CN117196776A - Cross-border electronic commerce product credit marking and after-sale system based on random gradient lifting tree algorithm - Google Patents
Cross-border electronic commerce product credit marking and after-sale system based on random gradient lifting tree algorithm Download PDFInfo
- Publication number
- CN117196776A CN117196776A CN202311167185.0A CN202311167185A CN117196776A CN 117196776 A CN117196776 A CN 117196776A CN 202311167185 A CN202311167185 A CN 202311167185A CN 117196776 A CN117196776 A CN 117196776A
- Authority
- CN
- China
- Prior art keywords
- credit
- cross
- border
- buyer
- model
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000004422 calculation algorithm Methods 0.000 title claims abstract description 56
- 238000000034 method Methods 0.000 claims abstract description 53
- 238000009826 distribution Methods 0.000 claims abstract description 9
- 238000011156 evaluation Methods 0.000 claims description 40
- 238000013528 artificial neural network Methods 0.000 claims description 38
- 230000008569 process Effects 0.000 claims description 31
- 238000003066 decision tree Methods 0.000 claims description 27
- 238000012549 training Methods 0.000 claims description 25
- 238000012545 processing Methods 0.000 claims description 19
- 238000013145 classification model Methods 0.000 claims description 17
- 238000004364 calculation method Methods 0.000 claims description 13
- 238000007477 logistic regression Methods 0.000 claims description 9
- 238000013210 evaluation model Methods 0.000 claims description 8
- 238000013459 approach Methods 0.000 claims description 6
- 238000005516 engineering process Methods 0.000 claims description 6
- 238000012797 qualification Methods 0.000 claims description 5
- 238000007476 Maximum Likelihood Methods 0.000 claims description 4
- 230000005477 standard model Effects 0.000 claims description 3
- 230000008901 benefit Effects 0.000 abstract description 9
- 238000012423 maintenance Methods 0.000 abstract description 4
- 230000006870 function Effects 0.000 description 34
- FGUUSXIOTUKUDN-IBGZPJMESA-N C1(=CC=CC=C1)N1C2=C(NC([C@H](C1)NC=1OC(=NN=1)C1=CC=CC=C1)=O)C=CC=C2 Chemical compound C1(=CC=CC=C1)N1C2=C(NC([C@H](C1)NC=1OC(=NN=1)C1=CC=CC=C1)=O)C=CC=C2 FGUUSXIOTUKUDN-IBGZPJMESA-N 0.000 description 18
- 238000012937 correction Methods 0.000 description 10
- 238000010586 diagram Methods 0.000 description 7
- 238000004590 computer program Methods 0.000 description 5
- 238000010276 construction Methods 0.000 description 5
- 238000012360 testing method Methods 0.000 description 4
- 210000002569 neuron Anatomy 0.000 description 3
- 238000005457 optimization Methods 0.000 description 3
- 238000004088 simulation Methods 0.000 description 3
- 238000003860 storage Methods 0.000 description 3
- 230000009286 beneficial effect Effects 0.000 description 2
- 230000015572 biosynthetic process Effects 0.000 description 2
- 230000004069 differentiation Effects 0.000 description 2
- 230000005284 excitation Effects 0.000 description 2
- 238000011478 gradient descent method Methods 0.000 description 2
- 238000010801 machine learning Methods 0.000 description 2
- 230000001737 promoting effect Effects 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 238000003786 synthesis reaction Methods 0.000 description 2
- 230000004888 barrier function Effects 0.000 description 1
- 238000013500 data storage Methods 0.000 description 1
- 238000003745 diagnosis Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000010354 integration Effects 0.000 description 1
- 238000002372 labelling Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000003909 pattern recognition Methods 0.000 description 1
- 230000000644 propagated effect Effects 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
Landscapes
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
A cross-border e-commerce product credit marking and after-sales system based on a random gradient lifting tree algorithm comprises a cross-border e-commerce product cloud platform module, a cross-border e-commerce product credit grade classification module and a cross-border e-commerce product after-sales module, and aims at the problems that the manual operation is complex, the identity is different, the system distribution is inflexible and the product credit is difficult to distinguish in the existing international cross-border e-commerce integrated operation and maintenance problem, and the cloud platform and the intelligent algorithm are combined with the cross-border e-commerce to realize the integral system control of the cross-border e-commerce and the product credit tracking in different areas. The method has the advantages of wide application range and low economic cost, can be popularized to social application, enriches a cross-border free trade system, and brings good social and economic benefits.
Description
Technical Field
The invention relates to the field of cross-border electronic commerce, in particular to a cross-border electronic commerce product credit marking and after-sales system based on a random gradient lifting tree algorithm.
Background
The cross-border e-commerce refers to transaction subjects belonging to different environments, achieves transaction and electronic payment settlement through an e-commerce platform, and achieves goods through cross-border e-commerce logistics and off-site storage, thereby completing the international business activity of the transaction. The country deploys the layout construction of the propulsion logistics hub greatly, thereby promoting the improvement of the national economy running quality and efficiency. Cross-border electronic commerce is a technical foundation for promoting economic integration and trade globalization, and has very important strategic significance. The cross-border electronic commerce breaks through the barriers among countries, so that international trade moves to non-national trade, and simultaneously, the cross-border electronic commerce is also causing great transformation of world economic trade. For enterprises, the open, multidimensional and three-dimensional multi-side trade cooperation mode constructed by cross-border electronic commerce greatly widens the path for entering the international market and greatly promotes the optimal configuration of multi-side resources and the mutual win-win between enterprises; for consumers, cross-border e-commerce makes it very easy for them to obtain information from other countries and to purchase good and inexpensive goods. However, the problems of complex manual operation, difference in identity, inflexible system distribution, indistinguishable product credit and the like in the existing international cross-border e-commerce integrated operation and maintenance problem still exist. The method has the advantages of wide application range and low economic cost, can be popularized to social application, enriches a cross-border free trade system, and brings good social and economic benefits.
Disclosure of Invention
The invention aims to provide a cross-border e-commerce product credit marking and after-sales system based on a random gradient lifting tree algorithm so as to solve the problems in the background technology.
In order to achieve the above purpose, the cross-border e-commerce product credit marking and after-sales system based on the random gradient promotion tree algorithm comprises a cross-border e-commerce product cloud platform module, a cross-border e-commerce product credit level classification module and a cross-border e-commerce product after-sales module.
S1, tracing information of manufacturers and agent sellers of cross-border electronic commerce products, and primarily evaluating credit states of the manufacturers;
s2, marking sales comparison information of the same product in different international areas, constructing a system database of cross-border products, constructing a distributed cloud platform and access rights of cross-border e-commerce network agents based on the system database, a user interface and cloud services, and guaranteeing information security;
s3, recording the product processing problems in the transaction process in a system database, and classifying the transaction evaluation information of the same product in a unit time period based on a random gradient lifting tree algorithm to carry out multi-angle credit classification of the product manufacturer;
s4, based on the credit classification result of the product manufacturer, a dynamic credit assessment model is built by combining with the buyer evaluation grade, the buyer information is recorded in a system database, after-sales evaluation and speaking are tracked, the matching degree of the credit classification condition of the product manufacturer and the credit assessment model is verified, the buyer information is inverted and communicated with the buyer, if the malicious evaluation phenomenon of the buyer exists, the corresponding record is removed, and the relatively objective credit information is ensured;
s5, updating system data in real time, updating information in a unit time period to a user interface, providing purchasing basis for buyers, and providing credit constraint and selling service for cross-border electronic commerce;
further, the information tracing in the S1 includes one or more combinations of registered trademark, brand information and operation qualification of manufacturers and agent sellers of cross-border e-commerce products;
furthermore, the system database in the step S2 is mainly constructed by a MongoDB technology for processing non-relational and relational databases;
further, the transaction evaluation information of the same product in the unit time period in S3 is classified by the manufacturer credit of the product in multiple angles based on a random gradient lifting tree algorithm, and the detailed process is as follows:
using improved random gradient lifting tree algorithm (Grandient Boosting Decision Tree GBDT) for short, i.e. gradient lifting decision tree, machine learning algorithm for regression and classification research to obtain a model to make its predicted value F (x) of input variable approximate to its true value y, adopting greedy strategy to train a weak classification model each time to make the predicted value h of each base model m (x) Approximating the partial true value it needs to predict, thenCombining the predictors of the base models in a weighted manner; the structure of each weak classification model is a binary decision tree, and in the process of training the weak classification model, the model learns the true value y and the predicted value F after the previous iteration m-1 (x) The difference, i.e. the fit residual, residual y-F m-1 (x) Is the inverse gradient of the cubic loss function, as shown in the following equation:
wherein x represents attribute variable of the product database, and after residual error is fitted, previous iteration predicted value F m-1 (x) H adding the fitting residual error of the present round m (x) F obtained m (x) The square error loss function can be reduced, the predicted value F (x) of the integral model can be finally made to approach the true value y, at the moment, each basic model is fitted with the inverse gradient of the loss function, and the standard model is obtained after input, training and iteration. The present invention uses an improved random gradient lifting tree algorithm, residual y-F m-1 (x) And compared with the traditional random gradient lifting tree algorithm, the inverse gradient of the cubic loss function is used, the simulation predicted value iterates faster, and the algorithm efficiency is improved.
In the gradient descent method solving process, the learning rate v is set, the true value is easier to approach, and when v is reduced, the number of base models is correspondingly increased, so that the accuracy of the models is improved. The training algorithm for the gradient boosting decision tree is as follows:
inputting a given training setA differentiable loss function L (y, F (x)), the number of iterations M (i.e., the number of decision trees/the number of base models);
(1) With a constant gamma 0 Initializing a model:
wherein F is 0 (x) To initialize the model;is->Taking a variable point x corresponding to the minimum value; y is i For elements in a given training set.
(2) For m=1 to M:
calculating pseudo residual r min The following are provided:
according to training setConstructing a weak learner h m (x) Fitting the pseudo-residual;
the multiplier gamma is calculated by one-dimensional optimization as follows m ：
v represents the learning rate, and the calculation formula of the update model is as follows:
F m ＝F m-1 (x)+vγ m h m (x) 0<v≤1
(3) Output F M (x) Representing predictions of a strong classifier combined from a series of weak decision tree models.
And in the model construction process, combining a plurality of learners f (x) to form a hierarchical classifier G (x), and connecting a plurality of positive case output results of G (x) in series to form the cascaded classifier G (x) by using a GBDT cascade classification prediction model facing unbalanced data.
The cascade classifier sets different thresholds in each layer of classifier to divide samples and performs classification training. If the test sample passing through the classifier of the previous layer meets the threshold standard of the next layer, the next layer of classifier test can be entered, and so on.
The input to the classification model is the raw dataset s= { x, y }, where x is the buyerCredit rating, x= { d, h, l, c }, y is a two-class label of buyer credit manifestation. Each layer of output of the classification model is used for predicting product manufacturer creditz is the synthesis level output->And obtaining the credit of the predicted product manufacturer.
Transaction evaluation information of the same product is divided into a negative sample and a positive sample according to a hierarchical time standard. The evaluation data showed problems with tag imbalance. The positive and negative sample balance is realized by adopting a random undersampling method, and the specific method is as follows: for each level i (i.e. [1, m)]) Randomly undersampling the data to obtain k mutually independent positive and negative sample balanced data sets, wherein each data set is denoted as s i,j (j is the data set number), training to obtain k GBDT learners f i,j (x)(i∈[1,m],j∈[1,k]Combining the results of the k GBDT learners yields a final classification result.
GBDT is adopted as a learner f (x), and is an integrated learning model based on CART (classification and regression decision tree). The model trains a 1-group weak learner (CART decision tree) serially, and gradually fits and approximates the predicted delay time to the true value. And for the classification model, positive and negative classification is carried out on the sample, and a sigmoid function is adopted to calculate the classification.
Input sample set S i,j ＝(x,y i )i∈[1,m],j∈[1,k]. Where x is the input feature, x= { d, h, l, c }; y is i Is the actual delay tag corresponding to sample x. The learner is trained on a dataset having n samples at the i-th level. The construction step f (x) of the GBDT model is as follows:
step 1, initializing a learner f (x), and adjusting decision tree parameters to make the loss function L (y) by using a logarithmic loss function i F (x)) is minimized.
Wherein,predicting delay labels of the model on samples x; θ 1 Is a decision tree parameter;
step 2, negative gradient gamma using loss function i Fitting the residual, adjusting the parameter targets of the decision tree to minimize the loss function, and updating the model f (x).
Wherein θ 2 Is a decision tree parameter;
and 3, repeating the step 2 to finish L-1 times of iteration, and finishing probability calculation through a sigmoid function to realize category discrimination:
wherein σ is a constant; θ 1 Is a decision tree parameter;
GBDT multi-classification is the classification of samples containing multiple classification tags. Compared with GBDT multi-classification, the GBDT cascade classification prediction model performs balance processing of class data at each level, and each level is independent of each other and can be trained simultaneously. When the credit of the predicted product manufacturer is greater than the level standard, the level classifier predicts the output positive example, and enters the next level classifier for prediction; if the hierarchical classifier predicts an output negative case, the computation is terminated. And (5) connecting the prediction results of the hierarchical classifiers in series through the positive examples to obtain the credit of the predicted product manufacturer.
Further, in S4, a dynamic credit evaluation model is built in combination with the buyer evaluation level, and the detailed process is as follows:
constructing a dynamic credit assessment model by using a Logistic regression model, wherein in the practice of enterprise credit assessment, a linear structure contained in the model enables the model to have good stability and interpretability; assume that in a cross-border e-commerce product transaction, there are n buyers, x= (X) 1 ,x 2 ,…,x m ) Characteristic variables influencing the credit performance of the buyers, wherein m is the number of variables, each buyer is marked with a label v according to the credit performance of the buyer, and v represents a binary variable of whether the credit performance of the buyer is normal or not; if the credit of a buyer is to be evaluated to show the probability of occurrence of a violation, the prediction result of the calculation model is v=1 occurrence probability p, p=f (v= 1|x 1 ,x 2 ,…,x m ) The logistic regression logic (p) has the following specific expression:
wherein (beta) 0 ,β 1 ,β 2 ,…,β m ) To be the coefficient to be determined of the model, exp (beta) 0 +β 1 x 1 +β 2 x 2 +…+β m x m ) Representing a desired calculation;
the probability of occurrence p (v= 1|X) of the credit expression of the buyer can be obtained by obtaining the undetermined coefficient in the solution by using the maximum likelihood estimation method:
p(v＝0|X)＝1-θ
the probability function of v can be combined as:
according to the bernoulli distribution, the maximum likelihood function l (β, X) is written as:
the log likelihood function lnl (β, X)) is:
wherein θ i Representing the unknown parameters to be calculated, for Logistic regression, ln (l (β, X)):
wherein x is 1 ，x 2 ，x i Is a Bernoulli distribution parameter.
In the actual problem of constructing a dynamic credit evaluation model by a buyer evaluation level, most transaction process records hardly meet the premise assumption of probability distribution, but the Logistic regression model has good interpretability without excessively strict limiting conditions.
Further, in the step S4, the matching degree between the credit classification condition of the product manufacturer and the credit evaluation model is verified, and the buyer evaluation information is inverted, and the detailed process is as follows:
the improved BP neural network algorithm is used for inverting the buyer information, the method can approximate to any function, meanwhile, the number of intermediate layers, the number of processing units of each layer, the learning factors of the network and the like can be set according to actual conditions, the flexibility is high, and the BP neural network algorithm is used for inverting the buyer information:
let training sample set be x= [ X ] 1 ,X 2 ,…,X j ]For any sample set X k ＝[X k1 ,X k2 ,…,X km ]Obtain the actual output Y k ＝[Y k1 ,Y k2 ,…,Y kp ] T Its expected output is D k ＝[d k1 ,d k2 ,…,d kp ] T 。
In forward propagation of the BP neural network, there may be, from initial input to layer I:
wherein,representing logical computationsThe propagation process from layer I to layer J is:
wherein,is output data when BP neural network forward propagates to I layer, x km Is the sample value, w mi To be a propagation path from initial input to layer I, w ij For the propagation path from layer I to layer J, -/->For the forward propagation speed of the BP neural network from initial input to layer I,/>And f is the inversion result of the BP neural network algorithm on the buyer information, and the following layers are similar.
The propagation process from the J layer to the P layer is:
then a predicted output can be obtained:
wherein,initial output data for BP neural network, +.>As output data when the BP neural network propagates forward to the I layer,x km is the sample value, w io For initial input propagation path, w mi For the propagation path from the initial input to the I-layer, < > for>The forward propagation speed of the BP neural network from initial input to the I layer is obtained, and f is the inversion result of the BP neural network algorithm on the buyer information; the reverse error signal propagation of the BP neural network is realized through multiple error calculation and network iteration, the buyer evaluation data set is trained according to the principle, the inversion buyer information result is obtained, and if the phenomenon of malicious evaluation of the buyer exists, the corresponding record is removed, so that the relatively objective credit information is ensured;
the error is as follows:
e kp (n)＝d kp (n)-y kp (n)
wherein y is kp (n) is the actual output of the sample, d kp (n) defining error energy as the desired output of the sample
Then the error sum of the neurons is:
the forward process in the BP neural network is a process in which the signal propagates forward, and in order to minimize the error it gets, the error signal needs to propagate backward, and the model is fed back. The error back propagation process is as follows: correction amount between hidden layer J and output layer P in BP neural network, correction amount of weight in BP algorithm is proportional to partial differentiation of weight of error, namely:
where E (n) is the sum of the errors of the neurons, w IO (n) is a propagation path from the I layer to the O layer, and λ is a constant.
And has the following steps:
wherein y is kp (n) is the actual output of the sample, e kp (n) is an error, and,is output data when the BP neural network is propagated forward to the O layer.
The method can obtain the following steps:
the gradient is as follows:
from the calculation of the excitation function f (x):
the method can obtain the following steps:
thus, it is possible to obtain:
therefore, correction amount w ko (n) can be expressed as:
from this, an iterative formula can be derived:
w io (n+1)＝w io (n+Δw io (n)
similarly, the weight correction from the hidden layer I to the hidden layer J can be obtained as follows:
w ij (n+1)＝w ij (n)+Δw ij (n)
the weight correction from the hidden layer M to the hidden layer I is as follows:
w mi (n+1)＝w mi (n)+Δw mi (n)
the back error signal propagation of the BP neural network can be realized. Training the buyer evaluation data set according to the principle to obtain an inversion buyer information result, and if the phenomenon of malicious evaluation of the buyer exists, removing the corresponding record to ensure relatively objective credit information.
The invention has the beneficial effects that: the invention discloses a cross-border electronic commerce product credit marking and after-sales system based on a random gradient lifting tree algorithm, which comprises a cross-border electronic commerce product cloud platform module, a cross-border electronic commerce product credit grade classification module and a cross-border electronic commerce product after-sales module, wherein information tracing is carried out on manufacturers and agent sellers of the cross-border electronic commerce products, the information comprises registered trademarks, brand information, operation qualification and the like of the manufacturers and agent sellers of the cross-border electronic commerce products, the credit states of the manufacturers are primarily assessed, sales comparison information of the same products in different international areas is marked, a system database of the cross-border products is built by MongoDB technology for processing non-relational and relational databases, the access rights of a distributed cloud platform and a cross-border electronic commerce network agent are built based on the system database, a user interface and cloud service, the information security is ensured, the product processing problems in the transaction process are recorded in the system database, the transaction evaluation information of the same product in a unit time period is classified by multi-angle product manufacturers based on an improved GBDT algorithm, and the residual error y-F is obtained m-1 (x) The inverse gradient of the cubic loss function is used, compared with the traditional random gradient lifting tree algorithm, the simulation predicted value iterates faster, the algorithm efficiency is improved, and the Logistic regression model is used based on the credit classification result of a product manufacturerThe method comprises the steps of constructing a dynamic credit evaluation model by combining a buyer evaluation level, recording buyer information in a system database, tracking after-sale evaluation and speaking, verifying the matching degree of credit classification conditions of product manufacturers and the credit evaluation model, inverting the buyer information by using a BP neural network algorithm, inverting the buyer information, communicating with the buyer, eliminating corresponding records if the phenomenon of malicious evaluation of the buyer exists, ensuring relatively objective credit information, updating system data in real time, updating information in a unit time period to a user interface, providing purchasing basis for the buyer, and providing credit constraint and selling service for cross-border electronic commerce. Aiming at the problems of complex manual operation, difference in identity, inflexible system distribution, indistinguishable product credit and the like in the existing international cross-border e-commerce integrated operation and maintenance problem, the cloud platform and intelligent algorithm are combined with the cross-border e-commerce, and the integral system control of the cross-border e-commerce and the product credit tracking in different areas are realized. The method has the advantages of wide application range and low economic cost, can be popularized to social application, enriches a cross-border free trade system, and brings good social and economic benefits.
Drawings
The invention will be further described with reference to the accompanying drawings, in which embodiments do not constitute any limitation on the invention, and other drawings can be obtained by one of ordinary skill in the art without undue effort from the following drawings.
Fig. 1 is a schematic diagram of the structure of the present invention.
Detailed Description
The invention is further described in connection with the following examples.
Referring to fig. 1, the present invention is directed to providing a cross-border e-commerce product credit labeling and after-sales system based on a random gradient promotion tree algorithm, so as to solve the problems set forth in the above-mentioned background art.
In order to achieve the above purpose, the cross-border e-commerce product credit marking and after-sales system based on the random gradient promotion tree algorithm comprises a cross-border e-commerce product cloud platform module, a cross-border e-commerce product credit level classification module and a cross-border e-commerce product after-sales module.
S1, tracing information of manufacturers and agent sellers of cross-border electronic commerce products, and primarily evaluating credit states of the manufacturers, wherein the credit states comprise one or more combinations of registered trademarks, brand information and operation qualification of the manufacturers and agent sellers of the cross-border electronic commerce products;
s2, marking sales comparison information of the same product in different international areas, constructing a system database of the cross-border product by using a MongoDB technology for processing non-relational and relational databases, and constructing access rights of a distributed cloud platform and a cross-border e-commerce network agent based on the system database, a user interface and a cloud service to ensure information security;
s3, recording the product processing problems in the transaction process in a system database, classifying the transaction evaluation information of the same product in a unit time period based on a random gradient lifting tree algorithm by multi-angle product manufacturer credit, wherein the detailed process is as follows:
a random gradient lifting tree algorithm (Grandient Boosting Decision Tree is GBDT) is used, namely a gradient lifting decision tree is a machine learning algorithm for regression and classification research, and is formed by combining a series of integrated weak classification models, each weak classification model respectively gives a predicted value, and the predicted values are frustrated together according to a certain weight to form a final predicted value, so that a strong classifier model can be finally obtained. Generally, the goal of the training is to find a model that approximates its predicted value F (x) to its true value y for the input variable, and the GBDT algorithm uses a greedy strategy to train only one weak classification model (weak classifier) at a time, i.e., let the predicted value h of each base model m (x) Approximating the partial true values it needs to predict, and then weighting the predicted values of these base models.
The structure of each weak classification model is a binary decision tree, and in the process of training the weak classification model, the model learns the true value y and the predicted value F after the previous iteration m-1 (x) The difference, i.e. the fit residual, residual y-F m-1 (x) Is the inverse gradient of the cubic loss function, as shown in the following equation:
wherein x represents attribute variable of the product database, and after residual error is fitted, previous iteration predicted value F m-1 (x) H adding the fitting residual error of the present round m (x) F obtained m (x) The square error loss function can be reduced, the predicted value F (x) of the integral model can be finally made to approach the true value y, at the moment, each basic model is fitted with the inverse gradient of the loss function, and the standard model is obtained after input, training and iteration.
In the gradient descent method solving process, the learning rate v is set, the true value is easier to approach, and when v is reduced, the number of base models is correspondingly increased, so that the accuracy of the models is improved. The training algorithm for the gradient boosting decision tree is as follows:
inputting a given training setA differentiable loss function L (y, F (x)), the number of iterations M (i.e., the number of decision trees/the number of base models);
(1) With a constant gamma 0 Initializing a model:
(2) For m=1 toM:
calculating pseudo residual r min The following are provided:
according to training setConstructing a weak learner h m (x) Fitting the pseudo-residual;
the multiplier gamma is calculated by one-dimensional optimization as follows m ：
v represents the learning rate, and the calculation formula of the update model is as follows:
F m ＝F m-1 (x)+vγ m h m (x) 0<v≤1
(3) Output F M (x) Representing predictions of a strong classifier combined from a series of weak decision tree models.
In the invention, a GBDT cascade classification prediction model facing unbalanced data is used, a plurality of learners f (x) are combined to form a hierarchical classifier G (x) in the model construction process, and positive case output results of a plurality of G (x) are connected in series to form a cascade classifier G (x).
The cascade classifier sets different thresholds in each layer of classifier to divide samples and performs classification training. If the test sample passing through the classifier of the previous layer meets the threshold standard of the next layer, the next layer of classifier test can be entered, and so on.
The input of the classification model is the original dataset s= { x, y }, where x is the buyer credit rating level, x= { d, h, l, c }, and y is the buyer credit performance's classification label. Each layer of output of the classification model is used for predicting product manufacturer creditz is the synthesis level output->And obtaining the credit of the predicted product manufacturer.
Transaction evaluation information of the same product is divided into a negative sample and a positive sample according to a hierarchical time standard. The evaluation data showed problems with tag imbalance. The positive and negative sample balance is realized by adopting a random undersampling method, and the specific method is as follows: for each level i (i.e. [1, m)]) Randomly undersampling the data to obtain k mutually independent positive and negative sample balanced data sets, wherein each data set is denoted as s i,j (j is the data set number), training to obtain k GBDT learners f i,j (x)(i∈[1,m],j∈[1,k]Combining the results of the k GBDT learners yields a final classification result.
GBDT is adopted as a learner f (x), and is an integrated learning model based on CART (classification and regression decision tree). The model trains a 1-group weak learner (CART decision tree) serially, and gradually fits and approximates the predicted delay time to the true value. And for the classification model, positive and negative classification is carried out on the sample, and a sigmoid function is adopted to calculate the classification.
Input sample set s i,j ＝(x,y i )i∈[1,m],j∈[1,k]. Where x is the input feature, x= { d, h, l, c }; y is i Is the actual delay tag corresponding to sample x. The learner is trained on a dataset having n samples at the i-th level. The construction step f (x) of the GBDT model is as follows:
step 1, initializing a learner f (x), and adjusting decision tree parameters to make the loss function L (y) by using a logarithmic loss function i F (x)) is minimized.
Wherein,predicting delay labels of the model on samples x; θ 1 Is a decision tree parameter.
Step 2, negative gradient gamma using loss function i Fitting the residual, adjusting the parameter targets of the decision tree to minimize the loss function, and updating the model f (x).
Wherein θ 2 Is a decision tree parameter.
And step 3, repeating the step 2 to finish L-1 times of iteration, and finishing probability calculation through a sigmoid function to realize class discrimination.
Wherein θ 1 Is a decision tree parameter.
GBDT multi-classification is the classification of samples containing multiple classification tags. Compared with GBDT multi-classification, the GBDT cascade classification prediction model performs balance processing of class data at each level, and each level is independent of each other and can be trained simultaneously. When the credit of the predicted product manufacturer is greater than the level standard, the level classifier predicts the output positive example, and enters the next level classifier for prediction; if the hierarchical classifier predicts an output negative case, the computation is terminated. The prediction results of all the layers of classifiers are connected in series through positive examples, so that the credit of a predicted product manufacturer is obtained;
s4, based on the credit classification result of the product manufacturer, a dynamic credit assessment model is built by combining with the buyer evaluation grade, the buyer information is recorded in a system database, after-sales evaluation and speaking are tracked, the matching degree of the credit classification condition of the product manufacturer and the credit assessment model is verified, the buyer information is inverted and communicated with the buyer, if the malicious evaluation phenomenon of the buyer exists, the corresponding record is removed, and the relatively objective credit information is ensured, wherein the detailed process is as follows:
the BP neural network algorithm is used for inverting the buyer information, has most of the advantages of an artificial neural network, is a supervised self-adaptive learning algorithm in the artificial neural network, can be also called an error back propagation algorithm, and has the characteristics of good nonlinear mapping performance and can approach any function theoretically, and the basic structure of the BP neural network is composed of nonlinear variable units; meanwhile, the number of middle layers, the number of processing units of each layer, the learning factors of a network and the like can be set according to actual conditions, the method has high flexibility, and has wide application prospects in the aspects of optimization, signal processing, pattern recognition, intelligent control, fault diagnosis and the like, so that the method uses BP neural network algorithm to invert buyer information:
let training sample set be x= [ X ] 1 ,X 2 ,…,X j ]For any sample set X k ＝[X k1 ,X k2 ,…,X km ]Obtain the actual output Y k ＝[Y k1 ,Y k2 ,…,Y kp ] T Its expected output is D k ＝[d k1 ,d k2 ,…,d kp ] T 。
In forward propagation of the BP neural network, there may be, from initial input to layer I:
wherein,representing logically calculated weights, the propagation process from layer I to layer J is: />
The propagation process from the J layer to the P layer is:
then a predicted output can be obtained:
wherein,initial output data for BP neural network, +.>Is output data when BP neural network forward propagates to I layer, x km Is the sample value, w io For initial input propagation path, w mi For the propagation path from the initial input to the I-layer, < > for>The forward propagation speed of the BP neural network from initial input to the I layer is obtained, and f is the inversion result of the BP neural network algorithm on the buyer information; the reverse error signal propagation of the BP neural network is realized through multiple error calculation and network iteration, the buyer evaluation data set is trained according to the principle, the inversion buyer information result is obtained, and if the phenomenon of malicious evaluation of the buyer exists, the corresponding record is removed, so that the relatively objective credit information is ensured;
the error is as follows:
e kp (n)＝d kp (n)-y kp (n)
defining the error energy as
Then the error sum of the neurons is:
the forward process in the BP neural network is a process in which the signal propagates forward, and in order to minimize the error it gets, the error signal needs to propagate backward, and the model is fed back. The error back propagation process is as follows: correction amount between hidden layer J and output layer P in BP neural network, correction amount of weight in BP algorithm is proportional to partial differentiation of weight of error, namely:
and has the following steps:
the method can obtain the following steps:
/>
the gradient is as follows:
from the calculation of the excitation function f (x):
the method can obtain the following steps:
thus, it is possible to obtain:
therefore, correction amount w ko (n) can be expressed as:
from this, an iterative formula can be derived:
w io (n+1)＝w io (n)+Δw io (n)
similarly, the weight correction from the hidden layer I to the hidden layer J can be obtained as follows:
w ij (n+1)＝w ij (n)+Δw ij (n)
the weight correction from the hidden layer M to the hidden layer I is as follows:
w mi (n+1)＝w mi (n)+Δw mi (n)
the back error signal propagation of the BP neural network can be realized. Training the buyer evaluation data set according to the principle to obtain an inversion buyer information result;
s5, updating system data in real time, updating information in a unit time period to a user interface, providing purchasing basis for buyers, and providing credit constraint and selling service for cross-border electronic commerce.
The invention has the beneficial effects that: the invention discloses a cross-border electronic commerce product credit marking and after-sales system based on a random gradient lifting tree algorithm, which comprises a cross-border electronic commerce product cloud platform module, a cross-border electronic commerce product credit grade classification module and a cross-border electronic commerce product after-sales module, wherein information tracing is carried out on manufacturers and agent sellers of the cross-border electronic commerce products, the information comprises registered trademarks, brand information, operation qualification and the like of the manufacturers and agent sellers of the cross-border electronic commerce products, the credit states of the manufacturers are primarily assessed, sales comparison information of the same products in different international areas is marked, a system database of the cross-border products is built by MongoDB technology for processing non-relational and relational databases, the access rights of a distributed cloud platform and a cross-border electronic commerce network agent are built based on the system database, a user interface and cloud service, the information security is ensured, the product processing problems in the transaction process are recorded in the system database, the transaction evaluation information of the same product in a unit time period is classified by multi-angle product manufacturers based on an improved GBDT algorithm, and the residual error y-F is obtained m-1 (x) The inverse gradient of the cubic loss function is used, compared with a traditional random gradient lifting tree algorithm, the simulation predicted value iterates faster, algorithm efficiency is improved, based on the credit classification result of a product manufacturer, a Logistic regression model is used for constructing a dynamic credit evaluation model in combination with the buyer evaluation level, buyer information is recorded in a system database, after-sales evaluation and speaking are tracked, matching degree of the credit classification condition of the product manufacturer and the credit evaluation model is verified, the buyer information is inverted by using a BP neural network algorithm, the buyer information is inverted and communicated with the buyer, if the malicious evaluation phenomenon of the buyer exists, corresponding records are removed, the relatively objective credit information is guaranteed, system data is updated in real time, information in a unit time period is updated to a user interface, a purchasing basis is provided for the buyer, and credit constraint and selling service are provided for a cross-border electronic commerce. The invention aims at the cross-border electrons in the prior internationalThe problems of complex manual operation, difference in identity, inflexible system distribution, indistinguishable product credit and the like in the business integrated operation and maintenance problem are solved, and the cloud platform and the intelligent algorithm are combined with the cross-border electronic business to realize the integral system control of the cross-border electronic business and the product credit tracking in different areas. The method has the advantages of wide application range and low economic cost, can be popularized to social application, enriches a cross-border free trade system, and brings good social and economic benefits.
The present invention also provides a computer readable storage medium having stored therein at least one instruction that is loaded and executed by a processor to implement the above-described method. The computer readable storage medium may be, among other things, ROM, random access memory, CD-ROM, magnetic tape, floppy disk, optical data storage device, etc. The instructions stored therein may be loaded by a processor in the terminal and perform the methods described above.
The present invention is described with reference to flowchart illustrations and/or block diagrams of methods, terminal devices (systems), and computer program products according to embodiments of the invention. It will be understood that each flow and/or block of the flowchart illustrations and/or block diagrams, and combinations of flows and/or blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, embedded processor, or other programmable data processing terminal device to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing terminal device, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks. These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
The previous description of the disclosed embodiments is provided to enable any person skilled in the art to make or use the present invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other embodiments without departing from the spirit or scope of the invention. Thus, the present invention is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.
Claims (6)
1. A cross-border e-commerce product credit marking and after-sales system based on a random gradient lifting tree algorithm comprises a cross-border e-commerce product cloud platform module, a cross-border e-commerce product credit level classification module and a cross-border e-commerce product after-sales module;
s1, tracing information of manufacturers and agent sellers of cross-border electronic commerce products, and primarily evaluating credit states of the manufacturers;
s2, marking sales comparison information of the same product in different international areas, constructing a cross-border e-commerce product database, and constructing access rights of a distributed cloud platform and a cross-border e-commerce network agent based on a system database, a user interface and cloud services to ensure information security;
s3, recording the product processing problems in the transaction process in a system database, and classifying the transaction evaluation information of the same product in a unit time period based on a random gradient lifting tree algorithm to carry out multi-angle credit classification of the product manufacturer;
s4, based on the credit classification result of the product manufacturer, a dynamic credit assessment model is built by combining with the buyer evaluation grade, the buyer information is recorded in a system database, after-sales evaluation and speaking are tracked, the matching degree of the credit classification condition of the product manufacturer and the credit assessment model is verified, the buyer information is inverted and communicated with the buyer, if the malicious evaluation phenomenon of the buyer exists, the corresponding record is removed, and the relatively objective credit information is ensured;
s5, updating system data in real time, updating information in a unit time period to a user interface, providing purchasing basis for buyers, and providing credit constraint and selling service for cross-border electronic commerce.
2. The cross-border e-commerce product credit marking and after-sales system based on the random gradient promotion tree algorithm of claim 1, wherein the information tracing in S1 comprises a combination of registered trademarks, brand information and operation qualification of manufacturers and agent sellers of the cross-border e-commerce products.
3. The cross-border e-commerce product credit marking and after-sales system based on the random gradient promotion tree algorithm of claim 1, wherein the system database in S2 is mainly built by a MongoDB technology for processing non-relational and relational databases.
4. The cross-border e-commerce product credit marking and after-sales system based on the random gradient promotion tree algorithm of claim 1, wherein the transaction evaluation information of the same product in the unit time period in S3 is classified by multi-angle product manufacturer credit based on the improved GBDT algorithm, and the detailed process is as follows:
training a weak classification model each time by adopting a greedy strategy, and enabling the predicted value h of each base model m (x) Approximating the partial true values that it needs to predict, and then weighting and combining the predicted values of the base models; the structure of each weak classification model is a binary decision tree, and in the process of training the weak classification model, the model learns the true value y and the predicted value F after the previous iteration m-1 (x) The difference, i.e. the fit residual, residual y-F m-1 (x) Is the inverse gradient of the cubic loss function, x is the input manufacturer's credit argument, as shown in the following equation:
wherein x represents attribute variable of the product database, and after residual error is fitted, previous iteration predicted value F m-1 (x) Adding the predicted value h obtained by the round m (x) F obtained m (x) The square error loss function is reduced, the predicted value F (x) of the integral model is finally enabled to approach the true value y, at the moment, each basic model is fitted with the inverse gradient of the loss function, and a standard model is obtained after input, training and iteration; residual y-F using improved GBDT algorithm m-1 (x) An inverse gradient of the cubic loss function is used.
5. The cross-border e-commerce product credit marking and after-sales system based on the random gradient promotion tree algorithm of claim 1, wherein the dynamic credit assessment model is constructed by combining the buyer evaluation level in S4, and the detailed process is as follows:
constructing a dynamic credit assessment model by using a Logistic regression model, wherein in the practice of enterprise credit assessment, a linear structure contained in the model enables the model to have good stability and interpretability; assume that in a cross-border e-commerce product transaction, there are n buyers, x= (X) 1 ,x 2 ,…,x m ) Characteristic variables influencing the credit performance of the buyers, wherein m is the number of variables, each buyer is marked with a label v according to the credit performance of the buyer, and v represents a binary variable of whether the credit performance of the buyer is normal or not; if the credit of a buyer is to be evaluated to show the probability of occurrence of a violation, the prediction result of the calculation model is v=1 occurrence probability p, p=f (v= 1|x 1 ,x 2 ,…,x m ) The logistic regression logic (p) has the following specific expression:
wherein (beta) 0 ,β 1 ,β 2 ,…,β m ) To be the coefficient to be determined of the model, exp (beta) 0 +β 1 x 1 +β 2 x 2 +…+β m x m ) Representing a desired calculation;
the probability of occurrence p (v= 1|X) of the credit expression of the buyer can be obtained by obtaining the undetermined coefficient in the solution by using the maximum likelihood estimation method:
p(v＝0|X)＝1-θ
the probability function of v can be combined as:
according to the bernoulli distribution, the maximum likelihood function l (β, X) is written as:
the log-likelihood function ln (l (β, X)) is:
wherein θ i Representing the unknown parameters to be calculated, for Logistic regression, ln (l (β, X)):
wherein x is 1 ，x 2 ，x i Is a Bernoulli distribution parameter.
6. The cross-border e-commerce product credit marking and after-sales system based on the random gradient lifting tree algorithm of claim 1, wherein the step S4 is to verify the matching degree of the credit classification condition and the credit evaluation model of the product manufacturer, invert the buyer evaluation information, and comprises the following detailed procedures:
inversion is carried out on buyer information by using BP neural network algorithm, and a training sample set is made to be X= [ X ] 1 ,X 2 ,…,X j ]For any sample set X k ＝[X k1 ,X k2 ,…,X km ]Obtain the actual output Y k ＝[Y k1 ,Y k2 ,…,Y kp ] T In forward propagation of the BP neural network, there is from initial input to layer I:
wherein,representing the logically calculated weights, then the predicted output is obtained:
wherein,initial output data for BP neural network, +.>Is output data when BP neural network forward propagates to I layer, x km Is the sample value, w io For initial input propagation path, w mi For the propagation path from the initial input to the I-layer, < > for>The forward propagation speed of the BP neural network from initial input to the I layer is obtained, and f is the inversion result of the BP neural network algorithm on the buyer information; and (3) realizing reverse error signal propagation of the BP neural network through repeated error calculation and network iteration, training a buyer evaluation data set according to the principle to obtain an inversion buyer information result, and if the phenomenon of malicious evaluation of the buyer exists, eliminating corresponding records to ensure relatively objective credit information.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311167185.0A CN117196776A (en) | 2023-09-09 | 2023-09-09 | Cross-border electronic commerce product credit marking and after-sale system based on random gradient lifting tree algorithm |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311167185.0A CN117196776A (en) | 2023-09-09 | 2023-09-09 | Cross-border electronic commerce product credit marking and after-sale system based on random gradient lifting tree algorithm |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN117196776A true CN117196776A (en) | 2023-12-08 |
Family
ID=89004768
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311167185.0A Pending CN117196776A (en) | 2023-09-09 | 2023-09-09 | Cross-border electronic commerce product credit marking and after-sale system based on random gradient lifting tree algorithm |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117196776A (en) |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111652710A (en) * | 2020-06-03 | 2020-09-11 | 北京化工大学 | Personal credit risk assessment method based on ensemble tree feature extraction and Logistic regression |
| WO2022088606A1 (en) * | 2020-10-29 | 2022-05-05 | 平安科技（深圳）有限公司 | Gbdt and lr fusion method and apparatus based on federated learning, device, and storage medium |
| CN114707699A (en) * | 2022-03-04 | 2022-07-05 | 易视腾科技股份有限公司 | Network television order unsubscribe prediction method based on machine learning |
| CN115439180A (en) * | 2022-08-26 | 2022-12-06 | 中国银联股份有限公司 | Target object determination method and device, electronic equipment and storage medium |
| CN116091206A (en) * | 2023-01-31 | 2023-05-09 | 金电联行（北京）信息技术有限公司 | Credit evaluation method, credit evaluation device, electronic equipment and storage medium |
| CN116596570A (en) * | 2023-05-11 | 2023-08-15 | 广东德澳智慧医疗科技有限公司 | Information comparison system of same product in different E-commerce platforms based on big data analysis algorithm |
-
2023
- 2023-09-09 CN CN202311167185.0A patent/CN117196776A/en active Pending
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111652710A (en) * | 2020-06-03 | 2020-09-11 | 北京化工大学 | Personal credit risk assessment method based on ensemble tree feature extraction and Logistic regression |
| WO2022088606A1 (en) * | 2020-10-29 | 2022-05-05 | 平安科技（深圳）有限公司 | Gbdt and lr fusion method and apparatus based on federated learning, device, and storage medium |
| CN114707699A (en) * | 2022-03-04 | 2022-07-05 | 易视腾科技股份有限公司 | Network television order unsubscribe prediction method based on machine learning |
| CN115439180A (en) * | 2022-08-26 | 2022-12-06 | 中国银联股份有限公司 | Target object determination method and device, electronic equipment and storage medium |
| CN116091206A (en) * | 2023-01-31 | 2023-05-09 | 金电联行（北京）信息技术有限公司 | Credit evaluation method, credit evaluation device, electronic equipment and storage medium |
| CN116596570A (en) * | 2023-05-11 | 2023-08-15 | 广东德澳智慧医疗科技有限公司 | Information comparison system of same product in different E-commerce platforms based on big data analysis algorithm |
Non-Patent Citations (1)
| Title |
|---|
| 马百皓;: "基于BP神经网络的农产品跨境电商人才培养方案设计与应用", 微型电脑应用, no. 05 * |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Popat et al. | A survey on credit card fraud detection using machine learning | |
| CN109978538B (en) | Method and device for determining fraudulent user, training model and identifying fraudulent risk | |
| Hajizadeh et al. | Application of data mining techniques in stock markets: A survey | |
| CN107330741A (en) | Graded electron-like certificate uses Forecasting Methodology, device and electronic equipment | |
| Samal et al. | Optimization of variable demand fuzzy economic order quantity inventory models without and with backordering | |
| CN111352965B (en) | Training method of sequence mining model, and processing method and equipment of sequence data | |
| US20210350393A1 (en) | Behavioral data analytics platform | |
| Wu et al. | Wavelet fuzzy cognitive maps | |
| CN113283936A (en) | Sales forecasting method, sales forecasting device and electronic equipment | |
| CN114240659A (en) | Block chain abnormal node identification method based on dynamic graph convolutional neural network | |
| CN113144624B (en) | Data processing method, device, equipment and storage medium | |
| Zhang | Prediction of Purchase Volume of Cross‐Border e‐Commerce Platform Based on BP Neural Network | |
| Mohbey et al. | Credit card fraud prediction using XGBoost: an ensemble learning approach | |
| US20240161117A1 (en) | Trigger-Based Electronic Fund Transfers | |
| CN117196776A (en) | Cross-border electronic commerce product credit marking and after-sale system based on random gradient lifting tree algorithm | |
| Vidanelage et al. | Study on machine learning techniques with conventional tools for payment fraud detection | |
| CN116303786A (en) | Block chain financial big data management system based on multidimensional data fusion algorithm | |
| Peng | Research on credit risk identification of internet financial enterprises based on big data | |
| Jackson et al. | Automl approach to classification of candidate solutions for simulation models of logistic systems | |
| CN113379456A (en) | Credit interest rate differential pricing method, device, equipment and storage medium | |
| Raximov et al. | The importance of loss function in artificial intelligence | |
| Wang et al. | Risk assessment of supply chain based on BP neural network | |
| Petelin et al. | Financial modeling using Gaussian process models | |
| CN111179070A (en) | Loan risk timeliness prediction system and method based on LSTM | |
| Peng et al. | Intelligent Optimization Model of Enterprise Financial Account Receivable Management |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |