Commodity risk identification method and device and electronic equipment

Abstract

The application provides a commodity risk identification method and device and electronic equipment. The method comprises: obtaining commodity information of a plurality of to-be-identified commodities; inputting the commodity information of each to-be-identified commodity into each prediction model in a plurality of pre-trained prediction models to obtain a prediction risk probability output by each prediction model for each to-be-identified commodity; sorting a plurality of prediction risk probabilities corresponding to a plurality of to-be-identified commodities on the same prediction model to determine an initial sorting result of each to-be-identified commodity, and repeating the step until a plurality of initial sorting results corresponding to a plurality of prediction models for each to-be-identified commodity are obtained; calculating a risk sorting result of each to-be-identified commodity according to the plurality of initial sorting results of each to-be-identified commodity; and identifying the risk of a plurality of to-be-identified commodities and associated stores according to the risk sorting result. The above method can improve the identification coverage, speed and accuracy of commodity risk.

Description

Technical Field

[0001] This application belongs to the field of computer science, specifically the field of risk management technology, and in particular relates to a method, apparatus, and electronic device for identifying commodity risks. Background Technology

[0002] Products sold on e-commerce platforms may include counterfeit or illegal items. Currently, the identification of risky products mainly relies on the experience of professionals. However, due to the vast variety and quantity of products, the identification capabilities of professionals are limited, resulting in low coverage, accuracy, and speed of risk identification.

[0003] If the technical problems of low coverage, poor accuracy, and slow speed of product risk identification cannot be resolved, it will affect the normal operation of stores and e-commerce platforms. Summary of the Invention

[0004] In view of the above, it is necessary to provide a commodity risk identification method, device and electronic device that can solve the technical problems of low identification coverage, poor identification accuracy and poor identification speed of commodity risks.

[0005] On the one hand, this application provides a product risk identification method, the method comprising: acquiring product information of multiple products to be identified; inputting the product information of each product to be identified into each of multiple pre-trained prediction models to obtain the predicted risk probability output by each prediction model for each product to be identified; sorting the multiple predicted risk probabilities of multiple products to be identified on the same prediction model; determining an initial sorting result for each product to be identified; repeating this step until multiple initial sorting results of each product to be identified in multiple prediction models are obtained; calculating the risk sorting result of each product to be identified based on the multiple initial sorting results of each product to be identified; and identifying the risks of multiple products to be identified based on the risk sorting results.

[0006] In some embodiments, before inputting the product information of each product to be identified into each of the pre-trained multiple prediction models, the method further includes: for each product listed within a preset observation period, determining a time interval based on the preset observation period and the preset performance period corresponding to the preset observation period, and obtaining the historical transaction records of each product within the time interval; determining the first high-risk product and the first low-risk product among all listed products within the preset observation period based on the historical transaction records; generating multiple training datasets based on the product information of the first high-risk product and the first low-risk product; matching a machine learning algorithm in a preset network framework for each training dataset based on the data distribution of each training dataset; and training the matched machine learning algorithm using each training dataset to obtain multiple prediction models.

[0007] In some embodiments, historical transaction records include the number of transactions and the number of customer complaints corresponding to the number of transactions. Determining the first high-risk product and the first low-risk product among all listed products within a preset observation period based on historical transaction records includes: determining the customer complaint ratio of each product based on the number of customer complaints and the number of transactions; identifying products with a customer complaint ratio greater than a preset threshold as the first high-risk product among all listed products; and identifying products with a customer complaint ratio less than or equal to the preset threshold as the first low-risk product among all listed products.

[0008] In some embodiments, generating multiple training datasets based on the product information of the first high-risk product and the first low-risk product includes: constructing a high-risk dataset based on the product information of the first high-risk product, constructing a low-risk dataset based on the product information of the first low-risk product, counting the first quantity of the first high-risk product, dividing the low-risk dataset into multiple sub-datasets based on the first quantity and a preset first ratio coefficient, and combining the high-risk dataset with any one of the sub-datasets to obtain a training dataset.

[0009] In some embodiments, calculating the risk ranking result for each product to be identified based on multiple initial ranking results for each product to be identified includes: determining the risk ranking result based on the multiple initial ranking results for each product to be identified and the number of multiple initial ranking results; or, determining a first statistical value based on each initial ranking result for each product to be identified and the corresponding predicted risk probability, and determining the risk ranking result based on the second statistical value corresponding to the multiple predicted risk probabilities and the first statistical value.

[0010] In some embodiments, risk identification of multiple products to be identified based on risk ranking results includes: selecting a preset number of products to be identified corresponding to the top-ranked risk ranking results from the risk ranking results sorted from low to high as target high-risk products, and pushing the list of risk products of the target high-risk products to a preset terminal device.

[0011] In some embodiments, the method further includes: risk identification of stores associated with multiple products to be identified based on risk ranking results.

[0012] In some embodiments, risk identification of stores associated with multiple products to be identified based on risk ranking results includes: calculating the number of high-risk products based on a preset second ratio coefficient and a second quantity of multiple products to be identified; calculating the number of low-risk products based on a preset third ratio coefficient and the second quantity; selecting products corresponding to the top-ranked risk ranking results from the risk ranking results sorted from low to high based on the number of high-risk products, as the second high-risk products; and selecting products corresponding to the bottom-ranked risk ranking results from the risk ranking results sorted from low to high based on the number of low-risk products, as the second low-risk products; if multiple products to be identified are associated with multiple stores, calculating a first proportion of the second high-risk products in each associated store, and calculating a second proportion of the second low-risk products in each associated store; and performing risk identification for each associated store based on the first proportion and / or the second proportion.

[0013] On the other hand, this application provides a commodity risk identification device, operating in an electronic device. The device includes: an acquisition unit for acquiring commodity information of multiple commodities to be identified; an input unit for inputting the commodity information of each commodity to be identified into each of a plurality of pre-trained prediction models to obtain the predicted risk probability output by each prediction model for each commodity to be identified; a sorting unit for sorting the plurality of predicted risk probabilities of the multiple commodities to be identified corresponding to the same prediction model, determining an initial sorting result for each commodity to be identified, and repeating this step until multiple initial sorting results of each commodity to be identified corresponding to the multiple prediction models are obtained; a calculation unit for calculating the risk sorting result of each commodity to be identified based on the multiple initial sorting results of each commodity to be identified; and a risk identification unit for identifying the risks of the multiple commodities to be identified based on the risk sorting results.

[0014] On the other hand, this application provides an electronic device, which includes: a memory storing at least one instruction; and a processor executing at least one instruction to implement the commodity risk identification method.

[0015] Through the above implementation method, since multiple prediction models are trained using different training datasets, there are parameter differences among the multiple prediction models. Using multiple prediction models for product risk identification can increase the number of products identified and the identification coverage. By using multiple prediction models to identify the risks of multiple products, it is no longer necessary to rely on a large amount of manual time for product identification, thus improving the speed of product risk identification. Furthermore, since the risk ranking result for each product to be identified integrates multiple initial ranking results corresponding to multiple prediction models, the risk ranking result for each product has higher representativeness and accuracy, thus improving the accuracy of product risk identification. Attached Figure Description

[0016] Figure 1 This is a structural diagram of an electronic device provided in an embodiment of this application.

[0017] Figure 2 This is a flowchart of a product risk identification method provided in an embodiment of this application.

[0018] Figure 3 This is a flowchart of a training method for multiple prediction models provided in an embodiment of this application.

[0019] Figure 4 This is a flowchart of a method for generating multiple training datasets provided in an embodiment of this application.

[0020] Figure 5 This is a flowchart of a method for risk identification of associated stores provided in an embodiment of this application.

[0021] Figure 6 This is a functional block diagram of a commodity risk identification device provided in an embodiment of this application. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of this application clearer, the application will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0023] It should be noted that in this application, "at least one" means one or more, and "more than one" means two or more. "And / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, or B alone, where A and B can be singular or plural. The terms "first," "second," "third," "fourth," etc. (if present) in the specification, claims, and drawings of this application are used to distinguish similar objects, not to describe a specific order or sequence.

[0024] In the embodiments of this application, the terms "exemplary" or "for example" are used to indicate that something is an example, illustration, or description. Any embodiment or design that is described as "exemplary" or "for example" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or design. Specifically, the use of the terms "exemplary" or "for example" is intended to present the relevant concepts in a specific manner.

[0025] The number of products listed and sold in various stores on e-commerce platforms is enormous, sometimes reaching tens or hundreds of millions. Merchants on e-commerce platforms are listing, delisting, and updating countless products every moment. If risky products are not identified and dealt with in a timely manner after being listed, more refunds and complaints will be generated as more orders are placed, which will have an adverse impact on the operation of e-commerce platforms and normal stores on e-commerce platforms, resulting in significant losses for both e-commerce platforms and normal stores.

[0026] Based on the above, it is necessary to promptly identify the risks of products on e-commerce platforms. Currently, the identification of risky products mainly relies on the experience of professionals (e.g., risk control personnel). Due to the large number of products, manual identification has low coverage and cannot conduct timely and accurate risk screening, resulting in significant delays in the manual identification process. Furthermore, current methods for identifying product risks are too simplistic and cannot meet actual usage needs, leading to low accuracy and uncontrollable risks, thus hindering effective product risk management.

[0027] To address the aforementioned technical problems, this application provides a product risk identification method, apparatus, and electronic device, which can improve the identification coverage, speed, and accuracy of product risks. The product risk identification method provided in this application can be applied to one or more electronic devices.

[0028] like Figure 1 The diagram shown is a structural diagram of an electronic device provided in one embodiment of this application. The electronic device 10 can be a mobile phone, tablet computer, laptop computer, self-moving device, or other electronic device. This application embodiment does not impose any restrictions on the specific type of electronic device.

[0029] like Figure 1 As shown, the electronic device 10 may include a communication module 101, a memory 102, a processor 103, an input / output (I / O) interface 104, and a bus 105. The processor 103 is coupled to the communication module 101, the memory 102, and the input / output interface 104 via the bus 105.

[0030] Communication module 101 may include a wired communication module and / or a wireless communication module. The wired communication module may provide one or more wired communication solutions such as Universal Serial Bus (USB) and Controller Area Network (CAN). The wireless communication module may provide one or more wireless communication solutions such as Wireless Fidelity (Wi-Fi), Bluetooth (BT), mobile communication networks, frequency modulation (FM), near-field communication (NFC), and infrared (IR).

[0031] Memory 102 may include one or more random access memories (RAM) and one or more non-volatile memories (NVM). The RAM can be directly read and written by the processor 103, and can be used to store executable programs (e.g., machine instructions) of other running programs, as well as user and application data. The RAM may include static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDR SDRAM), etc.

[0032] Non-volatile memory can also store executable programs and user and application data, and can be pre-loaded into random access memory for direct reading and writing by the processor 110. Non-volatile memory can include disk storage devices and flash memory.

[0033] The memory 102 is used to store one or more computer programs. The one or more computer programs are configured to be executed by the processor 103. The one or more computer programs include multiple instructions that, when executed by the processor 103, can implement a commodity risk identification method that is executed on the electronic device 10.

[0034] In other embodiments, such as Figure 1The electronic device 10 shown also includes an external memory interface for connecting to an external memory to expand the storage capacity of the electronic device 10.

[0035] Processor 103 may include one or more processing units, such as an application processor (AP), a modem processor, a graphics processing unit (GPU), an image signal processor (ISP), a controller, a video codec, a digital signal processor (DSP), and / or a neural network processing unit (NPU). These different processing units may be independent devices or integrated into one or more processors.

[0036] The processor 103 provides computing and control capabilities; for example, the processor 103 is used to execute computer programs stored in the memory 102 to implement the commodity risk identification method described above.

[0037] The input / output interface 104 is used to provide a channel for user input or output. For example, the input / output interface 104 can be used to connect various input / output devices, such as a mouse, keyboard, touch device, display screen, etc., so that users can enter information or visualize information.

[0038] Bus 105 is used at least to provide a channel for communication between communication modules 101, memory 102, processor 103, and input / output interface 104 in electronic device 10.

[0039] It is understood that the structures illustrated in the embodiments of this application do not constitute a specific limitation on the electronic device 10. In other embodiments of this application, the electronic device 10 may include more or fewer components than illustrated, or combine some components, or split some components, or have different component arrangements. The illustrated components may be implemented in hardware, software, or a combination of software and hardware.

[0040] like Figure 2 The diagram shown is a flowchart of a product risk identification method provided in an embodiment of this application. Depending on different needs, the order of the steps in this flowchart can be adjusted according to actual requirements, and some steps can be omitted. The method is executed by an electronic device, for example... Figure 1 The electronic device 10 shown.

[0041] S11, Obtain product information for multiple products to be identified.

[0042] In some embodiments, electronic devices can obtain product information of the product to be identified in a variety of ways. For example, an electronic device can use all products listed on an e-commerce platform during a predetermined observation period (hereinafter referred to as the "predetermined observation period") as products to be identified, and obtain product information of each product to be identified from the e-commerce platform's database.

[0043] In one embodiment of this application, the predetermined observation period can refer to the listing time corresponding to multiple products to be identified. There can be multiple predetermined observation periods, and these periods can be set or updated independently; this application does not impose any restrictions on this. For example, the predetermined observation period may include January 1, 2023, January 2, 2023, and January 3, 2023. The electronic device can identify all products listed on the e-commerce platform within these three days as products to be identified.

[0044] Product information may include, but is not limited to: basic information, transaction information, customer complaint information, and prohibited / restricted sales information for each product to be identified and the store associated with each product.

[0045] The basic information for each product to be identified and its associated store includes, but is not limited to: product category, all listing times, discount rate, whether shipping is included, shipping cost when shipping is included, the time interval between the first listing date and the planned observation period (in days), the time interval between the most recent listing date and the planned observation period (in days), and the cumulative number of listings. It also includes the number of products listed by the store within the preset number of days closest to the planned observation period, the total number of products in the store, and the number of products delisted by the store within the preset number of days closest to the planned observation period. The product category includes multi-level classifications for each product to be identified (e.g., 1-3 level classifications). For example, a men's shirt could have a 1-3 level classification as: Clothing - Men's Clothing - Men's Shirts.

[0046] The transaction information for each product to be identified and its associated store includes, but is not limited to: the time interval between the first transaction and the predetermined observation period (in units of: / day), the cumulative number of transactions, the average transaction amount, the discount rate of the product's listing price compared to the historical cumulative transaction amount, the time interval between the first transaction and the predetermined observation period (in units of: / day), and the time interval between the most recent transaction and the predetermined observation period (in units of: / day). It also includes the store's cumulative transaction amount, the store's transaction amount within the preset number of days closest to the predetermined observation period, the store's cumulative number of transaction days, the time interval between the store's first transaction and the predetermined observation period (in units of: / day), the average transaction price of similar products on the platform, and the discount rate of the product's listing price compared to the average price of similar products.

[0047] Customer complaint information for each product to be identified and its associated store includes, but is not limited to: the number of historical customer complaints for each product to be identified and its associated store, the total amount of historical customer complaints, the number of customer complaints within the most recent preset number of days before the predetermined observation period, the total amount of customer complaints within the most recent preset number of days before the predetermined observation period, and the time interval between the most recent customer complaint and the predetermined observation period (unit: /

[0048] The data includes the number of customer complaints (days), the time interval between the first complaint and the planned observation period (unit: / day), and changes in the number of complaints / complaint amount.

[0049] The preset number of days can be set by the user, and this application embodiment does not impose any restrictions on this. For example, the preset number of days can be 1 day, 7 days, 30 days, or 365 days.

[0050] The restricted and prohibited sales information for each product to be identified and its associated store includes, but is not limited to: whether each product to be identified and its associated store infringes on copyright, the infringement information at the time of infringement, and the restricted and prohibited sales labels, etc.

[0051] In this embodiment, in order for each prediction model in the following text to fully understand each product to be identified, it is necessary to obtain sufficient product information for each product to be identified. The product information and associated store information mentioned above can be used as a reference, but in actual applications, they are not limited to the examples mentioned above.

[0052] S12, input the product information of each product to be identified into each of the pre-trained prediction models to obtain the predicted risk probability of each prediction model for each product to be identified.

[0053] In some embodiments, the multiple prediction models can be binary classification models, including, but not limited to, models such as XGBoost, LightGBM, CatBoost, RandomForest, and k-Nearest Neighbors. For example, the number of multiple prediction models can be 13, and in practice, it can be more or fewer. A binary classification model is a machine learning model used to classify input data into two categories. Binary classification models are a form of supervised learning where the training data can be labeled, and the labels can be represented using labels or categories, such as "yes" and "no".

[0054] In one embodiment of this application, the product information of each product to be identified can be vectorized to obtain an information vector corresponding to the product information of each product to be identified. Different numbers or letters can be used to represent different product information in each information vector. Then, each information vector is input into each prediction model to obtain the predicted risk probability output by each prediction model for each product to be identified.

[0055] In other embodiments of this application, if a prediction model is an XGBoost model, which includes a feature extraction layer, a fully connected layer, and a classification layer, the electronic device can directly input the product information of each product to be identified into each of the pre-trained prediction models to obtain the predicted risk probability output by each prediction model for each product to be identified. Specifically, the electronic device uses the feature extraction layer to extract features from the product information of each product to be identified, obtaining a feature vector. Then, the electronic device uses the fully connected layer to perform a fully connected operation on the feature vector to reduce its dimensionality, obtaining a classification vector. Finally, the classification layer is used to classify the classification vector to obtain the predicted risk probability of the product to be identified.

[0056] The feature extraction layer can be a convolutional layer, and the classification layer can be a Softmax classifier.

[0057] In this embodiment, by using multiple prediction models to identify risks in multiple products, it is no longer necessary to rely on manual identification of products, which takes a lot of time. Therefore, the efficiency of identifying product risks can be improved.

[0058] S13, sort the multiple predicted risk probabilities of multiple products to be identified on the same prediction model, and determine an initial sorting result for each product to be identified.

[0059] In some embodiments, the electronic device may sort multiple predicted risk probabilities in ascending or descending order to determine an initial sorting result for each item to be identified. Since each item to be identified corresponds to one initial sorting result for each prediction model, after repeated sorting, each item to be identified will have multiple initial sorting results corresponding to multiple prediction models. The number of multiple initial sorting results is the same as the number of multiple prediction models.

[0060] For example, if there are four items to be identified, namely the first item, the second item, the third item, and the fourth item (e.g., the first item is a water cup, the second item is a washing machine, the third item is a mobile phone, and the fourth item is a television), and there are three prediction models: the first prediction model, the second prediction model, and the third prediction model. The prediction probability of the first item corresponding to the first prediction model is 0.76, the prediction probability of the second item is 0.85, the prediction probability of the third item is 0.9, and the prediction probability of the fourth item is 0.82. After sorting the four predicted risk probabilities of the first, second, third, and fourth products in the first prediction model in descending order, we obtain probabilities of 0.9, 0.85, 0.82, and 0.76, respectively. Therefore, the initial ranking result of the first product in the first prediction model is 4, the second product is 2, the third product is 1, and the fourth product is 3. Similarly, the initial ranking results of the first, second, third, and fourth products in the first and third prediction models can be obtained.

[0061] In this embodiment, each predicted risk probability is converted into an initial ranking result. Using each initial ranking result is a smoother approach compared to directly using each predicted risk probability to measure the risk level of each product to be identified.

[0062] S14, determine whether multiple initial ranking results of multiple prediction models corresponding to each product to be identified have been obtained.

[0063] In some embodiments, before multiple initial ranking results of multiple prediction models corresponding to each product to be identified are obtained, the process returns to step S13, and the electronic device continues to execute the step of determining an initial ranking result for each product to be identified until multiple initial ranking results of multiple prediction models corresponding to each product to be identified are obtained. When multiple initial ranking results of multiple prediction models corresponding to each product to be identified are obtained, the electronic device executes step S15.

[0064] S15, calculate the risk ranking result for each product to be identified based on the multiple initial ranking results for each product to be identified.

[0065] In some embodiments, the electronic device calculates the risk ranking result for each item to be identified based on multiple initial ranking results for each item to be identified, including: the electronic device determines the risk ranking result based on the multiple initial ranking results for each item to be identified and the number of multiple initial ranking results.

[0066] In one embodiment, the electronic device can determine the risk ranking result for each item to be identified by the ratio between the sum of multiple initial ranking results and the number of multiple initial ranking results.

[0067] For example, continuing with the above embodiments, if the multiple products to be identified include a first product to be identified, a second product to be identified, a third product to be identified, and a fourth product to be identified, and the number of prediction models is three (namely, a first prediction model, a second prediction model, and a third prediction model), and the initial ranking result of the first product to be identified in the first prediction model is 4, the initial ranking result of the first product to be identified in the second prediction model is 3, and the initial ranking result of the first product to be identified in the third prediction model is 2, then the sum of the three initial ranking results of the first product to be identified is 4 + 3 + 2 = 9, and the risk ranking result of the first product to be identified is the sum 9 / number of models 3 = 3. Similarly, the risk ranking results of the second, third, and fourth products to be identified can be obtained respectively.

[0068] In this embodiment, since the risk ranking result of each product to be identified integrates multiple initial ranking results corresponding to multiple prediction models, the risk ranking result of each product to be identified has higher representativeness and accuracy.

[0069] In another embodiment of this application, the electronic device calculates the risk ranking result of each product to be identified based on multiple initial ranking results of each product to be identified, including: the electronic device determines a first statistical value based on each initial ranking result of each product to be identified and the corresponding predicted risk probability, and determines the risk ranking result based on the second statistical value corresponding to multiple predicted risk probabilities and the first statistical value.

[0070] In some embodiments, the electronic device calculates the probability product between each initial ranking result of each item to be identified and its corresponding predicted risk probability, and determines the sum of the multiple probability products corresponding to multiple initial ranking results as a first statistical value. A second statistical value is the sum of multiple predicted risk probabilities. The electronic device determines the risk ranking result for each item to be identified by the ratio between the first statistical value and the second statistical value.

[0071] For example, continuing with the above embodiment, if the predicted risk probability of the first product to be identified in the first prediction model is 0.76 and the initial ranking result is 4, if the predicted risk probability of the first product to be identified in the second prediction model is 0.72 and the initial ranking result is 3, and if the predicted risk probability of the first product to be identified in the third prediction model is 0.8 and the initial ranking result is 2, then the probability product between the predicted risk probability of the first product to be identified in the first prediction model (0.76) and the initial ranking result (4) is 0.76 * 4 = 3.04, the probability product between the predicted risk probability of the first product to be identified in the second prediction model (0.76) and the initial ranking result (4) is 0.72 * 3 = 2.16, and the probability product between the predicted risk probability of the first product to be identified in the third prediction model (0.8) and the initial ranking result (2) is 0.8 * 2 = 1.6. The first statistical value is...

[0072] 3.04 + 2.16 + 1.6 = 6.8, the second statistical value is 0.76 + 0.72 + 0.8 = 2.28, and the ratio between the first statistical value 6.8 and the second statistical value 2.28 is approximately 3. Therefore, the risk ranking result for the first product to be identified is 3. Similarly, the risk ranking results for the second, third, and fourth products to be identified can be obtained.

[0073] In other embodiments of this application, the electronic device can also perform a weighted average calculation on multiple initial ranking results of each product to be identified according to preset weights to obtain a risk ranking result for each product to be identified. The preset weights can be set independently, and this application does not impose any restrictions on them. For example, the weights of models with high prediction accuracy can be set higher, and the weights of models with low prediction accuracy can be set lower.

[0074] S16. Based on the risk ranking results, identify the risks of multiple products to be identified.

[0075] In some embodiments, the electronic device sorts multiple risk ranking results of multiple products to be identified, and selects a preset number (e.g., 100) of the top-ranked products from the risk ranking results in descending order as target high-risk products. The list of target high-risk products is then pushed to a preset terminal device, such as a pre-defined communication device. In other embodiments, the list of high-risk products can also be pushed to a preset email address or stored in a preset storage space, such as a cloud drive.

[0076] The preset quantity can be set and / or modified according to the needs of identifying high-risk products, and can also be set based on the daily workload of the e-commerce platform's risk control personnel. For example, if the daily workload of risk control personnel is 100 items, then the preset quantity is 100 items. A list of high-risk products can be generated based on the names of multiple target high-risk products.

[0077] The electronic device can also identify the risk of multiple stores associated with the products to be identified based on the risk ranking results. The specific method for identifying the risk of associated stores will be described in detail in steps S161-S164 below.

[0078] In the embodiments of this application, since the product information of each product to be identified includes multiple types of information from different dimensions such as customer complaint information and infringement information, when the product information of each product to be identified is input into each prediction model, each prediction model can fully understand the abnormal transaction situation such as customer complaint situation and refund situation of each product to be identified, and accurately predict the risk of each product based on the abnormal transaction situation it understands. When there are many abnormal transaction situations of a product to be identified, the prediction model will consider that the product to be identified has a high risk, thereby enabling the initial screening of target high-risk products that are determined to have a greater risk by multiple prediction models from multiple products to be identified.

[0079] In this embodiment, by pushing a list of high-risk products to preset terminal devices, it is beneficial to remind users (e.g., risk control personnel) to further confirm the high-risk products in a timely manner. This allows for the identification of product risks through multiple predictive models, eliminating the need for extensive manual screening and identification of product information, thus improving the efficiency of risk product identification. When the user (risk control personnel) confirms that the target high-risk product pushed by the model is indeed high-risk, the user-confirmed high-risk product is promptly dealt with (e.g., reminding the seller, removing the product from the platform, and suspending payment), thereby ensuring accurate and timely handling of high-risk products. Specific handling measures can be determined based on factors such as the number of high-risk products. Furthermore, the user-confirmed high-risk products can also be used to assess the risk level of associated stores, facilitating timely handling of high-risk stores. By promptly handling high-risk products and high-risk stores, the ability to handle high-risk products is effectively improved, and more customer complaints and other issues are avoided, thereby enhancing the security of e-commerce transactions.

[0080] Through the above implementation method, since multiple prediction models are trained using different training datasets, there are parameter differences among the multiple prediction models. Using multiple prediction models for product risk identification can increase the number of products identified and the identification coverage. By using multiple prediction models to identify the risks of multiple products, it is no longer necessary to rely on a large amount of manual time for product identification, thus improving the speed of product risk identification. Furthermore, since the risk ranking result for each product to be identified integrates multiple initial ranking results corresponding to multiple prediction models, the risk ranking result for each product has higher representativeness and accuracy, thus improving the accuracy of product risk identification.

[0081] Before using multiple prediction models to identify multiple items, multiple machine learning algorithms need to be trained to obtain the multiple prediction models. For example... Figure 3 The diagram shown is a flowchart of a training method for multiple prediction models provided in an embodiment of this application, including the following steps:

[0082] S21. For each product listed within the preset observation period, a time interval is determined based on the preset observation period and the preset performance period corresponding to the preset observation period, and the historical transaction records of each product within the time interval are obtained.

[0083] In some embodiments, there can be multiple preset observation periods and preset performance periods, and the preset performance periods corresponding to multiple preset observation periods can be the same or different.

[0084] For example, if multiple preset observation periods are January 1, 2023, and January 2, 2023, the preset performance period corresponding to the preset observation period January 1, 2023 can be 30 days, and the preset performance period corresponding to the preset observation period January 2, 2023 can be 29 days. When the preset performance period corresponding to the preset observation period January 1, 2023 is 30 days, the time interval between the preset observation period January 1, 2023, and the preset performance period of 30 days is January 1, 2023 - January 30, 2023. When the preset performance period corresponding to the preset observation period January 2, 2023 is 29 days, the time interval between the preset observation period January 2, 2023, and the preset performance period of 29 days is January 2, 2023 - January 30, 2023. Alternatively, if multiple preset observation periods are January 1, 2023 and January 2, 2023, the preset performance period corresponding to both the preset observation periods of January 1, 2023 and January 2, 2023 can be 30 days. The time interval corresponding to the preset observation period of January 1, 2023 and the preset performance period of 30 days is January 1, 2023 to January 30, 2023. The time interval corresponding to the preset observation period of January 2, 2023 and the preset performance period of 30 days is January 2, 2023 to January 31, 2023.

[0085] In some embodiments, in order to enrich the multiple training datasets described below, there may be multiple preset observation periods, not limited to the examples above. For example, multiple preset observation periods may include each day from January 1, 2023 to March 31, 2023.

[0086] In some embodiments, for each product listed on the e-commerce platform within a preset observation period, the electronic device retrieves the historical transaction records of each product within the time interval from the e-commerce platform's database. Alternatively, the electronic device may also receive information uploaded by the user as historical transaction records.

[0087] S22, based on historical transaction records, determine the highest-risk and lowest-risk products among all listed products within the preset observation period.

[0088] In some embodiments, historical transaction records include the number of transactions and the number of customer complaints corresponding to the number of transactions. The electronic device determines the first high-risk product and the first low-risk product among all listed products within the preset observation period based on the historical transaction records by: determining the customer complaint ratio of each product based on the number of customer complaints and the number of transactions; identifying products with a customer complaint ratio greater than a preset threshold as the first high-risk product among all listed products; and identifying products with a customer complaint ratio less than or equal to the preset threshold as the first low-risk product among all listed products.

[0089] The customer complaint percentage can be the ratio between the number of customer complaints and the number of transactions. For example, if the transaction volume of any product is 1000 and the number of customer complaints for any product is 10, then the customer complaint percentage for any product is 10 / 1000*100% = 1%. The preset threshold can be set by the user, and this application does not impose any restrictions on it. For example, the preset threshold can be 20%.

[0090] S23, generate multiple training datasets based on the product information of the first high-risk product and the first low-risk product.

[0091] In some embodiments, the electronic device can obtain product information for a first high-risk product and a first low-risk product from the database of an e-commerce platform. Furthermore, embodiments of this application will perform data preprocessing (e.g., data cleaning) on ​​the obtained product information for the first high-risk product and the first low-risk product. For example, product information corresponding to products with both a transaction volume and a customer complaint volume of 1 can be filtered out.

[0092] In some embodiments, under normal circumstances, the number of first-highest-risk products is typically far less than the number of first-lowest-risk products. For example, in most cases, the proportion of first-highest-risk products in all listed products is usually 1%, and even when there are a sufficient number of listed products, the proportion of first-highest-risk products is only 0.1%. Therefore, the product information for first-highest-risk products is also far less than that for first-lowest-risk products, resulting in an extreme imbalance between the amount of product information for first-highest-risk products and first-lowest-risk products.

[0093] Directly using the product information of the most high-risk and least high-risk goods in an extremely imbalanced manner to train multiple machine learning algorithms will result in low prediction accuracy of the trained models. Alternatively, sampling the product information of the least high-risk goods and combining it with the information of the most high-risk goods as training samples, then using these samples to train multiple machine learning algorithms, will lead to insufficient learning of the data distribution of the least high-risk goods in the training samples. This will result in poor prediction accuracy for the multiple models. Specifically, when using multiple prediction models to predict the goods to be identified, a large number of low-risk goods will be misclassified as high-risk goods, resulting in low efficiency for risk control personnel in identifying high-risk goods recommended by multiple prediction models.

[0094] To address this issue, this embodiment of the application divides the product information of the first low-risk product according to a preset first proportion coefficient. Then, it combines each portion of the first low-risk product information with the product information of the first high-risk product to construct multiple training datasets. Each training dataset is then used to train a corresponding prediction model. Dividing the product information of the first low-risk product according to the preset first proportion coefficient ensures that the proportion of the first high-risk product information in each training dataset falls within the range corresponding to the first proportion coefficient. This first proportion coefficient is obtained through multiple verifications and tests, enabling each prediction model to fully learn the product information of the low-risk products, thereby improving the prediction accuracy of each model.

[0095] S24. Based on the data distribution of each training dataset, a machine learning algorithm is matched for each training dataset in a preset network framework. The matched machine learning algorithm is trained using each training dataset to obtain multiple prediction models.

[0096] In some embodiments, the default network framework is the GoogleAutoGluon framework.

[0097] GoogleAutoGluon is an open-source machine learning platform that supports various algorithms and frameworks, such as TensorFlow and PyTorch, and can automatically tune hyperparameters to optimize model performance. GoogleAutoGluon analyzes and learns the data distribution of the input training datasets and automatically selects appropriate machine learning algorithms for training based on the data distribution of each training dataset, resulting in a predictive model for each dataset. Furthermore, GoogleAutoGluon provides a visual interface to display the data distribution of each input training dataset, and its machine learning algorithm search function also allows users to manually select machine learning algorithms.

[0098] In other embodiments of this application, after training multiple prediction models, their predictive performance needs to be verified before they are put into use. To evaluate the predictive performance of the trained prediction models, the electronic device can acquire a verification dataset and input it into each of the prediction models. A preset metric is used to evaluate the predictive performance of each prediction model on the verification dataset. Only when the predictive performance of the multiple prediction models meets preset conditions will the multiple prediction models be used to predict the risk of the product to be identified.

[0099] The validation dataset consists of product information for multiple products listed outside of several pre-defined observation periods. Pre-defined metrics include, but are not limited to, the Area Under the ROC Curve (AUC) and the Kolmogorov-Smirnov distance (KS). AUC is a metric that measures the model's prediction accuracy. The ROC curve is a curve with the F1 score on the x-axis and the True Positive Rate (TPR) on the y-axis, where TPR represents the proportion of samples that were actually positive (low-risk products) but were correctly predicted as positive (low-risk products). The AUC value ranges from [0, 1], and the closer the AUC value is to 1, the better the prediction performance of the model.

[0100] The Kolmogorov-Smirnov distance can be used to compare the degree of difference between the model's predictions and the true labels. The value of KS ranges from [0, 1]. The closer the value of KS is to 0, the better the prediction performance of the model.

[0101] The preset conditions can be set by the user, and this application does not impose any restrictions on them. For example, when the AUC value of a prediction model is greater than 0.8, the electronic device determines that the prediction model can perform risk prediction for the product to be identified, and / or, when the Kolmogorov-Smirnov distance of a prediction model is less than 0.2, the electronic device determines that the prediction model can perform risk prediction for the product to be identified.

[0102] As mentioned above, multiple training datasets need to be generated based on the product information of the highest-risk and lowest-risk products. Figure 4 The diagram shown is a flowchart of a method for generating multiple training datasets according to an embodiment of this application, including the following steps:

[0103] S231, construct a high-risk dataset based on the product information of the first high-risk product, and construct a low-risk dataset based on the product information of the first low-risk product.

[0104] In some embodiments, if there are multiple first high-risk products, the electronic device uses a dataset consisting of the product information of the multiple first high-risk products as the high-risk dataset. The construction process of the low-risk dataset is basically the same as that of the high-risk dataset, so it will not be described again in this application. Typically, the number of first low-risk products is much greater than the number of first high-risk products, therefore the data volume of the low-risk dataset is also much greater than the data volume of the high-risk dataset.

[0105] S232, count the first quantity of the first high-risk commodity, and divide the low-risk dataset into multiple sub-datasets based on the first quantity and the preset first ratio coefficient.

[0106] In some embodiments, the first scaling factor can be obtained through multiple tests and verifications. For example, the first scaling factor can be 5%.

[0107] In other embodiments of this application, the first scaling factor can be set by the user, and this application does not impose any restrictions on it.

[0108] In some embodiments, each subset contains a first low-risk item. The electronic device divides the low-risk dataset into multiple subsets based on a first quantity and a preset first ratio coefficient. This includes: the electronic device calculating the ratio of the first quantity to the first ratio coefficient, using the difference between the calculated ratio and the first quantity as the estimated quantity of items in each subset, and then completely splitting the low-risk dataset based on the estimated quantity of items in each subset to obtain multiple subsets. The proportion of the first quantity in the total quantity of the training dataset composed of the high-risk dataset and any subset is approximately equal to the first ratio coefficient.

[0109] For example, if the first proportion coefficient is 5%, the first quantity of the first high-risk item is 10, and the total quantity of the first low-risk item in the low-risk dataset is 1330, the ratio of the first quantity 10 to the first proportion coefficient 5% is 200. Therefore, the estimated quantity of items in each subset is 200 - 10 = 190. To ensure that the 1330 first low-risk items in the low-risk dataset can be completely split and that the ratio of each subset to the first quantity 10 is approximately 5%, we calculate the result of dividing 1330 by 190, which is an integer 7. Therefore, the low-risk dataset can be divided into 7 subsets, each containing 190 items.

[0110] In some embodiments, due to the uncertainty of the products listed on e-commerce platforms, it is difficult to equally split the low-risk dataset based on the first quantity of the first high-risk product. Therefore, the ratio of the data volume of each sub-dataset to the first quantity is unlikely to perfectly match the first proportional coefficient. Thus, this embodiment only needs to ensure that the proportion of the first quantity in the total number of the training dataset composed of the high-risk dataset and any of the split sub-datasets is approximately the first proportional coefficient. In practical applications, the splitting of the low-risk dataset may encounter various situations; therefore, the low-risk dataset can be flexibly split according to actual needs.

[0111] S233 combines the high-risk dataset with any subset of datasets to obtain a training dataset.

[0112] In some embodiments, the electronic device combines the high-risk dataset with each subset to obtain multiple training datasets.

[0113] In this embodiment, since the first proportion coefficient can be obtained through multiple tests and verifications, and multiple training datasets are used to train the corresponding machine learning algorithms to obtain the prediction models corresponding to the machine learning algorithms, it is ensured that the proportion of the first quantity in the training dataset composed of the high-risk dataset and any subset dataset obtained by splitting is approximately the first proportion coefficient. This enables each machine learning algorithm to fully learn the product information of the first high-risk product and the product information of the first low-risk product, thereby improving the prediction effect of each prediction model.

[0114] like Figure 5 The diagram shown is a flowchart of a method for risk identification of associated stores according to an embodiment of this application, including the following steps:

[0115] S161, calculate the number of high-risk goods based on the preset second ratio coefficient and the second quantity of multiple goods to be identified, and calculate the number of low-risk goods based on the preset third ratio coefficient and the second quantity.

[0116] In some embodiments, the second and third proportionality coefficients are obtained through multiple verifications and tests. For example, the second proportionality coefficient can be 30% (0.3), and the third proportionality coefficient can be 50%. The second quantity is the number of multiple items to be identified.

[0117] In this embodiment, the electronic device can use the product of the second proportional coefficient and the second quantity as the quantity of high-risk goods, and the product of the third proportional coefficient and the second quantity as the quantity of low-risk goods.

[0118] S162, based on the number of high-risk goods, select the goods to be identified corresponding to the top-ranked risk ranking results from the risk ranking results sorted from low to high as the second highest-risk goods, and based on the number of low-risk goods, select the goods to be identified corresponding to the bottom-ranked risk ranking results from the risk ranking results sorted from low to high as the second lowest-risk goods.

[0119] In some embodiments, in the risk ranking results sorted from low to high, the higher the risk of the ranking result, the higher the risk. The electronic device selects a first number (e.g., the number of high-risk goods) of the top-ranked risk ranking results from the low-to-high risk ranking results as the second high-risk goods, and selects a second number (e.g., the number of low-risk goods) of the bottom-ranked risk ranking results from the low-to-high risk ranking results as the second low-risk goods.

[0120] For example, the plurality of items to be identified is seven, namely, the first item to be identified, the second item to be identified, the third item to be identified, ..., the seventh item to be identified. The risk ranking result of the first item to be identified is 3, the risk ranking result of the second item to be identified is 4, the risk ranking result of the third item to be identified is 1, the risk ranking result of the fourth item to be identified is 2, the risk ranking result of the fifth item to be identified is 6, the risk ranking result of the sixth item to be identified is 7, and the risk ranking result of the seventh item to be identified is 5. When the number of high-risk items is three, the three items to be identified corresponding to the top three risk ranking results from the risk ranking results from low to high are selected as the second high-risk items. Therefore, the second high-risk items are the third item to be identified, the fourth item to be identified, and the first item to be identified. When the number of low-risk items is two, the two items to be identified corresponding to the bottom two risk ranking results from the risk ranking results from low to high are selected as the second low-risk items. Therefore, the second low-risk items are the fifth item to be identified and the sixth item to be identified. In some embodiments, there are multiple second high-risk items and multiple second low-risk items.

[0121] In this embodiment, since the second and third proportional coefficients are obtained through verification and testing, selecting the second high-risk product by the quantity of high-risk products and selecting the second low-risk product by the quantity of low-risk products can ensure the accuracy of the second high-risk product and the second low-risk product to a certain extent.

[0122] S163, if multiple products to be identified are associated with multiple stores, calculate the first proportion of the second highest-risk products in each associated store, and calculate the second proportion of the second lowest-risk products in each associated store.

[0123] In some embodiments, since there are typically multiple stores listing products on an e-commerce platform during the predetermined observation period, multiple products to be identified are usually associated with multiple stores. The electronic device determines the stores associated with each second highest-risk product and each second lowest-risk product, using the ratio of the total number of products listed in each associated store during the predetermined observation period to the number of second highest-risk products in that store as a first percentage, and the ratio of the total number of products listed in each associated store during the predetermined observation period to the number of second lowest-risk products in that store as a second percentage.

[0124] For example, if each associated store has a total of 1,000 products listed during the planned observation period, and the store has 1 product with the second highest risk, and the store has 999 products with the second highest risk, then the first percentage is 1 / 999.

[0125] In this embodiment, a larger first percentage corresponds to more second-highest-risk products in each associated store, and a smaller second percentage corresponds to more second-lowest-risk products in each associated store. Therefore, the first percentage and the second percentage can be used as two key dimensions to measure the risk level of each associated store.

[0126] S164, risk identification is performed for each associated store based on the first percentage and / or the second percentage.

[0127] In some embodiments, the electronic device may use various methods to identify the risk of each associated store based on a first percentage and / or a second percentage. For example, the electronic device may compare the first percentage of each associated store with a first preset value and compare the second percentage of that store with a second preset value. If the first percentage is greater than the first preset value and the second percentage is less than or equal to the second preset value, the electronic device determines that the store is a high-risk store. Alternatively, if the first percentage of each associated store is greater than the first preset value, the electronic device determines that the store is a high-risk store. Alternatively, if the second percentage of each associated store is less than the second preset value, the electronic device determines that the store is a high-risk store.

[0128] The examples above are merely illustrations; in reality, electronic devices can use many more flexible methods to conduct risk assessments for each associated store.

[0129] In this embodiment, identifying high-risk stores on e-commerce platforms is beneficial for taking timely measures to deal with them (such as sending warning messages to high-risk stores to make rectifications or close the stores), thereby maintaining a good operating environment for e-commerce platforms and reducing losses caused by customer complaints.

[0130] like Figure 6 The diagram shown is a functional block diagram of a commodity risk identification device provided in an embodiment of this application. The commodity risk identification device 11 includes an acquisition unit 110, an input unit 111, a sorting unit 112, a calculation unit 113, and a risk identification unit 114. The module / unit referred to in this application refers to a module / unit capable of being... Figure 1 The processor 103 in the middle acquires a series of computer-readable instruction segments that are capable of performing a fixed function, and these segments are stored in Figure 1 The memory 102 is used for this purpose. In this embodiment, the functions of each module / unit will be described in detail in subsequent embodiments.

[0131] The acquisition unit 110 is used to acquire product information of multiple products to be identified.

[0132] In some embodiments, the acquisition unit is further configured to, for each product listed within a preset observation period, determine a time interval based on the preset observation period and the preset performance period corresponding to the preset observation period, acquire historical transaction records for each product within the time interval, determine the first high-risk product and the first low-risk product among all listed products within the preset observation period based on the historical transaction records, generate multiple training datasets based on the product information of the first high-risk product and the first low-risk product, match a machine learning algorithm for each training dataset in a preset network framework based on the data distribution of each training dataset, and train the matched machine learning algorithm using each training dataset to obtain multiple prediction models.

[0133] In some embodiments, historical transaction records include the number of transactions and the number of customer complaints corresponding to the number of transactions. The acquisition unit 110 further includes a determination subunit and a generation subunit. The determination subunit is used to determine the customer complaint ratio of each product based on the number of customer complaints and the number of transactions. Products with a customer complaint ratio greater than a preset threshold are identified as first high-risk products among all listed products, and products with a customer complaint ratio less than or equal to the preset threshold are identified as first low-risk products among all listed products. The generation subunit is used to construct a high-risk dataset based on the product information of the first high-risk products and a low-risk dataset based on the product information of the first low-risk products. The generation subunit is used to count the first quantity of the first high-risk products and divide the low-risk dataset into multiple sub-datasets based on the first quantity and a preset first ratio coefficient. The high-risk dataset is combined with any one of the sub-datasets to obtain a training dataset.

[0134] The input unit 111 is used to input the product information of each product to be identified into each of the pre-trained prediction models to obtain the predicted risk probability of each prediction model for each product to be identified.

[0135] The sorting unit 112 is used to sort the multiple predicted risk probabilities of multiple products to be identified on the same prediction model, determine an initial sorting result for each product to be identified, and repeat this step until multiple initial sorting results for each product to be identified on multiple prediction models are obtained.

[0136] The calculation unit 113 is used to calculate the risk ranking result of each product to be identified based on multiple initial ranking results of each product to be identified.

[0137] In some embodiments, the calculation unit 113 is further configured to determine a risk ranking result based on multiple initial ranking results and the number of multiple initial ranking results for each product to be identified, or to determine a first statistical value based on each initial ranking result of each product to be identified and the corresponding predicted risk probability, and to determine a risk ranking result based on a second statistical value and a first statistical value corresponding to multiple predicted risk probabilities.

[0138] The risk identification unit 114 is used to identify risks in multiple products to be identified based on the risk ranking results.

[0139] In some embodiments, the risk identification unit is further configured to select a preset number of products to be identified corresponding to the top-ranked risk ranking results from the risk ranking results sorted from low to high, as target high-risk products, and push the list of risk products of the target high-risk products to a preset terminal device.

[0140] In some embodiments, the risk identification unit is further configured to identify risks associated with multiple stores for products to be identified based on risk ranking results.

[0141] In some embodiments, the risk identification unit is further configured to calculate the number of high-risk goods based on a preset second ratio coefficient and a second quantity of multiple goods to be identified, and calculate the number of low-risk goods based on a preset third ratio coefficient and the second quantity; select goods to be identified corresponding to the top-ranked risk ranking results from the risk ranking results sorted from low to high based on the number of high-risk goods, as the second high-risk goods; and select goods to be identified corresponding to the bottom-ranked risk ranking results from the risk ranking results sorted from low to high based on the number of low-risk goods, as the second low-risk goods; if multiple goods to be identified are associated with multiple stores, calculate the first proportion of the second high-risk goods in each associated store, and calculate the second proportion of the second low-risk goods in each associated store; and perform risk identification for each associated store based on the first proportion and / or the second proportion.

[0142] In the above embodiments, the descriptions of each embodiment have different focuses. For parts that are not described in detail or recorded in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0143] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0144] In the embodiments provided in this application, it should be understood that the disclosed devices / electronic devices and methods can be implemented in other ways. For example, the device / electronic device embodiments described above are merely illustrative. For instance, the division of modules or units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the displayed or discussed mutual couplings or direct couplings or communication connections may be through some interfaces; indirect couplings or communication connections between devices or units may be electrical, mechanical, or other forms.

[0145] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0146] The above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. A commodity risk identification method, characterized by, The method includes: Obtain product information for multiple products to be identified; The product information of each product to be identified is input into each of the pre-trained prediction models to obtain the predicted risk probability output by each prediction model for each product to be identified. The multiple predicted risk probabilities of the multiple products to be identified on the same prediction model are sorted to determine an initial sorting result for each product to be identified. This step is repeated until multiple initial sorting results of each product to be identified on the multiple prediction models are obtained. Calculating a risk ranking result for each product to be identified based on the plurality of initial ranking results for each product to be identified includes: determining the risk ranking result based on the plurality of initial ranking results for each product to be identified and the number of the plurality of initial ranking results; or, determining a first statistical value based on each initial ranking result for each product to be identified and the corresponding predicted risk probability, and determining the risk ranking result based on the second statistical value corresponding to the plurality of predicted risk probabilities and the first statistical value. Based on the risk ranking results, risk identification is performed on the plurality of products to be identified; The training method for the multiple prediction models includes: constructing a high-risk dataset based on the product information of the first high-risk product among the products listed during the preset observation period; constructing a low-risk dataset based on the product information of the first low-risk product among the products listed during the preset observation period; counting the first quantity of the first high-risk product; dividing the low-risk dataset into multiple sub-datasets based on the first quantity and a preset first ratio coefficient; combining the high-risk dataset with any one of the sub-datasets to obtain a training dataset; forming multiple training datasets; matching a machine learning algorithm to each training dataset in a preset network framework based on the data distribution of each training dataset; and training the matched machine learning algorithm using each training dataset to obtain the multiple prediction models.

2. The merchandise risk identification method of claim 1, wherein, The methods for determining the first high-risk commodity and the first low-risk commodity include: For each product listed within the preset observation period, a time interval is determined based on the preset observation period and the preset performance period corresponding to the preset observation period, and the historical transaction records of each product within the time interval are obtained. Based on the historical transaction records, the first high-risk product and the first low-risk product are determined from all listed products within the preset observation period.

3. The merchandise risk identification method of claim 2, wherein, The historical transaction records include the number of transactions and the number of customer complaints corresponding to the number of transactions. The step of determining the first high-risk product and the first low-risk product from all listed products within the preset observation period based on the historical transaction records includes: The proportion of customer complaints for each product is determined based on the number of customer complaints and the number of transactions. Products with a customer complaint rate greater than a preset threshold are identified as the first high-risk products among all listed products, while products with a customer complaint rate less than or equal to the preset threshold are identified as the first low-risk products.

4. The commodity risk identification method as described in claim 1, characterized in that, The step of identifying the risks of the plurality of goods to be identified based on the risk ranking results includes: Select a preset number of products corresponding to the top-ranked risk ranking results from the risk ranking results from low to high as target high-risk products, and push the list of risk products of the target high-risk products to the preset terminal device.

5. The commodity risk identification method as described in claim 1, characterized in that, The method further includes: Based on the risk ranking results, risk identification is performed on the stores associated with the multiple products to be identified.

6. The commodity risk identification method as described in claim 5, characterized in that, The step of identifying the risk of the stores associated with the multiple products to be identified based on the risk ranking results includes: The number of high-risk goods is calculated based on a preset second ratio coefficient and a second quantity of the plurality of goods to be identified, and the number of low-risk goods is calculated based on a preset third ratio coefficient and the second quantity. Based on the quantity of high-risk goods, select the top-ranked risk ranking results from the risk ranking results sorted from low to high as the second high-risk goods, and based on the quantity of low-risk goods, select the bottom-ranked risk ranking results from the risk ranking results sorted from low to high as the second low-risk goods. If the multiple products to be identified are associated with multiple stores, calculate the first proportion of the second high-risk products in each associated store, and calculate the second proportion of the second low-risk products in each associated store. Risk identification is performed on each of the associated stores based on the first percentage and / or the second percentage.

7. A commodity risk identification device, operating in an electronic device, said device being used to implement the commodity risk identification method as described in any one of claims 1 to 6, characterized in that, The device includes: The acquisition unit is used to acquire product information for multiple products to be identified; The input unit is used to input the product information of each product to be identified into each of the pre-trained prediction models to obtain the predicted risk probability output by each prediction model for each product to be identified. The sorting unit is used to sort the multiple predicted risk probabilities of the multiple products to be identified on the same prediction model, determine an initial sorting result for each product to be identified, and repeat this step until multiple initial sorting results of each product to be identified on the multiple prediction models are obtained. The calculation unit is used to calculate the risk ranking result of each product to be identified based on the plurality of initial ranking results of each product to be identified; The risk identification unit is used to identify the risks of the plurality of products to be identified based on the risk ranking results.

8. An electronic device, characterized in that, The electronic device includes: Memory, storing at least one instruction; and The processor executes the at least one instruction to implement the commodity risk identification method as described in any one of claims 1 to 6.

Images