Cross-border e-commerce transaction data management system

Abstract

The invention discloses a cross-border e-commerce transaction data management system, which relates to the field of data management, and is characterized in that a data acquisition unit acquires data commodities, preprocesses the data commodities and marks the data commodities, and a data processing module is used for acquiring data types and initial information total quantity contained in the data commodities and sending the data types and the initial information total quantity to a server; comprising an AdaBoost optimization model and a counting unit, data commodities are imported into the AdaBoost optimization model, data types included in the data commodities are obtained, the counting unit is used for obtaining the total amount of initial information in each type, and a calculation unit is used for calculating direct benefits obtained by all suppliers according to the provided data commodities. And the quantity updating unit is used for obtaining the total data information quantity after the homogenized data information is removed, obtaining the real total benefit obtained by the supplier through an RBF neural network model, realizing the accurate prediction of the total benefit obtained by the supplier investment, and providing data support for the benefit distribution of the project.

Description

Technical Field

[0001] The present invention relates to the field of data management, and more specifically, to a cross-border e-commerce transaction data management system. Background Art

[0002] The scale of cross-border e-commerce continues to expand, and the amount and categories of factor markets continue to expand. With the vigorous development of new-generation information technology, the economic value of various data information is also increasing. Therefore, the circulation and profit analysis of data commodities has also become an important economic consideration. Since cross-border e-commerce involves multiple local entities, it is more cautious and focused on profit in the development and cooperation of data commodities. When cross-border e-commerce conducts data information transactions, due to various reasons such as unclear cooperation value and distribution rules, cooperative development and transactions are difficult to carry out. Especially for multi-party data collaboration, when multiple data suppliers carry out data cooperative development through data products, the final data product benefits are difficult to explore and cannot be expected. The joint efforts of multiple parties increase the sample size and data dimensions covered by the data, making the profit value difficult to calculate and estimate, affecting developers' confidence in cooperation, and is not conducive to the development and trading of data information in cross-border e-commerce.

[0003] A cross-border e-commerce transaction system with publication number CN111784466A includes a cross-border retail platform, a trade customs clearance service platform, a bonded warehouse WMS system, and a consumer client. The cross-border retail platform has a platform management unit, a store management unit, and a three-order information management unit. The trade customs clearance service platform has a filing management unit, an account book management unit, and a three-order management unit. The trade customs clearance service platform has an information management unit, a warehousing management unit, a warehouse management unit, and a warehouse-out management unit. The payment information, commodity information, and logistics information of overseas commodities can be timely counted, but the profit prediction and distribution of digital commodities have never been involved. Therefore, the present invention is innovatively designed to address the above problems. Summary of the invention

[0004] In view of the problems existing in the prior art, the purpose of the present invention is to provide a cross-border e-commerce transaction data management system to provide intuitive data support for digital commodity transactions and profit sharing.

[0005] To achieve the above-mentioned purpose, the present invention provides the following technical solution: a cross-border e-commerce transaction data management system, the cross-border e-commerce transaction data management system comprising:

[0006] Data collection module: includes data collection unit and division unit. The data collection unit obtains data commodities and performs preprocessing. Each supplier is set to provide a data commodity and mark the data commodity, which is recorded as X. i ; The division unit divides the data information involved in the project into m data types;

[0007] Among them, X i represents the data commodity provided by the i-th supplier, and the data commodity includes multiple data information;

[0008] Data processing module: used to obtain data product X i The data type and total amount of initial information contained in the AdaBoost optimization model and counting units are converted into data commodity X i Import into AdaBoost optimization model and obtain data product X i The types of data included, the counting unit is used to obtain the initial amount of information in each type;

[0009] Profit calculation module: includes a calculation unit and a quantity update unit. The calculation unit is used to calculate the direct profit obtained by each supplier based on the data products provided. The quantity update unit is used to obtain the total amount of data information after the homogeneous data information is removed. ;

[0010] Profit prediction module: used to predict the actual distributed benefits that each supplier can obtain, obtain the real total benefits obtained by the supplier through the RBF neural network model, and make accurate predictions on the real total benefits that the supplier can obtain.

[0011] Preferably, the AdaBoost optimization model includes:

[0012] Obtain a training set and a validation set, wherein the validation set and the training set include N groups of training data respectively, each group of training data includes n training samples, and each training sample includes a data information and its corresponding data type;

[0013] Initialize the weight of each training sample in the training data, then the weight distribution of the data information sample of the first weak classifier is:

[0014] D(t)=(ω 1，1 ,…,ω 1，g ,…,ω 1，n );

[0015] Where g = 1, 2, ..., n, g is a label, and ω 1，1 ,ω 1，g and ω 1，n Respectively represent the weight coefficients of the 1st, gth and nth data information samples of the first weak classifier;

[0016] Set the update of weight coefficients, base classifier coefficients and normalization factor to get each weak classifier:

[0017] The final classifier of the optimized AdaBoost optimization model is obtained by combining the weak classifiers as follows:

[0018]

[0019] Among them, α t represents the base classifier coefficient after t updates, G t (x g ) represents the base classifier.

[0020] The data information in the training set is imported into the AdaBoost optimization model, and the function value of the mean square error loss function MSELoss is used as the supervision signal. The parameters in the framework network are updated through back propagation to minimize the function value of the mean square error loss function MSELoss, and the trained AdaBoost optimization model is obtained.

[0021] The data information in the validation set is input into the trained AdaBoost optimization model in turn to verify the effect. When the function value of the mean square error loss function MSELoss no longer decreases, the training reaches the expected result and the training ends, and the trained AdaBoost optimization model is obtained.

[0022] Preferably, the weight coefficient of the AdaBoost optimization model is updated as follows:

[0023]

[0024] Among them, ω t+1,g represents the weight coefficient of the g-th data information sample after t+1 updates, ω t,g represents the weight coefficient of the g-th data information sample after t updates, Z t represents the normalization factor after t updates, α t represents the base classifier coefficient after t updates, y g is a label and y g ∈{-1,1},G t (x g ) represents the base classifier;

[0025] Set the classification error rate ξ of the AdaBoost optimization model t for:

[0026]

[0027] In the formula, ω t,g I(G t (x g )≠y g ) represents the model G t (x g)The weight coefficient of misclassified samples.

[0028] Preferably, the loss function of the AdaBoost optimization model is the mean square error loss MSELoss, and the mean square error loss function is as follows:

[0029]

[0030] Among them, ξ d is the type of prediction data generated by the data information, It is the actual data type corresponding to the data information.

[0031] Preferably, the quantity updating unit is used to process the homogenized data information in the total amount of initial information acquired by the data processing module, and to update the total amount of data information in each data type, and to obtain the updated total amount of data information as

[0032] Traverse and calculate the correlation coefficient for each type of data information:

[0033]

[0034] In the formula, Cov(c h ,c j ) indicates data information c h With data information c j The covariance of Var[c h ] indicates data information c h The variance of Var[c j ] indicates data information c j The variance of .

[0035] Preferably, the quantity updating unit further includes:

[0036] Set a threshold a, if the correlation coefficient r j ≥a, it means that the similarity between the data information is high, and the correlation coefficient r j The number of data information ≥ a is recorded as 1, that is, only the correlation coefficient r between the data information and all data information of the same type j <a can be recorded as a data quantity, and then obtain each data type

[0037] —The total amount of data after homogeneous data is eliminated is se.

[0038] Preferably, the calculation unit of the benefit calculation module includes:

[0039] Pre-calculation is performed based on the benefit formula. The benefit function formula is as follows:

[0040] Q i =ui +Δu i ;

[0041]

[0042] Among them, Q i is the actual total benefit that the i-th supplier can obtain, u i is the direct benefit that the i-th supplier can obtain; Δu i is the indirect benefit that the ith supplier can obtain, ψe is the weight factor, f is the total number of suppliers, and |s| is the number of suppliers that can provide the same data type.

[0043] Preferably, the RBF neural network model includes:

[0044] Build the RBF neural network framework and set the network output The weight vector is updated through the Gaussian basis function to obtain the mapping relationship between input and output;

[0045] The loss function of the RBF neural network model is set to the optimized adversarial loss function;

[0046] Construct a sample set and a measured set. The sample set and the measured set include kl groups of historical data. Each group of historical data includes the direct benefits that the supplier can obtain and the corresponding real total benefits. The direct benefits of each group of historical data in the sample set are used as the input of the RBF neural network model, and the corresponding real total benefits are used as the output. The RBF neural network model is trained to obtain the trained RBF neural network model.

[0047] The measured data set is imported into the trained RBF neural network model, and the trained RBF neural network model is obtained when the resistance loss function is minimum and does not change.

[0048] Preferably, the adversarial loss function is:

[0049]

[0050] Among them, J(D,G) is the objective function of the entire generative adversarial network, where D is the discriminator network and G is the generator network; rec (G(y),x) is the reconstruction loss, which measures the reconstruction error between the predicted total benefit G(y) generated by the generator and the true total benefit x; l meas (y,o(G(y))) is the metric loss, which measures the metric error between the predicted total benefit G(y) generated by the generator and its output o(G(y)); H GAN (G(y),D) is the generative adversarial loss.

[0051] Preferably, the vending machine management system is applied to a microcomputer, including a human-computer interaction interface, which is used for information display, and suppliers access the system through a client, using an integrated information processing system and a multi-client information architecture.

[0052] Compared with the prior art, the present invention has the following beneficial effects:

[0053] 1. In the present invention, the data types in the data products provided by each supplier and the total amount of initial information included in each data type are obtained through the AdaBoost optimization model, so as to achieve clear segmentation of chaotic data or bulk data, provide data support for the contribution rate calculation of each data type, achieve better and faster division of data products, and improve data processing efficiency.

[0054] 2. In the present invention, homogeneous data information is eliminated through the quantity updating unit, the total amount of data information is updated to improve the accuracy of calculation, the direct benefits of each supplier's data product are obtained through the calculation unit, and the real total benefit obtained by the supplier is obtained through the RBF neural network model. The real total benefit that the supplier can obtain is accurately predicted, providing a data basis for the supplier's profit distribution, and enabling the supplier to intuitively feel the benefits brought by the invested data products, thereby realizing the digital and intuitive display of the data product benefits. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Figure 1 A schematic diagram of the structure of a cross-border e-commerce transaction data management system is proposed for the present invention;

[0056] Figure 2 Schematic diagram of the method steps in the present invention. DETAILED DESCRIPTION

[0057] Below, the exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the exemplary embodiments described here.

[0058] Embodiment 1

[0059] Reference Figure 1 and Figure 2 Embodiment 1 further illustrates a cross-border e-commerce transaction data management system proposed by the present invention.

[0060] When cross-border e-commerce conducts data information transactions, cooperative development and transactions are difficult to carry out due to various reasons such as unclear cooperation value and distribution rules. Especially for multi-party data collaboration, when multiple data suppliers conduct data cooperative development through data products, the final data product benefits are difficult to explore and cannot be predicted. The joint efforts of multiple parties increase the sample size and data dimensions covered by the data, making the value of the benefits difficult to calculate and estimate, affecting developers' confidence in cooperation, and is not conducive to the development and trading of data information in cross-border e-commerce. Moreover, when distributing benefits, the benefit distribution mechanism is unclear due to the difficulty in determining the contribution rate, which in turn leads to the problem of vague benefit division. Therefore, a clear benefit division is obtained through relevant design.

[0061] Data collection module: includes data collection unit and division unit. The data collection unit obtains data commodities and performs preprocessing. Each supplier is set to provide a data commodity and mark the data commodity, which is recorded as X. i ;X i represents the data commodity provided by the i-th supplier, the data commodity includes multiple data information, and the division unit divides the data information involved in the project into m data types;

[0062] Data commodities are the transaction objects of cross-border e-commerce. Data commodities include data information that can generate economic benefits. They can be the data development results of suppliers, or the value information related to a certain business or production, or the collection and summary of information in a certain related field. In short, data commodities contain multiple categories or multiple types of data information in multiple fields. When multiple suppliers jointly develop a project across borders, each supplier provides a data commodity. Based on the economic benefits generated by the data commodities provided in the project, the reasonable prediction of the distribution of benefits of each supplier through the data commodity exchange through the project is evaluated, and a transaction reference is provided for each supplier.

[0063] Data processing module: used to obtain data product X i The data type and total amount of initial information contained in the AdaBoost optimization model and counting units are converted into data commodity X i Import into AdaBoost optimization model and obtain data product X i The types of data included, the counting unit is used to obtain the initial amount of information in each type;

[0064] The AdaBoost algorithm is a boosting algorithm. In classification problems, the algorithm learns multiple classifiers by changing the weights of training samples. It can combine multiple weak classifiers into a strong classifier and improve the accuracy of the classifier by weighting and repeating the training of the data set. Its basic principle is that the samples that were misclassified by the previous basic classifier will be strengthened (that is, they will get higher weights), and all weighted samples will be used again to train the next basic classifier. At the same time, a new weak classifier is added in each round until a predetermined small enough error rate is reached or the pre-specified maximum number of iterations is reached, and the algorithm stops.

[0065] AdaBoost optimization models include:

[0066] Get the training set and validation set. The validation set and training set include N groups of training data respectively. Each group of training data includes n training samples. Each training sample includes a data information and its corresponding data type. The purpose of the Adaboost algorithm is to learn a series of weak classifiers or base classifiers from the training data, and then combine these weak classifiers into a strong classifier.

[0067] Initialize the weight of each training sample in the training data. Since each set of training data is set to include n training samples, the weight of the data information in each training sample is initialized as Then the weight distribution of the data information sample of the first weak classifier is:

[0068] D(t)=(ω 1，1 , ..., ω 1，g , ..., ω 1，n );

[0069] Among them, g = 1, 2, ..., n, g are labels, and ω 1，1 ,ω 1，g and ω 1，n Respectively represent the weight coefficients of the 1st, gth and nth data information samples of the first weak classifier;

[0070] The update of the weight coefficient is set as:

[0071]

[0072] Among them, ω t+1,g represents the weight coefficient of the g-th data information sample after t+1 updates, ω t,g represents the weight coefficient of the g-th data information sample after t updates, Z t represents the normalization factor after t updates, α t represents the base classifier coefficient after t updates, y g is a label and yg ∈{-1,1},G t (x g ) represents the base classifier;

[0073] The update formula for setting the base classifier coefficient and the normalization factor is:

[0074]

[0075] Set the classification error rate ξ of the AdaBoost optimization model t for:

[0076]

[0077] In the formula, ω t,g I(G t (x g )≠y g ) represents the model G t (x g )The weight coefficient of misclassified samples;

[0078] The final classifier of the optimized AdaBoost optimization model is obtained by combining the weak classifiers as follows:

[0079]

[0080] The loss function of the AdaBoost optimization model is set to mean square error loss MSELoss. The mean square error loss function is as follows:

[0081]

[0082] Among them, ξ d is the type of prediction data generated by the data information, It is the actual data type corresponding to the data information.

[0083] The data information in the training set is imported into the AdaBoost optimization model, that is, the input end of the AdaBoost optimization model is the data information, and the output end is the data type; the function value of the mean square error loss function MSELoss is used as the supervision signal, and the parameters in the framework network are updated through back propagation to minimize the function value of the mean square error loss function MSELoss, thereby obtaining the trained AdaBoost optimization model;

[0084] The data information in the validation set is input into the trained AdaBoost optimization model in turn to verify the effect. When the function value of the mean square error loss function MSELoss no longer decreases, the training reaches the expected result and the training ends, and the trained AdaBoost optimization model is obtained.

[0085] In this embodiment, the AdaBoost optimization model is used to obtain the data type in the data products provided by each supplier, as well as the total amount of initial information included in each data type, to achieve clear segmentation of chaotic data or bulk data, provide data support for the contribution rate calculation of each data type, achieve better and faster division of data products, and improve data processing efficiency.

[0086] Embodiment 2

[0087] Reference Figure 1 and Figure 2 , Example 2 further illustrates a cross-border e-commerce transaction data management system proposed by the present invention.

[0088] The profit distribution model of data cooperative development results needs to evaluate the data value of data suppliers. Since the data products provided by different suppliers are different or the information focuses are different, it is impossible to unify the price of data products and the value of data assets. In the process of project development, multi-faceted and multi-type data information needs to interact with each other, and the value that each aspect or type of data information can provide is relatively easier to estimate. Therefore, by calculating the value that data in each data information type can provide, it is easier to obtain the overall value provided by each data commodity, and it is more efficient and accurate to estimate the expected distribution of benefits through the value of data commodities.

[0089] Profit calculation module: includes a calculation unit and a quantity update unit. The calculation unit is used to calculate the direct profit obtained by each supplier based on the data products provided. The quantity update unit is used to obtain the total amount of data information after the homogeneous data information is removed.

[0090] The quantity updating unit is used to process the homogenized data information in the total amount of initial information obtained by the data processing module, and to update the total amount of data information in each data type, and obtain the updated total amount of data information as

[0091] Traverse and calculate the correlation coefficient for each type of data information:

[0092]

[0093] In the formula, Cov(c h ,c j ) indicates data information c h With data information c j The covariance of Var[c h ] indicates data information c h The variance of Var[c j ] indicates data information c j The variance of

[0094] Set a threshold a, if the correlation coefficient r j ≥a, it means that the similarity between the data information is high, there is a high degree of information overlap, and the contribution to value does not meet expectations. Therefore, the correlation coefficient r j The number of data information ≥ a is recorded as 1, that is, only the correlation coefficient r between the data information and all data information of the same type j <a can be recorded as a data quantity, and then the total amount of data information after removing homogeneous data information in each data type is obtained as

[0095] The data processing module obtains the data types and the total amount of information in each data type in the data products provided by each supplier, and pre-calculates according to the profit formula. The profit function formula is as follows:

[0096] Q i =u i +Δu i ;

[0097]

[0098] Among them, Q i is the actual total benefit that the i-th supplier can obtain, u i is the direct benefit that the i-th supplier can obtain; Δu i is the indirect benefit that the ith supplier can obtain, ψe is the weight factor, f is the total number of suppliers, and |s| is the number of suppliers that can provide the same data type.

[0099] Indirect benefit Δu i Due to the interference of external uncontrollable factors, it is impossible to quantify it, so a profit prediction module is needed to obtain the total profit Q i The specific value of .

[0100] Profit prediction module: used to predict the actual distributed benefits that each supplier can obtain, obtain the real total benefits obtained by the supplier through the RBF neural network model, and make accurate predictions on the real total benefits that the supplier can obtain

[0101] Construct a sample set and a measured set. The sample set and the measured set include kl groups of historical data. Each group of historical data includes the direct benefits that the supplier can obtain and the corresponding real total benefits. The direct benefits of each group of historical data in the sample set are used as the input of the RBF neural network model, and the corresponding real total benefits are used as the output to train the RBF neural network model.

[0102] Setting Network Output The weight vector is updated through the Gaussian basis function to obtain the mapping relationship between input and output;

[0103] In the RBF neural network, the input layer receives input data and passes it to the hidden layer; the hidden layer is composed of a series of radial basis functions, which calculate the output based on the distance between the input data and their respective center points; each basis function corresponds to a center point and a deviation parameter, and by adjusting these parameters, the basis function can adapt to the characteristics of the input data; the output layer calculates the final prediction result based on the output of the hidden layer; usually, the output layer uses a linear activation function to obtain continuous prediction values.

[0104] The present invention adopts RBF neural network to express the relationship between direct benefits and corresponding real total benefits, solves the generalization problem of traditional BP network prediction and evaluation, and the RBF neural network adopts the random approximation method to obtain better prediction results.

[0105] In the training set, direct benefits are taken as input, and the direct benefits are set as cc, and the number of direct benefits cc is ds, then the training set is represented as Cc = {cc1,......,cc ds} T ,{bx1,......,bx al} T is the center point of the network radial basis function, where al represents the number of center points of the basis function,

[0106] The Gaussian basis function is;

[0107]

[0108] So the network output The weight corresponding to the σth group of real-time feature signals is the weight vector, and κ is the basis width parameter of the node corresponding to the σth group of data.

[0109] When setting the loss function of the RBF neural network model, adversarial loss is used to minimize the distribution difference between the predicted total benefit generated by the generator and the true total benefit through the mutual game between the generator and the discriminator to achieve the purpose of being closer to reality. The goal of this loss function is to minimize the probability that the predicted total benefit generated by the generator is judged as false by the discriminator, that is, to make the generated predicted total benefit more difficult for the discriminator to distinguish between true and false.

[0110] The adversarial loss function is:

[0111]

[0112] Among them, J(D,G) is the objective function of the entire generative adversarial network, where D is the discriminator network and G is the generator network; rec(G(y),x) is the reconstruction loss, which measures the reconstruction error between the predicted total benefit G(y) generated by the generator and the true total benefit x; l meas (y,o(G(y))) is the metric loss, which measures the metric error between the predicted total benefit G(y) generated by the generator and its output o(G(y)); H GAN (G(y), D) is the generative adversarial loss, which is used to make the predicted total benefits generated by the generator more difficult for the discriminator to distinguish. This formula reflects the game process in which the generator strives to generate real total benefits close to reality, and the discriminator strives to accurately distinguish the true from the false.

[0113] In this embodiment, homogeneous data information is eliminated through the quantity updating unit, and the total amount of data information is updated to improve the accuracy of calculation. The direct benefits of each supplier's data product are obtained through the calculation unit, and the real total benefit obtained by the supplier is obtained through the RBF neural network model. The real total benefit that the supplier can obtain is accurately predicted, providing a data basis for the supplier's profit distribution, allowing the supplier to intuitively feel the benefits brought by the invested data products, and realizing the digital and intuitive display of the data product benefits.

[0114] Embodiment 3

[0115] Reference Figure 1 and Figure 2 , Example 3 further illustrates a cross-border e-commerce transaction data management system proposed by the present invention.

[0116] A cross-border e-commerce transaction data management system, the cross-border e-commerce transaction data management system comprising the following steps:

[0117] Step S1: Obtain data products provided by various suppliers;

[0118] Step S2: Import the data products provided by each supplier into the AdaBoost optimization model to obtain the data type and total amount of initial information included in each data product;

[0119] Step S3: importing the total amount of initial information of each data commodity into the quantity update unit to obtain the total amount of data information after the homogeneous data information is eliminated;

[0120] Step S4: obtaining the direct benefits of each supplier by calculating the total amount of information processed by the data type and quantity updating unit in the data commodities of each supplier;

[0121] Step S5: Import the direct benefits of each supplier into the RBF neural network model to obtain the total benefits of each supplier.

[0122] A cross-border e-commerce transaction data management system, comprising:

[0123] Data collection module: includes data collection unit and division unit. The data collection unit obtains data commodities and performs preprocessing. Each supplier is set to provide a data commodity and mark the data commodity, which is recorded as X. i ; The division unit divides the data information involved in the project into m data types;

[0124] Among them, X i represents the data commodity provided by the i-th supplier, and the data commodity includes multiple data information;

[0125] Data processing module: used to obtain data product X i The data type and total amount of initial information contained in the AdaBoost optimization model and counting units are converted into data commodity X i Import into AdaBoost optimization model and obtain data product X i The types of data included, the counting unit is used to obtain the initial amount of information in each type;

[0126] Profit calculation module: includes calculation unit and quantity update unit. The calculation unit is used to calculate the direct profit obtained by each supplier based on the data products provided, and the quantity update unit is used to obtain the homogeneous data.

[0127] —The total amount of data after information elimination is se;

[0128] Profit prediction module: used to predict the actual distributed benefits that each supplier can obtain, obtain the real total benefits obtained by the supplier through the RBF neural network model, and make accurate predictions on the real total benefits that the supplier can obtain.

[0129] Since the electronic device introduced in this embodiment is an electronic device used to implement a cross-border e-commerce transaction data management method in the embodiment of this application, based on the cross-border e-commerce transaction data management method introduced in the embodiment of this application, the technical personnel of this field can understand the specific implementation of the electronic device of this embodiment and its various variations, so how the electronic device implements the method in the embodiment of this application is not described in detail here. As long as the technical personnel of this field implement the electronic device used in the cross-border e-commerce transaction data management method in the embodiment of this application, it belongs to the scope of protection of this application.

[0130] The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters and thresholds in the formula are set by technicians in this field according to actual conditions.

[0131] The above is only a preferred embodiment of the present invention, and the protection scope of the present invention is not limited to the above embodiments. All technical solutions under the concept of the present invention belong to the protection scope of the present invention. It should be pointed out that for ordinary technical users in this technical field, some improvements and modifications without departing from the principle of the present invention should also be regarded as the protection scope of the present invention.

Claims

1. A cross-border e-commerce transaction data management system, characterized in that: The cross-border e-commerce transaction data management system includes: Data collection module: includes data collection unit and division unit. The data collection unit obtains data commodities and performs preprocessing. Each supplier is set to provide a data commodity and mark the data commodity, which is recorded as X. i ; The division unit divides the data information involved in the project into m data types; Among them, X i represents the data commodity provided by the i-th supplier, and the data commodity includes multiple data information; Data processing module: used to obtain data product X i The data type and total amount of information contained in the AdaBoost optimization model and counting units are converted into data commodity X i Import into AdaBoost optimization model and obtain data product X i The types of data included, the counting unit is used to obtain the initial amount of information in each type; Profit calculation module: includes a calculation unit and a quantity update unit. The calculation unit is used to calculate the direct profit obtained by each supplier based on the data products provided, and the quantity update unit is used to obtain homogeneous data. The total amount of data information after eliminating the information is se; Profit prediction module: used to predict the actual distributed benefits that each supplier can obtain, obtain the real total benefits obtained by the supplier through the RBF neural network model, and make accurate predictions on the real total benefits that the supplier can obtain.

2. A cross-border e-commerce transaction data management system according to claim 1, characterized in that: The AdaBoost optimization model includes: Obtain a training set and a validation set, wherein the validation set and the training set include N groups of training data respectively, each group of training data includes n training samples, and each training sample includes a data information and its corresponding data type; Initialize the weight of each training sample in the training data, then the weight distribution of the data information sample of the first weak classifier is: D(t)=(ω 1，1 ,...,oh 1，g ,...,oh 1，n ); Among them, g = 1, 2, ..., n, g are labels, and ω 1，1 ,ω 1，g and ω 1，n Respectively represent the weight coefficients of the 1st, gth, and nth data information samples of the first weak classifier; Set the update of weight coefficients, base classifier coefficients and normalization factor to get each weak classifier: The final classifier of the optimized AdaBoost optimization model is obtained by combining the weak classifiers as follows: Among them, α t represents the base classifier coefficient after t updates, G t (x g ) represents the base classifier. The data information in the training set is imported into the AdaBoost optimization model, and the function value of the mean square error loss function MSELoss is used as the supervision signal. The parameters in the framework network are updated through back propagation to minimize the function value of the mean square error loss function MSELoss, and the trained AdaBoost optimization model is obtained. The data information in the validation set is input into the trained AdaBoost optimization model in turn to verify the effect. When the function value of the mean square error loss function MSELoss no longer decreases, the training reaches the expected result and the training ends, and the trained AdaBoost optimization model is obtained.

3. A cross-border e-commerce transaction data management system according to claim 2, characterized in that: The weight coefficient of the AdaBoost optimization model is updated as follows: Among them, ω t+1,g represents the weight coefficient of the g-th data information sample after t+1 updates, ω t,g represents the weight coefficient of the g-th data information sample after t updates, Z t represents the normalization factor after t updates, α t represents the base classifier coefficient after t updates, y g is a label and y g ∈{-1,1},G t (x g ) represents the base classifier; Set the classification error rate ξ of the AdaBoost optimization model t for: In the formula, ω t,g I(G t (x g )≠y g ) represents the model G t (x g )The weight coefficient of misclassified samples.

4. A cross-border e-commerce transaction data management system according to claim 3, characterized in that: The loss function of the AdaBoost optimization model is the mean square error loss MSELoss, and the mean square error loss function is as follows: Among them, ξ d is the type of prediction data generated by the data information, It is the actual data type corresponding to the data information.

5. A cross-border e-commerce transaction data management system according to claim 4, characterized in that: The quantity updating unit is used to process the homogenized data information in the initial information total amount obtained by the data processing module, and to update the total amount of data information in each data type to obtain the updated total amount of data information— For se, Traverse and calculate the correlation coefficient for each type of data information: In the formula, Cov(c h ,c j ) indicates data information c h With data information j The covariance of Var[c h ] indicates data information c h The variance of Indicates data information c j The variance of .

6. A cross-border e-commerce transaction data management system according to claim 5, characterized in that: The quantity updating unit further includes: Set a threshold a, if the correlation coefficient r j ≥a, it means that the similarity between the data information is high, and the correlation coefficient r j The number of data information ≥ a is recorded as 1, that is, only the correlation coefficient r between the data information and all data information of the same type j <a can be recorded as a data quantity, and then obtain each data type— The total amount of data after homogeneous data is eliminated is se.

7. A cross-border e-commerce transaction data management system according to claim 6, characterized in that: The calculation unit of the benefit calculation module includes: Pre-calculation is performed based on the benefit formula. The benefit function formula is as follows: Q i =u i +Δu i ; Among them, Q i is the actual total benefit that the i-th supplier can obtain, u i is the direct benefit that the i-th supplier can obtain; Δu i is the indirect benefit that the ith supplier can obtain, ψe is the weight factor, f is the total number of suppliers, and |s| is the number of suppliers that can provide the same data type.

8. A cross-border e-commerce transaction data management system according to claim 7, characterized in that: The RBF neural network model includes: Build the RBF neural network framework and set the network output The weight vector is updated through the Gaussian basis function to obtain the mapping relationship between input and output; The loss function of the RBF neural network model is set to the optimized adversarial loss function; Construct a sample set and a measured set. The sample set and the measured set include kl groups of historical data. Each group of historical data includes the direct benefits that the supplier can obtain and the corresponding real total benefits. The direct benefits of each group of historical data in the sample set are used as the input of the RBF neural network model, and the corresponding real total benefits are used as the output. The RBF neural network model is trained to obtain the trained RBF neural network model. The measured data set is imported into the trained RBF neural network model, and the trained RBF neural network model is obtained when the resistance loss function is minimum and does not change.

9. A cross-border e-commerce transaction data management system according to claim 8, characterized in that: The adversarial loss function is: Among them, J(D,G) is the objective function of the entire generative adversarial network, where D is the discriminator network and G is the generator network; rec (G(y),x) is the reconstruction loss, which measures the reconstruction error between the predicted total benefit G(y) generated by the generator and the true total benefit x; l meas (y,o(G(y))) is the metric loss, which measures the metric error between the predicted total benefit G(y) generated by the generator and its output o(G(y)); H GAN (G(y),D) is the generative adversarial loss.

10. A cross-border e-commerce transaction data management system according to claim 9, characterized in that: The automatic vending machine management system is applied to a microcomputer and includes a human-computer interaction interface, which is used for information display. Suppliers access the system through a client and adopt an integrated information processing system and a multi-client information architecture.

Images