A method and device for processing orders based on neural network model

Abstract

The present invention provides an order processing method based on a neural network model, the method comprising: obtaining a first category order containing a value correction parameter and a second category order containing a value correction parameter; determining feature data of the obtained order, and using the feature data of the obtained order to train a pre-built neural network model with an embedding layer to obtain an order recognition model with an embedding layer; receiving an order to be recognized containing a value correction parameter, inputting the feature data of the order to be recognized into the order recognition model, obtaining a feature vector of the order to be recognized output by the embedding layer of the order recognition model and a category measurement value of the order to be recognized output by the order recognition model; determining whether the order to be recognized is a first category order according to the category measurement value and the feature vector of the order to be recognized, and when the order to be recognized is a first category order, sending the order to be recognized to an audit platform for audit.

Description

Technical Field

[0001] The present invention relates to the technical field of e-commerce, and in particular to an order processing method and device based on a neural network model. Background Art

[0002] E-commerce websites often launch a variety of promotional activities (coupons, discounts, gifts, etc.). Some promotional activities support overlapping use. When these promotional activities are not set up reasonably or there are certain loopholes, they may be exploited by black market groups to conduct large-scale arbitrage, causing huge economic losses to merchants.

[0003] There are currently two ways to prevent the above situation from occurring: one is to discover loopholes in promotional activities through manual data analysis and experience, and set business rules accordingly; the other is to train machine learning models through historical data and use the trained models to identify arbitrage orders.

[0004] The above two methods can only be used to determine whether a single order has arbitrage risks, but they cannot link them together, and cannot distinguish between malicious orders for the purpose of arbitrage by group behavior and normal orders for the purpose of arbitrage by non-group behavior. Summary of the invention

[0005] In view of this, the purpose of the present invention is to provide an order processing method and device based on a neural network model, which can be used to identify and process malicious orders for the purpose of arbitrage from group behavior.

[0006] In order to achieve the above object, the present invention provides the following technical solutions:

[0007] A method for identifying group behavior orders, comprising:

[0008] Get the first type of orders with value correction parameters and the second type of orders with value correction parameters;

[0009] Determine the feature data of the acquired order, and use the feature data of the acquired order to train a pre-built neural network model with an embedding layer to obtain an order recognition model with an embedding layer;

[0010] Receiving an order to be identified including a value correction parameter, inputting feature data of the order to be identified into the order identification model, obtaining a feature vector of the order to be identified output by an embedding layer of the order identification model and a category metric value of the order to be identified output by the order identification model;

[0011] Whether the order to be identified is a first-category order is determined according to the category measurement value and feature vector of the order to be identified. When the order to be identified is a first-category order, the order to be identified is sent to the review platform for review.

[0012] An order processing device based on a neural network model, comprising:

[0013] A collecting unit, used for obtaining the first type of orders including the value modification parameters and the second type of orders including the value modification parameters;

[0014] A training unit, used to determine the characteristic data of the order acquired by the collection unit, and use the acquired characteristic data of the order to train a pre-built neural network model with an embedding layer to obtain an order recognition model with an embedding layer;

[0015] a processing unit, configured to receive an order to be identified including a value correction parameter, input feature data of the order to be identified into the order identification model, obtain a feature vector of the order to be identified output by the embedding layer of the order identification model and a category metric value of the order to be identified output by the order identification model;

[0016] The decision unit is used to determine whether the order to be identified is a first-category order according to the category measurement value and feature vector of the order to be identified. When the order to be identified is a first-category order, the order to be identified is sent to the review platform for review.

[0017] An electronic device comprises: at least one processor, and a memory connected to the at least one processor via a bus; the memory stores one or more computer programs executable by the at least one processor; and the at least one processor implements the steps in the above-mentioned group behavior order recognition method when executing the one or more computer programs.

[0018] A computer-readable storage medium stores one or more computer programs, which, when executed by a processor, implement the steps in the above-mentioned group behavior order identification method.

[0019] It can be seen from the above technical solution that in the present invention, by obtaining the first category orders and the second category orders containing value correction parameters, and using the feature data of these orders to train to obtain an order recognition model with an embedding layer, the order recognition model is used to determine the category metric value and feature vector of the order to be recognized, and when it is determined that the order to be recognized is a first category order according to the category metric value and feature vector of the order to be recognized, the order to be recognized is sent to the review platform for review. The present invention can be used to identify and process malicious orders for the purpose of arbitrage of group behavior. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0021] Figure 1 This is a flow chart of a method for identifying group behavior orders according to a first embodiment of the present invention;

[0022] Figure 2 is a flow chart of a method for identifying group behavior orders according to Embodiment 2 of the present invention;

[0023] Figure 3 is a schematic diagram of a neural network model with an embedding layer provided in an embodiment of the present invention;

[0024] Figure 4 is a flow chart of a method for identifying group behavior orders according to Embodiment 3 of the present invention;

[0025] Figure 5 is a flow chart of a method for identifying group behavior orders according to a fourth embodiment of the present invention;

[0026] Figure 6 is a flow chart of a method for identifying group behavior orders according to a fifth embodiment of the present invention;

[0027] Figure 7 It is a structural schematic diagram of a device for identifying group behavior orders according to an embodiment of the present invention;

[0028] Figure 8 It is a schematic diagram of the structure of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0029] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0030] See also Figure 1 , Figure 1 : is a flow chart of an order processing method based on a neural network model according to Embodiment 1 of the present invention, such as Figure 1 As shown, the method mainly includes the following steps:

[0031] Step 101, obtaining the first type of order and the second type of order containing value modification parameters;

[0032] Step 102: determining feature data of the acquired order, and using the acquired feature data of the order to train a pre-built neural network model with an embedding layer, to obtain an order recognition model with an embedding layer;

[0033] In this embodiment, step 101 to step 102 may be performed only once.

[0034] Step 103: receiving an order to be identified including a value correction parameter, inputting feature data of the order to be identified into the order identification model, obtaining a feature vector of the order to be identified output by the embedding layer of the order identification model and a category metric value of the order to be identified output by the order identification model;

[0035] Step 104: determine whether the order to be identified is a first-category order based on the category metric and feature vector of the order to be identified. If the order to be identified is a first-category order, send the order to be identified to the review platform for review.

[0036] In this embodiment, step 103 to step 104 can be executed cyclically.

[0037] from Figure 1 It can be seen from the method shown that in this embodiment, by obtaining the first category orders and the second category orders containing value correction parameters, and using the feature data of these orders to train to obtain an order recognition model with an embedding layer, the order recognition model is used to determine the category metric value and feature vector of the order to be recognized, and when it is determined that the order to be recognized is a first category order according to the category metric value and feature vector of the order to be recognized, the present invention can be used to identify and process malicious orders of group behavior.

[0038] See also Figure 2 , Figure 2 is a flow chart of a method for identifying group behavior orders according to Embodiment 2 of the present invention. Figure 2 As shown, the method mainly includes the following steps:

[0039] Step 201, obtaining the first type of order and the second type of order containing value modification parameters;

[0040] In this embodiment, the value correction parameter is a parameter used to correct the value of an order, and may specifically correspond to a promotion in an e-commerce platform, such as a 100 yuan discount for purchases over 199 yuan.

[0041] In this embodiment, malicious orders that contain value correction parameters and belong to group behavior are called first-class orders, and non-malicious orders that contain value correction parameters and do not belong to group behavior are called second-class orders. Here, a group refers to a group of people who are connected in a certain way to achieve specific goals and meet common needs. Group behavior refers to the common behavior of multiple people formed by the stimulation or influence of a certain factor. Orders placed by a group under the influence of the same factor (such as a certain promotional activity), through the same ordering channel, and using the same payment method are orders that belong to group behavior.

[0042] In this embodiment, after obtaining the first category of orders and the second category of orders containing value correction parameters, the feature data of these orders can be used to train a pre-built neural network model with an embedding layer.

[0043] In existing implementations, a neural network model usually includes an input layer, a hidden layer, and an output layer. In this embodiment, an embedding layer is added between the input layer and the hidden layer to build a neural network model such as Figure 3 The neural network model with an embedding layer is shown. The hidden layer of this neural network model with an embedding layer may include one or more hidden layers, each hidden layer may be a fully connected layer, wherein the activation function of the fully connected layer can use Relu, the activation function of the output layer can use the sigmoid function, and the sigmoid function expects the output to be 0 and 1. In addition, the loss function of the neural network model with an embedding layer can use the cross entropy loss, and the model weights are updated through the stochastic gradient descent strategy to complete the model training.

[0044] Step 2021: Determine the characteristic data of the acquired order;

[0045] In this embodiment, the feature data of an order may include multiple features as shown in the following table:

[0046]

[0047]

[0048]

[0049] Table 1

[0050] Step 2022: One-hot encode the acquired feature data of the order to obtain a sparse matrix;

[0051] Step 2023: input the sparse matrix into the embedding layer through the input layer of the neural network model for dimensionality reduction processing;

[0052] Step 2024: input the sparse matrix after dimensionality reduction processing into the hidden layer for feature extraction, and output the feature extraction result to the output layer, thereby realizing the training of the neural network model.

[0053] The above steps 2022 to 2024 are Figure 1 The specific details of "using the acquired feature data of the order to train the pre-built neural network model with the embedding layer" in step 102 are shown.

[0054] Step 2025, using the trained model as an order recognition model.

[0055] The above steps 2021 to 2025 are Figure 1 The specific refinement of step 102 is shown.

[0056] The neural network model with an embedding layer includes an input layer, an embedding layer, a hidden layer and an output layer;

[0057] In this embodiment, step 201 to step 202 may be performed only once.

[0058] Step 203: receiving an order to be identified that includes a value correction parameter, inputting feature data of the order to be identified into the order identification model, obtaining a feature vector of the order to be identified output by the embedding layer of the order identification model and a category metric value of the order to be identified output by the order identification model;

[0059] In this implementation, after the order to be identified is input into the order identification model, the data output by the embedding layer of the order identification model is input into the hidden layer for feature extraction, and the data output by the embedding layer of the order identification model is also output as a feature vector of the order to be identified.

[0060] Step 204: determine whether the order to be identified is a first-category order based on the category measurement value and feature vector of the order to be identified. When the order to be identified is a first-category order, send the order to be identified to the review platform for review.

[0061] In this embodiment, step 203 to step 204 may be executed cyclically.

[0062] from Figure 2 It can be seen from the method shown that in this embodiment, the trained order recognition model with an embedding layer is used to determine the category metric value and feature vector of the order to be recognized, so that it can be determined whether the order to be recognized is a first-class order based on the category metric value and feature vector of the order to be recognized. The present invention can be used to identify and process malicious orders of group behavior.

[0063] See also Figure 4 , Figure 4is a flow chart of a method for identifying group behavior orders according to Embodiment 3 of the present invention. Figure 4 As shown, the method mainly includes the following steps:

[0064] Step 401, obtaining the first type of order and the second type of order containing value modification parameters;

[0065] Step 402: determining feature data of the acquired order, and using the acquired feature data of the order to train a pre-built neural network model with an embedding layer, to obtain an order recognition model with an embedding layer;

[0066] Step 403: receiving an order to be identified that includes a value correction parameter, inputting feature data of the order to be identified into the order identification model, obtaining a feature vector of the order to be identified output by the embedding layer of the order identification model and a category metric value of the order to be identified output by the order identification model;

[0067] Step 4041: when the category metric value of the order to be identified is greater than a preset category metric threshold, determining the cool-down period of the order to be identified according to the category metric value of the order to be identified, and determining whether the order to be identified is a first category order according to the feature vector of the order to be identified and the identified orders whose cool-down period has not ended;

[0068] In this embodiment, when the category metric value of the order to be identified is not greater than a preset category metric threshold, it is determined that the order to be identified is not a first category order.

[0069] In this embodiment, the cool-down time of the order to be identified is determined according to the category measurement value of the order to be identified, specifically including:

[0070] Using the formula Calculate the cool-down time of the order to be identified; where t is the cool-down time of the order to be identified, p is the preset category measurement threshold, γ and λ are two coefficient values ​​obtained by linear fitting of multiple sample data collected in advance, and the sample data is the corresponding relationship between t value and p value.

[0071] In this embodiment, the order includes the item number. Determining whether the order to be identified is a first-class order based on the feature vector of the order to be identified and the identified orders whose cool-down period has not ended specifically includes:

[0072] S11, grouping the orders to be identified and the identified orders whose cooling-off period has not ended according to the item numbers;

[0073] In this embodiment, after determining that the category metric value of the order to be identified is greater than the preset category metric value threshold, the order to be identified can be placed in an order pool for storing identified orders whose cool-down period has not ended. All orders in the order pool are grouped according to the item number to achieve this step S11.

[0074] S12, for each order in each group, pairing the order with other orders in the group, and calculating the Euclidean distance between the feature vectors of two orders in each paired order;

[0075] When the item numbers of two orders are the same, the smaller the Euclidean distance between the feature vectors of the two orders, the more similar the feature vectors of the two orders are, so as to distinguish whether the ordering users of the two orders belong to the same group.

[0076] S13, sorting the paired orders from small to large according to the Euclidean distance between the feature vectors of two orders in the paired orders, and determining whether there is a paired order containing the order to be identified among the paired orders of a first preset proportion that are ranked first; if so, determining that the order to be identified is a first-category order.

[0077] The above step 4041 is Figure 1 The specific refinement of "determining whether the order to be identified is a first category order based on the category metric value and feature vector of the order to be identified" in step 104 is shown.

[0078] Step 4042: When it is determined that the order to be identified is a first-category order, the order to be identified is sent to the review platform for review.

[0079] In this embodiment, when it is determined that the order to be identified is not a first-category order, the order to be identified can be transmitted to the warehouse for production according to the normal processing flow.

[0080] The above steps 4041 and 4042 are Figure 1 The specific refinement of step 104 is shown.

[0081] from Figure 4 It can be seen from the method shown that in this embodiment, all orders whose cool-down period has not ended are grouped according to the item number, and the Euclidean distance between the orders in the group is calculated, so as to find out the malicious orders that are most likely to belong to group behavior. If the order to be identified is also among these malicious orders, it is confirmed that the order to be identified is a first-class order. The application of the present invention can identify malicious orders of group behavior.

[0082] See also Figure 5 , Figure 5 is a flow chart of a method for identifying group behavior orders according to a fourth embodiment of the present invention. Figure 5 As shown, the method mainly includes the following steps:

[0083] Step 501, obtaining the first type of order and the second type of order containing value modification parameters;

[0084] Step 502: determining feature data of the acquired order, and using the acquired feature data of the order to train a pre-built neural network model with an embedding layer, to obtain an order recognition model with an embedding layer;

[0085] Step 503: receiving an order to be identified that includes a value correction parameter, inputting feature data of the order to be identified into the order identification model, obtaining a feature vector of the order to be identified output by the embedding layer of the order identification model and a category metric value of the order to be identified output by the order identification model;

[0086] Step 5041: Determine whether the order to be identified is a first-category order based on the category metric value and feature vector of the order to be identified.

[0087] Step 5042: When it is determined that the order to be identified is a first-category order, an alarm is issued for the order to be identified, and the order to be identified is sent to the review platform for manual review or review based on preset rules.

[0088] In this embodiment, if it is determined that the order to be identified is a first-class order, an alarm can be issued for the order to be identified. The order that has been alarmed can be subsequently reviewed manually or according to preset rules. Only after the review is passed can it be allowed to continue to be transmitted to the warehouse for production.

[0089] The above step 5042 is Figure 1 The specific details of the step "when it is determined that the order to be identified is a first-category order, sending the order to be identified to the review platform for review" are provided.

[0090] The above steps 5041 and 5042 are Figure 1 The specific refinement of step 104 is shown.

[0091] from Figure 5 It can be seen from the method shown that in this embodiment, after determining that the order to be identified is a first-category order, an alarm is processed for the order to be identified, and the order to be identified is sent to the review platform for manual review or review based on preset rules, thereby further ensuring the accuracy of the first-category order identification.

[0092] See also Figure 6 , Figure 6 is a flow chart of a method for identifying group behavior orders according to Embodiment 5 of the present invention. Figure 6 As shown, the method mainly includes the following steps:

[0093] Step 6011, obtaining historical orders containing value correction parameters within a first preset time period before the current time;

[0094] In this embodiment, the order includes user behavior information and order attribute information; the user behavior information includes value modification parameters, order channel, and payment method, etc.; the order attribute information includes: order date, order number, user account, item number, and number of items, etc.;

[0095] In this embodiment, the historical orders obtained within the first preset time period before the current time may be specifically shown in Table 2:

[0096]

[0097] Table 2

[0098] It should be noted that an order can usually be divided into multiple sub-orders. Each row in Table 2 is a sub-order, and sub-orders with the same order number belong to the same order. For example, the three sub-orders with order number 10001 above actually belong to the historical order with order number 10001.

[0099] Step 6012: for each historical order, search for other historical orders whose order time is within the second preset time period of the order time of the historical order and have the same user behavior information, the same item number and the same number of items as the historical order; pair the historical order with each other historical order according to the item number and value correction parameter to generate sub-order pairing information; the sub-order pairing information includes two user accounts, value correction parameters and order date.

[0100] In this embodiment, the historical order is paired with each other historical order as a sub-order based on the item number and value correction parameter to generate sub-order pairing information, specifically including: for each sub-order of the historical order, searching the sub-orders of the other historical orders for a sub-order with the same item number and value correction parameter as the sub-order of the historical order, and merging the order number and user account in the sub-order of the historical order with the found sub-order to obtain the corresponding sub-order pairing information.

[0101] In this embodiment, based on the data provided in Table 2 above, the sub-order pairing information shown in Table 3 can be obtained:

[0102]

[0103]

[0104] Table 3

[0105] Step 6013: Count the number of sub-order pairing information with the same two user accounts, value correction parameters, and order dates, and count the number of historical orders including the order date and value correction parameters, and determine the quotient of the number of sub-order pairing information and the number of historical orders as the support degree of the two users placing orders based on the value correction parameters on the order date;

[0106] In this embodiment, based on the data provided in Table 3 above, the number of sub-order pairing information with the same two user accounts, value modification parameters, and order dates is counted, and the statistical results shown in Table 4 can be obtained:

[0107]

[0108] Table 4

[0109] Assuming that the order date is 2020-01-05, and the number of historical orders with value correction parameters 567 and 568 are 10 and 20 respectively, it can be calculated that the support for Jack and Lili placing an order based on the value correction parameter 567 on the date 2020-01-05 is 4 / 10=0.4, and the support for placing an order based on the value correction parameter 568 is 2 / 20=0.1.

[0110] Step 6014: sort the sub-order pairing information from large to small according to the order placement dates of the two user accounts in the sub-order pairing information in the sub-order pairing information based on the support for order placement by the value correction parameter in the sub-order pairing information, and determine the sub-order pairing information with a second preset ratio (e.g., 1%) that is ranked first and the sub-order pairing information with a third preset ratio (e.g., 5%) that is ranked last;

[0111] Step 6015: for each sub-order pairing information in the second preset ratio of sub-order pairing information ranked first, two historical orders used when performing sub-order pairing to generate the sub-order pairing information are determined as first-category orders containing value correction parameters;

[0112] Step 6016: for each sub-order pairing information in the third preset proportion of sub-order pairing information ranked later, two historical orders used in generating the sub-order pairing information are determined as second-category orders containing value correction parameters.

[0113] The above steps 6011 to 6016 are Figure 1 The specific refinement of step 101 is shown.

[0114] Step 602: determining feature data of the first category orders and the second category orders, and using the obtained feature data of the first category orders and the second category orders to train a pre-built neural network model with an embedding layer to obtain an order recognition model with an embedding layer;

[0115] Step 603: receiving an order to be identified that includes a value correction parameter, inputting feature data of the order to be identified into the order identification model, obtaining a feature vector of the order to be identified output by the embedding layer of the order identification model and a category metric value of the order to be identified output by the order identification model;

[0116] Step 604: determine whether the order to be identified is a first-category order based on the category measurement value and feature vector of the order to be identified. When the order to be identified is a first-category order, send the order to be identified to the review platform for review.

[0117] from Figure 6 It can be seen from the method shown that in this embodiment, by obtaining the first-class orders and the second-class orders containing the value correction parameters, and using the feature data of these orders to train to obtain an order recognition model with an embedding layer, the order recognition model is used to determine the category metric value and feature vector of the order to be recognized, and according to the category metric value and feature vector of the order to be recognized, it is determined whether the order to be recognized is a first-class order, and if the order to be recognized is a first-class order, the order to be recognized is further reviewed. The application of the present invention can identify malicious orders of group behavior.

[0118] The embodiment of the present invention also provides a device for identifying group behavior orders, such as Figure 7 As shown, the device comprises:

[0119] A collecting unit 701 is used to obtain the first type of orders and the second type of orders containing value modification parameters;

[0120] The training unit 702 is used to determine the characteristic data of the order acquired by the collection unit 701, and use the acquired characteristic data of the order to train a pre-built neural network model with an embedding layer to obtain an order recognition model with an embedding layer;

[0121] The processing unit 703 is used to receive an order to be identified including a value correction parameter, input feature data of the order to be identified into the order recognition model, and obtain a feature vector of the order to be identified output by the embedding layer of the order recognition model and a category metric value of the order to be identified output by the order recognition model;

[0122] The decision unit 704 is used to determine whether the order to be identified is a first-category order according to the category measurement value and feature vector of the order to be identified. When the order to be identified is a first-category order, the order to be identified is sent to the review platform for review.

[0123] Figure 7 In the device shown,

[0124] The neural network model with an embedding layer includes an input layer, an embedding layer, a hidden layer and an output layer;

[0125] The training unit 702 uses the acquired feature data of the order to train the pre-built neural network model with the embedding layer, including:

[0126] Perform one-hot encoding on the feature data of the acquired orders to obtain a sparse matrix;

[0127] Inputting the sparse matrix into the embedding layer through the input layer of the neural network model for dimensionality reduction processing;

[0128] The sparse matrix after dimension reduction processing is input into the hidden layer for feature extraction, and the feature extraction result is output to the output layer, thereby realizing the training of the neural network model.

[0129] Figure 7 In the device shown,

[0130] The decision unit 704 determines whether the order to be identified is a first-category order according to the category metric value and the feature vector of the order to be identified, including:

[0131] When the category metric value of the order to be identified is greater than the preset category metric threshold, the cool-down period of the order to be identified is determined according to the category metric value of the order to be identified, and whether the order to be identified is a first category order is determined according to the feature vector of the order to be identified and the identified orders whose cool-down period has not ended.

[0132] Figure 7 In the device shown,

[0133] The decision unit 704 determines the cool-down period of the order to be identified according to the category metric value of the order to be identified, including:

[0134] Using the formula Calculate the cool-down time of the order to be identified; where t is the cool-down time of the order to be identified, p is the preset category measurement threshold, γ and λ are two coefficient values ​​obtained by linear fitting of multiple sample data collected in advance, and the sample data is the corresponding relationship between t value and p value.

[0135] Figure 7 In the device shown,

[0136] Include the item number in your order;

[0137] The decision unit 704 determines whether the order to be identified is a first-category order according to the feature vector of the order to be identified and the identified orders whose cool-down period has not ended, including:

[0138] Group the pending orders and the identified orders whose cool-down period has not ended by item number;

[0139] For each order in each group, pair the order with other orders in the group, and calculate the Euclidean distance between the feature vectors of the two orders in each paired order;

[0140] Sort the paired orders in ascending order according to the Euclidean distance between the feature vectors of the two orders in the paired orders;

[0141] It is determined whether there is a paired order containing the order to be identified among the paired orders of the first preset ratio that are ranked top. If so, it is determined that the order to be identified is a first-category order.

[0142] Figure 7 In the device shown,

[0143] The decision unit 704 sends the order to be identified to the review platform for review, including:

[0144] Alerts are issued for orders to be identified, and the orders to be identified are sent to the review platform for manual review or review based on preset rules.

[0145] Figure 7 In the device shown,

[0146] The order includes user behavior information and order attribute information; the user behavior information includes value modification parameters, order channel, and payment method; the order attribute information includes: order date, order number, user account, item number, and item quantity;

[0147] The collecting unit 701 obtains the first type of order including the value modification parameter and the second type of order including the value modification parameter, including:

[0148] Get the historical orders containing value correction parameters within the first preset time before the current time;

[0149] For each historical order, search for other historical orders whose order time is within the second preset time period of the order time of the historical order and have the same user behavior information, the same item number and the same number of items as the historical order, and perform sub-order pairing of the historical order with each other historical order according to the item number and the value correction parameter to generate sub-order pairing information; the sub-order pairing information includes two user accounts, the value correction parameter and the order date;

[0150] Count the number of sub-order pairing information for which two user accounts, value correction parameters, and order dates are the same, and count the number of historical orders that include the order date and value correction parameters, and determine the quotient of the number of sub-order pairing information and the number of historical orders as the support degree for the two users to place orders based on the value correction parameters on the order date;

[0151] Sort the sub-order pairing information from large to small according to the order placement dates of the two user accounts in the sub-order pairing information in the sub-order pairing information based on the support for order placement by the value correction parameter in the sub-order pairing information, and determine the sub-order pairing information with a second preset ratio ranked first and the sub-order pairing information with a third preset ratio ranked last;

[0152] For each sub-order pairing information in the second preset ratio of sub-order pairing information ranked first, two historical orders used when performing sub-order pairing to generate the sub-order pairing information are determined as first-category orders containing value correction parameters;

[0153] For each sub-order pairing information in the third preset proportion of sub-order pairing information ranked lower, two historical orders used in performing sub-order pairing to generate the sub-order pairing information are determined as second-category orders containing value correction parameters.

[0154] Figure 7 In the device shown,

[0155] The collecting unit 701 performs sub-order pairing of the historical order with each other historical order according to the item number and the value correction parameter to generate sub-order pairing information, including:

[0156] For each sub-order of the historical order, search the sub-orders of other historical orders for a sub-order with the same item number and value modification parameters as the sub-order of the historical order, merge the order number and user account in the sub-order of the historical order with the found sub-order to obtain the corresponding sub-order matching information.

[0157] The embodiment of the present invention further provides an electronic device, such as Figure 8 As shown, the electronic device includes: at least one processor 801, and a memory 802 connected to the at least one processor 801 through a bus; the memory 802 stores one or more computer programs that can be executed by the at least one processor 801; when the at least one processor 801 executes the one or more computer programs, Figure 1 , 2 , 4, 5, 6, or any of the steps in the method shown in the flowchart.

[0158] The embodiment of the present invention further provides a computer-readable storage medium, wherein the computer-readable storage medium stores one or more computer programs, and when the one or more computer programs are executed by a processor, Figure 1 , 2 , 4, 5, 6, or any of the steps in the method shown in the flowchart.

[0159] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. An order processing method based on a neural network model, characterized in that: The method includes: Obtaining a first type of order containing a value correction parameter and a second type of order containing a value correction parameter; the first type of order includes malicious orders belonging to group behavior, and the second type of order includes non-malicious orders that do not belong to group behavior; Determine the feature data of the acquired order, and use the feature data of the acquired order to train a pre-built neural network model with an embedding layer to obtain an order recognition model with an embedding layer; Receiving an order to be identified including a value correction parameter, inputting feature data of the order to be identified into the order identification model, obtaining a feature vector of the order to be identified output by an embedding layer of the order identification model and a category metric value of the order to be identified output by the order identification model; Determine whether the order to be identified is a first-category order according to the category measurement value and feature vector of the order to be identified. If the order to be identified is a first-category order, send the order to be identified to the review platform for review. Determining whether the order to be identified is a first-category order according to the category measurement value and feature vector of the order to be identified includes: When the category metric value of the order to be identified is greater than the preset category metric threshold, the cool-down period of the order to be identified is determined according to the category metric value of the order to be identified, and whether the order to be identified is a first category order is determined according to the feature vector of the order to be identified and the identified orders whose cool-down period has not ended; The cool-down period of the order to be identified is determined according to the category measurement value of the order to be identified, including: Using the formula Calculate the cool-down time of the order to be identified; where, The cool-down time for the order to be identified. is the preset category measurement threshold, and are two coefficient values ​​obtained by linear fitting of a plurality of sample data collected in advance, wherein the sample data is Value and The corresponding relationship of values.

2. The method according to claim 1, characterized in that The neural network model with an embedding layer includes an input layer, an embedding layer, a hidden layer and an output layer; The pre-built neural network model with an embedding layer is trained using the feature data of the acquired orders, including: Perform one-hot encoding on the feature data of the acquired orders to obtain a sparse matrix; Inputting the sparse matrix into the embedding layer through the input layer of the neural network model for dimensionality reduction processing; The sparse matrix after dimension reduction processing is input into the hidden layer for feature extraction, and the feature extraction result is output to the output layer, thereby realizing the training of the neural network model.

3. The method according to claim 1, characterized in that Include the item number in your order; Determine whether the order to be identified is a first-class order based on the feature vector of the order to be identified and the identified orders whose cool-down period has not ended, including: Group the pending orders and the identified orders whose cool-down period has not ended by item number; For each order in each group, pair the order with other orders in the group, and calculate the Euclidean distance between the feature vectors of the two orders in each paired order; Sort the paired orders in ascending order according to the Euclidean distance between the feature vectors of the two orders in the paired orders; It is determined whether there is a paired order containing the order to be identified among the paired orders of the first preset ratio that are ranked top. If so, it is determined that the order to be identified is a first-category order.

4. The method according to claim 1, characterized in that Sending the order to be identified to the review platform for review includes: Alerts are issued for orders to be identified, and orders to be identified are sent to the review platform for manual review or review based on preset rules.

5. The method according to claim 1, characterized in that The order includes user behavior information and order attribute information; the user behavior information includes value modification parameters, order channel, and payment method; the order attribute information includes: order date, order number, user account, item number, and item quantity; Get first-class orders with value modification parameters and second-class orders with value modification parameters, including: Get the historical orders containing value correction parameters within the first preset time before the current time; For each historical order, search for other historical orders whose order time is within the second preset time period of the order time of the historical order and have the same user behavior information, the same item number and the same number of items as the historical order, and perform sub-order pairing of the historical order with each other historical order according to the item number and the value correction parameter to generate sub-order pairing information; the sub-order pairing information includes two user accounts, the value correction parameter and the order date; Count the number of sub-order pairing information for which two user accounts, value correction parameters, and order dates are the same, and count the number of historical orders that include the order date and value correction parameters, and determine the quotient of the number of sub-order pairing information and the number of historical orders as the support degree for the two users to place orders based on the value correction parameters on the order date; Sort the sub-order pairing information from large to small according to the order placement dates of the two user accounts in the sub-order pairing information in the sub-order pairing information based on the support for order placement by the value correction parameter in the sub-order pairing information, and determine the sub-order pairing information with a second preset ratio ranked first and the sub-order pairing information with a third preset ratio ranked last; For each sub-order pairing information in the second preset ratio of sub-order pairing information ranked first, two historical orders used when performing sub-order pairing to generate the sub-order pairing information are determined as first-category orders containing value correction parameters; For each sub-order pairing information in the third preset proportion of sub-order pairing information ranked lower, two historical orders used in performing sub-order pairing to generate the sub-order pairing information are determined as second-category orders containing value correction parameters.

6. The method according to claim 5, characterized in that The historical order is matched with each other historical order according to the item number and the value correction parameter to generate sub-order matching information, including: For each sub-order of the historical order, search the sub-orders of other historical orders for a sub-order with the same item number and value modification parameters as the sub-order of the historical order, merge the order number and user account in the sub-order of the historical order with the found sub-order to obtain the corresponding sub-order matching information.

7. An order processing device based on a neural network model, characterized in that: The device includes: A collecting unit, used to obtain a first type of order containing a value modification parameter and a second type of order containing a value modification parameter; the first type of order includes malicious orders belonging to group behavior, and the second type of order includes non-malicious orders that do not belong to group behavior; A training unit, used to determine the characteristic data of the order acquired by the collection unit, and use the acquired characteristic data of the order to train a pre-built neural network model with an embedding layer to obtain an order recognition model with an embedding layer; a processing unit, configured to receive an order to be identified including a value correction parameter, input feature data of the order to be identified into the order identification model, obtain a feature vector of the order to be identified output by the embedding layer of the order identification model and a category metric value of the order to be identified output by the order identification model; A decision unit, used for determining whether the order to be identified is a first-category order according to the category measurement value and feature vector of the order to be identified, and when the order to be identified is a first-category order, sending the order to be identified to the review platform for review; The decision unit determines whether the order to be identified is a first-category order according to the category measurement value and the feature vector of the order to be identified, including: When the category metric value of the order to be identified is greater than the preset category metric threshold, the cool-down period of the order to be identified is determined according to the category metric value of the order to be identified, and whether the order to be identified is a first category order is determined according to the feature vector of the order to be identified and the identified orders whose cool-down period has not ended; The decision unit is used to determine the cool-down time of the order to be identified according to the category measurement value of the order to be identified, including: Using the formula Calculate the cool-down time of the order to be identified; where, The cool-down time for the order to be identified. is the preset category measurement threshold, and are two coefficient values ​​obtained by linear fitting of a plurality of sample data collected in advance, wherein the sample data is Value and The corresponding relationship of values.

8. An electronic device comprising: at least one processor, and a memory connected to the at least one processor via a bus; The memory stores one or more computer programs that can be executed by the at least one processor; it is characterized in that when the at least one processor executes the one or more computer programs, the steps in the method described in any one of claims 1-6 are implemented.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores one or more computer programs, and when the one or more computer programs are executed by a processor, the steps in the method of any one of claims 1 to 6 are implemented.

Images