matched for the first store group can be filtered according to a preset threshold. Specifically, the highest semantic similarity of each online store in the first store group is compared with the preset threshold. When the highest semantic similarity is lower than the preset threshold, the corresponding similar store group can be ignored to ensure that all similar store groups matched for the first store group have high confidence, thereby improving the accuracy of matching with physical store groups.

[0070] Step S1300: Determine the vote rate of each similar store group based on the number of similar store groups. When the vote rate of a similar store group exceeds a preset threshold, determine that similar store group as the same entity store group of the first store group.

[0071] Based on the above explanation, it is easy to understand that each online store in the first store group may correspond to a similar store group. Thus, the first store group may match multiple similar store groups. However, so far, this matching relationship only considers the semantic association between the store profiles of two online stores, and does not consider the semantic association between multiple online stores in the first store group and multiple online stores in the similar store group. In order to further improve the accuracy of the final matched same-entity store group, a voting mechanism can be adopted. The number of similar store groups matched by each online store in the first store group can be used to count the vote rate of each similar store group, and then the same-entity store group of the first store group can be determined based on the vote rate.

[0072] In one embodiment, the voting rate of each similar store group is determined by dividing the number of similar store groups matched in the first store group by the total number of similar store groups matched in the first store group. Accordingly, the voting rates of each similar store group can be normalized to the same dimension for comparison.

[0073] To further enhance the credibility of similar store groups, a preset threshold can be provided for the selection of voting rates. The voting rates of each similar store group are compared with the preset threshold. When the voting rate of a similar store group is lower than the preset threshold, the similar store group is ignored. When the voting rate of a similar store group exceeds the preset threshold, the similar store group can be identified as a similar store group of the first store group, and the two are regarded as groups controlled by the same operating entity.

[0074] Following the above process, a first store group may also identify multiple stores with the same physical entity. These stores with the same physical entity and the first store group are all regarded as being owned by the same operating entity. In fact, the technical solution of this application realizes the clustering of all online stores in the e-commerce platform and identifies the stores with the same physical entity that have a community relationship.

[0075] Of course, any other store group on the e-commerce platform that has not been identified as a group of physical stores can be regarded as the first store group and identified in accordance with the process of this application. Through continuous iteration, the clustering of all online stores on the entire platform can eventually be achieved.

[0076] Building upon the ability to cluster and group online stores across the entire platform, further control can be applied to groups of stores with the same physical entity. For example, each online store within these groups can be assigned a tag corresponding to the same entity.

[0077] As can be seen from the above embodiments, this application has numerous technical advantages, including but not limited to:

[0078] First, this application divides different store groups based on the communication characteristics of merchant registration. Then, it determines the semantic similarity between the store profiles of each pair of online stores in each store group. Next, taking each online store in the first store group as the standard, it determines the store group containing the online store with the highest semantic similarity to each online store as the similar store group. Finally, it determines some similar store groups as the same entity store groups of the first store group based on the vote rate of the similar store groups, thus realizing the batch identification of same entity store groups across the entire platform, which is very efficient.

[0079] Secondly, in the process of identifying similar store groups of the first store, this application has already divided the online stores into groups in advance through communication features. On this basis, it mainly uses multiple specific sampling features in the store profile of the online stores to perform semantic matching to determine the similarity relationship between the online stores. This method can intelligently determine the similarity between any two online stores based on the semantic understanding of specific sampling features, thereby improving the accurate identification ability of whether any two online stores are similar. On this basis, it is more accurate to determine whether two store groups belong to the same entity.

[0080] Furthermore, this application is economical, practical, and highly efficient in improving the ability to identify similar stores on e-commerce platforms and enhancing their risk control capabilities, thereby effectively maintaining the operational order of e-commerce platforms.

[0081] Based on any embodiment of this application, obtain the store profiles of online stores in all store groups, including:

[0082] Step S1110: Based on the communication characteristics registered by all online stores, divide the online stores that have registered the same communication characteristics into the same store group. The communication characteristics include any one of telephone number, email address, and communication address.

[0083] In the process of obtaining the store profiles of all store groups in the e-commerce platform, the communication features in the registration information of each online store in the e-commerce platform can be read first. The communication features refer to the feature information that has a relatively unique pointing to the operating entity of the online store. In this embodiment, the communication features can be the phone number in the registration information of the online store, its email address, or its registration address.

[0084] Based on whether the communication characteristics are consistent, all online stores on the e-commerce platform are grouped. Multiple online stores with the same communication characteristics are grouped into the same store group, thereby dividing all online stores into multiple store groups.

[0085] In one embodiment, multiple communication features can be used simultaneously to associate and group stores. Specifically, a first communication feature, such as a phone number, is used to complete the initial grouping. Then, a second communication feature, such as an email address, is used to find whether different store groups use the same second communication feature. When this is the case, different store groups with member online stores using the same second communication feature are merged into the same store group. This process is repeated to maximize the use of prior knowledge of communication features and to group all online stores on the e-commerce platform in the most thorough way. This ensures that each store group contains as many online stores as possible owned by the same business entity, and each store group contains various generalized features of the same business entity, making subsequent similarity matching between different store groups more accurate.

[0086] Step S1120: For each online store, obtain multiple specific sampling features to form a store profile. The multiple specific sampling features include product dynamic information features and merchant static information features.

[0087] After grouping all online stores on the e-commerce platform, feature collection is required according to the store profile of the online store. As mentioned earlier, the store profile consists of multiple specific sampling features, which may include, but are not limited to, category coverage features, operation coverage features, email address, network address, etc.

[0088] In this embodiment, specific sampling features for the store profile are selected from two dimensions: dynamic product information features and static merchant information features. Dynamic product information features, as the name suggests, mainly refer to features corresponding to product items that undergo frequent daily changes during the online store's daily operations, such as category coverage features and operation coverage features. Since merchants frequently operate on product items, these features can be summarized as dynamic product information features. Static information features mainly refer to features closely related to the merchant but relatively stable during the online store's daily operations. For example, email addresses are usually relatively stable information used by merchants to contact customers, and network addresses are the network IP addresses used by merchants when logging into the online store's backend. Since merchants typically use dedicated lines to manage their online stores, these are generally quite stable. The characteristic of dynamic product information features is their discreteness, and they are primarily in text form.

[0089] It's easy to understand that constructing a store profile using dynamic product information features and static merchant information features means that the former implicitly contains behavioral semantics corresponding to the daily operations of the online store, while the latter implicitly contains relational semantics of the merchant's relatively unchanging information. As a result, the store profile can comprehensively represent the implicit operational characteristics of the operating entity behind the online store. Based on this, similarity identification of online stores can point to the operating entity behind the online store, effectively achieving matching with physical stores. Compared to simply relying on individual registration information, such as the communication features mentioned above, to determine the same physical store group, the level of intelligence is significantly higher, and the accuracy is higher.

[0090] Based on any embodiment of this application, please refer to Figure 3 Determine the semantic similarity between the online store profiles of the first store group and any other store group in the entire store group. For each online store in the first store group, determine the other arbitrary store groups containing the online store with the highest semantic similarity, including:

[0091] Step S1210: Determine the pairs of online stores between the first store group and any other store group based on the Cartesian product, forming the store pairs to be compared.

[0092] After dividing all online stores on the e-commerce platform into multiple store groups, in order to process efficiently, for each online store in the first store group, it is paired with any online store in any other store group to construct a store pair to be compared.

[0093] Specifically, both the first and second store groups contain multiple online stores, effectively forming two sets of online stores. Therefore, by using a Cartesian product (direct product) on these two sets, we can obtain a pair of stores to be compared, formed by combining any online store from the first group with any online store from the second group. If the first group contains m online stores and the second group contains n online stores, then we can obtain m*n pairs of stores to be compared.

[0094] Step S1220: Convert the store profile of each online store in the store pair to a profile feature representation.

[0095] For each pair of stores to be compared, the online stores in the first store group can be called the first online stores, and the online stores in the second store group can be called the second online stores. The store profiles of the first online stores and the second online stores are obtained accordingly, and their store profiles are converted into corresponding profile feature representations. The profile feature representations are obtained by extracting deep semantic information based on the encoded information obtained after encoding each specific sampling feature of the store profile.

[0096] In one embodiment, the dynamic information features of the products of the first online store and the second online store can be encoded separately, and then the deep semantic information can be extracted separately and concatenated into the same dynamic feature representation. Similarly, the static information features of the merchants of the first online store and the second online store can be encoded separately, and then the deep semantic information can be extracted separately and concatenated into the same static feature representation. Then, the dynamic feature representation and the static feature representation can be concatenated into the same profile feature representation.

[0097] In another embodiment, the product dynamic information features and merchant static information features of the first online store and the second online store can be encoded separately according to different stores, and deep semantic information can be extracted and then concatenated into a same store feature representation. Then, the two same store feature representations can be concatenated into the same profile feature representation.

[0098] In summary, the portrait feature representation is a joint feature representation of the deep semantic information of the store portraits of two online stores in the same store pair to be compared, which can represent the deep semantic features of the underlying business entities of the two online stores.

[0099] Step S1230: Map the portrait feature representation of the stores to be compared to the classification space to determine the semantic similarity between two online stores.

[0100] Once the profile feature representation of a store to be compared is determined, a classifier is used. Through the fully connected layer in the classifier, the profile feature representation is mapped to a preset classification space. The classification probability corresponding to the positive classification in the classification space is determined. This classification probability can be used as the semantic similarity between the two online stores in the store comparison pair.

[0101] In one embodiment, the semantic similarity between two online stores in a pair of stores to be compared can be determined by using a pre-trained store similarity recognition model to predict their semantic similarity. The network architecture of the store similarity recognition model includes a feature representation network and a classifier. The classification space of the classifier can be set to include two or more categories, one of which is a positive category and the others are negative categories. The store similarity recognition model is pre-trained to a convergent state using corresponding training samples, enabling it to determine the corresponding profile feature representations based on the store profiles of the two online stores, and map these profile feature representations to the classification space to obtain the classification probability corresponding to the positive category as the semantic similarity.

[0102] During the training phase of the store similarity recognition model, when predicting the classification probability of each category in the classification space for a training sample labeled as a positive sample, the loss value is calculated using the positive sample label. Similarly, when predicting the classification probability of each category in the classification space for a training sample labeled as a negative sample, the loss value is calculated using the negative sample label. The gradient of the store similarity recognition model is updated based on the loss value. The model is iteratively trained using a large number of training samples until it reaches convergence, allowing it to be used in the online inference phase to predict the semantic similarity between the store profiles of two online stores in a pair of stores to be compared.

[0103] It is easy to understand that each pair of stores to be compared can have its corresponding semantic similarity determined by the store similarity recognition model. This semantic similarity represents the degree of semantic similarity between the business entities pointed to by the store profiles of the two online stores in the pair of stores to be compared.

[0104] Step S1240: Traverse all online stores in the first store group. For each online store, determine the store group containing the online store with the highest semantic similarity as the similar store group.

[0105] For each first online store in the first store group, since each first online store has a pair of second online stores belonging to other store groups, by traversing each first online store in the first store group, the semantic similarity of each pair of stores containing the first online store is determined. Thus, for the first store group, the semantic similarity of each online store in it to each online store in other store groups is determined.

[0106] Generally, the highest semantic similarity between a first online store in a first store group and any other store group represents the highest semantic similarity between the first store group and that other store group. Therefore, regardless of the number of other store groups with which the first online store has varying degrees of semantic similarity, the store group most likely to constitute a true same-entity store group is the one with the highest semantic similarity to the first online store. Based on this principle, in this embodiment, for each first online store in the first store group, the store pair corresponding to the highest semantic similarity among all its comparable store pairs is determined. Then, based on the second online store corresponding to the first online store in that comparable store pair, the store group to which this second online store belongs is determined as a similar store group for the first store group determined based on the first online store. Thus, based on each member online store in the first store group, a similar store group corresponding to the highest semantic similarity can be determined. In other words, multiple similar store groups are determined for the first store group, and each similar store group has a highest semantic similarity with the first store group determined based on a first online store in the first store group.

[0107] As can be seen from the above embodiments, this application compares the similarity of the operating entities of two store groups to the similarity of the store profiles of online stores in each pair of store groups. Furthermore, for the first store group, although the semantic similarity is determined based on the pairs of stores to be compared, only the store groups with the highest semantic similarity are selected as similar store groups for each online store in the first store group. In fact, it focuses on the group similarity information between different store groups. Moreover, the determination of this group similarity information is not satisfied with simple registration information, but is based on specific sampling features in the store profiles corresponding to the operating entities. Therefore, the determined similar store groups are more accurate, taking into account both common information and individual association information, and the grouping results are more accurate.

[0108] Based on any embodiment of this application, please refer to Figure 4 The store profiles of each online store in the comparison store pair are converted into profile feature representations, including:

[0109] Step S1221: Compile the product dynamic information features in the store profiles of the two online stores in the store pair to be compared into one-hot encoded vectors and construct a joint representation of dynamic features;

[0110] When encoding the dynamic product information features of two online stores in a comparison pair, considering the discreteness of these features, a one-hot vector encoding can be constructed. Specifically, for the category coverage feature in the dynamic product information features, based on the product categories covered by the product items published by the online stores, assuming all product categories are {c1, c2, ..., c...} k Merchant m1 sells products covering the three categories {c1, c2, c3}, which can be converted into [1, 1, 1, 0, 0, ..., 0]. This forms the category coverage feature. Similarly, for the operation coverage feature in the product dynamic information feature, assuming all operation types are {b1, b2, ..., b...},... k Merchant m1's actions cover the three categories {b1, b2, b3}, which can be converted into [1, 1, 1, 0, 0, ..., 0]. This forms the action coverage feature.

[0111] Let the category coverage characteristic of the first online store in the same pair of stores to be compared be Input. cate，first The second online store's category coverage is characterized by Input cate，second The category coverage characteristics of the two online stores can be encoded using the following process:

[0112] V cate，first =Embedding(Input) cate，first )

[0113] V cate，second =Embedding(Input) cate，second )

[0114] V cate =Concat(V cate，first V cate，second )

[0115] F cate =FullyConnect(V cate )

[0116] The above encoding process shows that the category coverage features of the first and second online stores should be encoded separately first. cate，first Input cate，secondAfter encoding, the two are concatenated and then fully connected. This process can be implemented in the feature representation network of the store similarity recognition model in this application.

[0117] Let the operational coverage feature of the first online store in the same pair of stores to be compared be Input. behav，first The second online store's operational coverage features are Input behav，second Similarly, the coding can be implemented according to the following process:

[0118] V behav，first =Embedding(Input) behav，first )

[0119] V behav，secand =Embedding(Input) behav，second )

[0120] V behav =Concat(V behav，first V behav，second )

[0121] F behav =FullyConnect(V behav )

[0122] It can be seen that the encoding process for the operation coverage features of two online stores is similar to the encoding process for the category coverage features, and can also be implemented in the feature representation network of the store similarity recognition model in this application.

[0123] After encoding the specific features of the product dynamic information features, such as the category coverage feature and operation coverage feature mentioned above, the dynamic feature joint representation of the product dynamic information features of the two online stores is further constructed according to the following process:

[0124] F merge =Concat(F cate F behav )

[0125] It is evident that the dynamic information features of the products jointly constituted by the two online stores in the comparison store pair will eventually be transformed into a joint representation of the same dynamic features, thereby realizing the feature representation of the dynamic information features of the products.

[0126] Step S1222: Compile the merchant static information features in the store profiles of the two online stores in the store pair to be compared into text feature vectors and construct a joint representation of static features;

[0127] For the merchant static information features of the two online stores in the store pair to be compared, since they are mainly text content, they can be encoded into text feature vectors, and the text feature vectors of the two online stores can be constructed into a joint representation of the same static features. The process is as follows:

[0128] First, the joint static feature representations of the two online stores are segmented and corresponding model labels are added to construct a text sequence. For example, for the email addresses hi1@163.com and hi2@163.com, the following text sequence can be obtained:

[0129] [[CLS],h,i,1,@,1,6,3,.,c,o,m,[SEP],h,i,2,@,1,6,3,.,c,o,m,[SEP]]

[0130] [cls] and [sep] are labels used to instruct text encoders like BERT to perform classification tasks and to indicate the corresponding segmentation, respectively.

[0131] Similarly, for the network addresses of two online stores, 127.0.0.1 and 127.0.0.2, their word segmentation sequence is represented as: [[CLS], 1, 2, 7, ., 0, ., 0, ., 1, [SEP], 1, 2, 7, ., 0, ., 0, ., 2, [SEP]].

[0132] Then, for the text sequence of email addresses, feature representation can be performed as follows:

[0133] V email =Embedding(Input) email )

[0134] T email =TransformerEncoder(V email )

[0135] F email =FullyConnect(T email )

[0136] As can be seen, the text sequence of email addresses is encoded by the dictionary of the reference text encoder (TransformerEncoder), then input into the text encoder to extract deep semantics, and finally passed through a fully connected mapping to obtain the corresponding feature representation. The text encoder used in this application can also be other models implemented using recurrent neural networks, such as BERT or LSTM.

[0137] Similarly, the feature representation process of a text sequence of network addresses is as follows:

[0138] V ip =Embedding(Input) ip )

[0139] T ip =TransformerEncoder(V ip )

[0140] F ip =FullyConnect(T ip )

[0141] It can be seen that both email addresses and web addresses are treated as text and processed for feature representation, and the feature representation process is the same.

[0142] Finally, the characteristics of an email address can be represented as F email and the characteristic representation of network address F ip They are then further concatenated into a joint representation of static features.

[0143] Step S1223: Construct the joint representation of dynamic features and the joint representation of static features into a portrait feature representation.

[0144] After obtaining the joint dynamic feature representation and the joint static feature representation of the store pair to be compared, a concatenation layer is further used to concatenate the two into a single representation, which constitutes the profile feature representation of the store pair to be compared. This profile feature representation realizes a comprehensive representation of various features of the store profiles of the two online stores in the store pair to be compared, which can be used to perform classification mapping and determine the degree of semantic association between the business entities behind the two online stores in the store pair to be compared.

[0145] The entire process of this embodiment can be implemented using the store similarity recognition model of this application. The encoding process can be controlled by the weights learned during the training process of the store similarity recognition model, so that the final portrait feature representation can effectively highlight the potential semantic association between the business entities behind the two online stores, so that the semantic similarity corresponding to this semantic association can be accurately predicted through classification mapping.

[0146] The above embodiments illustrate a method for jointly representing various features of the store profiles of two online stores in a comparison store pair through an exemplary encoding process. It can be seen that by adapting to different types of information features in the store profile and performing corresponding encoding, various information features can be effectively represented, enabling the corresponding store similarity model to perform efficient semantic mining, and ensuring that the semantic similarity between the two online stores can be predicted by the classifier in the future, which serves as a reliable basis for realizing store similarity recognition.

[0147] Based on any embodiment of this application, please refer to Figure 5The voting rate of each similar store group is determined based on the number of similar store groups. When the voting rate of a similar store group exceeds a preset threshold, that similar store group is determined as the same entity store group of the first store group, including:

[0148] Step S1310: Construct a store group and message, which includes the store profiles of the first store group and its corresponding physical store groups;

[0149] Once a first store group is identified and its corresponding physical store groups are determined, the review mechanism can be triggered. First, a store group merge message is constructed. To facilitate review users quickly understanding the characteristics of each store group, this store group merge message includes store profiles of the first store group and its corresponding physical store groups. Specifically, this includes data such as dynamic product information features and static merchant information features.

[0150] Step S1320: Send the combined store message to the review interface;

[0151] The e-commerce platform has a pre-set review interface. Sending the store group merger message to the review interface can send the store group merger message to the corresponding review user. After reading the store group merger message, the review user can use their human experience to confirm or reject the first store group and the same entity store group. When they confirm, they can send back the corresponding confirmation instruction through the review interface; otherwise, they can send back a rejection instruction or not respond.

[0152] Step S1330: In response to the confirmation instruction returned by the audit interface, merge the same entity store group and its first store group into the same store group.

[0153] Since the confirmation instruction contains corresponding user instructions, the same entity store group and its first store group will be merged into the same store group only after the confirmation instruction returned by the audit interface is received. This will merge all online stores in different groups under the control of the same business entity into the same online store group, which is actually a further clustering of different groups of stores and a fine mining of the same entity store community.

[0154] As can be seen from the above embodiments, by confirming whether the store group belongs to the same business entity through human-computer interaction with the backend user via the audit interface, the identification effect of the similarities and differences of business entities can be further ensured, the identification result of the same entity store group in the e-commerce platform can be more accurate, and the business order of the e-commerce platform can be effectively maintained.

[0155] Based on any embodiment of this application, please refer to Figure 6The voting rate of each similar store group is determined based on the number of similar store groups. When the voting rate of a similar store group exceeds a preset threshold, that similar store group is determined as the same entity store group of the first store group, including:

[0156] Step S1400: Exclude the first store group and its same entity store group from all store groups;

[0157] In an e-commerce platform, if a group of stores is designated as the first group of stores, and after the process described in the above embodiments is followed to identify the other stores in the same entity group, the first group of stores and its other stores in the same entity group can be excluded from all the store groups on the e-commerce platform. This ensures that the store groups identified as having the same business entity will no longer participate in the subsequent identification process.

[0158] Step S1500: Continue to use any store group that has not been excluded from all store groups as the first store group, and iterate the above process to determine the same entity store group for it.

[0159] After excluding the identified store groups with the same business entity, a new first store group is determined from the remaining store groups on the e-commerce platform. Then, the prior steps in any embodiment of this application are iteratively executed to further determine the new first store group and its same entity store group. This process is repeated until all store groups on the e-commerce platform have been traversed, thus achieving the grouping of all store groups on the e-commerce platform and dividing all store groups on the e-commerce platform into multiple same entity store communities according to different business entities.

[0160] As can be seen from the above embodiments, by efficiently repeating the operation, iteratively identifying the same physical store groups of the first store group with a store group as the first store group, the entire store group of the e-commerce platform can be processed by social groups, achieving low cost but accurate identification effect, and playing an effective role in maintaining the business order of the e-commerce platform.

[0161] Please see Figure 7According to one aspect of this application, a device for identifying physical stores includes a profile acquisition module 1100, a similarity identification module 1200, and a difference identification module 1300. The profile acquisition module 1100 is configured to acquire store profiles of online stores in all store groups. Each store group includes a store profile corresponding to at least one online store. Multiple online stores in the same store group have registered the same communication features, and each online store's store profile includes multiple specific sampling features. The similarity identification module 1200 is configured to determine the semantic similarity between the store profiles of a first store group and any other store group, and to identify any other store group containing the online store with the highest semantic similarity to each online store in the first store group as a similar store group. The difference identification module 1300 is configured to determine the voting rate of each similar store group based on the number of similar store groups. When the voting rate of a similar store group exceeds a preset threshold, the similar store group is identified as a physical store group of the first store group.

[0162] Based on any embodiment of this application, the profile acquisition module 1100 includes: a store grouping unit, configured to divide online stores that have registered the same communication characteristics into the same store group according to the communication characteristics registered by all online stores, wherein the communication characteristics include any one of telephone number, email address, and communication address; and a feature extraction unit, configured to acquire multiple specific sampling features for each online store to form a store profile, wherein the multiple specific sampling features include product dynamic information features and merchant static information features.

[0163] Based on any embodiment of this application, the similarity recognition module 1200 includes: a store pairing unit, configured to determine the pairs of online stores between the first store group and any other store group based on the Cartesian product, forming store pairs to be compared; a feature conversion unit, configured to convert the store profile of the online store in each store pair to a profile feature representation; a similarity operation unit, configured to map the profile feature representation of the store to be compared to a classification space to determine the semantic similarity between two online stores; and a similarity determination unit, configured to traverse each online store in the first store group, and for each online store, determine the store group containing the online store with the highest semantic similarity as the similar store group.

[0164] Based on any embodiment of this application, the feature conversion unit includes: a dynamic information processing unit configured to compile the product dynamic information features in the store profiles of the two online stores in the store pair to be compared into a one-hot encoded vector and construct a dynamic feature joint representation; a static information processing unit configured to compile the merchant static information features in the store profiles of the two online stores in the store pair to be compared into a text feature vector and construct a static feature joint representation; and a feature joint representation unit configured to construct a profile feature representation by combining the dynamic feature joint representation and the static feature joint representation.

[0165] Based on any embodiment of this application, the product dynamic information features include category coverage features and / or operation coverage features, and the merchant static information features include email address and / or network address.

[0166] Based on any embodiment of this application, the same-entity store identification device of this application includes: a message construction module, configured to construct a store group merging message, which includes store profiles of a first store group and its same-entity store group; a message sending module, configured to send the store group merging message to an audit interface; and an audit merging module, configured to respond to the confirmation instruction returned by the audit interface and merge the same-entity store group and its first store group into the same store group.

[0167] Based on any embodiment of this application, the same-entity store identification device of this application includes: a store exclusion module, configured to exclude the first store group and its same-entity store group from all store groups; and an iterative identification module, configured to continue to use any store group that has not been excluded from all store groups as the first store group, and iterate the above process to determine the same-entity store group for it.

[0168] Another embodiment of this application also provides a device for identifying physical stores. For example... Figure 8 The diagram shows the internal structure of a physical store identification device. This device includes a processor, a computer-readable storage medium, a memory, and a network interface connected via a system bus. The computer-readable, non-volatile storage medium stores an operating system, a database, and computer-readable instructions. The database stores information sequences, and when executed by the processor, these computer-readable instructions enable the processor to implement a physical store identification method.

[0169] The processor of the same-store identification device provides computing and control capabilities to support the operation of the entire device. The memory of the same-store identification device can store computer-readable instructions, which, when executed by the processor, cause the processor to perform the same-store identification method of this application. The network interface of the same-store identification device is used for communication with a terminal.

[0170] Those skilled in the art will understand that Figure 8 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the physical store identification device to which the present application is applied. The specific physical store identification device may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0171] In this embodiment, the processor is used to execute... Figure 7 The specific functions of each module are described, and the memory stores the program code and various data required to execute the above modules or sub-modules. 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 physical store identification device of this application, and the server can call the server's program code and data to execute the functions of all modules.

[0172] This application also provides a non-volatile readable storage medium storing computer-readable instructions, which, when executed by one or more processors, cause the one or more processors to perform the steps of the same physical store identification method of any embodiment of this application.

[0173] This application also provides a computer program product, including a computer program / instructions that, when executed by one or more processors, implement the steps of the method described in any embodiment of this application.

[0174] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments of this application can be implemented by a computer program instructing related hardware. This computer program can be stored in a non-volatile readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. The aforementioned storage medium can be a computer-readable storage medium such as a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM).

[0175] In conclusion, this application is economical, practical, and highly efficient in improving the ability to identify similar stores on e-commerce platforms and enhancing their risk control capabilities, thus effectively maintaining the operational order of e-commerce platforms.

Claims

1. A method of identifying a physical store, characterized by, The method comprises the following steps: obtaining the store images of all online stores in all store groups, each store group comprising at least one store image corresponding to an online store, and the online stores in the same store group being registered with the same communication feature, and each store image of an online store comprising a plurality of specific sampling features, wherein the plurality of specific sampling features comprise commodity dynamic information features and merchant static information features, the commodity dynamic information features comprising features corresponding to commodity items that change frequently in the daily operation of the online store, and the static information features comprising features closely related to the merchant but relatively stable in the daily operation of the online store; determining the semantic similarity between the store images of the online stores in the first store group and any other store group, and determining the similar store group of each online store in the first store group as the store group in which the online store with the highest semantic similarity is located; determining the vote rate of each similar store group according to the number of similar store groups, and determining the similar store group as the same entity store group of the first store group when the vote rate of the similar store group exceeds a preset threshold, wherein the vote rate of each similar store group is counted by using the number of similar store groups hit by each online store in the first store group.

2. The same entity store identification method according to claim 1, characterized by, Obtaining the store images of all online stores in all store groups comprises: dividing the online stores registered with the same communication feature into the same store group according to the communication feature registered by all online stores, wherein the communication feature comprises any one of a telephone number, an email address, and a communication address; for each online store, obtaining a plurality of specific sampling features to form a store image.

3. The same entity store identification method according to claim 2, characterized by, determining the semantic similarity between the store images of the online stores in the first store group and any other store group, and determining the similar store group of each online store in the first store group as the store group in which the online store with the highest semantic similarity is located, comprising: determining the two online stores between the first store group and any other store group based on Cartesian product to form a to-be-compared store pair; converting the store images of the online stores in each to-be-compared store pair into image feature representations; mapping the image feature representations of the to-be-compared store pair to a classification space to determine the semantic similarity between the two online stores; traversing each online store in the first store group, and for each online store, determining the store group in which the online store with the highest semantic similarity is located as the similar store group.

4. The same entity store identification method according to claim 3, characterized by, converting the store images of the online stores in each to-be-compared store pair into image feature representations comprises: encoding the commodity dynamic information features in the store images of the two online stores in the to-be-compared store pair into a one-hot encoding vector and constructing a dynamic feature joint representation; encoding the merchant static information features in the store images of the two online stores in the to-be-compared store pair into a text feature vector and constructing a static feature joint representation; constructing the dynamic feature joint representation and the static feature joint representation into an image feature representation.

5. The same entity store identification method according to claim 2, wherein, The commodity dynamic information features include category coverage features and / or operation coverage features, and the merchant static information features include mailbox addresses and / or network addresses.

6. The method according to any one of claims 1 to 5, wherein After determining that the similar store group is the same entity store group of the first store group according to the number of similar store groups and the vote rate of each similar store group, the method further comprises: constructing a store group combination message, wherein the store group combination message comprises a store image of the first store group and the same entity store group of the first store group; sending the store group combination message to an audit interface; combining the same entity store group and the first store group into the same store group according to a confirmation instruction returned by the audit interface.

7. The method according to any one of claims 1 to 5, wherein After determining that the similar store group is the same entity store group of the first store group according to the number of similar store groups and the vote rate of each similar store group, the method further comprises: excluding the first store group and the same entity store group from all store groups; iterating the above process to determine the same entity store group for the first store group by using any store group that is not excluded from all store groups as the first store group.

8. A same store identification device, characterized by, The method comprises: an image acquisition module configured to acquire a store image of an online store in all store groups, each store group comprising a store image corresponding to at least one online store, a plurality of online stores in the same store group being registered with the same communication feature, the store image of each online store comprising a plurality of specific sampling features, the plurality of specific sampling features comprising commodity dynamic information features and merchant static information features, the commodity dynamic information features comprising features corresponding to commodity items that change frequently in the daily operation of the online store, and the static information features comprising features that are closely related to the merchant but relatively stable in the daily operation of the online store; a similarity identification module configured to determine a semantic similarity between store images of two online stores in a first store group and any other store group, and determine, for each online store in the first store group, the other store group in which the online store constitutes the highest semantic similarity as a similar store group; a same-different identification module configured to determine a vote rate of each similar store group according to the number of similar store groups, and determine that the similar store group is the same entity store group of the first store group when the vote rate of the similar store group exceeds a preset threshold, wherein the number of similar store groups hit by each online store in the first store group is used to calculate the vote rate of each similar store group.

9. A same entity store identification device comprising a central processing unit and a memory, characterized by, The central processing unit is configured to invoke a 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 by The computer program is stored in the form of computer readable instructions and is implemented according to the method of any one of claims 1 to 7, and when the computer program is invoked and run by a computer, the steps included in the corresponding method are executed.

Images