Risk assessment model training method, risk assessment method and related equipment

Abstract

The invention provides a risk assessment model training method, a risk assessment method and related equipment. The method comprises the steps of determining risk feature vectors of a sample article supplier under multiple risk features according to logistics side business data and platform side business data of the sample article supplier; the risk features are used for evaluating the reliability of the article supplier and the articles thereof; processing the risk feature vector under each risk feature through a risk assessment model to obtain a predicted risk tag of the sample article supplier; obtaining a reliability checking result of the sample article supplier and further determining an actual risk tag of the sample article supplier; and training a risk assessment model according to the predicted risk tag and the actual risk tag to obtain a trained risk assessment model. According to the method, the risk feature vector of the text comparison class can be extracted in combination with the logistics side business data and the platform side business data of the article supplier so as to train the risk assessment model, and whether the article supplier and the articles thereof have risks or not is assessed.

Description

TECHNICAL FIELD

[0001] The present disclosure relates to the technical field of computer, and particularly relates to a risk assessment model training method, a risk assessment method and related equipment. BACKGROUND

[0002] With the development of network technology and computer technology, the application of online mall platforms based on the Internet is more and more widely used. Merchants in the online mall platform can input platform virtual goods on the platform for sale.

[0003] In the related art, the risk of platform virtual goods and online service behavior of the merchant is usually assessed, and whether there is a physical goods matching the platform virtual goods is not checked, so it is difficult to ensure the consistency of online goods selling service and offline goods transportation, thereby destroying the normal operation of goods supply business.

[0004] It should be noted that the information disclosed in the above background section is only used to strengthen the understanding of the background of the present disclosure, and therefore can include information that does not constitute prior art known to those of ordinary skill in the art. SUMMARY

[0005] The purpose of the present disclosure is to provide a risk assessment model training method, a risk assessment method and related equipment, which can extract a text comparison type risk feature vector by combining logistics side business data and platform side business data of a goods supplier to train a risk assessment model, and use the model to assess whether the goods supplier and the goods provided by the goods supplier are at risk.

[0006] Other characteristics and advantages of the present disclosure will become apparent from the following detailed description, or will be learned by practice of the present disclosure.

[0007] According to one aspect of the present disclosure, a risk assessment model training method is provided, comprising: obtaining logistics side business data and platform side business data of a plurality of sample goods suppliers, determining risk feature vectors of each sample goods supplier under a plurality of risk features according to the logistics side business data and the platform side business data; wherein the risk feature is used to assess the reliability of the goods supplier and the goods provided by the goods supplier; processing the risk feature vectors of the sample goods suppliers under each risk feature by a risk assessment model to obtain a predicted risk label of the sample goods supplier; obtaining reliability verification results of the plurality of sample goods suppliers, determining actual risk labels of the sample goods suppliers according to the reliability verification results; training the risk assessment model according to the predicted risk labels and the actual risk labels of the sample goods suppliers to obtain a trained risk assessment model.

[0008] In an embodiment of the present disclosure, the risk feature includes item description text similarity; wherein the risk feature vector of each sample item supplier under multiple risk features is determined according to the logistics-side business data and the platform-side business data, including: extracting the description text of the inventory items stored in the warehouse of the sample item supplier from the logistics-side business data, and extracting the description text of the platform virtual items provided by the sample item supplier on the platform from the platform-side business data; for a current inventory item in the inventory items, performing similarity calculation processing on the description text of the current inventory item and the description text of each platform virtual item to obtain the text similarity between the current inventory item and each platform virtual item; determining the maximum value in the text similarity as the target similarity of the current inventory item compared with the platform side; and determining the risk feature vector of the sample item supplier under the item description text similarity according to the target similarity of all inventory items of the sample item supplier compared with the platform side.

[0009] In an embodiment of the present disclosure, the item description text similarity includes at least one of the following: item name similarity, item primary category similarity, item secondary category similarity, and item tertiary category similarity.

[0010] In an embodiment of the present disclosure, the platform virtual items include a current virtual item; wherein, for a current inventory item in the inventory items, performing similarity calculation processing on the description text of the current inventory item and the description text of each platform virtual item to obtain the text similarity between the current inventory item and each platform virtual item, includes: performing word segmentation processing on the description text of the current inventory item to obtain a first word set; performing word segmentation processing on the description text of the current virtual item to obtain a second word set; merging the first word set and the second word set to obtain a merged word set; comparing the words in the first word set with the words in the merged word set to obtain a first word vector composed of 0 elements and 1 elements; wherein, the word at the position corresponding to the 0 element exists in the merged word set but does not exist in the first word set, and the word at the position corresponding to the 1 element exists in the merged word set and exists in the first word set; comparing the words in the second word set with the words in the merged word set to obtain a second word vector composed of 0 elements and 1 elements; wherein, the word at the position corresponding to the 0 element exists in the merged word set but does not exist in the second word set, and the word at the position corresponding to the 1 element exists in the merged word set and exists in the second word set; calculating the cosine similarity of the first word vector and the second word vector, and taking the cosine similarity as the text similarity between the current inventory item and the current virtual item.

[0011] In an embodiment of the present disclosure, the risk features include attribute change risk features and text description change risk features; wherein the risk feature vectors of each sample commodity supplier under multiple risk features are determined according to the logistics-side business data and the platform-side business data, including: extracting the attribute change times and the text description change times of each inventory commodity stored in the warehouse of the sample commodity supplier from the logistics-side business data; determining the risk feature vector of the sample commodity supplier under the attribute change risk features according to the attribute change times of all inventory commodities of the sample commodity supplier; determining the risk feature vector of the sample commodity supplier under the text description change risk features according to the text description change times of all inventory commodities of the sample commodity supplier.

[0012] In an embodiment of the present disclosure, the weight parameters of each risk feature are configured in the risk assessment model; wherein the predicted risk label of the sample commodity supplier is obtained by processing the risk feature vectors of the sample commodity supplier under each risk feature through the risk assessment model, including: obtaining the predicted sub-risk label under the corresponding risk feature by processing the risk feature vector of the sample commodity supplier under the corresponding risk feature through the risk assessment model; obtaining the predicted risk assessment value of the sample commodity supplier by weighting the predicted sub-risk labels under multiple risk features through the weight parameters in the risk assessment model; determining the predicted risk label of the sample commodity supplier according to the predicted sub-risk label under each risk feature and the predicted risk assessment value.

[0013] In an embodiment of the present disclosure, the reliability verification result of the sample commodity supplier includes the sub-risk verification label and the risk verification value of the sample commodity supplier under each risk feature; the risk assessment model is trained according to the predicted risk label and the actual risk label of the sample commodity supplier, including: constructing multiple first loss functions according to the predicted sub-risk label and the sub-risk verification label under each risk feature; constructing a second loss function according to the predicted risk assessment value and the risk verification value; training the risk assessment model by using the multiple first loss functions and the second loss function.

[0014] In an embodiment of the present disclosure, before the risk feature vectors of the sample commodity supplier under each risk feature are processed through the risk assessment model, the training method of the risk assessment model further includes: obtaining a set of abnormal commodity suppliers, which contains multiple historical abnormal commodity suppliers providing commodities with low reliability; obtaining common risk features and initial weights of each common risk feature through common analysis of the logistics-side business data and the platform-side business data of the historical abnormal commodity suppliers; obtaining the initial weight of the text comparison type risk feature; determining the multiple risk features as the common risk features and the text comparison type risk features, and configuring the risk assessment model based on the multiple risk features and their initial weights.

[0015] In one embodiment of the present disclosure, after obtaining the reliability verification results of the plurality of sample commodity providers, the training method of the risk assessment model further comprises: determining a sample commodity provider whose reliability verification result satisfies the risk condition as a target abnormal commodity provider; updating the set of abnormal commodity providers according to the target abnormal commodity provider; re-performing commonality analysis according to the updated set of abnormal commodity providers to obtain updated commonality risk features and initial weights of each updated commonality risk feature; and updating the risk assessment model based on the updated commonality risk features and the initial weights thereof.

[0016] According to another aspect of the present disclosure, a risk assessment method is provided, comprising: obtaining to-be-evaluated logistics-side business data and to-be-evaluated platform-side business data of a to-be-evaluated commodity provider, determining a risk feature vector of the to-be-evaluated commodity provider under a plurality of risk features according to the to-be-evaluated logistics-side business data and the to-be-evaluated platform-side business data; wherein the risk features are used to evaluate the reliability of the commodity provider and the commodity provided by the commodity provider; processing the risk feature vector of the to-be-evaluated commodity provider under each risk feature by a trained risk assessment model to obtain a target risk label of the to-be-evaluated commodity provider; wherein the trained risk assessment model is trained by the training method of the risk assessment model as described above; obtaining a risk condition, and comparing the target risk label with the risk condition to determine a risk result of the to-be-evaluated commodity provider.

[0017] According to still another aspect of the present disclosure, a training device of a risk assessment model is provided, comprising: a first feature determination module configured to obtain logistics-side business data and platform-side business data of a plurality of sample commodity providers, and determine a risk feature vector of each sample commodity provider under a plurality of risk features according to the logistics-side business data and the platform-side business data; wherein the risk features are used to evaluate the reliability of the commodity provider and the commodity provided by the commodity provider; a first processing module configured to process the risk feature vector of the sample commodity provider under each risk feature by a risk assessment model to obtain a predicted risk label of the sample commodity provider; a first obtaining module configured to obtain reliability verification results of the plurality of sample commodity providers, and determine an actual risk label of the sample commodity provider according to the reliability verification results; and a training module configured to train the risk assessment model according to the predicted risk label and the actual risk label of the sample commodity provider to obtain a trained risk assessment model.

[0018] According to still another aspect of the present disclosure, there is provided a risk assessment device, comprising: a second feature determination module configured to obtain to-be-assessed logistics-side business data and to-be-assessed platform-side business data of a to-be-assessed commodity provider, and determine a risk feature vector of the to-be-assessed commodity provider under a plurality of risk features according to the to-be-assessed logistics-side business data and the to-be-assessed platform-side business data; wherein the risk features are used to assess the reliability of the commodity provider and the commodities provided by the commodity provider; a second processing module configured to process the risk feature vector of the to-be-assessed commodity provider under each risk feature by using a trained risk assessment model to obtain a target risk label of the to-be-assessed commodity provider; wherein the trained risk assessment model is obtained by using the training method of the risk assessment model as described above; and an assessment module configured to obtain a risk condition, compare the target risk label with the risk condition, and determine a risk result of the to-be-assessed commodity provider.

[0019] According to still another aspect of the present disclosure, there is provided a computer readable storage medium having stored thereon a computer program, which, when executed by a processor, implements the training method of the risk assessment model or the risk assessment method as described above.

[0020] According to still another aspect of the present disclosure, there is provided an electronic device, comprising: a processor; and a memory configured to store executable instructions of the processor; wherein the processor is configured to execute the training method of the risk assessment model or the risk assessment method as described above by executing the executable instructions.

[0021] The training method of the risk assessment model provided by the embodiments of the present disclosure can extract a risk feature vector under a risk feature used to assess the reliability of a commodity provider and the commodities provided by the commodity provider from logistics-side business data and platform-side business data of a sample commodity provider, then process the risk feature vector of the sample commodity provider under each risk feature by using a risk assessment model to obtain a predicted risk label of the sample commodity provider, determine an actual risk label of the sample commodity provider according to an actual reliability verification result of the sample commodity provider, and train the risk assessment model based on the predicted risk label and the actual risk label. It can be seen that the present solution can obtain a risk feature vector related to the reliability of a commodity provider and the commodities provided by the commodity provider from logistics-side business data and platform-side business data of the commodity provider, process such a risk feature vector by using a risk assessment model to obtain a predicted risk label, and train the risk assessment model by using the predicted risk label and an actual reliability verification result of the commodity provider, so that the trained risk assessment model learns how to identify whether the commodity provider and the commodities provided by the commodity provider are at risk, so as to enable a related system to further analyze or check the commodity provider at risk, and avoid unnecessary losses caused by the commodity provider providing commodities that are not authentic.

[0022] It should be understood that the general description and detailed description, which follow below, are merely exemplary and explanatory, and are not restrictive of the present disclosure. BRIEF DESCRIPTION OF DRAWINGS

[0023] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present disclosure and serve to explain the principles of the present disclosure. It is readily apparent to one skilled in the art that the following description of the drawings is merely exemplary and explanatory of some embodiments of the present disclosure, and other drawings can be obtained from these drawings without creative labor.

[0024] Figure 1 A schematic diagram showing an exemplary system architecture of a training method of a risk assessment model to which embodiments of the present disclosure can be applied is shown;

[0025] Figure 2 A flowchart showing a training method of a risk assessment model of one embodiment of the present disclosure is shown;

[0026] Figure 3 A flowchart showing determination of a risk feature vector in a training method of a risk assessment model of one embodiment of the present disclosure is shown;

[0027] Figure 4 A flowchart showing determination of a text similarity between an inventory item and a platform virtual item in a training method of a risk assessment model of one embodiment of the present disclosure is shown;

[0028] Figure 5 A flowchart showing configuration of a risk assessment model in a training method of a risk assessment model of one embodiment of the present disclosure is shown;

[0029] Figure 6 A flowchart showing a risk assessment method of one embodiment of the present disclosure is shown;

[0030] Figure 7 A schematic diagram showing a training method and an application method of a risk assessment model of one embodiment of the present disclosure is shown;

[0031] Figure 8 A block diagram showing a training device 800 of a risk assessment model of one embodiment of the present disclosure is shown;

[0032] Figure 9 A block diagram showing a risk assessment device of one embodiment of the present disclosure is shown; and

[0033] Figure 10 A structural block diagram of a computer device suitable for use in implementing exemplary embodiments of the present disclosure is shown in an embodiment of the present disclosure. DETAILED DESCRIPTION

[0034] Example implementations are now described with reference to the following drawings. The example implementations, can, however, be implemented in many different forms and should not be construed as limited to the examples set forth herein; rather, these implementations are provided so that this disclosure will be thorough and complete, and will fully convey the inventive aspects to those skilled in the art. The described features, structures, or characteristics can be combined in one or more implementations.

[0035] In addition, the accompanying drawings are included to provide a further understanding of the present disclosure and are incorporated in and constitute a part of this specification. The drawings illustrate examples of the present disclosure and, as such, a change in the size or proportion of some of the elements in the drawings can be exaggerated for clarity.

[0036] In addition, the terms "first", "second", and the like, are used only to describe the features and do not imply or suggest relative importance or a number of the indicated technical features. Thus, the features defined with "first", "second" can explicitly or implicitly include one or more of the features. In the description of the present disclosure, the meaning of "a plurality of" is at least two, for example, two, three, etc., unless otherwise explicitly specified.

[0037] Figure 1 A schematic diagram of an exemplary system architecture illustrating a training method of a risk assessment model to which embodiments of the present disclosure can be applied is shown.

[0038] As Figure 1 shown, the system architecture can include a server 101, a network 102, and a client 103. The network 102 is a medium to provide a communication link between the client 103 and the server 101. The network 102 can include various connection types, such as wired, wireless communication links, or fiber optic cables, etc.

[0039] In an exemplary embodiment, the client 103, which performs data transmission with the server 101, can include, but is not limited to, electronic devices such as a smartphone, a desktop computer, a tablet, a notebook computer, a smart speaker, a digital assistant, an AR (Augmented Reality) device, a VR (Virtual Reality) device, a smart wearable device, etc. Optionally, an operating system running on the electronic device can include, but is not limited to, an Android system, an IOS system, a Linux system, a Windows system, etc.

[0040] The server 101 can be a standalone physical server, a server cluster composed of multiple physical servers, or a distributed system, and can also be a cloud server providing cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN (Content Delivery Network), and basic cloud computing services such as big data and artificial intelligence platforms. In some actual applications, the server 101 can also be a server of a network platform, for example, a transaction platform, a live broadcast platform, a social platform, or a music platform, and the like, and the embodiments of the present disclosure do not limit the same. The server can be a server or a cluster formed by multiple servers, and the specific architecture of the server is not limited in the present disclosure.

[0041] In the example embodiments, the server 101 or the client 103 can obtain the logistics-side business data and the platform-side business data of the plurality of sample commodity suppliers in response to a training instruction of the risk assessment model, and then train the risk assessment model by using the data through the training method of the risk assessment model provided by the present disclosure to obtain the trained risk assessment model. In some actual applications, the logistics-side business data and the platform-side business data of the plurality of sample commodity suppliers can be obtained from one or more servers 101 or clients 103 for model training.

[0042] In the example embodiments, the server 101 can implement the process of the training method of the risk assessment model as follows: the server 101 obtains the logistics-side business data and the platform-side business data of the plurality of sample commodity suppliers, and determines the risk feature vectors of each sample commodity supplier under a plurality of risk features according to the logistics-side business data and the platform-side business data; the risk features are used to evaluate the reliability of the commodity suppliers and the commodities provided by the commodity suppliers; the server 101 processes the risk feature vectors of the sample commodity suppliers under each risk feature by using the risk assessment model to obtain the predicted risk labels of the sample commodity suppliers; the server 101 obtains the reliability verification results of the plurality of sample commodity suppliers, and determines the actual risk labels of the sample commodity suppliers according to the reliability verification results; and the server 101 trains the risk assessment model according to the predicted risk labels and the actual risk labels of the sample commodity suppliers to obtain the trained risk assessment model.

[0043] In addition, it should be noted that, Figure 1 The above is only one application environment of the training method of the risk assessment model provided by the present disclosure. Figure 1 The number of servers 101, networks 102, and clients 103 in the above is only illustrative, and any number of clients, networks, and servers can be provided according to actual needs.

[0044] In order for those skilled in the art to better understand the technical solutions of the present disclosure, the training method of the risk assessment model in the example embodiments of the present disclosure will be described in more detail below in combination with the accompanying drawings and examples.

[0045] Figure 2 A flowchart of the training method of the risk assessment model of one embodiment of the present disclosure is shown. The method provided by the present embodiment can be executed by the server 101 or the client 103 as shown in Figure 1 but the present disclosure is not limited thereto.

[0046] In the following example, the server 101 is taken as the execution subject for example.

[0047] As shown in Figure 2 The training method of the risk assessment model provided by the present embodiment can include the following steps.

[0048] In step S201, the logistics-side business data and the platform-side business data of a plurality of sample item suppliers are obtained, and the risk feature vectors of each sample item supplier under a plurality of risk features are determined according to the logistics-side business data and the platform-side business data; wherein the risk features are used to evaluate the reliability of the item suppliers and the items provided by them.

[0049] In this step, the item supplier can be a merchant who sells goods through an online transaction platform (such as an online mall type application) and stores the goods sold by him in the warehouse, and the sample item supplier can be considered as a sample merchant for training the risk assessment model; the item provided by the item supplier can be at least one of the goods sold by the merchant and the items existing in the warehouse of the merchant.

[0050] In the following example, the item supplier is taken as a merchant for example.

[0051] The logistics-side business data may, for example, be the warehouse records, inventory information, transportation data, and quantity records of the inventory items entered by the merchant in the warehouse for sale, and the platform-side business data may, for example, be the item information, item attributes, and transaction data of the virtual items to be sold entered by the merchant in the platform. In some practical applications, the same identification code (such as an EMG code) is used when the merchant enters the inventory item information in the warehouse and the platform virtual item information in the mall, so the same identification code can be used to obtain the logistics-side business data and the platform-side business data of the same merchant.

[0052] The actual result of the risk feature (i.e., the risk feature vector) can be used to evaluate whether the merchant is consistent between the platform virtual goods provided by the platform and the inventory goods stored in the warehouse for sale, or to evaluate whether the merchant is reliable in the platform virtual goods provided by the platform or the inventory goods stored in the warehouse, or to evaluate whether the merchant itself is reliable.

[0053] Specifically, the risk feature can be a feature for judging whether the names, categories or volume weights of the goods on the logistics side and the platform side are consistent, for example, if the name of a certain good on the logistics side and the name of the good on the platform side are not similar, it can be considered that the good of the merchant is unreliable; if the platform virtual goods sold on the platform side are not similar to all the names of the inventory goods stored in the warehouse, it can be considered that the good of the merchant is unreliable; if there are a large number of inconsistent goods, it can be considered that the merchant is unreliable.

[0054] The risk feature can also be a feature for judging whether the merchant modifies the name of the good too frequently, a feature for judging whether the number of goods in the storage area is abnormal, or a feature for judging whether the merchant itself is abnormal, for example, if the merchant modifies the name or attributes of a certain good too many times (such as 10 times a month), it can be considered that the good of the merchant is unreliable; if the merchant modifies the name or attributes (such as the type of business) of the merchant on the platform side too many times (such as 10 times a month), it can be considered that the merchant is unreliable.

[0055] In some practical applications, the risk feature can include a text comparison type risk feature or a data change frequency type risk feature. For example, the specified type of text in the logistics side business data and the platform side business data can be compared and processed, and the risk feature vector of the text comparison type risk feature is determined according to the comparison and processing result; the change frequency of certain type of data of the merchant can also be counted, and the risk feature vector of the change frequency type risk feature is determined according to the change frequency. When the risk feature is used to evaluate the reliability of the goods provided by the merchant, the risk feature vector thereof can be processed by a risk evaluation model to determine whether the goods provided by the merchant are reliable or unreliable, the degree of reliability or the degree of risk.

[0056] In some practical applications, the risk feature vector can be composed of multiple quantified data of multiple goods provided by the merchant under the corresponding risk feature.

[0057] The "determining, according to the logistics-side business data and the platform-side business data, a risk feature vector of each sample commodity supplier under a risk feature of a text comparison type" in step S201 can include: extracting, from the logistics-side business data of the sample merchant, a logistics-side description text corresponding to the risk feature of the text comparison type, and extracting, from the platform-side business data of the sample merchant, a platform-side description text corresponding to the risk feature of the text comparison type; and performing similarity calculation processing on the logistics-side description text and the platform-side description text to determine the risk feature vector of the sample merchant under the risk feature of the text comparison type.

[0058] In step S203, the risk feature vector of the sample commodity supplier under each risk feature is processed by the risk assessment model to obtain a predicted risk label of the sample commodity supplier.

[0059] In some embodiments, the risk assessment model is configured with weight parameters of each risk feature.

[0060] Based on this, step S203 can include: processing the risk feature vector of the sample commodity supplier under the corresponding risk feature by the risk assessment model to obtain a predicted sub-risk label under the corresponding risk feature; weighting the predicted sub-risk labels under the multiple risk features by the weight parameters in the risk assessment model to obtain a predicted risk assessment value of the sample commodity supplier; and determining the predicted risk label of the sample commodity supplier according to the predicted sub-risk labels under each risk feature and the predicted risk assessment value.

[0061] In some actual applications, the risk assessment model can further include multiple sub-assessment networks corresponding to the multiple risk features, and a risk weighting network; the risk weighting network is configured with the weight parameters of the above-mentioned risk features.

[0062] Based on this, step S203 can include: processing the risk feature vector of the sample commodity supplier under the corresponding risk feature by each sub-assessment network to obtain a predicted sub-risk label under the corresponding risk feature; weighting the predicted sub-risk labels under the multiple risk features by the weight parameters in the risk weighting network to obtain a predicted risk assessment value of the sample commodity supplier; and determining the predicted risk label of the sample commodity supplier according to the predicted sub-risk labels under each risk feature and the predicted risk assessment value.

[0063] The weight parameters can be positively correlated with the importance or frequency of occurrence of each risk feature, for example, the weight parameter of a risk feature that appears more commonly in merchants can be higher. In some actual applications, in the training of the risk assessment model, the weight parameters of each risk feature can be trained and adjusted so that the adjusted weight parameters obtain more accurate prediction results after weighting processing of the predicted sub-risk labels.

[0064] In the present application, the predicted sub-risk labels of the sample merchant under each wind direction feature can be determined by the risk assessment model. The predicted sub-risk label can be 1 or 0, indicating whether the sample merchant has risk under the wind direction feature. Specifically, the output predicted sub-risk label is 1, which means that the sample merchant is predicted to have risk under the wind direction feature. The output predicted sub-risk label is 0, which means that the sample merchant is predicted to have no risk under the wind direction feature.

[0065] In some practical applications, if the risk feature vector is used to represent the plurality of quantified data of the plurality of items provided by the merchant under the corresponding risk feature, the predicted sub-risk label output by the risk assessment model for the risk feature can include sub-risk prediction information for the plurality of items; and the predicted risk assessment value of the sample merchant output by the risk assessment model after weighting calculation can include risk assessment values for the plurality of items in the sample merchant.

[0066] For example, assuming that there are two risk features P and Q, and a sample merchant has five items, the sample merchant can have two risk feature vectors under the two risk features: the risk feature vector (a, b, c, d, e) of the risk feature P and the risk feature vector (f, g, h, i, j) of the risk feature Q; for the two risk features, the predicted sub-risk label of the risk feature P (0, 0, 0, 1, 0) can be output, and the predicted sub-risk label of the risk feature Q (1, 0, 0, 1, 0) can be output; this means that under the risk feature P, the fourth item is predicted to have risk, and the remaining items are predicted to have no risk, and under the risk feature Q, the first and fourth items are predicted to have risk, and the remaining items are predicted to have no risk. Further, the two predicted sub-risk labels can be weighted calculated with the weight parameters of the risk features P and Q. Assuming that the weight parameters of the risk features P and Q are 4 and 3 respectively, for the five items of the sample merchant, the predicted risk assessment value can be calculated as (0+3, 0+0, 0+0, 4+3, 0+0), i.e. (3, 0, 0, 7, 0). That is, the predicted sub-risk label output by the risk assessment model can include prediction information about whether the five items of the sample merchant have risk, and the predicted risk assessment value output by the risk assessment model can include risk assessment values for the five items of the sample merchant.

[0067] In the present application, the predicted sub-risk labels of the sample merchant under each wind direction feature can be determined by the risk assessment model. The predicted sub-risk label can be 1 or 0, indicating whether the sample merchant has risk under the wind direction feature. Specifically, the output predicted sub-risk label is 1, which means that the sample merchant is predicted to have risk under the wind direction feature. The output predicted sub-risk label is 0, which means that the sample merchant is predicted to have no risk under the wind direction feature.

[0068] In the process of processing each risk feature vector by the risk assessment model to obtain a predicted sub-risk label in this embodiment, the risk feature vector can be directly processed to obtain the predicted sub-risk label, or a risk threshold parameter for each risk feature can be maintained, which is used for comparison with an element in the risk feature vector to obtain a predicted sub-risk label according to a comparison result; wherein in the training of the risk assessment model, the risk threshold parameters of various risk features can be trained and adjusted, so that the adjusted risk threshold parameters obtain more accurate (closer to the actual risk label) predicted sub-risk label results after comparison with the elements in the risk feature vector. In some actual applications, the above risk threshold parameters can be configured in the sub-evaluation network.

[0069] Through step S203, the risk feature vector obtained from the logistics-side business data and the platform-side business data of the sample merchant can be input into the risk assessment model, and a multi-aspect predicted result is output as a predicted risk label for the training of the risk assessment model, so that the finally trained risk assessment model can be more in line with the actual risk label in multiple aspects.

[0070] In step S205, the reliability verification results of a plurality of sample commodity suppliers are obtained, and the actual risk label of the sample commodity supplier is determined according to the reliability verification results.

[0071] The reliability verification result of the sample merchant can be a result obtained after on-site verification of the sample merchant itself or the goods stored in the warehouse of the sample merchant.

[0072] In some embodiments, the reliability verification result of the sample commodity supplier can include a sub-risk verification label and a risk verification value of the sample commodity supplier under each risk feature. The risk verification value can be determined according to a pre-set merchant reliability evaluation algorithm. The sub-risk verification label can be a value indicating whether the merchant or the plurality of goods in the merchant is reliable under the corresponding risk feature, which can be identified by 0 or 1 (for example, 0 represents reliable and no risk, and 1 represents unreliable and risk). For example, if the verification finds that a certain inventory good of the sample merchant recorded for sale is indeed sold as a platform virtual good on the platform, the sub-risk verification label of the inventory good under the text comparison type risk feature can be determined as 0, i.e. no risk; if the verification finds that a certain inventory good of the sample merchant is not sold on the platform, the sub-risk verification label of the inventory good under the text comparison type risk feature can be determined as 1, i.e. risk.

[0073] In step S207, the risk assessment model is trained according to the predicted risk label and the actual risk label of the sample commodity supplier, and a trained risk assessment model is obtained.

[0074] In some embodiments, the step of "training the risk assessment model according to the predicted risk label and the actual risk label of the sample supplier" in step S207 can include: constructing a plurality of first loss functions according to the predicted sub-risk label and the sub-risk check label under each risk feature; constructing a second loss function according to the predicted risk assessment value and the risk check value; and training the risk assessment model by using the plurality of first loss functions and the second loss function.

[0075] The first loss function and / or the second loss function can be any one of a 0-1 loss function, an absolute value loss function, a logarithmic loss function, a square loss function, an exponential loss function, a perception loss function, a cross-entropy loss function, a mean square error loss function, etc., and the embodiments of the present disclosure are not limited thereto.

[0076] By the training method of the risk assessment model provided by the present disclosure, the risk feature vectors under the risk features related to the reliability of the suppliers and the goods provided by the suppliers can be extracted from the logistics-side business data and the platform-side business data of the sample suppliers, and then the risk feature vectors under each risk feature of the sample suppliers are processed by the risk assessment model to obtain the predicted risk label of the sample suppliers. The actual risk label of the sample suppliers is determined according to the actual reliability check result of the sample suppliers, and the risk assessment model is trained based on the predicted risk label and the actual risk label. It can be seen that the present solution can obtain the risk feature vectors related to the reliability of the suppliers and the goods provided by the suppliers from the logistics-side business data and the platform-side business data of the suppliers, obtain the predicted risk label by processing such risk feature vectors through the risk assessment model, and train the risk assessment model by using the predicted risk label and the actual reliability check result of the suppliers. The trained risk assessment model learns how to identify whether the suppliers and the goods provided by the suppliers have risks, so that the related system can further analyze or check the suppliers with risks to avoid unnecessary losses caused by the suppliers providing goods that are not original.

[0077] In some embodiments, the risk features can include goods description text similarity. The goods description text similarity can include at least one of the following: goods name similarity, goods primary category similarity, goods secondary category similarity, and goods tertiary category similarity.

[0078] Based on this, Figure 3 The flowchart of determining the risk feature vector in the training method of the risk assessment model of one embodiment of the present disclosure is shown in FIG. 3. Figure 3As shown, in some embodiments, the step of "determining the risk feature vector of each sample commodity supplier in a plurality of risk features according to the logistics-side business data and the platform-side business data" in step S201 can include the following steps.

[0079] Step S301, extracting the description text of the inventory commodity of the sample commodity supplier stored in the warehouse from the logistics-side business data, and extracting the description text of the platform virtual commodity provided by the sample commodity supplier on the platform from the platform-side business data.

[0080] Correspondingly, the description text can include name text, first-level category text, second-level category text, third-level category text, etc.

[0081] Step S303, for the current inventory commodity in the inventory commodities, performing similarity calculation processing on the description text of the current inventory commodity and the description text of each platform virtual commodity, to obtain the text similarity between the current inventory commodity and each platform virtual commodity.

[0082] For example, if the merchant has four platform virtual commodities for sale on the platform, a total of 4 text similarities between the current inventory commodity and the four platform virtual commodities can be obtained.

[0083] In this step, the name text of the current inventory commodity and the name text of each platform virtual commodity can be subjected to similarity calculation processing to obtain the name text similarity between the current inventory commodity and each platform virtual commodity; the first-level category text of the current inventory commodity and the first-level category text of each platform virtual commodity can be subjected to similarity calculation processing to obtain the first-level category text similarity between the current inventory commodity and each platform virtual commodity, and so on.

[0084] Figure 4 A flowchart for determining the text similarity between the inventory commodity and the platform virtual commodity in the training method of the risk assessment model of one embodiment of the present disclosure is shown. As shown in Figure 4 As shown, in some embodiments, the platform virtual commodity includes a current virtual commodity, and step S303 can further include the following steps.

[0085] Step S401, performing word segmentation processing on the description text of the current inventory commodity to obtain a first word set. For example, if the description text of the current inventory commodity is "ABCD", the first word set can be ['A', 'B', 'C', 'D'].

[0086] Step S403, performing word segmentation processing on the description text of the current virtual commodity to obtain a second word set. For example, if the description text of the current virtual commodity is "ABDEFGH", the first word set can be ['A', 'B', 'D', 'E', 'G', 'H'].

[0087] Step S405, deduplicate and merge the first word set and the second word set to obtain a merged word set. In the above example, the merged word set can be [‘A’, ‘B’, ‘C’, ‘D’, ‘E’, ‘F’, ‘G’].

[0088] Step S407, compare the words in the first word set and the words in the merged word set to obtain a first word vector composed of 0 elements and 1 elements; wherein the word at the position corresponding to the 0 element exists in the merged word set and does not exist in the first word set, and the word at the position corresponding to the 1 element exists in the merged word set and exists in the first word set.

[0089] Step S409, compare the words in the second word set and the words in the merged word set to obtain a second word vector composed of 0 elements and 1 elements; wherein the word at the position corresponding to the 0 element exists in the merged word set and does not exist in the second word set, and the word at the position corresponding to the 1 element exists in the merged word set and exists in the second word set.

[0090] In the above example, the first word vector can be [1, 1, 1, 1, 0, 0, 0], and the second word vector can be [1, 1, 0, 0, 1, 1, 1].

[0091] Step S411, calculate the cosine similarity of the first word vector and the second word vector, and take the cosine similarity as the text similarity between the current inventory item and the current virtual item.

[0092] In the above example, the cosine similarity of the first word vector [1, 1, 1, 1, 0, 0, 0] and the second word vector [1, 1, 0, 0, 1, 1, 1] can be calculated to obtain the text similarity between the current inventory item and the current virtual item.

[0093] Step S305, determine the maximum value in the text similarity as the target similarity of the current inventory item compared to the platform side.

[0094] In this step, the maximum text similarity between the current inventory item and each platform virtual item is determined as the target similarity, which can be regarded as the highest value of the similarity that the current inventory item can achieve on the platform side. The closer the target similarity of the inventory item is to 1, the more likely the current inventory item is to be sold on the platform, and the lower the risk of the inventory item under the risk characteristic of the item description text similarity. The closer the target similarity of the inventory item is to 0, the less likely the current inventory item is to be sold on the platform, and the higher the risk of the inventory item under the risk characteristic of the item description text similarity, i.e. the higher the possibility of the “wrong version” situation.

[0095] Similarly, the text similarity between each platform virtual item and each inventory item can also be obtained by the above-mentioned manner of determining the text similarity, and then the maximum value is determined as the target similarity of each platform virtual item compared with the logistics side. The closer the target similarity of the platform virtual item is to 1, the more likely it is that the platform virtual item has a corresponding inventory item stored in the warehouse, and the lower the risk of the platform virtual item under the risk characteristic of the item description text similarity. The closer the target similarity of the platform virtual item is to 0, the less likely it is that the platform virtual item has a corresponding inventory item stored in the warehouse, and the higher the risk of the platform virtual item under the risk characteristic of the item description text similarity, i.e., the higher the possibility of the "wrong version" situation.

[0096] In step S307, the risk characteristic vector of the sample item supplier under the item description text similarity is determined according to the target similarity of all inventory items of the sample item supplier compared with the platform side.

[0097] In the risk characteristic vector of the sample item supplier under the item description text similarity, in addition to the target similarity of the inventory items compared with the platform side, the target similarity of all platform virtual items of the sample item supplier compared with the logistics side can also be included.

[0098] Through the embodiments as shown in Figure 3 The embodiments provide a manner of determining the risk characteristic vector of the sample item supplier under the item description text similarity according to the logistics side business data and the platform side business data. Based on the text comparison manner of word segmentation processing and word vector calculation, the possibility that the inventory items provided by the item supplier are truly sold and the possibility that the platform virtual items provided by the item supplier truly have inventory can be determined. With this information as the basis for training and risk prediction of the risk assessment model, the prediction risk assessment value output by the risk assessment model can be more accurate and more in line with the actual situation.

[0099] In some embodiments, the risk characteristics can also include an attribute change risk characteristic and a text description change risk characteristic. Based on this, the "determining the risk characteristic vector of each sample item supplier under multiple risk characteristics according to the logistics side business data and the platform side business data" in step S201 can include: extracting the attribute change times and the text description change times of each inventory item stored in the warehouse of the sample item supplier from the logistics side business data; determining the risk characteristic vector of the sample item supplier under the attribute change risk characteristic according to the attribute change times of all inventory items of the sample item supplier; and determining the risk characteristic vector of the sample item supplier under the text description change risk characteristic according to the text description change times of all inventory items of the sample item supplier.

[0100] In this embodiment, the number of attribute changes and the number of text description changes of each platform virtual item provided by the sample item supplier on the platform can also be extracted from the platform-side business data. The risk feature vectors under the attribute change risk feature and the text description change risk feature can also include the number of changes extracted from the platform side.

[0101] The number of attribute changes can be, for example, the number of changes in the attributes of the item origin, item material, item function, weight, size, etc. The number of text description changes can be, for example, the number of changes in the text description of the name, primary category, secondary category, tertiary category, etc. of the item. The number of attribute changes or the number of text description changes can be counted within a predetermined time period (such as three months, one month, etc.). The higher the number, the more unstable the item information provided by the item supplier, the more likely it is to be a fake sale or a fake storage item, and the higher the risk under the attribute change risk feature and / or the text description change risk feature, i.e. the higher the possibility of the "wrong version" situation.

[0102] Figure 5 A flowchart of a method for training a risk assessment model of one embodiment of the present disclosure is shown. As shown in Figure 5 Before performing the step S203 of "processing the risk feature vectors of the sample item suppliers under each risk feature by the risk assessment model", the following steps can also be included in some embodiments.

[0103] Step S501, obtaining a set of abnormal item suppliers, the set of abnormal item suppliers containing a plurality of historical abnormal item suppliers providing items with low reliability.

[0104] The historical abnormal item supplier can be a merchant that has been verified to provide unreliable items. In addition, a certain number of historical abnormal item suppliers can be included as part of the sample item suppliers to ensure that there are item suppliers with low reliability in the sample item suppliers.

[0105] Step S503, performing commonality analysis on the logistics-side business data and the platform-side business data of the historical abnormal item suppliers to obtain common risk features and initial weights of each common risk feature.

[0106] In this step, multiple to-be-analyzed attributes in the stream-side business data and the platform-side business data can be extracted, the to-be-analyzed attributes can include text descriptions (such as name text, first-level category text, second-level category text, etc.), weight, density, number of times of change of an article attribute, text description similarity, etc., and then commonality analysis is performed on these to-be-analyzed attributes. The to-be-analyzed attributes with outstanding characteristics can be determined as commonality risk features. For example, if it is found that many historical abnormal article suppliers have articles with a large number of times of change of an article attribute, the number of times of change of an article attribute can be determined as a commonality risk feature. The initial weight of the commonality risk feature can be determined based on the universality of the commonality risk feature in the historical abnormal article suppliers. The higher the universality is, the higher the initial weight can be set.

[0107] In step S505, an initial weight of the text comparison type risk feature is obtained.

[0108] The initial weight of the text comparison type risk feature can be pre-set.

[0109] In step S507, the commonality risk features and the text comparison type risk feature are determined as multiple risk features, and a risk assessment model is configured based on the multiple risk features and the initial weights of the multiple risk features.

[0110] In some actual applications, the risk features can be used to configure a corresponding sub-evaluation network, and the initial weights can be configured as weight parameters of the risk features.

[0111] In some embodiments, after obtaining the reliability check results of the multiple sample article suppliers, the training method of the risk assessment model further includes: determining a sample article supplier whose reliability check result satisfies a risk condition as a target abnormal article supplier; updating the set of abnormal article suppliers according to the target abnormal article supplier; performing commonality analysis again according to the updated set of abnormal article suppliers to obtain updated commonality risk features and initial weights of the updated commonality risk features; and updating the risk assessment model based on the updated commonality risk features and the initial weights of the updated commonality risk features.

[0112] Through this embodiment, the set of abnormal article suppliers can be expanded, and the types of risk features can be continuously optimized based on the set of abnormal article suppliers, so that the risk assessment model can perform more accurate risk assessment.

[0113] The training method of the risk assessment model provided by the above embodiment of the present disclosure can extract a risk feature vector based on the name, weight, volume, category, density, number of times of changing the attribute of a commodity, number of times of changing the name of a commodity, name similarity, and the like of the in-warehouse commodity (i.e., inventory item) and mall commodity (i.e., platform virtual item) of a merchant in a warehouse, and train and adjust the weight parameters of different risk features, output a predicted risk label (including a predicted sub-risk label and a predicted risk assessment value), and also output a list of items or a list of merchants with a high predicted risk assessment value as a to-be-checked object. Then, the to-be-checked object is locally checked in the field, and the abnormal item supplier set is updated using the to-be-checked object that is actually found to have risks, and then the risk features and their weight parameters are continuously optimized to improve the accuracy of the risk assessment model training. After the accuracy of the risk assessment model reaches a certain level, the trained risk assessment model can be obtained, and the trained risk assessment model can be output to the loss prevention system for online checking.

[0114] Figure 6 A flowchart of a risk assessment method of an embodiment of the present disclosure is shown. The method provided by the embodiment of the present disclosure can be executed by the server 101 or the client 103 as shown in Figure 1 but the present disclosure is not limited thereto. In the following example, the server 101 is taken as an example to illustrate the execution subject.

[0115] As shown in Figure 6 the risk assessment method provided by the embodiment of the present disclosure can include the following steps.

[0116] Step S601, obtaining to-be-evaluated logistics-side business data and to-be-evaluated platform-side business data of a to-be-evaluated item supplier, and determining a risk feature vector of the to-be-evaluated item supplier under a plurality of risk features according to the to-be-evaluated logistics-side business data and the to-be-evaluated platform-side business data; wherein the risk feature is used to evaluate the reliability of the item supplier and the item provided by the item supplier.

[0117] The to-be-evaluated item supplier can be a merchant who has not yet known whether there is a risk.

[0118] Step S603, processing the risk feature vector of the to-be-evaluated item supplier under each risk feature by using the trained risk assessment model to obtain a target risk label of the to-be-evaluated item supplier; wherein the trained risk assessment model is obtained by training the risk assessment model provided by the present disclosure.

[0119] The target risk label can include a target sub-risk label of the to-be-evaluated item supplier under the corresponding risk feature, and a target risk assessment value.

[0120] In step S605, the risk condition is obtained, and the target risk label is compared with the risk condition to determine the risk result of the to-be-evaluated commodity supplier.

[0121] In this step, the risk condition can include a plurality of level thresholds related to a plurality of risk levels. For example, the plurality of risk levels and the plurality of level thresholds can be obtained by pre-classifying the distribution of the risk check values of the abnormal commodity suppliers in the set of abnormal commodity suppliers. The level thresholds can be 30, 70, and 120, for example. The plurality of risk levels can correspond to the ranges [0, 30), [30, 70), [70, 120), and [120, +∞) respectively. A higher risk level means that the merchant has a higher risk degree. Then, the target risk evaluation value is compared with each level threshold, and the target risk level to which the target risk evaluation value falls is taken as the risk result of the to-be-evaluated commodity supplier. Subsequently, the to-be-evaluated merchant can be processed using the risk processing mode corresponding to the target risk level, such as being ordered to check the goods not in accordance with the version, correcting the commodity information, prohibiting the provision of the commodity, and the like.

[0122] By using the risk evaluation method provided in the present disclosure, a plurality of risk feature vectors can be extracted from the to-be-evaluated logistics-side business data and the to-be-evaluated platform-side business data of the to-be-evaluated commodity supplier. The target risk label of the to-be-evaluated commodity supplier is obtained by processing the plurality of risk feature vectors using the trained risk evaluation model. Then, the final risk result of the to-be-evaluated commodity supplier is determined in combination with the risk condition. It can be seen that the logistics-side business data and the platform-side business data of the commodity supplier can be combined as the basis for processing by the risk evaluation model, and the risk result of the to-be-evaluated commodity supplier can be accurately obtained. This is helpful for timely processing of the commodity suppliers (such as merchants) with risks in the commodity supply business, avoiding adverse effects on the commodity supply business line and commodity demanders (such as shopping consumers), and avoiding incorrect evaluation of the commodity suppliers due to the failure to identify the “goods not in accordance with the version”.

[0123] Figure 7 The architecture schematic diagram of the training method and the application method of the risk evaluation model of one embodiment of the present disclosure is shown in FIG. 7. Figure 7 As shown in FIG. 7, the architecture schematic diagram includes a data lake 701, a risk feature extraction module 702, a risk evaluation model training module 703, a risk verification module 704, and a grading processing module 705. The data lake 701, the risk feature extraction module 702, the risk evaluation model training module 703, and the risk verification module 704 are applied to the offline training stage, and the grading processing module 705 is applied to the online use stage.

[0124] Firstly, the logistics-side business data and the platform-side business data of the sample merchant can be summarized to the data lake 701, and the data lake 701 can specifically include merchant main data, platform store main data, platform commodity operation log, waybill data, attribute operation log, weight and volume data, etc. Then the waybill dimension, merchant dimension, user behavior operation data, volume data, and billing data of the sample merchant are extracted from the data lake 701, wherein the extracted waybill quantity, merchant name, merchant address, change item, and change content are non-sensitive data.

[0125] Secondly, the data lake 701 can send multiple to-be-analyzed attributes of the sample merchant to the risk feature extraction module 702 to clean and process the multiple to-be-analyzed attribute data from the mall side and the logistics side to obtain multiple to-be-processed attribute data. The cleaning and processing methods can include: de-duplication, default value processing, format and type conversion, data integration (hierarchical view), etc.

[0126] In the risk feature extraction module 702, multiple risk features can also be determined by using the black sample data 7021 (i.e., the abnormal item supplier set) in advance, and the risk feature vectors 7022 under each risk feature can be determined according to the multiple to-be-processed attribute data. The risk features can include item description text similarity. In the process of obtaining the risk feature vector of the item description text similarity, a pre-configured similarity determination algorithm 7023 can be used to process the related to-be-processed attribute data (such as the name text of the inventory item and the name text of the platform virtual item).

[0127] Then, the risk feature extraction module 702 can transmit the risk feature vectors 7022 under each risk feature to the risk assessment model training module 703. In the risk assessment model training module 703, the risk assessment model processes the risk feature vectors under each risk feature to obtain the predicted risk label of the sample merchant. The predicted risk label can include a predicted sub-risk label and a predicted risk assessment value 7034. The sub-label matrix 7032 can also be composed according to the predicted sub-risk label of each risk feature to speed up the processing speed of the computer.

[0128] In the process of processing the risk feature vectors by the risk assessment model, a threshold switch 7031 can be set to control whether to start training the risk assessment model. When it is closed, the training iteration can be stopped; when it is opened, it means that the training does not stop. In the training of the risk assessment model, the risk threshold parameters of each risk feature can be trained and adjusted to make the risk assessment model output a predicted risk label that is closer to the actual situation by using the adjusted risk threshold parameters. Specifically, the elements in the risk feature vector can be compared with the risk threshold parameters to obtain the predicted sub-risk label.

[0129] For example, for the risk feature of item description text similarity, the risk threshold parameter can be set as a number between 0 and 1. If the element in the risk feature vector is greater than or equal to the risk threshold parameter, it means that the text description of the inventory item is similar enough to the text description of the platform virtual item, and the corresponding predicted risk label is 0. If the element in the risk feature vector is less than the risk threshold parameter, it means that the text description of the inventory item is not similar enough to the text description of the platform virtual item, and the corresponding predicted risk label is 1. For example, for the risk feature of attribute change, the risk threshold parameter can be set as a positive integer. If the attribute change frequency of a certain item of the sample merchant is less than or equal to the positive integer, it means that the attribute change frequency of the item is within a reasonable range, and the corresponding predicted risk label is 0. If the attribute change frequency of a certain item of the sample merchant is greater than the positive integer, it means that the attribute change frequency of the item has exceeded the reasonable range, and the corresponding predicted risk label is 0.

[0130] In addition, the risk assessment model can also be configured with weight parameters 7033 of each risk feature for weighted processing with the sub-label matrix 7032 or the predicted risk label of each risk feature to obtain the predicted risk assessment value 7034. In the training of the risk assessment model, the weight parameters of each risk feature can also be trained and adjusted to make the prediction result of the risk assessment model more accurate.

[0131] Then, the risk verification module 704 can select the top N sample merchants with the highest predicted risk assessment value 7034 from all sample merchants as the to-be-verified merchants (here, the preset white list merchants can also be obtained, and the white list merchants in the to-be-verified merchants are removed), and use the verification system 7041 to conduct on-site verification on these to-be-verified merchants, make reliability judgment on the items provided by the to-be-verified merchants, and use 0 or 1 to identify the sub-risk verification label of the to-be-verified merchants under each risk feature, and determine the risk verification value of the to-be-verified merchants; wherein the sub-risk verification label under each risk feature and the risk verification value can be used as the actual risk label of the to-be-verified merchants. In addition, the to-be-verified merchants can also be verified according to the pre-divided regions to improve the verification efficiency. Then, the actual risk label of the to-be-verified merchants can be fed back to the risk assessment model training module 703 for training of the risk assessment model, and the to-be-verified merchants satisfying the risk condition (such as the risk verification value exceeding the condition threshold) can also be determined as target abnormal merchants, and the target abnormal merchants are fed back to the black sample data 7021 to update the black sample data 7021 using the target abnormal merchants, so as to optimize the type and weight parameters of the risk features used in the risk assessment model.

[0132] In combination with the risk assessment model training module 703 and the risk verification module 704, in the configuration process of the risk assessment model, commonalities of risk features can be extracted for black samples (abnormal merchants) and white samples (white list merchants), and the modeling configuration of the risk assessment model can be completed in combination with the risk performance analyzed from past risk cases (abnormal item supplier set). In the training process of the risk assessment model, a risk feature vector of item description text similarity can be obtained based on a cosine similarity algorithm. Specifically, the similarity algorithm can be used for the data on the merchant side and the data on the logistics side to obtain the relationship intimacy between the two sides of data, so as to identify abnormal data (the less similar, the higher the risk). In this process, the business scenario is improved, and the risk assessment model can be optimized based on the results of on-site verification, so it is very helpful to improve the accuracy of the model.

[0133] In addition, in the training iteration process of the risk assessment model, the to-be-verified merchant output by the risk assessment model in the training process can be used as a risk clue, and the risk clue can be issued offline to the regional loss prevention system (i.e., the verification system 7041) for processing, so as to complete the logical iteration of data verification and model training. After the accuracy of the risk assessment model is high and the verification process is solidified, the trained risk assessment model can be connected to the pre-trial system for online closed loop of the clue, and then put into online use.

[0134] After obtaining the trained risk assessment model, the risk assessment model can be put into online use, and the to-be-evaluated logistics side business data and the to-be-evaluated platform side business data of the to-be-evaluated merchant can be processed based on the risk assessment model to output a target risk label of the to-be-evaluated merchant, and the target risk label can include a target risk evaluation value. The hierarchical processing module 705 can be used to perform hierarchical processing on the to-be-evaluated merchant based on the target risk evaluation value. The risk level and the level threshold can be obtained by pre-processing the distribution of the risk verification value of the abnormal merchant in the black sample data 7021, then comparing the target risk evaluation value with each level threshold to determine which risk level the target risk evaluation value falls into, and then using the risk processing mode corresponding to the risk level to process the to-be-evaluated merchant.

[0135] It should be noted that the above-described figures are only schematic representations of the processes included in the method according to the exemplary embodiments of the present application, and are not intended to limit the purpose. It is easy to understand that the processes shown in the above-described figures do not indicate or limit the time sequence of these processes. In addition, it is also easy to understand that these processes can be executed synchronously or asynchronously in multiple modules, for example.

[0136] Furthermore, it should be noted that the collection, collection, updating, analysis, processing, use, transmission, and storage of user personal information involved in the technical solutions disclosed herein all comply with relevant laws and regulations, are used for legitimate purposes, and do not violate public order and good morals. Necessary measures are taken with respect to user personal information to prevent unauthorized access to user personal information data and to safeguard the security of user personal information, network security, and national security.

[0137] Figure 8 FIG. 8 is a block diagram showing a training device 800 for a risk assessment model according to an embodiment of the present disclosure; Figure 8 As shown, it includes: a first feature determination module 801, which is used to obtain logistics-side business data and platform-side business data of multiple sample item suppliers, and determine the risk feature vectors of each sample item supplier under multiple risk features based on the logistics-side business data and the platform-side business data; wherein the risk features are used to evaluate the reliability of the item supplier and the items it provides; a first processing module 802, which is used to process the risk feature vectors of the sample item supplier under various risk features through a risk assessment model to obtain a predicted risk label of the sample item supplier; a first acquisition module 803, which is used to obtain reliability verification results of multiple sample item suppliers, and determine the actual risk label of the sample item supplier based on the reliability verification results; a training module 804, which is used to train the risk assessment model based on the predicted risk label and actual risk label of the sample item supplier to obtain a trained risk assessment model.

[0138] The risk assessment model training device provided by the present disclosure can extract risk feature vectors under risk characteristics used to assess the reliability of the item supplier and the items it provides from the logistics-side business data and platform-side business data of the sample item supplier. The risk feature vectors under various risk characteristics of the sample item supplier are then processed using the risk assessment model to obtain a predicted risk label for the sample item supplier. The actual risk label of the sample item supplier is then determined based on the actual reliability verification results of the sample item supplier. The risk assessment model is then trained based on the predicted risk label and the actual risk label. Thus, this solution can combine the logistics-side business data and platform-side business data of the item supplier to obtain risk feature vectors related to the reliability of the item supplier and the items it provides. These risk feature vectors are processed using the risk assessment model to obtain predicted risk labels. The risk assessment model is then trained using the predicted risk labels and the actual reliability verification results of the item supplier. This allows the trained risk assessment model to learn how to identify whether an item supplier and the items it provides pose risks, enabling the relevant system to conduct further analysis or inventory of risky item suppliers, thereby avoiding unnecessary losses caused by the item supplier providing items that do not match the specifications.

[0139] In some embodiments, the risk feature includes item description text similarity; wherein the first feature determination module 801 determines the risk feature vector of each sample item supplier under multiple risk features according to the logistics side business data and the platform side business data, including: extracting the description text of the inventory items stored in the warehouse of the sample item supplier from the logistics side business data, and extracting the description text of the platform virtual items provided by the sample item supplier on the platform from the platform side business data; for a current inventory item in the inventory items, performing similarity calculation processing on the description text of the current inventory item and the description text of each platform virtual item to obtain the text similarity between the current inventory item and each platform virtual item; determining the maximum value in the text similarity as the target similarity of the current inventory item compared to the platform side; and determining the risk feature vector of the sample item supplier under the item description text similarity according to the target similarity of all inventory items of the sample item supplier compared to the platform side.

[0140] In some embodiments, the item description text similarity includes at least one of the following: item name similarity, item primary category similarity, item secondary category similarity, and item tertiary category similarity.

[0141] In some embodiments, the platform virtual items include a current virtual item; wherein the first feature determination module 801 performs similarity calculation processing on the description text of the current inventory item and the description text of each platform virtual item for a current inventory item in the inventory items to obtain the text similarity between the current inventory item and each platform virtual item, including: performing word segmentation processing on the description text of the current inventory item to obtain a first word set; performing word segmentation processing on the description text of the current virtual item to obtain a second word set; merging the first word set and the second word set to obtain a merged word set; comparing the words in the first word set and the words in the merged word set to obtain a first word vector composed of 0 elements and 1 elements; wherein the word at the position corresponding to the 0 element exists in the merged word set but does not exist in the first word set, and the word at the position corresponding to the 1 element exists in the merged word set and exists in the first word set; comparing the words in the second word set and the words in the merged word set to obtain a second word vector composed of 0 elements and 1 elements; wherein the word at the position corresponding to the 0 element exists in the merged word set but does not exist in the second word set, and the word at the position corresponding to the 1 element exists in the merged word set and exists in the second word set; and calculating the cosine similarity of the first word vector and the second word vector, taking the cosine similarity as the text similarity between the current inventory item and the current virtual item.

[0142] In some embodiments, the risk features include an attribute change risk feature and a text description change risk feature; wherein the first feature determination module 801 determines the risk feature vector of each sample commodity supplier under multiple risk features according to the logistics-side business data and the platform-side business data, including: extracting the attribute change times and the text description change times of each inventory commodity stored in the warehouse of the sample commodity supplier from the logistics-side business data; determining the risk feature vector of the sample commodity supplier under the attribute change risk feature according to the attribute change times of all inventory commodities of the sample commodity supplier; determining the risk feature vector of the sample commodity supplier under the text description change risk feature according to the text description change times of all inventory commodities of the sample commodity supplier.

[0143] In some embodiments, the risk assessment model is configured with weight parameters of each risk feature; wherein the first processing module 802 processes the risk feature vector of the sample commodity supplier under each risk feature through the risk assessment model to obtain the predicted risk label of the sample commodity supplier, including: processing the risk feature vector of the sample commodity supplier under the corresponding risk feature through the risk assessment model to obtain the predicted sub-risk label under the corresponding risk feature; weighting the predicted sub-risk labels under multiple risk features through the weight parameters in the risk assessment model to obtain the predicted risk assessment value of the sample commodity supplier; determining the predicted risk label of the sample commodity supplier according to the predicted sub-risk label under each risk feature and the predicted risk assessment value.

[0144] In some embodiments, the reliability verification result of the sample commodity supplier includes a sub-risk verification label and a risk verification value of the sample commodity supplier under each risk feature; the training module 804 trains the risk assessment model according to the predicted risk label and the actual risk label of the sample commodity supplier, including: constructing multiple first loss functions according to the predicted sub-risk label and the sub-risk verification label under each risk feature; constructing a second loss function according to the predicted risk assessment value and the risk verification value; training the risk assessment model by using the multiple first loss functions and the second loss function.

[0145] In some embodiments, the training device of the risk assessment model also includes a configuration module 805. Before the first processing module 802 processes the risk feature vector of the sample item supplier under various risk features through the risk assessment model, the configuration module 805 is used to: obtain a set of abnormal item suppliers, wherein the set of abnormal item suppliers includes multiple historical abnormal item suppliers with low reliability in providing items; perform commonality analysis based on the logistics-side business data and platform-side business data of the historical abnormal item suppliers to obtain common risk features and initial weights of each common risk feature; obtain the initial weights of text comparison risk features; determine the common risk features and the text comparison risk features as the multiple risk features, and configure the risk assessment model based on the multiple risk features and their initial weights.

[0146] In some embodiments, the training device of the risk assessment model also includes an updating module 806. After the first acquisition module 803 obtains the reliability verification results of multiple sample item suppliers, the updating module 806 is used to: determine the sample item suppliers whose reliability verification results meet the risk conditions as target abnormal item suppliers; update the abnormal item supplier set based on the target abnormal item suppliers; re-perform commonality analysis based on the updated abnormal item supplier set to obtain updated common risk characteristics and initial weights of each updated common risk characteristic; and update the risk assessment model based on the updated common risk characteristics and their initial weights.

[0147] Figure 8 For other contents of the embodiment, reference can be made to the other embodiments mentioned above.

[0148] Figure 9 FIG. 1 is a block diagram of a risk assessment device 900 according to an embodiment of the present disclosure; FIG. Figure 9 As shown, it includes: a second feature determination module 901, which is used to obtain the logistics-side business data to be evaluated and the platform-side business data to be evaluated of the supplier of the item to be evaluated, and determine the risk feature vector of the supplier of the item to be evaluated under multiple risk features based on the logistics-side business data to be evaluated and the platform-side business data to be evaluated; wherein, the risk feature is used to evaluate the reliability of the item supplier and the items it provides; a second processing module 902, which is used to process the risk feature vector of the supplier of the item to be evaluated under various risk features through a trained risk assessment model to obtain a target risk label of the supplier of the item to be evaluated; wherein, the trained risk assessment model is obtained by training the risk assessment model according to the training method of the above-mentioned risk assessment model; an assessment module 903, which is used to obtain risk conditions and compare the target risk label with the risk conditions to determine the risk result of the supplier of the item to be evaluated.

[0149] The risk assessment device provided by the present disclosure can extract a plurality of risk feature vectors from the to-be-evaluated logistics-side business data and the to-be-evaluated platform-side business data of the to-be-evaluated commodity supplier, process the plurality of risk feature vectors using the trained risk assessment model to obtain a target risk label of the to-be-evaluated commodity supplier, and then determine a final risk result of the to-be-evaluated commodity supplier in combination with a risk condition. It can be seen that the present solution can combine the logistics-side business data and the platform-side business data of the commodity supplier as the basis for processing by the risk assessment model, accurately obtain the risk result of the to-be-evaluated commodity supplier, help to timely process the commodity suppliers (such as merchants) with risks in the commodity supply business, help to avoid adverse effects on the commodity supply business line and commodity demanders (such as shopping consumers), and help to avoid incorrect evaluation of the commodity suppliers due to failure to identify the 'wrong goods' of the commodity suppliers.

[0150] Figure 9 Other contents of the embodiments can refer to the above-mentioned other embodiments.

[0151] Those skilled in the art can understand that each aspect of the present disclosure can be implemented as a system, a method or a program product. Therefore, each aspect of the present disclosure can be specifically implemented as follows: a complete hardware embodiment, a complete software embodiment (including firmware, microcode, etc.), or an embodiment combining hardware and software, which can be collectively referred to as "circuitry", "module" or "system".

[0152] Figure 10 A structural block diagram of a computer device suitable for implementing the exemplary embodiments of the present disclosure is shown. It should be noted that the electronic device shown is only an example and should not limit the functions and use range of the embodiments of the present disclosure.

[0153] The electronic device 1000 according to the embodiment of the present disclosure will be described below with reference to Figure 10 Figure 10 The electronic device 1000 shown is only an example and should not limit the functions and use range of the embodiments of the present disclosure.

[0154] As shown in Figure 10 The components of the electronic device 1000 can include but are not limited to the at least one processing unit 1010, the at least one storage unit 1020, and a bus 1030 connecting different system components, including the storage unit 1020 and the processing unit 1010.

[0155] ​The storage unit stores program codes which can be executed by the processing unit 1010, so that the processing unit 1010 performs the steps described in the above "Exemplary Methods" section according to various exemplary embodiments of the present application. For example, the processing unit 1010 can perform the method as shown in the above Figure 2

[0156] The storage unit 1020 can include a readable medium in the form of volatile storage such as a random access memory (RAM) 10201 and / or cache memory 10202, and also can further include a read-only memory (ROM) 10203.

[0157] The storage unit 1020 can further include program / utility 10204 having a set of programs / modules 10205, including an operating system, one or more application programs, other programs, and programmatic data, each or some combination thereof, which can include implementation of a network environment.

[0158] The bus 1030 can represent one or more of several types of bus structures, including a storage bus or bus controller, a peripheral bus, a graphics acceleration port, a processor or local bus using any of a variety of bus architectures.

[0159] The electronic device 1000 can also communicate with one or more external devices 1100 such as a keyboard or pointing device, a Bluetooth device, etc.; user interfaces and / or peripheral devices such as a scanner, or other devices that enable a user to interact with the electronic device 1000; and / or any devices (e.g., modems, network cards, etc.) that enable the electronic device 1000 to communicate with one or more other computing devices. Such communication can occur via Input / Output (I / O) interface(s) 1050. Still yet, the electronic device 1000 can communicate with one or more networks (such as a local area network (LAN), a wide area network (WAN), and / or the Internet) through a network adapter 1060. As depicted, the network adapter 1060 communicates with the other components of the electronic device 1000 via the bus 1030. It should be appreciated that although the network adapter 1060 is depicted as a single component, the network adapter 1060 can comprise two or more components that work together to facilitate communications with one or more other computing devices.

[0160] ​In the exemplary embodiments of the present disclosure, a computer readable storage medium is also provided, on which a program product capable of implementing the above-mentioned method of the present specification is stored. In some possible implementations, various aspects of the present application can also be implemented in the form of a program product, which includes program codes for causing a terminal device to perform the steps described in the above-mentioned "Exemplary Method" section of the present specification according to various exemplary embodiments of the present application when the program product is run on the terminal device.

[0161] The program product for implementing the above-mentioned method according to the embodiments of the present application can take a portable compact disc read-only memory (CD-ROM) and include program codes, and can be run on a terminal device, such as a personal computer. However, the program product of the present application is not limited to this, and in the present document, the readable storage medium can be any tangible medium containing or storing a program, which can be used by or in conjunction with an instruction execution system, apparatus or device.

[0162] The program product can take any combination of one or more readable media. The readable medium can be a readable signal medium or a readable storage medium. The readable storage medium may, for example, but is not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus or device, or any combination of the above. More specific examples (non-exhaustive list) of readable storage media include an electrical connection having one or more wires, a portable disc, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above.

[0163] The computer readable signal medium can include a data signal propagated in a baseband or as a carrier wave in a propagated data signal, in which readable program codes are carried. Such a propagated data signal can take various forms, including but not limited to an electromagnetic signal, an optical signal or any suitable combination of the above. The readable signal medium can also be any readable medium other than the readable storage medium, which can send, propagate or transmit a program for use by or in conjunction with an instruction execution system, apparatus or device.

[0164] The program codes contained on the readable medium can be transmitted by any suitable medium, including but not limited to wireless, wired, optical cable, RF, etc., or any suitable combination of the above.

[0165] Computer program code for carrying out operations of the present application can be written in any combination of one or more programming languages, including an object oriented programming language such as Java, C++ or the like, and conventional procedural programming languages, such as the "C" programming language or similar programming languages. The program code can execute entirely on the user's computing device, partly on the user's device, as a stand-alone software package, partly on the user's computing device and partly on a remote computing device or entirely on the remote computing device or server. In the latter scenario, the remote computing device can be connected to the user's computing device through any type of network, including a local area network (LAN) or a wide area network (WAN), or the connection can be made to an external computing device, such as through the Internet using an Internet Service Provider.

[0166] It should be noted that, although several modules or units of the devices for action execution are mentioned in the above detailed description, such division is not mandatory. Indeed, according to embodiments of the present disclosure, features and functionalities of two or more modules or units described above can be embodied in one module or unit. Conversely, features and functionalities of one module or unit described above can be further divided into embodied by multiple modules or units.

[0167] Furthermore, although the various steps of the methods of the present disclosure are described in a particular order in the figures, this is not required or implied as to the order of execution of the steps, nor is it required that all of the steps be executed to achieve the desired result. Additionally or alternatively, certain steps can be omitted, multiple steps can be combined into one step, one step can be broken into multiple steps, etc.

[0168] From the above description of embodiments, those skilled in the art will readily perceive that the example embodiments described herein can be implemented by software and / or by hardware and / or by software in combination with hardware. Thus, the technical solutions according to the embodiments of the present disclosure can be embodied in the form of a software product. The software product can be stored in a non-volatile storage medium, such as a CD-ROM, a USB flash drive, a mobile hard disk, etc., or a network, and includes a number of instructions for causing a computing device (such as a personal computer, a server, a mobile terminal, or a network device, etc.) to execute the methods according to the embodiments of the present disclosure.

[0169] According to an aspect of the present disclosure, a computer program product or computer program including computer instructions stored in a computer-readable storage medium is provided. A processor of a computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions to cause the computer device to perform the method provided in various optional implementations of the above-described embodiments.

[0170] Other embodiments of the present disclosure will be apparent to those skilled in the art from consideration of the specification and practice of the features disclosed herein. It is intended that the present disclosure cover any and all variations of the present disclosure including those variations that are now deemed to fall within the general principles of the present disclosure and including those variations that are deemed to fall within the patentably distinct field of this technology. The specification and examples are to be considered exemplary only, with the true scope and spirit of the present disclosure being indicated by the appended claims.

Claims

1. A training method for a risk assessment model, characterized in that: The method comprises: obtaining logistics-side business data and platform-side business data of a plurality of sample commodity providers, and determining a risk feature vector of each sample commodity provider under a plurality of risk features according to the logistics-side business data and the platform-side business data; wherein the risk features are used to evaluate the reliability of the commodity providers and the commodities provided by the commodity providers; processing the risk feature vector of the sample commodity provider under each risk feature through a risk assessment model to obtain a predicted risk label of the sample commodity provider; obtaining a reliability verification result of the plurality of sample commodity providers, and determining an actual risk label of the sample commodity provider according to the reliability verification result; training the risk assessment model according to the predicted risk label and the actual risk label of the sample commodity provider to obtain a trained risk assessment model.

2. The method of claim 1, wherein, The risk features comprise commodity description text similarity. The method comprises: extracting description texts of inventory commodities stored in a warehouse of the sample commodity provider from the logistics-side business data, and extracting description texts of platform virtual commodities provided by the sample commodity provider on a platform from the platform-side business data; performing similarity calculation processing on the description text of the current inventory commodity and the description texts of the platform virtual commodities to obtain text similarity between the current inventory commodity and each platform virtual commodity; determining a target similarity of the current inventory commodity compared with the platform side as a maximum value in the text similarity; determining the risk feature vector of the sample commodity provider under the commodity description text similarity according to the target similarity of all inventory commodities of the sample commodity provider compared with the platform side.

3. The method of claim 2, wherein, The commodity description text similarity comprises at least one of the following: commodity name similarity, commodity primary category similarity, commodity secondary category similarity, and commodity tertiary category similarity.

4. The method of claim 2, wherein, The platform virtual commodities comprise a current virtual commodity. The method comprises: performing word segmentation processing on the description text of the current inventory commodity to obtain a first word set; performing word segmentation processing on the description text of the current virtual commodity to obtain a second word set; performing de-duplication and merging on the first word set and the second word set to obtain a merged word set; comparing the words in the first word set with the words in the merged word set to obtain a first word vector composed of 0 elements and 1 elements; wherein the word corresponding to the 0 element exists in the merged word set but does not exist in the first word set, and the word corresponding to the 1 element exists in the merged word set and exists in the first word set; and performing similarity calculation processing on the description text of the current inventory commodity and the description texts of the platform virtual commodities to obtain text similarity between the current inventory commodity and each platform virtual commodity. comparing the words in the second word set with the words in the merged word set to obtain a second word vector composed of 0 elements and 1 elements; wherein a 0 element corresponds to a word that exists in the merged word set but does not exist in the second word set, and a 1 element corresponds to a word that exists in both the merged word set and the second word set; calculating the cosine similarity of the first word vector and the second word vector, and taking the cosine similarity as the text similarity between the current inventory item and the current virtual item.

5. The method of claim 1, wherein, The risk features include attribute change risk features and text description change risk features. The risk feature vector of each sample item supplier under each risk feature is determined according to the logistics-side business data and the platform-side business data, including: extracting the attribute change times and the text description change times of each inventory item stored in the warehouse of the sample item supplier from the logistics-side business data; determining the risk feature vector of the sample item supplier under the attribute change risk feature according to the attribute change times of all inventory items of the sample item supplier; determining the risk feature vector of the sample item supplier under the text description change risk feature according to the text description change times of all inventory items of the sample item supplier.

6. The method of claim 1, wherein, The risk assessment model is configured with weight parameters of each risk feature; The risk assessment model processes the risk feature vector of the sample item supplier under each risk feature to obtain a predicted risk label of the sample item supplier, including: The risk assessment model processes the risk feature vector of the sample item supplier under the corresponding risk feature to obtain a predicted sub-risk label under the corresponding risk feature; The risk assessment model processes the risk feature vector of the sample item supplier under each risk feature to obtain a predicted risk label of the sample item supplier, including: The risk assessment model processes the risk feature vector of the sample item supplier under each risk feature to obtain a predicted risk label of the sample item supplier, including:

7. The method of claim 6, wherein, The reliability verification result of the sample item supplier includes a sub-risk verification label and a risk verification value of the sample item supplier under each risk feature; The risk assessment model is trained according to the predicted risk label and the actual risk label of the sample item supplier, including: A plurality of first loss functions are constructed according to the predicted sub-risk label and the sub-risk verification label under each risk feature; A second loss function is constructed according to the predicted risk assessment value and the risk verification value; The risk assessment model is trained using the plurality of first loss functions and the second loss function.

8. The method of claim 1, wherein, Before the risk assessment model processes the risk feature vector of the sample item supplier under each risk feature, the method further includes: obtaining a set of abnormal item suppliers, which includes a plurality of historical abnormal item suppliers that provide items with low reliability; performing commonality analysis on the logistics-side business data and the platform-side business data of the historical abnormal item suppliers to obtain common risk features and initial weights of each common risk feature; obtaining initial weights of the text comparison type risk features; determining the common risk features and the text comparison type risk features as the plurality of risk features, and configuring the risk assessment model based on the plurality of risk features and the initial weights thereof.

9. The method of claim 8, wherein, After obtaining the reliability verification results of the plurality of sample commodity suppliers, the method further comprises: determining a sample commodity supplier whose reliability verification result meets a risk condition as a target abnormal commodity supplier; updating the set of abnormal commodity suppliers according to the target abnormal commodity supplier; re-performing commonality analysis according to the updated set of abnormal commodity suppliers to obtain updated common risk features and initial weights of each of the updated common risk features; updating the risk assessment model based on the updated common risk features and the initial weights thereof.

10. A method of risk assessment, characterized by, comprises: obtaining to-be-evaluated logistics side business data and to-be-evaluated platform side business data of a to-be-evaluated commodity supplier, and determining a risk feature vector of the to-be-evaluated commodity supplier under a plurality of risk features according to the to-be-evaluated logistics side business data and the to-be-evaluated platform side business data; wherein the risk features are used to evaluate the reliability of a commodity supplier and commodities provided by the commodity supplier; processing the risk feature vector of the to-be-evaluated commodity supplier under each risk feature by using the trained risk assessment model to obtain a target risk label of the to-be-evaluated commodity supplier; wherein the trained risk assessment model is obtained by training the method according to any one of claims 1 to 9; obtaining a risk condition, and comparing the target risk label with the risk condition to determine a risk result of the to-be-evaluated commodity supplier.

11. A training device for a risk assessment model, characterized in that: comprises: a first feature determination module, configured to obtain logistics side business data and platform side business data of a plurality of sample commodity suppliers, and determine a risk feature vector of each sample commodity supplier under a plurality of risk features according to the logistics side business data and the platform side business data; wherein the risk features are used to evaluate the reliability of a commodity supplier and commodities provided by the commodity supplier; a first processing module, configured to process the risk feature vector of a sample commodity supplier under each risk feature by using a risk assessment model to obtain a predicted risk label of the sample commodity supplier; a first obtaining module, configured to obtain reliability verification results of the plurality of sample commodity suppliers, and determine actual risk labels of the sample commodity suppliers according to the reliability verification results; a training module, configured to train the risk assessment model according to the predicted risk labels and the actual risk labels of the sample commodity suppliers to obtain a trained risk assessment model.

12. A risk assessment apparatus, characterized by, comprises: a second feature determination module, configured to obtain to-be-evaluated logistics side business data and to-be-evaluated platform side business data of a to-be-evaluated commodity supplier, and determine a risk feature vector of the to-be-evaluated commodity supplier under a plurality of risk features according to the to-be-evaluated logistics side business data and the to-be-evaluated platform side business data; wherein the risk features are used to evaluate the reliability of a commodity supplier and commodities provided by the commodity supplier; a second processing module, configured to process a risk feature vector of the to-be-evaluated commodity supplier under each risk feature by using a trained risk evaluation model to obtain a target risk label of the to-be-evaluated commodity supplier; wherein the trained risk evaluation model is obtained by using the method in any one of claims 1 to 9; an evaluation module, configured to obtain a risk condition, and compare the target risk label with the risk condition to determine a risk result of the to-be-evaluated commodity supplier.

13. A computer readable storage medium having stored thereon a computer program, the program, when executed by a processor, implementing the method in any one of claims 1 to 10.

14. An electronic device, comprising: comprising: one or more processors; a storage device for storing one or more programs, which, when executed by the one or more processors, cause the one or more processors to implement the method in any one of claims 1 to 10.

Images