CN115860482A

CN115860482A - Shop risk identification method and device, equipment, medium and product thereof

Info

Publication number: CN115860482A
Application number: CN202211700148.7A
Authority: CN
Inventors: 刘涛
Original assignee: Guangzhou Huanju Shidai Information Technology Co Ltd
Current assignee: Guangzhou Huanju Shidai Information Technology Co Ltd
Priority date: 2022-12-28
Filing date: 2022-12-28
Publication date: 2023-03-28

Abstract

The application relates to a shop risk identification method and a device, equipment, medium and product thereof, wherein the method comprises the following steps: acquiring a sample data set, wherein the sample data set comprises a plurality of shop samples, and each shop sample comprises a shop portrait of a single online shop and a risk label of the shop portrait; classifying all shop samples in the sample data set into a plurality of sample subsets by adopting a decision tree model trained by the sample data set to obtain a rule set corresponding to each sample subset; screening out a plurality of rule sets according to the statistical indexes of the sample subsets corresponding to the rule sets to serve as target rule sets; and configuring a risk strategy of the store wind control system by using the target rule set, so that the store wind control system identifies a risk label corresponding to the store image of any online store according to the risk strategy. Based on the decision tree model, risk strategies are excavated by utilizing the shop portrait, so that the accuracy of the strategies can be ensured, the strategy has stronger interpretability, and the online speed is higher.

Description

Shop risk identification method and device, equipment, medium and product thereof

Technical Field

The application relates to an e-commerce information processing technology, in particular to a shop risk identification method and a device, equipment, medium and product thereof.

Background

The e-commerce platform usually has a large number of online shops, and the online shops are actually operated by different operation entities, so that different operators and operation modes thereof enable the corresponding online shops to present different risk levels. The risk level of the online shop refers to an abstract description concept of the operational health condition of the online shop, and is generally reflected in the level of the operational credit of the online shop, the online shop with the high risk level has better operational credit, and the online shop with the low risk level, i.e. the risky shop, has relatively poor operational credit.

Therefore, the risk level of the online stores of the e-commerce platform basically determines the survival condition of the platform, if the concentration of the risk stores in the e-commerce platform is too high, the platform is difficult to develop healthily, various resources of the platform can be seized by the risk stores, the operating pressure of good stores is increased, and the platform faces huge risks under the vicious circle. Therefore, how to effectively, accurately and quickly identify and dispose the air-out shop is a big problem in the wind control of the e-commerce platform, and the adverse effect of the air-out shop on the platform is reduced.

In the existing situation, the risk shop investigation mainly depends on two parts, one is that the meat monitoring of the operators is depended on, or the group management and control is carried out through various statistical indexes, and the method is not efficient and strongly depends on the personal experience of the operators. Secondly, outputting the shop risk scores of the shops on the line through various models, and carrying out manual auditing and control on the high-risk-score shops; the model is mainly divided into a scoring card and an integrated tree, and the interpretability and the accuracy cannot be considered at the same time, wherein the model of the scoring card has stronger interpretability, but the precision of the model is reduced due to variable binning; the accuracy of the integrated tree model is high, but the model result is not strong in interpretability.

In view of this, it is necessary to improve the related art to optimize the store risk identification capability of the e-commerce platform and ensure the healthy operation of the platform.

Disclosure of Invention

The present application is directed to solving the above-mentioned problems and providing a store risk identification method, and a corresponding apparatus, device, non-volatile readable storage medium, and computer program product.

According to one aspect of the application, a shop risk identification method is provided, which includes the following steps:

acquiring a sample data set, wherein the sample data set comprises a plurality of shop samples, and each shop sample comprises a shop portrait of a single online shop and a risk label of the shop portrait;

classifying all shop samples in the sample data set into a plurality of sample subsets by adopting a decision tree model trained by the sample data set to obtain a rule set corresponding to each sample subset;

screening out a plurality of rule sets according to the statistical indexes of the sample subsets corresponding to the rule sets to serve as target rule sets;

and configuring a risk strategy of the store wind control system by using the target rule set, so that the store wind control system identifies a risk label corresponding to a store image of any online store according to the risk strategy.

Optionally, configuring a risk policy of the store wind control system by using the target rule set, so that after the store wind control system identifies a risk label corresponding to a store image of an arbitrary online store according to the risk policy, the method includes:

constructing a shop portrait of an online shop identified by the shop wind control system and a corresponding risk label as a new shop sample, and adding the new shop sample into the sample data set;

responding to a timing arrival event triggered by a timing task, and adopting the sample data set to retrain the decision tree model to a convergence state;

and re-determining a target rule set corresponding to the sample data set by adopting a decision tree model retrained to be in a convergence state, and updating the risk strategy of the store wind control system.

judging whether the risk label of the online shop identified by the shop wind control system belongs to a target type or not by the shop wind control system, and marking the online shop as a risky shop when the risk label belongs to the online shop of the target type;

generating an alarm message corresponding to the risky shop and sending the alarm message to a specific auditing interface;

and acquiring a confirmation instruction returned by the auditing interface, and responding to the confirmation instruction to modify the operation activity authority of the corresponding online shop.

Optionally, obtaining the sample data set includes:

determining a target line shop on the e-commerce platform for extracting a shop sample;

correspondingly setting the risk label of the shop on the corresponding target line as a high risk type or a low risk type according to whether the evaluation index of the shop on the target line is higher than a preset threshold value;

acquiring various shop-level data characteristics of each target on-line shop as a shop portrait of the corresponding target on-line shop;

store the shop images of all the target online shops and the corresponding risk labels as corresponding shop samples in a sample data set.

Optionally, in the step of obtaining a plurality of store-level data features of each online target store as the store representation of the corresponding online target store, the plurality of store-level data features include any of a basic feature of the store, a transaction feature of the store, an illegal feature of the store, a behavior feature of the store, and a seller-related feature associated with the store.

Optionally, screening out a plurality of rule sets according to the statistical indexes of the sample subsets corresponding to the rule sets, where the screening is used as a target rule set, and the screening includes:

according to the sample subsets corresponding to the rule sets, counting the accuracy and recall rate of the risk labels corresponding to the target types in each sample subset as statistical indexes;

screening out a plurality of sample subsets with accuracy higher than a first preset threshold and recall rate higher than a second preset threshold, and taking corresponding rule sets as target rule sets.

Optionally, configuring a risk policy of a store wind control system by using the target rule set, so that the store wind control system identifies a risk label corresponding to a store image of an arbitrary online store according to the risk policy, including:

determining a strategy feature table according to the data features in the target rule set;

configuring the policy feature table as a risk policy in a store wind control system;

and controlling the store wind control system to start the risk strategy, scanning each online store in the e-commerce platform, and identifying a risk label corresponding to the store image of each online store according to the risk strategy.

According to another aspect of the present application, there is provided a store risk identification device including:

the system comprises a sample acquisition module, a risk identification module and a risk identification module, wherein the sample acquisition module is used for acquiring a sample set, and the sample set comprises a plurality of shop samples, and each shop sample comprises a shop portrait of a single online shop and a risk label thereof;

the rule analysis module is used for classifying all shop samples in the sample data set into a plurality of sample subsets by adopting a decision tree model trained by the sample data set to obtain a rule set corresponding to each sample subset;

the rule optimization module is set to screen out a plurality of rule sets according to the statistical indexes of the sample subsets corresponding to the rule sets to serve as target rule sets;

and the wind control configuration module is set to configure a risk strategy of the store wind control system by using the target rule set, so that the store wind control system can identify a risk label corresponding to a store image of any online store according to the risk strategy.

According to another aspect of the present application, there is provided a store risk identification apparatus comprising a central processor and a memory, the central processor being configured to invoke execution of a computer program stored in the memory to perform the steps of the store risk identification method of the present application.

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

According to another aspect of the present application, there is provided a computer program product comprising computer programs/instructions which, when executed by a processor, implement the steps of the method described in any one of the embodiments of the present application.

Compared with the prior art, the application has abundant technical advantages, including but not limited to:

firstly, the method comprises the steps of training a decision tree model by using the store portrait of the online store of the e-commerce platform and a risk label thereof as store samples, classifying the store samples by using the decision tree model, determining a rule set and a sample subset which are opposite to each classification path, optimizing a target rule set according to statistical indexes of the sample subset, configuring a risk strategy of the store air control system by using the optimized target rule set, enabling the store air control system to adapt to various changes of the online store, and correctly identifying whether each online store in the e-commerce platform belongs to the air control store according to the risk strategy.

Secondly, the important basic data for determining whether the online shop belongs to the risky shop is the shop portrait of the online shop, the shop portrait is usually derived from the data of the online shop in terms of transaction, risk, behavior and the like, and the data is the dominant data of the e-commerce platform, so that an endogenous wind control capacity improving mechanism is established for the shop wind control system of the e-commerce platform, and the maintenance means of the e-commerce platform for self healthy operation is perfected in a technical implementation mode.

In addition, the online identification method and the online identification system are based on the decision tree model, the risk strategy for identifying whether the online shop is a risk shop is mined by utilizing the shop portrait, the accuracy of the strategy can be ensured, the online identification method and the online identification system have high interpretability, and the online speed can be high, so that the online identification method and the online identification system can be more suitable for high-speed change of the business risk of the e-commerce.

Drawings

In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed to be used in the description of the embodiments are briefly introduced below, and it is obvious that the drawings in the following description are only some embodiments of the present application, and it is obvious for those skilled in the art to obtain other drawings based on these drawings without creative efforts.

Fig. 1 is a network architecture corresponding to an exemplary deployment of a store wind control system in an e-commerce platform according to the present application;

FIG. 2 is a schematic flow chart diagram illustrating one embodiment of a store risk identification method of the present application;

FIG. 3 is a schematic flow chart illustrating the application of a self-learning mechanism to update a risk policy in the practice of the present application;

FIG. 4 is a schematic flow chart illustrating control of an inauguration store according to an embodiment of the present application;

FIG. 5 is a schematic flow chart illustrating the construction of a sample data set according to an embodiment of the present application;

FIG. 6 is a schematic flow chart illustrating risk policy configuration in an embodiment of the present application;

FIG. 7 is a functional block diagram of a store risk identification device of the present application;

fig. 8 is a schematic structural view of a store risk identification device used in the present application.

Detailed Description

Referring to fig. 1, an exemplary network architecture of the present application is suitable for an e-commerce platform scenario and includes a terminal device 80, a security server 81, and a store server 82.

The security server 81 may be a main executive body of the store wind control system of the present application, which identifies whether an online store in an e-commerce platform belongs to a risky store according to a risk policy, which may be determined and configured according to a target rule set determined by the store risk identification method of the present application. The shop risk identification method can be implemented as a computer program product through programming, and runs in any computer equipment so as to output the target rule set.

The store server 82 may be used to deploy one or more online stores of an internet platform, and various network requests to access the online stores may be responded to by corresponding application services provided by the store server. Each online store has various basic feature data suitable for creating the store image, such as the basic features of the store, the transaction features of the store, the violation features of the store, the behavior features of the store, the relevant features of the seller associated with the store, and the like, and any of these feature data can be used for constructing the store image.

In addition, each online store forms various evaluation index data in the historical operation process, such as a customer complaint rate generated by a customer complaint event in each time period, a good rating rate obtained by a user purchasing a commodity item of the online store, and the like. These evaluation index data can be used to determine the risk label corresponding to the shop image of the corresponding online shop, so as to construct a shop sample corresponding to each online shop, which is used to train the decision tree model of the present application to serve for the determination of the target rule set.

The terminal device 80 may be used to trigger the network request to use various application services in the store server 82, such as to browse the online store, perform an order placement operation for an item at the online store, and so on.

The security server 81 can read the relevant basic feature data of the online shop in the shop server 82 to update the shop image of the online shop, and the shop wind control system can determine the risk label of the online shop according to the matching relation between various feature data in the shop image and the risk strategy of the shop wind control system based on a rule matching mode, so as to realize the identification of the risk level of the online shop.

Of course, the computer program product programmed according to the store risk identification method of the present application may also be run in the security server 81, and serve the store wind control system specifically, interact with the store wind control system in a coordinated manner, continuously acquire feature data from each online store through a self-learning mechanism to update a corresponding store portrait, continuously identify a risk tag to which each online store belongs, iteratively update a decision tree model with a training data set constructed by the store portrait and the risk tag, then update a target rule set with the decision tree model, and update a risk policy of the store wind control system according to the updated target rule set, so that the store wind control system continuously improves the risk level identification capability of whether each online store belongs to a risk store and continuously improves the wind control capability of the e-commerce platform for the risk store depending on the continuously updated risk policy.

Referring to fig. 2, according to an embodiment of a store risk identification method provided by the present application, the method includes the following steps:

step S1100, obtaining a sample data set, wherein the sample data set comprises a plurality of shop samples, and each shop sample comprises a shop portrait of a single online shop and a risk label thereof;

the sample data set may be prepared in advance so that a large number of store samples are contained, the store samples are constructed for each online store, and the online stores used for constructing the store samples may be online stores in the current e-commerce platform or online stores of other third-party e-commerce platforms as long as corresponding data required for constructing the store samples can be provided. In general, it is preferable to collect their corresponding store samples for online stores within the current e-commerce platform.

The shop sample comprises shop images of corresponding online shops and corresponding risk labels. The shop image in the shop sample is a set of feature data formed by extracting various feature data of the corresponding online shop, which can directly or indirectly reflect the operation credit of the online shop, and the specific feature data of the shop image is formed, and can be flexibly set according to actual situations.

In one embodiment, various store-level data characteristics of online stores may be extracted to construct their corresponding store portraits. The store-level data characteristics refer to data characteristics described from the level of an online store, and include but are not limited to basic characteristics of the store, transaction characteristics of the store, violation characteristics of the store, behavior characteristics of the store, relevant characteristics of a seller associated with the store, and the like.

Each online shop can correspondingly obtain a shop portrait, each shop portrait is determined according to the uniform characteristic data composition, and all the characteristic data adopted in the shop portrait can comprehensively reflect the operation credit of the corresponding online shop. Therefore, for each online shop, the corresponding shop image is adopted, and the mapping relation data is established between the corresponding shop image and the corresponding risk label, so that the shop sample corresponding to the online shop is formed.

The risk label can be at least two sources for the sample data set, wherein one source can be determined by using some evaluation indexes of an online shop at the initial preparation stage of the sample data set, and the other source can be determined by identifying the corresponding risk label according to the shop image of the online shop by the subsequent air-controlled shop system, and the shop image of the online shop and the corresponding risk label form a new shop sample of the sample data set or replace the original shop sample.

In one embodiment, the risk labels of the online stores can be divided into two types of labels, namely a high risk type and a low risk type, wherein the high risk type risk label refers to the online store with lower operation credit as the name implies, and the corresponding online store is a risk store which is relatively not in line with the operation specification of the e-commerce platform; the low-risk type risk label refers to an online shop with higher business credit, and the corresponding online shop is a normal shop which relatively accords with the business specification of an e-commerce platform. In other embodiments, the risk labels could also be theoretically divided into more than two types, which could be implemented by one skilled in the art according to the above principles given in this application.

For the sake of image understanding, the risk label belongs to a high risk type store sample and can be regarded as a black sample, and the risk label belongs to a low risk type store sample and can be regarded as a white sample.

Step S1200, classifying all shop samples in the sample data set into a plurality of sample subsets by adopting a decision tree model trained by the sample data set, and obtaining a rule set corresponding to each sample subset;

in order to extract a rule for identifying a risk label to which an online shop belongs by using the sample data set, the sample data set is adopted to train a decision tree model, the decision tree model (decisions ion tree) is a simple and efficient prediction model with strong interpretability and is generally generated from top to bottom, each decision or event (namely a natural state) can lead out two or more events and lead to different results, and the decision branch is drawn into a graph and is very similar to a branch of a tree, so the decision tree is called a decision tree. The alternative expression is to use a tree diagram to represent the expected value of each decision, and finally select the decision method with the maximum benefit and the minimum cost through calculation.

And a decision tree model trained by the sample data set can obtain sample subsets obtained by classifying and collecting all store samples in the sample data set at each leaf node of the decision tree model, each sample subset comprises part of store samples in the sample data set, corresponding classification paths are formed from a root node of the decision tree model to each leaf node of the decision tree model, the characteristics of internal nodes in each classification path correspond to the conditions of the rules, class labels of the leaf nodes correspond to the conclusions of the rules, and therefore each classification path correspondingly provides a corresponding rule set.

Common decision tree models, including but not limited to: CLS, ID 3, C4.5, CART, RF, random forest algorithm and the like can be used as the selection type of the decision tree model of the application.

In one embodiment, after determining the type of the decision tree model of the present application, the decision tree model of the present application can be prepared and used according to the following principles:

first, the decision tree model is trained:

building a two-classification model by applying a corresponding decision tree algorithm to form the decision tree model of the application, wherein the relevant parameter setting of the model is roughly determined according to the distribution of the shop samples in the sample data set, and then fine adjustment or parameter grid search can be carried out; and the features can be iteratively screened according to the importance of the variables, the features with the former importance are retained, and the features with the latter importance are removed until the technical indexes of the model are stable or the retained features meet the on-line requirements.

Secondly, the effect of the decision tree model is evaluated:

the technical evaluation index of the decision tree model generally uses AUC (area under curve) and KS (Ko l mogorov-Smi rnov, the slope of the ROC curve is the intercept of the tangent of 1). The distinguishing capacity of the decision tree model for black and white samples can be evaluated, and the greater AUC and KS are, the better the decision tree model is represented; in order to ensure the stability of the model, cross-time sample effect verification is preferably added during model evaluation.

Then, a rule set generated by the decision tree model is extracted:

as previously described, the decision tree model may extract a rule set. If a decision tree with a depth of 4 is constructed, because the decision trees are binary trees, then a maximum of 16 rule sets are generated (note: the maximum number of decision tree model rules is 2^ h, and h is the decision tree depth), and the total store samples can be divided into 16 sample subsets at most. Thus, each sample subset and the corresponding rule set obtained after the decision tree model classifies all the shop samples in the sample data set are obtained.

S1300, screening a plurality of rule sets according to the statistical indexes of the sample subsets corresponding to the rule sets to serve as target rule sets;

the sample subsets generated by the decision tree model correspond to the rule sets one by one, but the classification effects of the sample subsets are different, and the classification effects can be determined through corresponding statistical indexes of the sample subsets. The statistical indexes are common in accuracy, recall rate and the like. The screening of the plurality of rule sets is implemented by taking accuracy as an example, so as to screen out a representative rule set as a target rule set.

Assuming that the total number of shop samples is C, wherein the number of black samples is A, the number of white samples is B, and still taking a decision tree with a depth of 4 as an example, 16 rule sets are extracted at most, and each rule set contains 4 features at most, wherein X i represents a policy feature, and Ci represents a segmentation node of the feature; the 16 rule sets divide the full-size sample C into mutually exclusive sample subsets C1-C16, and the sample subsets C1-C16 can count the number of black samples and the number of white samples respectively, the accuracy calculation caliber is the proportion of black and white samples in the sample subset (Pi = Ai/bi), and the recall calculation caliber is the proportion of black samples in the sample subset to the total black samples in the sample data set (Ri = Ai/a).

The following is a schematic diagram of the calculations performed according to the above principles, and may be referenced:

it can be seen from the above schematic diagram that each rule set has its corresponding feature for decision making and the rule condition corresponding to the feature, and on the basis of statistics of each rule set, statistical indexes such as accuracy and recall rate can be determined, and the rule sets are sorted according to any one statistical index, so that the rule set with better sorting can be selected as a target rule set, and the target rule set is a description of a rule which can reflect the sorting capability better. It is easy to understand that when the priority is selected according to the accuracy, the accuracy searching capability of the corresponding rule set is stronger, and the black and white samples can be more accurately judged; when the preference is selected according to the recall rate, the searching capability of the corresponding rule set is stronger, and the capability of identifying different samples in a generalization manner is better.

In one embodiment, the target rule set may be optimized according to a statistical indicator, such as screening based on accuracy, by a predetermined threshold, which may be set to a value such as 90%. In a further embodiment, it may also be associated whether the recall rate meets the service requirement for screening. In this regard, those skilled in the art will be able to implement the above principles with flexibility.

And step S1400, configuring a risk policy of the store wind control system by using the target rule set, so that the store wind control system can identify a risk label corresponding to a store image of any online store according to the risk policy.

The determination of the plurality of target rule sets is actually a determination of a plurality of preferred rules for identifying whether the store representation of the online store belongs to a black or white sample, i.e., a determination of whether the corresponding online store belongs to a high risk category or a low risk category based on the store representation. Accordingly, these targeting rules can be used to configure the risk policies of the store wind control system of the present application. The risk policy of the store wind control system may be configured with a plurality of sets, for example, a corresponding risk policy corresponding to a target rule set.

In addition, in other embodiments, based on the target rule set, more features may be manually introduced and corresponding rule conditions may be appended to constrain the corresponding target rule set, thereby forming a risk policy revised based on the target rule set. In this regard, one skilled in the art can implement this as desired.

After the configuration of the risk strategies of the store wind control system is completed, the store wind control system can apply the risk strategies, after a store portrait of an online store is obtained, one or more risk strategies are used for carrying out rule matching on feature data in the store portrait, then a rule matching result is obtained, a corresponding risk label is determined, namely whether the online store is a high-risk type or a low-risk type is judged, and when the online store is the high-risk type, the online store is indicated to be a risk store, and corresponding marking can be carried out.

In one embodiment, after the store wind control system identifies the risk label corresponding to the online store according to the matching of the store image of the online store and each risk policy, the store image of the online store and the corresponding risk label can be constructed into the store sample of the online store, and the store sample can be stored in the sample data set, so that the store sample in the sample data set can be expanded. Subsequently, the sample data set with the expanded shop samples is used for iterative training of the decision tree model, and a new target rule set is generated by using the decision tree model to update the risk strategy of the shop air control system, so that a self-learning mechanism can be realized for the shop air control system, the shop air control system can be continuously iterated and continuously upgraded, and the risk shop identification capability of the shop air control system can be continuously improved.

As can be seen from the above embodiments, the present application has rich technical advantages, including but not limited to:

In addition to any embodiment of the present application, referring to fig. 3, after configuring a risk policy of a store wind control system by using the target rule set, and enabling the store wind control system to identify a risk label corresponding to a store image of an arbitrary online store according to the risk policy, the method includes:

step S2100, constructing a shop portrait of an online shop identified by the shop wind control system and a corresponding risk label as a new shop sample, and adding the new shop sample into the sample data set;

various basic data of the online store are dynamically changed, and the store image is actually changed continuously. Therefore, the store wind control system runs in real time, continuously obtains the store portrait of each online store in the e-commerce platform, and dynamically identifies the risk label of each online store, so as to find out the risk store with abnormal operation credit in time.

When the store wind control system identifies a risk label corresponding to a store portrait of a certain online store, the store portrait and the risk label can be constructed into a new store sample, and then the new store sample is added into the sample data set.

In an embodiment, when a new shop sample is added to a sample data set, an additional mode can be adopted to expand the shop sample of the sample data set, so that the sample characteristics of the sample data set are generalized, the decision tree model is more easily converged when the decision tree model is retrained subsequently, and the characteristic generalization capability of the decision tree model is improved.

In another embodiment, when a new store sample is added to the sample data set, the last determined store sample of the online store may be replaced with the new store sample, so as to optimize the store sample of the online store.

In one embodiment, after the store wind control system identifies the corresponding risk label of the online store, the system can be confirmed by a background user and then determine whether to add the risk label into the sample data set, and by combining a manual auditing mechanism, if the strategy is online for a period of time, a new risk mode which cannot be covered by the original strategy set appears; if the samples confirmed to be normal through manual examination are strategy mishits, risk strategy correction is needed to be carried out on the samples, so that a self-learning mechanism is needed to be used, the mishit data are reconstructed into shop samples to be added into the sample data set, strategy mining and updating are carried out timely, and the accuracy rate and the application effect of the risk strategy are improved.

Through the execution of the steps, it is easy to understand that the shop samples with the concentrated sample data are continuously expanded and/or updated.

Step S2200, responding to a timing arrival event triggered by a timing task, and adopting the sample data set to retrain the decision tree model to a convergence state;

the method is suitable for continuous upgrading of the sample data set, and the triggering of the learning behavior of the self-learning mechanism of the store wind control system can be controlled through a timing task. The timing task may set the trigger time period of the timing arrival event as needed, for example, set by day/week/month. When the corresponding period expires, a corresponding timing arrival event is triggered.

And in response to the timing arrival event, the upgraded sample data set can be adopted, the training of the decision tree model is restarted, and the decision tree model is trained to a convergence state again through each shop sample in the sample data set and the risk label thereof.

And S2300, adopting a decision tree model retrained to be in a convergence state to re-determine a target rule set corresponding to the sample data set and update the risk strategy of the store wind control system.

And (3) retraining the decision tree model to be in a convergence state, so that the classification capability of the decision tree model is upgraded, and the decision tree model can more accurately make a rule set. Therefore, the upgraded decision tree model is used for predicting the sample data set again, and the sample data set is reclassified to determine a plurality of sample subsets, so that the rule set corresponding to each sample subset is determined.

Similar to the implementation manner of the foregoing embodiment, on the basis of the rule set and the corresponding sample subset that are re-determined by the updated decision tree model, the target rule set having a better statistical index in each rule set can be further determined by using the statistical indexes corresponding to each sample subset, and the risk policies in the store wind control system can be further updated by using these target rule sets, thereby implementing the updating of the risk policies. After the risk strategy of the shop wind control system is upgraded, the ability of accurately identifying whether the on-line shop is a risk shop can be further improved, and by analogy, continuous iteration is performed, so that a self-learning mechanism for upgrading the shop wind control system is actually realized.

According to the embodiment, a self-learning mechanism is introduced into the mining process of the risk strategy of the store wind control system, and for a strategy mishit store sample, the sample label can be updated at regular time, the decision tree model is retrained, and the strategy rule is corrected; meanwhile, as time goes on, a new risk mode appears, so that the original risk strategy cannot be covered, the self-learning mechanism can regularly learn new shop risk data distribution, a new rule set is mined, the accuracy of the online rule set is constantly guaranteed, and the generalization of the shop risk strategy is increased. Therefore, under the action of the self-learning mechanism, the risk strategy in the store wind control system can be kept, and not only can the sudden risk be covered, but also the normal risk can be dealt with.

In addition to any embodiment of the present application, referring to fig. 4, after configuring a risk policy of a store wind control system by using the target rule set, and enabling the store wind control system to identify a risk label corresponding to a store image of an arbitrary online store according to the risk policy, the method includes:

step S3100, judging whether the risk label of the on-line shop identified by the shop air control system belongs to a target type, and marking the on-line shop as a risk shop when the risk label belongs to the target type of the on-line shop;

in order to improve the ability of the e-commerce platform to maintain the operational order in the platform, the store wind control system can perform corresponding processing according to the result determined by the e-commerce platform for the on-line stores.

Therefore, after the store wind control system identifies the risk label of a certain online store by using the risk strategy, whether the risk label belongs to a target type is firstly identified, the target type refers to a high-risk type by taking the above two classification conditions as an example, and when the risk label of the online store is the high-risk type, the corresponding online store is shown to be not in accordance with the operating specification of the current e-commerce platform, so that the online store can be directly marked as the risk store.

The operation of marking the online stores as the risky stores can include adding corresponding risky store identifications for the online stores at a data level, and can also include plain text marking the corresponding risky store identifications on the homepages of the online stores, so that the operation can be flexibly implemented by the technical personnel in the field.

Step S3200, generating an alarm message corresponding to the risk shop, and sending the alarm message to a specific checking interface;

in order to facilitate the background user to manually confirm the risky stores or make corresponding responses, further, for the risky stores, a corresponding warning message may be generated and sent to a corresponding auditing interface for the corresponding background user to perform further processing according to the warning message, for example, confirming the identification result of the store wind control system and sending back a corresponding confirmation instruction.

And S3300, obtaining a confirmation instruction returned by the audit interface, and modifying the operation activity authority of the corresponding online shop in response to the confirmation instruction.

And after the background user sends back a confirmation instruction, responding to the confirmation instruction, and performing corresponding authority control processing on the online stores belonging to the risky stores by the store wind control system according to preset business logic. In a typical embodiment, the operation authority of the online store may be controlled, where the operation authority is used to limit the operation capability of the online store, for example, the operation authority may be set by a person skilled in the art, and specifically, the person may set the operation authority to limit the operation capability of the online store, such as closing an access entrance of a user of the online store, and prohibiting the online store from issuing a sales activity.

According to the embodiment, the shop wind control system can process and respond in real time according to the recognition result of the shop on each online shop on the risk label, and after the shop on the line is recognized to belong to the risk shop, the system can interact with the auditing interface through the warning message to timely implement necessary right limiting processing on the risk shop, so that the healthy operation of the E-commerce platform is ensured, and the management order of the E-commerce platform is effectively maintained. Therefore, the risk strategy of the store wind control system is updated based on the technical scheme, so that the method has a plurality of advantages, corresponding criteria are more accurate, and the disposal efficiency is higher.

On the basis of any embodiment of the present application, please refer to fig. 5, which illustrates obtaining a sample data set, including:

step S1110, determining a target line shop for extracting shop samples in the E-commerce platform;

for example, some online stores are in a closed state for a long time, and corresponding characteristic data cannot effectively reflect daily operation conditions of the online stores, so that the online stores such as the online stores can not be considered for preparing store samples.

In any case, those skilled in the art can determine some target online stores, including the full amount of online stores, from the e-commerce platform, so as to prepare corresponding store samples on the basis of the target online stores.

Step 1120, correspondingly setting the risk label of the shop on the corresponding target line as a high risk type or a low risk type according to whether the evaluation index of the shop on the target line is higher than a preset threshold value;

for each target online store, firstly, the corresponding risk label is generated in an automatic mode to form labeling information of the store sample, and the increase of implementation cost caused by using a manpower to define the risk label is avoided.

When the evaluation indexes of the stores on the target lines are automatically identified, for example, the calibers of the risky stores are determined, namely the evaluation indexes are unified, if the stores with the customer complaint amount rate exceeding 10% in the last 3 months are risky stores, the customer complaint amount rates of the stores on the target lines in the last 3 months need to be calculated for the stores on the target lines, the risk label of the store with the customer complaint amount rate exceeding 10% is identified as a high-risk type, and the value is 1, otherwise, the risk label of the store is identified as a low-risk type, and the value is 0.

S1130, acquiring various shop-level data characteristics of each target online shop as shop images of corresponding target online shops;

correspondingly, various shop-level data characteristics of the shops on the target lines are acquired, and corresponding shop images are constructed. In the embodiment, five types of characteristics, namely, the basic characteristic of the shop of the target on-line shop, the transaction characteristic of the shop, the violation characteristic of the shop, the behavior characteristic of the shop and the related characteristic of the seller related to the shop, can be simultaneously obtained to construct the corresponding shop image. The following exemplary description of the composition of various types of features is provided by way of an exemplary acquisition scheme for the various types of features:

the basic characteristics of the store, for example, may include the store status, the number of registered days, the registered country, the number of paid days, the subscription status, and the subscription package of the store, and characteristics of the number of registered days of the store corresponding to the seller, the number of stores to which the seller belongs, and the number of stores to which the seller belongs to open business, which are used to describe the registration, business, subscription, and the like of the store.

The trading characteristics of the shop comprise historical order quantity, historical order amount, order quantity on a near x day, trading days on a near x day, maximum/minimum/average discount rate of orders on a near x day, average logistics cost ratio of orders on a near x day and other characteristics < x > 1/3/5/7/15/30/60/90/180/365 (unit: day) respectively taken according to slicing time, and are used for depicting the trading situation of the shop/seller in different time intervals.

The violation characteristics of the shop comprise the order conditions of the shop and the seller corresponding to the shop, such as ultralow price, forbidden sale, infringement commodity and reverse customer complaint, and mainly comprise the following steps: the method comprises the steps that x is respectively taken as 1/3/5/7/15/30/60/90/180/365 (unit: day) according to slicing time, and the number of the ultralow price/forbidden limit sale/infringement commodities on the shelves of the stores and the sellers on the near x days, the ratio of the ultralow price/forbidden limit sale/infringement commodities on the shelves of the stores and the sellers on the near x days, the number of the order/number of the order commodities on the shelves of the stores and the sellers on the near x days, the reverse amount generated on the last x days of the stores and the sellers, the ratio of the reverse amount generated on the last x days of the stores and the sellers to the deal amount and the like are used for depicting the risk characteristic situation of the stores/sellers in different time intervals.

The behavior characteristics of the shop are processed according to the shop behavior buried point data, the overall behaviors of the shop and the seller are depicted, and the overall behaviors comprise the number of times of login of the shop on the near x days, the number of times of putting commodities on shelves of the shop on the near x days, the number of times of single click of shop on the near x days, the number of times of newly-added payment modes of the shop on the near x days, the number of times of login of the shop on the country/city on the near x days, the number of days of latest login of the shop up to the present, the most frequently-logged-in country, the most frequently-logged-in city and other characteristics < x (unit: day) are respectively valued according to the slicing time, and the characteristics are used for depicting the behavior conditions of the shop/the seller in different time intervals.

As for the store-associated seller-related features, they have been included in the acquisition of the four features exemplified above.

According to the above exemplary shop image feature data sampling scheme, the feature composition of the shop image can be flexibly defined as required, the overall goal is to determine some feature data capable of describing the online shop business information on the whole, and the feature data is organized in order to form a corresponding feature set, wherein the feature set plays the role of the shop image.

It will be appreciated that each targeted online store may have its corresponding store representation determined in accordance with the principles of the above example.

Step S1140 is to store the store images of the respective online stores and the corresponding risk tags in a sample data set.

After the store portrait of each target online store and the corresponding risk label are determined according to the above process, the store portrait of each target online store and the risk label are mapped and stored in the sample data set to become the store sample in the sample data set, which can be used for implementing the training of the decision tree model, and can also be used for classifying the trained decision tree model to determine the rule set.

According to the embodiment, the shop images of the shops on the line are constructed by matching the shop feature data with rich dimensions, so that the risk level of the shops in each dimension can be favorably drawn from multiple directions, and the potential risk points can be more accurately captured.

On the basis of any embodiment of the present application, screening out a plurality of rule sets according to the statistical indexes of the sample subsets corresponding to the respective rule sets, as a target rule set, includes:

step 1310, according to the sample subsets corresponding to the rule sets, counting the accuracy and recall rate of the risk labels corresponding to the target types in each sample subset as statistical indexes;

in this embodiment, when the rule set is optimized based on the statistical indicators of the sample subsets determined after the classification by the decision tree model, such statistics can be implemented by using both the statistical indicators of accuracy and recall rate. The statistical methods for the accuracy and recall of each sample subset are the same as the embodiments disclosed in the previous embodiments of the present application, and are not repeated here. According to the methods, the accuracy and the recall rate corresponding to the target type indicating the target type belonging to the black sample, namely the risk label of the high risk type corresponding to the risk shop, are determined.

Step S1320, screening out a plurality of sample subsets having accuracy higher than a first preset threshold and recall rate higher than a second preset threshold, and using their corresponding rule sets as target rule sets.

In the embodiment, the accuracy is used as a preferential statistical index, a part of rule sets corresponding to a part of sample subsets with accuracy higher than a first preset threshold, for example, 90% are determined from all rule sets determined by a decision tree model, and then on the basis of the part of rule sets with high accuracy, a second preset threshold corresponding to a recall rate is used to screen out a target rule set of the part of rule sets, wherein the recall rate is higher than the second preset threshold, the accuracy statistical index is applied to ensure the accuracy of the store wind control system in risk store identification, and the recall rate statistical index is applied to ensure that the search capability of the corresponding sample subsets can cover business requirements.

According to the embodiments, the application adapts to specific business requirements, the target rule sets can be optimized by means of recall rate according to the specific business requirements under the condition that accuracy is considered preferentially, and the risk strategies generated by the target rule sets can meet the business requirements, ensure accurate identification of the risk labels of the online stores, and enable the store wind control system to have higher reliability.

In addition to any embodiment of the present application, referring to fig. 6, configuring a risk policy of a store wind control system by using the target rule set, so that the store wind control system identifies a risk label corresponding to a store image of an arbitrary online store according to the risk policy, includes:

step S1410, determining a strategy feature table according to the data features in the target rule set;

assuming that N standard rule sets are finally screened out to meet the online standard after the rule sets generated by the decision tree model are screened, offline features or real-time features can be further developed for the features used in the N rule sets, the original target rule sets are enriched, and a strategy feature table is correspondingly generated to be used for strategy online configuration so as to generate corresponding risk strategies in the store wind control system.

Step S1420, configuring the strategy feature table into a risk strategy in the store wind control system;

and further, docking the strategy feature table developed in the previous step into a risk strategy library of the store wind control system, then segmenting nodes according to all features of the target rule set, and configuring a risk strategy, so that if the risk strategy is hit by the store on a wire, an alarm record can be submitted in a wind control center.

And S1430, controlling the store wind control system to start the risk strategy, scanning each online store in the e-commerce platform, and identifying a risk label corresponding to the store image of each online store according to the risk strategy.

After the configuration of the risk strategies is completed, the store wind control system can be controlled to start a new risk strategy, so that the store wind control system can identify operation credit of any online store according to each risk strategy in the just-upgraded wind control strategy library, identify whether the online store belongs to a risk store according to store pictures of the online stores, and determine a risk label corresponding to the online store.

In one embodiment, considering that the risk strategy of the store wind control system is updated, the store wind control system can be controlled to re-identify the risk labels of all store portraits of online stores in the e-commerce platform one by one so as to update the identification result of the business credit of each online store in time.

According to the embodiment, when the risk strategy of the store wind control system is configured based on the target rule set, the characteristics can be further enriched according to the target rule set, the risk labels corresponding to all online stores in the e-commerce platform can be timely re-identified according to the upgraded risk strategy, the investigation of the risk stores of the whole platform is realized, the established risk strategy has stronger identification capacity, the identification capacity of the store wind control system to the risk stores is comprehensively improved, and the operating order of the e-commerce platform is effectively maintained.

Referring to fig. 7, a store risk identification apparatus according to an aspect of the present application includes a sample acquiring module 1100, a rule analyzing module 1200, a rule optimizing module 1300, and a wind control configuration module 1400, wherein the sample acquiring module 1100 is configured to acquire a sample data set including a plurality of store samples, each store sample including a store portrait of a single online store and a risk tag thereof; the rule analysis module 1200 is configured to classify all store samples in the sample data set into a plurality of sample subsets by using a decision tree model trained from the sample data set, and obtain a rule set corresponding to each sample subset; the rule optimization module 1300 is configured to select multiple rule sets as target rule sets according to the statistical indexes of the sample subsets corresponding to the rule sets; the wind control configuration module 1400 is configured to configure a risk policy of the store wind control system by using the target rule set, so that the store wind control system identifies a risk label corresponding to a store image of any online store according to the risk policy.

On the basis of any embodiment of the present application, the shop risk recognition device of the present application further includes: the sample construction module is used for constructing the shop images of the online shops identified by the shop wind control system and the corresponding risk labels thereof into new shop samples and adding the new shop samples into the sample data set; the model updating module is set to respond to a timing arrival event triggered by a timing task and adopt the sample data set to retrain the decision tree model to a convergence state; and the strategy updating module is set to adopt a decision tree model which is retrained to be in a convergence state to re-determine a target rule set corresponding to the sample data set and update the risk strategy of the store wind control system.

On the basis of any embodiment of the present application, the store risk recognition apparatus of the present application further includes: the identification execution module is set to judge whether the risk label of the online shop identified by the shop wind control system belongs to a target type or not, and when the risk label belongs to the online shop of the target type, the online shop is marked as a risk shop; the alarm execution module is set to generate an alarm message corresponding to the risky shop and send the alarm message to a specific audit interface; and the wind control execution module is used for acquiring a confirmation instruction returned by the audit interface and responding to the confirmation instruction to modify the corresponding operating activity authority of the online shop.

On the basis of any embodiment of the present application, the sample acquiring module 1100 includes: a target determining unit, which is arranged to determine a target line-on store for extracting a store sample in the e-commerce platform; the evaluation identification unit is set to correspondingly set the risk label of the shop on the corresponding target line as a high risk type or a low risk type according to whether the evaluation index of the shop on each target line is higher than a preset threshold value; a feature portrayal unit configured to acquire a plurality of shop-level data features of each of the on-target-line shops as shop portrayals of the corresponding on-target-line shops; and the sample construction unit is arranged to construct the shop images of all the target online shops and the corresponding risk labels into corresponding shop samples to be stored in the sample data set.

On the basis of any embodiment of the application, the multiple store-level data characteristics comprise any multiple of basic characteristics of the store, transaction characteristics of the store, violation characteristics of the store, behavior characteristics of the store and seller-related characteristics related to the store.

On the basis of any embodiment of the present application, the rule preference module 1300 includes: the index counting unit is set to count the accuracy and the recall rate of the risk label of the corresponding target type in each sample subset according to the sample subsets corresponding to the rule sets and take the accuracy and the recall rate as statistical indexes; and the rule screening unit is set to screen out a plurality of sample subsets with accuracy higher than a first preset threshold and recall rate higher than a second preset threshold, and takes the corresponding rule sets as target rule sets.

On the basis of any embodiment of the present application, the wind control configuration module 1400 includes: the conversion processing unit is used for determining a strategy characteristic table according to the data characteristics in the target rule set; the strategy configuration unit is used for configuring the strategy feature table into a risk strategy in the store wind control system; and the wind control restarting unit is used for controlling the store wind control system to start the risk strategy, scanning each online store in the e-commerce platform, and identifying the risk label corresponding to the store image of each online store according to the risk strategy.

Another embodiment of the present application also provides a store risk identification apparatus. As shown in fig. 8, the interior of the store risk identification device is schematically illustrated. The store risk identification device includes a processor, a computer-readable storage medium, a memory, and a network interface connected by a system bus. The computer-readable non-transitory readable storage medium of the shop risk identification device stores an operating system, a database and computer-readable instructions, wherein the database can store information sequences, and the computer-readable instructions, when executed by a processor, can enable the processor to realize a shop risk identification method.

The processor of the store risk identification device is used for providing calculation and control capability and supporting the operation of the whole store risk identification device. The memory of the store risk identification device may have stored therein computer readable instructions that, when executed by the processor, may cause the processor to perform the store risk identification method of the present application. The network interface of the shop risk identification equipment is used for connecting and communicating with the terminal.

Those skilled in the art will appreciate that the configuration shown in fig. 8 is a block diagram of only a portion of the configuration associated with the subject application and does not constitute a limitation on the store risk identification devices to which the subject application is applied, and that a particular store risk identification device may include more or fewer components than shown, or some components may be combined, or have a different arrangement of components.

In this embodiment, the processor is configured to execute specific functions of each module in fig. 7, and the memory stores program codes and various data required for executing the modules or sub-modules. The network interface is used for realizing data transmission between user terminals or servers. The nonvolatile readable storage medium in the present embodiment stores program codes and data necessary for executing all modules in the store risk recognition device of the present application, and the server can call the program codes and data of the server to execute the functions of all modules.

The present application also provides a non-transitory 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 store risk identification method of any of the embodiments of the present application.

The present application also provides a computer program product comprising computer programs/instructions which, when executed by one or more processors, implement the steps of the method as described in any of the embodiments of the present application.

It will be understood by those skilled in the art that all or part of the processes of the methods of the embodiments of the present application may be implemented by hardware related to instructions of a computer program, which may be stored in a non-volatile readable storage medium, and when executed, may include the processes of the embodiments of the methods as described above. The storage medium may be a magnetic disk, an optical disk, a Read-only Memory (ROM), or a Random Access Memory (RAM).

In summary, the online shop risk identification method and system are based on the decision tree model, the shop portrait is used for mining the risk strategy for identifying whether the online shop is the risk shop, accuracy of the strategy can be guaranteed, the online shop risk identification method and system can have strong interpretability, and meanwhile the online speed can be high, so that the online shop risk identification method and system can adapt to high-speed changes of business risks of e-commerce better.

Claims

1. An identification method for an inauguration store, characterized by comprising:

and configuring a risk strategy of the store wind control system by using the target rule set, so that the store wind control system identifies a risk label corresponding to the store image of any online store according to the risk strategy.

2. The store risk identification method according to claim 1, wherein the step of configuring a risk policy of a store pneumatic control system by using the target rule set, and after the store pneumatic control system identifies a risk label corresponding to a store image of an arbitrary online store according to the risk policy, comprises:

3. The store risk identification method according to claim 1, wherein a risk policy of a store wind control system is configured by using the target rule set, and after the store wind control system identifies a risk label corresponding to a store image of an arbitrary online store according to the risk policy, the method comprises:

4. The store risk identification method according to any one of claims 1 to 3, wherein acquiring a sample data set comprises:

5. The store risk identification method according to claim 4, wherein in the step of obtaining a plurality of store-level data characteristics of each of the on-target line stores as a store representation of the corresponding on-target line store, the plurality of store-level data characteristics comprise any of a base characteristic of the store, a transaction characteristic of the store, an infraction characteristic of the store, a behavior characteristic of the store, and a store-associated seller-related characteristic.

6. The store risk identification method according to any one of claims 1 to 3, wherein the screening of the plurality of rule sets as the target rule set based on the statistical indicator of the sample subset corresponding to each rule set includes:

7. The store risk identification method according to any one of claims 1 to 3, wherein a risk policy of a store pneumatic control system is configured by using the target rule set, and the store pneumatic control system identifies a risk label corresponding to a store image of an arbitrary online store according to the risk policy, and the method comprises the steps of:

and controlling a store wind control system to start the risk strategy, scanning each online store in the e-commerce platform, and identifying a risk label corresponding to the store image of each online store according to the risk strategy.

8. A store risk identification device, comprising:

9. Store risk identification device comprising a central processor and a memory, characterized in that the central processor is configured to invoke the execution of a computer program stored in the memory to perform the steps of the method according to any one of claims 1 to 7.

10. A non-transitory readable storage medium storing a computer program implemented according to the method of any one of claims 1 to 7 in the form of computer readable instructions, the computer program, when invoked by a computer, performing the steps included in the corresponding method.