CN117237114A - Financing trade compliance detection method based on twin evolution - Google Patents
Financing trade compliance detection method based on twin evolution Download PDFInfo
- Publication number
- CN117237114A CN117237114A CN202311492810.9A CN202311492810A CN117237114A CN 117237114 A CN117237114 A CN 117237114A CN 202311492810 A CN202311492810 A CN 202311492810A CN 117237114 A CN117237114 A CN 117237114A
- Authority
- CN
- China
- Prior art keywords
- trade
- data
- similarity
- full
- financing
- 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.)
- Granted
Links
- 238000001514 detection method Methods 0.000 title claims abstract description 39
- 239000013598 vector Substances 0.000 claims abstract description 71
- 239000011159 matrix material Substances 0.000 claims abstract description 62
- 238000000034 method Methods 0.000 claims abstract description 30
- 230000001537 neural effect Effects 0.000 claims abstract description 30
- 230000002068 genetic effect Effects 0.000 claims abstract description 9
- 230000004913 activation Effects 0.000 claims description 22
- 238000012545 processing Methods 0.000 claims description 12
- 230000008569 process Effects 0.000 claims description 9
- 230000009467 reduction Effects 0.000 claims description 9
- 238000002372 labelling Methods 0.000 claims description 6
- 230000001105 regulatory effect Effects 0.000 claims description 6
- 238000004364 calculation method Methods 0.000 claims description 5
- 238000005457 optimization Methods 0.000 claims description 4
- 230000035772 mutation Effects 0.000 claims description 3
- 238000012216 screening Methods 0.000 claims description 3
- 230000002195 synergetic effect Effects 0.000 claims description 3
- 230000006870 function Effects 0.000 description 55
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000007796 conventional method 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
- 238000011946 reduction process Methods 0.000 description 1
Landscapes
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The application relates to a financing trade compliance detection method based on twin evolution, which comprises the following steps: acquiring trade data and trade specification data; vectorizing the trade data and the trade specification data to obtain trade feature vectors and trade specification data matrixes respectively; constructing trade data matrixes based on trade feature vectors of different enterprises; inputting the trade data matrix and the trade specification data matrix into an optimized twin neural cooperative network, and outputting trade-specification characteristics; optimizing a comparison function through a genetic algorithm based on trade-specification characteristics, and determining the similarity between the target financing trade data and the trade specification data according to the optimized comparison function; and determining a compliance detection result based on the similarity. The method can efficiently and accurately calculate the similarity between the target financing trade and the normative clause, thereby realizing the automatic detection of the financing trade compliance.
Description
Technical Field
The application relates to the technical field of financing trade compliance detection, in particular to a method for detecting financing trade compliance based on twin evolution.
Background
Traditional financing trade compliance detection methods rely primarily on manual and empirical judgment and may lead to subjective errors. Because of the large and complex data involved in financing trade, it is difficult to integrate and analyze efficiently, and conventional methods have limitations in multidimensional data processing. In addition, the existing method cannot efficiently and accurately calculate the similarity degree between the target financing trade and the standard clause, so that the compliance detection result of the target financing trade cannot be accurately judged. Thus, there is a need for an efficient, accurate, automated method of financing trade compliance detection.
Disclosure of Invention
Based on the above, it is necessary to provide a detection method capable of efficiently and accurately calculating the similarity between the target financing trade and the normative terms, so as to automatically detect the financing trade compliance, in particular to a financing trade compliance detection method based on twin evolution.
The application provides a financing trade compliance detection method based on twin evolution, which comprises the following steps:
s1: acquiring trade data and trade specification data;
s2: vectorizing the trade data and the trade specification data to obtain trade feature vectors and trade specification data matrixes respectively; constructing trade data matrixes based on trade feature vectors of different enterprises;
s3: inputting the trade data matrix and the trade specification data matrix into an optimized twin neural cooperative network, and outputting trade-specification characteristics;
s4: optimizing a comparison function through a genetic algorithm based on trade-specification characteristics, and determining the similarity between the target financing trade data and the trade specification data according to the optimized comparison function;
s5: and determining a compliance detection result based on the similarity.
Preferably, the trade data includes transaction records, contracts and agreements, customer relationship management data, annual reports and logistic documents; the transaction record comprises transaction amount, transaction date and transaction counterpart; the contract is used to describe the terms and details of the transaction; the customer relationship management data includes customer ratings, customer feedback; the annual report and logistics document comprises profit data and a transportation route description; the trade specification data includes regulatory policies, trade specification data; the regulatory policy is used to describe the regulations and requirements of financing trade; the trade specification data includes a transaction limit and a compliant contract term paradigm.
Preferably, the vectorization process includes:
carrying out standardization processing on the transaction amount to obtain an amount value vector;
converting the transaction date into a date value vector;
the transaction counterpart is converted into a transaction personal text vector by Word2Vec Word embedding technology;
performing independent heat coding on the client rating to obtain a client rating numerical vector;
the client feedback is converted into a client feedback text vector through Word2Vec Word embedding technology;
carrying out standardization processing on the profit data to obtain a profit value vector;
the transportation route description is converted into a transportation route text vector through Word2Vec Word embedding technology;
the supervision policy is converted into a supervision policy text vector by Word2Vec Word embedding technology;
carrying out standardization processing on the transaction quota to obtain a quota numerical vector;
the compliant treaty clause example is converted into a compliant clause text vector by Word2Vec Word embedding technology.
Preferably, the monetary value vector, the date value vector, the transactor text vector, the customer level value vector, the customer feedback text vector, the profit value vector and the transportation route text vector are combined to obtain the trade feature vector;
and combining the supervision policy text vector, the quota value vector and the compliance term text vector to obtain the trade specification data matrix.
Preferably, in S3, the method further includes: before the trade data matrix and the trade specification data matrix are input to the optimized twin neural cooperative network, the dimension reduction is carried out on the trade data matrix and the trade specification data matrix respectively by adopting a PCA technology, and the dimension of the trade data matrix and the dimension of the trade specification data matrix are unified.
Preferably, the twin neural cooperative network comprises two identical multi-layer perceptrons; each multi-layer perceptron comprises three full-connection layers, and each full-connection layer is followed by a ReLU activation function;
the forward propagation of the first multi-layer perceptron is expressed as:
z E 1 =W E1 ·E+b E1 ；
a E 1 =ReLU(z E1 )；
z E 2 =W E2 ·a E1 +b E2 ；
a E 2 =ReLU(z E2 )；
z E 3 =W E3 ·a E2 +b E3 ；
h E =ReLU(z E3 )；
wherein,h E representing the trade data matrix after forward propagation;Erepresenting the trade data matrix after dimension reduction;z E1 、z E2 、z E3 respectively representing the output of a first full-connection layer, a second full-connection layer and a third full-connection layer of the first multi-layer perceptron; w (W) E1 、W E2 、W E3 Respectively representing weight matrixes of a first full-connection layer, a second full-connection layer and a third full-connection layer of the first multi-layer perceptron; b E1 、b E2 、b E3 Respectively representing the bias of a first full-connection layer, a second full-connection layer and a third full-connection layer of the first multi-layer perceptron;a E1 、a E2 respectively representing the output of a first ReLU activation function and a second ReLU activation function of a first multi-layer perceptron; reLU (·) represents a ReLU activation function;
the forward propagation of the second multi-layer perceptron is expressed as:
z G 1 =W G1 ·E+b G1 ；
a G 1 =ReLU(z G1 )；
z G 2 =W G2 ·a G1 +b G2 ；
a G 2 =ReLU(z G2 )；
z G 3 =W G3 ·a G2 +b G3 ；
h G =ReLU(z G3 )；
wherein,h G representing the trade specification data matrix after forward propagation;Grepresenting the trade specification data matrix after dimension reduction;z G1 、z G2 、z G3 respectively representing the output of a first full-connection layer, a second full-connection layer and a third full-connection layer of the second multi-layer perceptron; w (W) G1 、W G2 、W G3 Respectively representing weight matrixes of a first full-connection layer, a second full-connection layer and a third full-connection layer of the second multi-layer perceptron; b G1 、b G2 、b G3 Respectively representing the bias of a first full-connection layer, a second full-connection layer and a third full-connection layer of the second multi-layer perceptron;a G1 、a G2 respectively representing the output of a first ReLU activation function and a second ReLU activation function of a second multi-layer perceptron;
and carrying out tensor product operation on the trade data matrix after forward propagation and the trade specification data matrix after forward propagation to obtain initial trade-specification characteristics.
Preferably, the optimization process of the twin neural cooperative network includes:
step 1: obtaining a true similarity label corresponding to an initial trade-specification feature, and calculating a loss function based on the initial trade-specification feature and the true similarity label corresponding to the initial trade-specification feature, wherein the loss function expression is as follows:
；
wherein,L(. Cndot.) represents the loss function,Trepresenting the initial trade-specification characteristics of the trade,Ytrue similarity labeling representing initial trade-specification characteristics;T i represent the firstiInitial trade-specification characteristics of the individual samples;Y i represent the firstiTrue similarity labeling of initial trade-specification features of individual samples;nis the number of samples;
step 2: calculating the gradient of the loss function relative to network parameters by adopting a chained rule, and calculating a gradient value for each network parameter;
step 3: updating weight parameters and bias parameters of the twin neural cooperative network based on the calculated gradient values to minimize the loss function;
step 4: repeating the steps 1-3 until the loss function converges or the maximum iteration number is reached, and obtaining the optimized twin neural cooperative network.
Preferably, the process of optimizing the comparison function based on trade-specification characteristics and by genetic algorithm includes:
step 1: initializing a population, wherein each individual in the population represents a comparison function;
step 2: in each generation, two individuals are randomly selected to be crossed, a new comparison function is generated, and mutation is carried out on each individual;
step 3: calculating the similarity between the financing trade and the trade specification data corresponding to the trade-specification characteristics through the newly generated comparison function;
the calculation formula of the similarity is as follows:
D=c * (T’)；
c * (T’)=σ(w·T’+b)；
wherein,Din order for the degree of similarity to be the same,D∈[0,1]；c * (. Cndot.) is a comparison function;T’representing trade-specification characteristics;σ(. Cndot.) represents a sigmoid activation function;wthe weight parameters of the optimized twin neural cooperative network are obtained;bthe bias parameters of the optimized twin neural cooperative network are obtained;
step 4: acquiring a true similarity label of similarity between financing trade and trade specification data corresponding to trade-specification characteristics; selecting the new comparison function with the smallest difference to enter the next generation based on the difference between the similarity between the financing trade corresponding to the trade-specification feature and the real similarity label corresponding to the trade-specification data;
step 5: and (3) repeating the steps 2-4 until convergence, and obtaining the optimized comparison function.
Preferably, the determining the similarity between the target financing trade data and the trade specification data according to the optimized comparison function includes:
step 1: respectively corresponding a plurality of trade specification data with the target qualifying trade data to form a plurality of trade-specification data pairs, and passing each trade-specification data pair through steps S2 and S3 to obtain a detection trade-specification characteristic;
step 2: and calculating the similarity between the target financing trade data and each trade specification data through each detection trade-specification characteristic by using the optimized comparison function.
Preferably, the determining the compliance detection result based on the similarity includes:
screening the lowest similarity among the calculated target financing trade data and the similarity between each piece of trade specification data, comparing the lowest similarity with a threshold value, and determining the compliance detection result; the expression is:
；
wherein,Compliancerepresenting a compliance detection result;Compliantindicating compliance;Non-Compliantindicating non-compliance; min%D 1 ,D 2 ,…,D m ) Representing the lowest degree of similarity;D 1 representing a similarity between the target qualifying trade data and the 1 st said trade specification data;D m representing targeted qualifying trade datamSimilarity between the trade specification data;θrepresenting a threshold.
The beneficial effects are that: according to the method, the optimized twin neural cooperative network outputs trade-specification characteristics, so that the optimized twin neural cooperative network can capture the complex relationship between financing trade and trade specification data more accurately, and the accuracy of subsequent data comparison is improved. Based on trade-normative features, a comparison function is optimized through a genetic algorithm, and the similarity between the target financing trade and the trade normative data can be rapidly and accurately calculated by the optimized comparison function, so that the accuracy of final compliance detection is improved. The method can efficiently and accurately calculate the similarity between the target financing trade and the normative clause, thereby realizing the automatic detection of the financing trade compliance.
Drawings
In order to more clearly illustrate the embodiments of the application or the technical solutions in the prior art, the drawings that are required in the embodiments or the description of the prior art will be briefly described, it being obvious that the drawings in the following description are only some embodiments of the application, and that other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.
FIG. 1 is a flow chart of a method for detecting financing trade compliance based on twin evolution according to an embodiment of the present application.
Detailed Description
In order that the above objects, features and advantages of the application will be readily understood, a more particular description of the application will be rendered by reference to the appended drawings. In the following description, numerous specific details are set forth in order to provide a thorough understanding of the present application. The present application may be embodied in many other forms than described herein and similarly modified by those skilled in the art without departing from the spirit of the application, whereby the application is not limited to the specific embodiments disclosed below.
Furthermore, the terms "first," "second," and the like, are used for descriptive purposes only and are not to be construed as indicating or implying a relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defining "a first" or "a second" may explicitly or implicitly include at least one such feature. In the description of the present application, the meaning of "plurality" means at least two, for example, two, three, etc., unless specifically defined otherwise.
As shown in fig. 1, the embodiment provides a method for detecting financing trade compliance based on twin evolution, which comprises the following steps:
s1: trade data and trade specification data are obtained.
Specifically, the trade data includes transaction records, contracts and agreements, customer relationship management data, annual reports and logistic documents; the transaction record comprises transaction amount (numerical data), transaction date (time data) and transaction counterpart (text data); the contract (text data) is used to describe the terms and details of the transaction; the customer relationship management data includes customer ratings (numerical data), customer feedback (text data); the annual report and logistics document comprises profit data (numerical data), transportation route description (text data); the trade specification data includes regulatory policies (text data), trade specification data; the regulatory policy is used to describe the regulations and requirements of financing trade; the trade specification data includes trade limits (numerical data), compliant treaty clauses examples (text data).
S2: vectorizing the trade data and the trade specification data to obtain trade feature vectors and trade specification data matrixes respectively; a trade data matrix is constructed based on trade feature vectors of different enterprises.
Specifically, the vectorization processing includes:
carrying out standardization processing on the transaction amount to obtain an amount value vector;
converting the transaction date into a date value vector;
the transaction counterpart is converted into a transaction personal text vector by Word2Vec Word embedding technology;
performing independent heat coding or standardization processing on the client rating to obtain a client rating numerical vector;
the client feedback is converted into a client feedback text vector through Word2Vec Word embedding technology;
carrying out standardization processing on the profit data to obtain a profit value vector;
the transportation route description is converted into a transportation route text vector through Word2Vec Word embedding technology;
the supervision policy is converted into a supervision policy text vector by Word2Vec Word embedding technology;
carrying out standardization processing on the transaction quota to obtain a quota numerical vector;
the compliant treaty clause example is converted into a compliant clause text vector by Word2Vec Word embedding technology.
Further, combining the monetary value vector, the date value vector, the transactor text vector, the customer level value vector, the customer feedback text vector, the profit value vector and the transportation route text vector to obtain the trade feature vector;
and combining the supervision policy text vector, the quota value vector and the compliance term text vector to obtain the trade specification data matrix.
S3: and inputting the trade data matrix and the trade specification data matrix into an optimized twin neural cooperative network, and outputting trade-specification characteristics.
Specifically, in order to ensure that the two matrices are in the same feature space, before the trade data matrix and the trade specification data matrix are input to the optimized twin neural cooperative network, the dimension of the trade data matrix and the dimension of the trade specification data matrix are reduced by adopting a PCA technology, and the dimension of the trade data matrix and the dimension of the trade specification data matrix are unified.
In this embodiment, taking the trade data matrix as an example, the process of using PCA technology to reduce the dimension of the trade data matrix includes:
step 1: assuming that the trade data matrix is onevRow of linesuA matrix of columns, denoted asX；
Step 2: zero-equalizing each row of the trade data matrix, namely subtracting the average value of the row;
step 3: solving a mean square error matrix, wherein the calculation formula is as follows:
；
wherein,X 1 representing a zero-averaged trade data matrix,trepresenting a transpose;
step 4: obtaining eigenvalues and corresponding eigenvectors of the covariance matrix;
step 5: the eigenvectors are arranged into a matrix according to the corresponding eigenvalue size from top to bottom and the front of the matrix is takenkLine composition matrixP；
Step 6: trade data matrix after dimension reductionE=PX。
The trade specification data matrix can also be subjected to dimension reduction through the dimension reduction process.
In this embodiment, the twin neural synergistic network includes two identical multi-layer perceptrons; each of the multi-layer perceptrons includes three fully connected layers, and each fully connected layer is followed by a ReLU activation function to capture complex relationships between trade data and trade specification data.
The forward propagation of the first multi-layer perceptron is expressed as:
z E 1 =W E1 ·E+b E1 ；
a E 1 =ReLU(z E1 )；
z E 2 =W E2 ·a E1 +b E2 ；
a E 2 =ReLU(z E2 )；
z E 3 =W E3 ·a E2 +b E3 ；
h E =ReLU(z E3 )；
wherein,h E representing the trade data matrix after forward propagation;Erepresenting the trade data matrix after dimension reduction;z E1 、z E2 、z E3 respectively represent a first layer full-connection layer, a second layer full-connection layer and a third layer full-connection of the first multi-layer perceptronOutputting a connecting layer; w (W) E1 、W E2 、W E3 Respectively representing weight matrixes of a first full-connection layer, a second full-connection layer and a third full-connection layer of the first multi-layer perceptron; b E1 、b E2 、b E3 Respectively representing the bias of a first full-connection layer, a second full-connection layer and a third full-connection layer of the first multi-layer perceptron;a E1 、a E2 respectively representing the output of a first ReLU activation function and a second ReLU activation function of a first multi-layer perceptron; reLU (·) represents a ReLU activation function;
the forward propagation of the second multi-layer perceptron is expressed as:
z G 1 =W G1 ·E+b G1 ；
a G 1 =ReLU(z G1 )；
z G 2 =W G2 ·a G1 +b G2 ；
a G 2 =ReLU(z G2 )；
z G 3 =W G3 ·a G2 +b G3 ；
h G =ReLU(z G3 )；
wherein,h G representing the trade specification data matrix after forward propagation;Grepresenting the trade specification data matrix after dimension reduction;z G1 、z G2 、z G3 respectively representing the output of a first full-connection layer, a second full-connection layer and a third full-connection layer of the second multi-layer perceptron; w (W) G1 、W G2 、W G3 Respectively represent a first layer full-connection layer and a second layer full-connection layer of the second multi-layer perceptronA weight matrix of the third full-connection layer; b G1 、b G2 、b G3 Respectively representing the bias of a first full-connection layer, a second full-connection layer and a third full-connection layer of the second multi-layer perceptron;a G1 、a G2 respectively representing the output of a first ReLU activation function and a second ReLU activation function of a second multi-layer perceptron;
the ReLU activation function will change any negative value to 0 and the other values will remain unchanged.
Carrying out tensor product operation on the trade data matrix after forward propagation and the trade specification data matrix after forward propagation to obtain initial trade-specification characteristics, wherein a calculation formula is as follows:
；
wherein,Trepresenting an initial trade-specification feature;represents a tensor product operation that helps capture the correlation between the two matrices after forward propagation.
Further, the present embodiment provides an optimization process of the twin neural cooperative network, including:
step 1: obtaining a true similarity label corresponding to an initial trade-specification feature, and calculating a loss function based on the initial trade-specification feature and the true similarity label corresponding to the initial trade-specification feature, wherein the loss function expression is as follows:
；
wherein,L(. Cndot.) represents the loss function,Trepresenting the initial trade-specification characteristics of the trade,Ytrue similarity labeling representing initial trade-specification characteristics;T i represent the firstiInitial trade-specification characteristics of the individual samples;Y i represent the firstiTrue similarity labeling of initial trade-specification features of individual samples;nfor the number of samples。
Step 2: and calculating the gradient of the loss function relative to the network parameters by adopting a chained rule, and calculating a gradient value for each network parameter.
Step 3: and updating weight parameters and bias parameters of the twin neural cooperative network based on the calculated gradient values to minimize the loss function.
The weight parameter update formula is expressed as:
；
wherein,W new representing updated weight parameters;W old representing the original weight parameters;αrepresenting a learning rate;representing the gradient value of the loss function to the weight parameter.
The bias parameter update formula is expressed as:
；
wherein,b new representing the updated bias parameters;b old representing the original bias parameters;αrepresenting a learning rate;representing the gradient value of the loss function versus the bias parameter.
Step 4: repeating the steps 1-3 until the loss function converges or the maximum iteration number is reached, and obtaining the optimized twin neural cooperative network.
The twin neural cooperative network optimized by the optimization mode can more accurately extract trade-specification characteristics consistent with the true similarity annotation, and further improve the accuracy of subsequent specification comparison.
S4: and optimizing a comparison function through a genetic algorithm based on the trade-specification characteristics, and determining the similarity between the target financing trade data and the trade specification data according to the optimized comparison function.
Specifically, the process of optimizing the comparison function based on trade-specification characteristics and by genetic algorithm includes:
step 1: a population is initialized, each individual in the population representing a comparison function.
Step 2: in each generation, two individuals are randomly selected for crossover, a new comparison function is generated, and mutation is performed on each individual.
Step 3: and calculating the similarity between the financing trade and the trade specification data corresponding to the trade-specification characteristics through the newly generated comparison function.
The calculation formula of the similarity is as follows:
D=c * (T’)；
c * (T’)=σ(w·T’+b)；
wherein,Dd E [0,1 ] is similarity]；c * (. Cndot.) is a comparison function, which is a mapping used to map trade-specification features to a [0,1 ]]Similarity value of the interval;T’representing trade-specification characteristics;σ(. Cndot.) represents a sigmoid activation function;wthe weight parameters of the optimized twin neural cooperative network are obtained;band the bias parameters of the optimized twin neural cooperative network are obtained.
Step 4: acquiring a true similarity label of similarity between financing trade and trade specification data corresponding to trade-specification characteristics; and selecting the new comparison function with the smallest difference to enter the next generation based on the difference between the similarity between the financing trade corresponding to the trade-specification feature and the real similarity label corresponding to the trade-specification data.
Step 5: and (3) repeating the steps 2-4 until convergence, and obtaining the optimized comparison function.
Further, the determining the similarity between the target financing trade data and the trade specification data according to the optimized comparison function includes:
step 1: and respectively corresponding a plurality of the trade specification data with the target qualifying trade data to form a plurality of trade-specification data pairs, and passing each trade-specification data pair through steps S2 and S3 to obtain the detection trade-specification characteristic.
Step 2: and calculating the similarity between the target financing trade data and each trade specification data through each detection trade-specification characteristic by using the optimized comparison function.
S5: and determining a compliance detection result based on the similarity.
Specifically, the method comprises the following steps:
screening the lowest similarity among the calculated target financing trade data and the similarity between each piece of trade specification data, comparing the lowest similarity with a threshold value, and determining the compliance detection result; the expression is:
；
wherein,Compliancerepresenting a compliance detection result;Compliantindicating compliance;Non-Compliantindicating non-compliance; min%D 1 ,D 2 ,…,D m ) Representing the lowest degree of similarity;D 1 representing a similarity between the target qualifying trade data and the 1 st said trade specification data;D m representing targeted qualifying trade datamSimilarity between the trade specification data;θrepresenting a threshold.
Integrating the obtained similarity, compliance detection result, the best matched trade specification data or the unmatched trade specification data as a detection report of target fused trade data; the detection report will more intuitively present compliance detection of the target financing trade data, contributing to the protection of the interests of the enterprise.
According to the method, the optimized twin neural cooperative network outputs trade-specification characteristics, so that the optimized twin neural cooperative network can capture the complex relationship between financing trade and trade specification data more accurately, and the accuracy of subsequent data comparison is improved. Based on trade-normative features, a comparison function is optimized through a genetic algorithm, and the similarity between the target financing trade and the trade normative data can be rapidly and accurately calculated by the optimized comparison function, so that the accuracy of final compliance detection is improved. The method can efficiently and accurately calculate the similarity between the target financing trade and the normative clause, thereby realizing the automatic detection of the financing trade compliance.
The technical features of the above-described embodiments may be arbitrarily combined, and all possible combinations of the technical features in the above-described embodiments are not described for brevity of description, however, as long as there is no contradiction between the combinations of the technical features, they should be considered as the scope of the description.
The above examples illustrate only a few embodiments of the application, which are described in detail and are not to be construed as limiting the scope of the claims. It should be noted that it will be apparent to those skilled in the art that several variations and modifications can be made without departing from the spirit of the application, which are all within the scope of the application. Accordingly, the scope of protection of the present application is to be determined by the appended claims.
Claims (10)
1. The financing trade compliance detection method based on twin evolution is characterized by comprising the following steps:
s1: acquiring trade data and trade specification data;
s2: vectorizing the trade data and the trade specification data to obtain trade feature vectors and trade specification data matrixes respectively; constructing trade data matrixes based on trade feature vectors of different enterprises;
s3: inputting the trade data matrix and the trade specification data matrix into an optimized twin neural cooperative network, and outputting trade-specification characteristics;
s4: optimizing a comparison function through a genetic algorithm based on trade-specification characteristics, and determining the similarity between the target financing trade data and the trade specification data according to the optimized comparison function;
s5: and determining a compliance detection result based on the similarity.
2. The method for detecting financing trade compliance based on twin evolution according to claim 1, wherein the trade data comprises transaction records, contracts and agreements, customer relationship management data, annual reports and logistic documents; the transaction record comprises transaction amount, transaction date and transaction counterpart; the contract is used to describe the terms and details of the transaction; the customer relationship management data includes customer ratings, customer feedback; the annual report and logistics document comprises profit data and a transportation route description; the trade specification data includes regulatory policies, trade specification data; the regulatory policy is used to describe the regulations and requirements of financing trade; the trade specification data includes a transaction limit and a compliant contract term paradigm.
3. The method for detecting financing trade compliance based on twin evolution according to claim 2, wherein the vectorization process comprises:
carrying out standardization processing on the transaction amount to obtain an amount value vector;
converting the transaction date into a date value vector;
the transaction counterpart is converted into a transaction personal text vector by Word2Vec Word embedding technology;
performing independent heat coding on the client rating to obtain a client rating numerical vector;
the client feedback is converted into a client feedback text vector through Word2Vec Word embedding technology;
carrying out standardization processing on the profit data to obtain a profit value vector;
the transportation route description is converted into a transportation route text vector through Word2Vec Word embedding technology;
the supervision policy is converted into a supervision policy text vector by Word2Vec Word embedding technology;
carrying out standardization processing on the transaction quota to obtain a quota numerical vector;
the compliant treaty clause example is converted into a compliant clause text vector by Word2Vec Word embedding technology.
4. The method for detecting financing trade compliance based on twin evolution according to claim 3, wherein the trade feature vector is obtained by combining the monetary value vector, the date numeric vector, the transactor text vector, the customer level numeric vector, the customer feedback text vector, the profit numeric vector, and the transportation route text vector;
and combining the supervision policy text vector, the quota value vector and the compliance term text vector to obtain the trade specification data matrix.
5. The method for detecting financing trade compliance based on twin evolution according to claim 1, wherein in S3, further comprising: before the trade data matrix and the trade specification data matrix are input to the optimized twin neural cooperative network, the dimension reduction is carried out on the trade data matrix and the trade specification data matrix respectively by adopting a PCA technology, and the dimension of the trade data matrix and the dimension of the trade specification data matrix are unified.
6. The method for detecting financing trade compliance based on twin evolution according to claim 5, wherein the twin neural synergistic network comprises two identical multi-layer perceptrons; each multi-layer perceptron comprises three full-connection layers, and each full-connection layer is followed by a ReLU activation function;
the forward propagation of the first multi-layer perceptron is expressed as:
z E 1 =W E1 ·E+b E1 ；
a E 1 =ReLU(z E1 )；
z E 2 =W E2 ·a E1 +b E2 ；
a E 2 =ReLU(z E2 )；
z E 3 =W E3 ·a E2 +b E3 ；
h E =ReLU(z E3 )；
wherein,h E representing the trade data matrix after forward propagation;Erepresenting the trade data matrix after dimension reduction;z E1 、z E2 、z E3 respectively representing the output of a first full-connection layer, a second full-connection layer and a third full-connection layer of the first multi-layer perceptron; w (W) E1 、W E2 、W E3 Respectively representing weight matrixes of a first full-connection layer, a second full-connection layer and a third full-connection layer of the first multi-layer perceptron; b E1 、b E2 、b E3 Respectively representing the bias of a first full-connection layer, a second full-connection layer and a third full-connection layer of the first multi-layer perceptron;a E1 、a E2 respectively representing the output of a first ReLU activation function and a second ReLU activation function of a first multi-layer perceptron; reLU (·) represents a ReLU activation function;
the forward propagation of the second multi-layer perceptron is expressed as:
z G 1 =W G1 ·E+b G1 ；
a G 1 =ReLU(z G1 )；
z G 2 =W G2 ·a G1 +b G2 ；
a G 2 =ReLU(z G2 )；
z G 3 =W G3 ·a G2 +b G3 ；
h G =ReLU(z G3 )；
wherein,h G representing the trade specification data matrix after forward propagation;Grepresenting the trade specification data matrix after dimension reduction;z G1 、z G2 、z G3 respectively representing the output of a first full-connection layer, a second full-connection layer and a third full-connection layer of the second multi-layer perceptron; w (W) G1 、W G2 、W G3 Respectively representing weight matrixes of a first full-connection layer, a second full-connection layer and a third full-connection layer of the second multi-layer perceptron; b G1 、b G2 、b G3 Respectively representing the bias of a first full-connection layer, a second full-connection layer and a third full-connection layer of the second multi-layer perceptron;a G1 、a G2 respectively representing the output of a first ReLU activation function and a second ReLU activation function of a second multi-layer perceptron;
and carrying out tensor product operation on the trade data matrix after forward propagation and the trade specification data matrix after forward propagation to obtain initial trade-specification characteristics.
7. The method for detecting financing trade compliance based on twin evolution according to claim 6, wherein the optimization process of the twin neural synergistic network comprises:
step 1: obtaining a true similarity label corresponding to an initial trade-specification feature, and calculating a loss function based on the initial trade-specification feature and the true similarity label corresponding to the initial trade-specification feature, wherein the loss function expression is as follows:
；
wherein,L(. Cndot.) represents the loss function,Trepresenting the initial trade-specification characteristics of the trade,Ytrue similarity labeling representing initial trade-specification characteristics;T i represent the firstiInitial trade-specification characteristics of the individual samples;Y i represent the firstiTrue similarity labeling of initial trade-specification features of individual samples;nis the number of samples;
step 2: calculating the gradient of the loss function relative to network parameters by adopting a chained rule, and calculating a gradient value for each network parameter;
step 3: updating weight parameters and bias parameters of the twin neural cooperative network based on the calculated gradient values to minimize the loss function;
step 4: repeating the steps 1-3 until the loss function converges or the maximum iteration number is reached, and obtaining the optimized twin neural cooperative network.
8. The method for detecting financing trade compliance based on twin evolution according to claim 7, wherein the process of optimizing the comparison function based on trade-specification characteristics and by genetic algorithm comprises:
step 1: initializing a population, wherein each individual in the population represents a comparison function;
step 2: in each generation, two individuals are randomly selected to be crossed, a new comparison function is generated, and mutation is carried out on each individual;
step 3: calculating the similarity between the financing trade and the trade specification data corresponding to the trade-specification characteristics through the newly generated comparison function;
the calculation formula of the similarity is as follows:
D=c * (T’)；
c * (T’)=σ(w·T’+b)；
wherein,Din order for the degree of similarity to be the same,D∈[0,1]；c * (. Cndot.) is a comparison function;T’representing trade-specification characteristics;σ(. Cndot.) represents a sigmoid activation function;wthe weight parameters of the optimized twin neural cooperative network are obtained;bthe bias parameters of the optimized twin neural cooperative network are obtained;
step 4: acquiring a true similarity label of similarity between financing trade and trade specification data corresponding to trade-specification characteristics; selecting the new comparison function with the smallest difference to enter the next generation based on the difference between the similarity between the financing trade corresponding to the trade-specification feature and the real similarity label corresponding to the trade-specification data;
step 5: and (3) repeating the steps 2-4 until convergence, and obtaining the optimized comparison function.
9. The method for twin evolution-based financing trade compliance detection of claim 1, wherein the determining the similarity between the target financing trade data and the trade specification data according to the optimized comparison function comprises:
step 1: respectively corresponding a plurality of trade specification data with the target qualifying trade data to form a plurality of trade-specification data pairs, and passing each trade-specification data pair through steps S2 and S3 to obtain a detection trade-specification characteristic;
step 2: and calculating the similarity between the target financing trade data and each trade specification data through each detection trade-specification characteristic by using the optimized comparison function.
10. The method of claim 9, wherein determining a compliance detection result based on the similarity comprises:
screening the lowest similarity among the calculated target financing trade data and the similarity between each piece of trade specification data, comparing the lowest similarity with a threshold value, and determining the compliance detection result; the expression is:
；
wherein,Compliancerepresenting a compliance detection result;Compliantindicating compliance;Non-Compliantindicating non-compliance; min%D 1 ,D 2 ,…,D m ) Representing the lowest degree of similarity;D 1 representing a similarity between the target qualifying trade data and the 1 st said trade specification data;D m representing targeted qualifying trade datamSimilarity between the trade specification data;θrepresenting a threshold.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311492810.9A CN117237114B (en) | 2023-11-10 | 2023-11-10 | Financing trade compliance detection method based on twin evolution |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311492810.9A CN117237114B (en) | 2023-11-10 | 2023-11-10 | Financing trade compliance detection method based on twin evolution |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN117237114A true CN117237114A (en) | 2023-12-15 |
| CN117237114B CN117237114B (en) | 2024-03-08 |
Family
ID=89082883
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311492810.9A Active CN117237114B (en) | 2023-11-10 | 2023-11-10 | Financing trade compliance detection method based on twin evolution |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117237114B (en) |
Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109360105A (en) * | 2018-09-18 | 2019-02-19 | 平安科技（深圳）有限公司 | Product risks method for early warning, device, computer equipment and storage medium |
| CN113378090A (en) * | 2021-04-23 | 2021-09-10 | 国家计算机网络与信息安全管理中心 | Internet website similarity analysis method and device and readable storage medium |
| CN114429344A (en) * | 2021-12-07 | 2022-05-03 | 华中科技大学 | Transaction method and transaction system of digital currency |
| CN115544211A (en) * | 2022-10-28 | 2022-12-30 | 沈阳师范大学 | Method for external trade and external law indexing and industry risk assessment |
| CN115545712A (en) * | 2022-09-22 | 2022-12-30 | 中国农业银行股份有限公司 | Fraud prediction method, device, equipment and storage medium for transaction behaviors |
| CN116245645A (en) * | 2023-03-14 | 2023-06-09 | 中国科学院软件研究所 | Financial crime partner detection method based on graph neural network |
| US20230206261A1 (en) * | 2021-11-23 | 2023-06-29 | Strong Force TX Portfolio 2018, LLC | Transaction platforms where systems include sets of other systems |
| CN116452007A (en) * | 2023-06-15 | 2023-07-18 | 深圳市迪博企业风险管理技术有限公司 | Enterprise tax compliance risk assessment method based on capsule network |
| CN116542774A (en) * | 2023-06-27 | 2023-08-04 | 深圳市迪博企业风险管理技术有限公司 | Probability diffusion model-based method for detecting compliance of company-associated transactions on sale |
-
2023
- 2023-11-10 CN CN202311492810.9A patent/CN117237114B/en active Active
Patent Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109360105A (en) * | 2018-09-18 | 2019-02-19 | 平安科技（深圳）有限公司 | Product risks method for early warning, device, computer equipment and storage medium |
| CN113378090A (en) * | 2021-04-23 | 2021-09-10 | 国家计算机网络与信息安全管理中心 | Internet website similarity analysis method and device and readable storage medium |
| US20230206261A1 (en) * | 2021-11-23 | 2023-06-29 | Strong Force TX Portfolio 2018, LLC | Transaction platforms where systems include sets of other systems |
| CN114429344A (en) * | 2021-12-07 | 2022-05-03 | 华中科技大学 | Transaction method and transaction system of digital currency |
| CN115545712A (en) * | 2022-09-22 | 2022-12-30 | 中国农业银行股份有限公司 | Fraud prediction method, device, equipment and storage medium for transaction behaviors |
| CN115544211A (en) * | 2022-10-28 | 2022-12-30 | 沈阳师范大学 | Method for external trade and external law indexing and industry risk assessment |
| CN116245645A (en) * | 2023-03-14 | 2023-06-09 | 中国科学院软件研究所 | Financial crime partner detection method based on graph neural network |
| CN116452007A (en) * | 2023-06-15 | 2023-07-18 | 深圳市迪博企业风险管理技术有限公司 | Enterprise tax compliance risk assessment method based on capsule network |
| CN116542774A (en) * | 2023-06-27 | 2023-08-04 | 深圳市迪博企业风险管理技术有限公司 | Probability diffusion model-based method for detecting compliance of company-associated transactions on sale |
Also Published As
| Publication number | Publication date |
|---|---|
| CN117237114B (en) | 2024-03-08 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| JP7169369B2 (en) | Method, system for generating data for machine learning algorithms | |
| US6636862B2 (en) | Method and system for the dynamic analysis of data | |
| CN112070125A (en) | Prediction method of unbalanced data set based on isolated forest learning | |
| CN112966114B (en) | Literature classification method and device based on symmetrical graph convolutional neural network | |
| CN111695719A (en) | User value prediction method and system | |
| CN111582538A (en) | Community value prediction method and system based on graph neural network | |
| Mendonça et al. | Approximating network centrality measures using node embedding and machine learning | |
| CN116628597B (en) | Heterogeneous graph node classification method based on relationship path attention | |
| Kullaya Swamy et al. | Bank transaction data modeling by optimized hybrid machine learning merged with ARIMA | |
| Ahelegbey et al. | Factorial network models to improve P2P credit risk management | |
| US20230041927A1 (en) | Combined commodity mining method based on knowledge graph rule embedding | |
| Cano et al. | International market selection using fuzzy weighing and Monte Carlo simulation | |
| CN111639690A (en) | Fraud analysis method, system, medium, and apparatus based on relational graph learning | |
| CN112836750A (en) | System resource allocation method, device and equipment | |
| Eddy et al. | Credit scoring models: Techniques and issues | |
| US8364614B2 (en) | Method for building predictive models with incomplete data | |
| CN116401379A (en) | Financial product data pushing method, device, equipment and storage medium | |
| CN117237114B (en) | Financing trade compliance detection method based on twin evolution | |
| CN115712775A (en) | Product recommendation method and device, computer equipment and storage medium | |
| Amzile et al. | Assessment of Support Vector Machine performance for default prediction and credit rating | |
| CN115063143A (en) | Account data processing method and device, computer equipment and storage medium | |
| CN113962327A (en) | Data classification method and device and electronic equipment | |
| Mueller et al. | Feature selection for measurement models | |
| Yuan | Applications of machine learning: consumer credit risk analysis | |
| Kavitha et al. | Hybrid ML classifier for loan prediction system |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |