Bill intermediary identification method, electronic equipment, storage medium and product

Abstract

The invention discloses a bill intermediary identification method, an electronic device, a storage medium and a product, and relates to the technical field of financial science and technology, and the method comprises the steps: obtaining original business data of a to-be-evaluated enterprise, inputting the original business data into a bill intermediary identification model, and outputting a probability score that the to-be-evaluated enterprise is a risk bill intermediary; constructing a network diagram representing an enterprise associated network structure based on the original business data, and calculating enterprise evaluation indexes of the to-be-evaluated enterprise based on the network diagram; performing score calibration on the probability score to generate a target risk score; and generating an enterprise risk portrait based on the target risk score, the network diagram, the enterprise evaluation index and the original business data, and generating an evaluation result of the to-be-evaluated enterprise based on the enterprise risk portrait, the evaluation result being used for representing whether the to-be-evaluated enterprise is a legal bill intermediary. According to the method and the device, the technical problem of poor recognition effect in current bill intermediary recognition is solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of financial technology, and particularly relates to a bill intermediary identification method, an electronic device, a storage medium and a product. BACKGROUND

[0002] With the deepening of the digitization of financial business, the scale of bill business continues to expand, and its role in serving the real economy is increasingly important. Therefore, banks and other financial institutions need to effectively identify and control bill intermediaries that provide bill services. At present, bill intermediaries are mainly identified by manually checking by bank staff, querying external business information, and checking bill endorsement flow records, which not only consumes a lot of manpower and time, but also relies heavily on personal experience, and is prone to misjudgment or omission. Therefore, the current identification of bill intermediaries has the problem of poor identification effect.

[0003] The above content is only used to assist in understanding the technical solutions of the present application and does not represent the acknowledgement of the above content as prior art. SUMMARY

[0004] The main purpose of the present application is to provide a bill intermediary identification method, an electronic device, a storage medium and a product, which aims to solve the technical problem of poor identification effect in the current identification of bill intermediaries.

[0005] To achieve the above purpose, the present application provides a bill intermediary identification method, system, electronic device, storage medium and computer program product, the bill intermediary identification method comprising:

[0006] Obtaining original business data of an enterprise to be evaluated, inputting the original business data into a bill intermediary identification model, and outputting a probability score of the enterprise to be evaluated being a risk bill intermediary;

[0007] Building a network graph representing the enterprise association network structure based on the original business data, and calculating an enterprise evaluation index of the enterprise to be evaluated based on the network graph;

[0008] Score calibration is performed on the probability score to generate a target risk score;

[0009] Generating an enterprise risk portrait based on the target risk score, the network graph, the enterprise evaluation index and the original business data, and generating an evaluation result of the enterprise to be evaluated based on the enterprise risk portrait, wherein the evaluation result is used to represent whether the enterprise to be evaluated is a legal bill intermediary.

[0010] In an embodiment, the step of inputting the original business data into the bill intermediary identification model further comprises:

[0011] Pre-stored enterprise historical data is taken as a training set, and the training set is input into a preset gradient decision tree model for iterative training.

[0012] A loss function between a prediction value generated in each round of training and a corresponding true label is calculated, and a new decision tree is constructed based on a gradient direction of the loss function, wherein the new decision tree is used to fit a residual error between the prediction value and the corresponding true label.

[0013] The training is terminated when a variation of the residual error is lower than a preset threshold, and a bill mediation recognition model integrated by a plurality of gradient boosting decision trees is obtained.

[0014] In an embodiment, the step of constructing a network graph representing an enterprise association network structure based on the original business data comprises:

[0015] Each enterprise entity, each associated person and each network address in the original business data are extracted.

[0016] A first relationship between each enterprise entity, a second relationship between each enterprise entity and each associated person, and a third relationship between each enterprise entity and each network address are determined based on the original business data.

[0017] The each enterprise entity, the each associated person and the each network address are defined as network nodes, and the first relationship, the second relationship and the third relationship are defined as connection edges between the network nodes, to obtain a network graph representing an enterprise association network structure.

[0018] In an embodiment, the enterprise evaluation index comprises a sub-network dimension index and an enterprise dimension index, and the step of calculating the enterprise evaluation index of the enterprise to be evaluated based on the network graph comprises:

[0019] The network graph is divided into sub-networks, and a target sub-network to which the enterprise to be evaluated belongs is determined.

[0020] A sub-network dimension index of the target sub-network is calculated based on topological information of the target sub-network.

[0021] An enterprise dimension index of the enterprise to be evaluated is calculated based on network nodes and connection edges associated with a node corresponding to the enterprise to be evaluated in the network graph.

[0022] In an embodiment, the step of dividing the network graph into sub-networks comprises:

[0023] A sub-network label of each network node in the enterprise association network structure in the network graph is initialized.

[0024] performing multiple rounds of label propagation iterations, in each round of iteration, traversing all target network nodes with empty subnetwork labels, determining a high-frequency label with the highest occurrence frequency among adjacent nodes of the target network node, and updating a subnetwork label of the target network node as the high-frequency label;

[0025] terminating the iteration in a case where the subnetwork label of each network node in the network graph no longer changes, and grouping network nodes with the same subnetwork label in the network graph into the same subnetwork.

[0026] In an embodiment, the step of performing score calibration on the probability score to generate a target risk score comprises:

[0027] determining, based on a preset bonus item rule, a bonus item rule satisfied by the enterprise to be evaluated in combination with the original business data, and increasing a corresponding score value for the probability score based on the bonus item rule;

[0028] determining, based on a preset deduction item rule, a deduction item rule satisfied by the enterprise to be evaluated in combination with the original business data, and decreasing a corresponding score value for the probability score based on the deduction item rule;

[0029] performing upper and lower limit range constraints on the adjusted probability score, and outputting a target risk score.

[0030] In an embodiment, the step of generating an evaluation result of the enterprise to be evaluated based on the enterprise risk portrait comprises:

[0031] extracting a target risk score in the enterprise risk portrait, and comparing the target risk score with a preset risk threshold;

[0032] in a case where the target risk score is higher than the preset risk threshold, extracting an enterprise evaluation indicator in the enterprise risk portrait, comparing the enterprise evaluation indicator with a corresponding indicator threshold, determining an exceeding condition of the enterprise evaluation indicator and the corresponding indicator threshold, and generating an evaluation result of the enterprise to be evaluated based on the exceeding condition.

[0033] In addition, to achieve the above-mentioned purpose, the application further provides a bill intermediary identification system, which comprises:

[0034] a model evaluation module configured to acquire original business data of an enterprise to be evaluated, input the original business data into a bill intermediary identification model, and output a probability score of the enterprise to be evaluated as a risk bill intermediary;

[0035] an indicator calculation module configured to construct a network graph representing an enterprise association network structure based on the original business data, and calculate an enterprise evaluation indicator of the enterprise to be evaluated based on the network graph;

[0036] a score calibration module configured to perform score calibration on the probability score to generate a target risk score;

[0037] a result generation module configured to generate an enterprise risk profile based on the target risk score, the network graph, the enterprise evaluation indicator and the original business data, and generate an evaluation result of the enterprise to be evaluated based on the enterprise risk profile, wherein the evaluation result is used to indicate whether the enterprise to be evaluated is a legal bill intermediary.

[0038] In addition, to achieve the above-mentioned purpose, the present application further provides an electronic device, which comprises a memory, a processor and a computer program stored in the memory and executable on the processor, and the computer program is configured to implement the steps of the bill intermediary identification method as described above.

[0039] In addition, to achieve the above-mentioned purpose, the present application further provides a storage medium, which is a computer readable storage medium, and the storage medium stores a computer program, and the computer program is executed by a processor to implement the steps of the bill intermediary identification method as described above.

[0040] In addition, to achieve the above-mentioned purpose, the present application further provides a computer program product, which comprises a computer program, and the computer program is executed by a processor to implement the steps of the bill intermediary identification method as described above.

[0041] The present application provides a bill intermediary identification method, which comprises: obtaining original business data of an enterprise to be evaluated, inputting the original business data into a bill intermediary identification model, and outputting a probability score of the enterprise to be evaluated being a risk bill intermediary; constructing a network graph representing an enterprise association network structure based on the original business data, and calculating an enterprise evaluation indicator of the enterprise to be evaluated based on the network graph; performing score calibration on the probability score to generate a target risk score; generating an enterprise risk profile based on the target risk score, the network graph, the enterprise evaluation indicator and the original business data, and generating an evaluation result of the enterprise to be evaluated based on the enterprise risk profile, wherein the evaluation result is used to indicate whether the enterprise to be evaluated is a legal bill intermediary.

[0042] Since the current identification of bill intermediaries mainly relies on manual investigation by bank staff, external business information query and bill endorsement flow record review, it not only consumes a lot of manpower and time, but also depends heavily on personal experience, which is easy to lead to misjudgment or omission. The present application automatically processes data through the model, quickly outputs the probability score of the to-be-evaluated enterprise as a risk bill intermediary, greatly improves the identification efficiency, reduces the time cost of manual operation, and the model is trained based on a large amount of data, which can more objectively and accurately mine potential risk features in the data, reduce the risk of misjudgment or omission caused by insufficient personal experience, and can deeply mine the potential association and risk transmission path between enterprises by constructing a network graph representing the enterprise association network structure and calculating enterprise evaluation indicators, thereby evaluating the risk status of the to-be-evaluated enterprise as a whole, discovering risk features that are difficult to detect through single enterprise data, improving the accuracy and comprehensiveness of identification, and generating a target risk score through score calibration, which can more accurately reflect the actual risk level of the to-be-evaluated enterprise, reduce the deviation of the model output result, provide a more reliable basis for subsequent risk assessment and decision-making, and generate an enterprise risk portrait, which can comprehensively and comprehensively evaluate the enterprise from multiple angles, more accurately depict the risk features of the enterprise, improve the identification accuracy of the bill intermediary, and effectively solve the problem of poor identification effect in the current bill intermediary identification. BRIEF DESCRIPTION OF DRAWINGS

[0043] The accompanying drawings, which are incorporated into and form part of the specification, illustrate embodiments consistent with the present application and, together with the specification, serve to explain the principles of the application.

[0044] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the accompanying drawings needed to be used in the embodiments or prior art description will be briefly introduced. Obviously, for those skilled in the art, other drawings can also be obtained without creative labor based on these drawings.

[0045] Figure 1 The flowchart provided for the bill intermediary identification method embodiment one of the present application;

[0046] Figure 2 The enterprise association network structure diagram of the bill intermediary identification method provided for the first embodiment of the present application;

[0047] Figure 3 The flowchart provided for the bill intermediary identification method embodiment two of the present application;

[0048] Figure 4 The overall flowchart of the bill intermediary identification method provided for the second embodiment of the present application;

[0049] Figure 5A module structure schematic diagram of a bill intermediary identification system in an embodiment of the present application;

[0050] Figure 6 A device structure schematic diagram of a hardware operating environment involved in a bill intermediary identification method in an embodiment of the present application.

[0051] The object implementation, functional features and advantages of the present application will be further described with reference to the embodiments and the accompanying drawings. DETAILED DESCRIPTION

[0052] It should be understood that the first embodiment described herein is only used to explain the technical solutions of the present application, and is not used to limit the present application.

[0053] In order to better understand the technical solutions of the present application, the following will be described in detail in combination with the drawings and specific embodiments of the specification.

[0054] The main solution of the first embodiment of the present application is: obtaining original business data of an enterprise to be evaluated, inputting the original business data into a bill intermediary identification model, and outputting a probability score of the enterprise to be evaluated as a risk bill intermediary; constructing a network graph representing the network structure of the enterprise based on the original business data, and calculating an enterprise evaluation index of the enterprise to be evaluated based on the network graph; score calibration is performed on the probability score to generate a target risk score; generating an enterprise risk portrait based on the target risk score, the network graph, the enterprise evaluation index and the original business data, and generating an evaluation result of the enterprise to be evaluated based on the enterprise risk portrait, wherein the evaluation result is used to represent whether the enterprise to be evaluated is a legal bill intermediary.

[0055] In the first embodiment, the bill intermediary identification system is taken as the execution subject for description for convenience.

[0056] Since the prior art mainly relies on manual investigation by bank staff, external business information query and bill endorsement flow record review to identify bill intermediaries, it not only consumes a lot of manpower and time, but also seriously depends on personal experience, which is easy to lead to misjudgment or omission.

[0057] The application provides a solution that automatically processes data through a model, quickly outputs a probability score of an enterprise to be evaluated as a risk bill intermediary, greatly improves the identification efficiency, reduces the time cost of manual operation, and the model is trained based on a large amount of data, can more objectively and accurately mine potential risk features in the data, reduces the risk of misjudgment or omission caused by insufficient personal experience, can deeply mine potential correlations and risk transmission paths between enterprises by constructing a network graph representing the correlation network structure of the enterprise and calculating the enterprise evaluation index, and can evaluate the risk status of the enterprise to be evaluated as a whole, thereby discovering risk features that are difficult to detect through single enterprise data, improving the accuracy and comprehensiveness of identification, and the target risk score generated through score calibration can more accurately reflect the actual risk level of the enterprise to be evaluated, reduce the deviation of the model output result, provide a more reliable basis for subsequent risk assessment and decision-making, generate an enterprise risk portrait, can comprehensively and comprehensively evaluate the enterprise from multiple angles, more accurately depict the risk features of the enterprise, improve the identification accuracy of the bill intermediary, and assist the identification of the bill intermediary, thereby effectively solving the problem of poor identification effect in the current bill intermediary identification.

[0058] It should be noted that the execution subject of the first embodiment can be a computing service device with data processing, network communication and program running functions, such as a tablet computer, a personal computer, a mobile phone, or an electronic device, or a system, an application, a program, etc. capable of realizing the above functions. The first embodiment and the following embodiments will be described below with reference to a bill intermediary identification system (hereinafter referred to as "system").

[0059] All actions of obtaining signals, information or data in this application are carried out in accordance with the corresponding data protection regulations and policies of the country where the device is located, and with the authorization of the corresponding device owner.

[0060] Based on this, the application embodiment provides a bill intermediary identification method, which refers to Figure 1 , Figure 1 The flowchart of the first embodiment of the bill intermediary identification method of the application is shown in the figure.

[0061] In this embodiment, the bill intermediary identification method comprises steps S01-S04:

[0062] Step S01, obtaining the original business data of the enterprise to be evaluated, inputting the original business data into the bill intermediary identification model, and outputting a probability score of the enterprise to be evaluated as a risk bill intermediary;

[0063] It should be noted that the enterprise to be evaluated refers to an enterprise entity that needs to be evaluated for bill intermediary risk. The original business data is all relevant data of the enterprise to be evaluated, including enterprise basic data (such as credit limit, whether listed, whether state-owned enterprise, etc.), enterprise business data (such as establishment time, registered capital, etc.), bill behavior data (such as endorsement, discount amount, etc.), transaction behavior data (such as transfer out, transfer in amount, etc.), device login data (such as login IP, MAC address, etc.), and enterprise associated person data (such as associated enterprise, associated person, etc.). The bill intermediary identification model learns the feature patterns of bill intermediaries by learning a large amount of known bill intermediaries and normal enterprises, so as to be able to judge the input enterprise data to be evaluated and predict whether it is a bill intermediary. It can be constructed based on the XGBoost algorithm. The risk bill intermediary refers to an enterprise that obtains benefits through illegal means (such as manipulating bill circulation, conducting false transactions, etc.) in bill business, which poses a potential threat to the financial market order. The probability score is a value between 0 and 1 output by the bill intermediary identification model after receiving the original business data of the enterprise to be evaluated, indicating the likelihood of the enterprise to be evaluated being a risk bill intermediary.

[0064] Exemplarily, the system collects multi-dimensional original business data of the enterprise to be evaluated based on a preset time interval (for example, every week), pre-processes the collected original business data (such as data cleaning, feature extraction, etc.), and inputs the enterprise basic data, enterprise business data, bill behavior data, and transaction behavior data in the original business data into the pre-trained bill intermediary identification model. The model outputs a probability score of the enterprise to be evaluated being a risk bill intermediary, which reflects the likelihood of the enterprise being identified as a bill intermediary.

[0065] It can be understood that, since the current bill intermediary identification mainly relies on manual investigation by bank staff, external business information query, and bill endorsement circulation record review, it is not only time-consuming and labor-intensive, but also heavily dependent on personal experience, making it difficult to quickly and accurately identify risk bill intermediaries from massive and complex data. Therefore, step S01 is performed to automatically obtain original business data and input the bill intermediary identification model, realizing fast processing and analysis of data, reducing the time cost of manual operation, and the model is trained based on a large amount of data, which can more objectively and accurately mine potential risk features in the data and reduce the risk of misjudgment or omission caused by insufficient personal experience.

[0066] Step S02, based on the original business data, a network graph representing the enterprise association network structure is constructed, and the enterprise evaluation index of the enterprise to be evaluated is calculated based on the network graph;

[0067] It should be noted that the network graph is constructed according to enterprise entity, associated person and network address and other enterprise relationship information, and can be used to represent the enterprise association network structure, wherein the enterprise entity, the associated person and the network address are taken as network nodes, and the relationship between them is taken as a connection edge, so that the association relationship between the enterprise entity, the associated person and the network address can be displayed in a graphical manner. The enterprise evaluation index is used to measure the risk degree of the enterprise to be evaluated in the association network, and can include indexes such as sub-network concentration, average clustering coefficient, abnormal association concentration, black list customer proportion and black list node internal density.

[0068] Exemplarily, the network graph is constructed by using the enterprise relationship information in the original business data, based on the constructed network graph, a sub-network discovery algorithm (such as LPA label propagation algorithm) is used to divide the enterprise entities in the network graph into sub-networks, and the enterprise evaluation index of the sub-network where the enterprise to be evaluated is located is calculated.

[0069] It can be understood that, since the traditional identification method often ignores the association relationship between enterprises, and important risk clues may be hidden in these association relationships, step S02 is performed to construct a network graph representing the enterprise association network structure, so as to deeply mine the potential association and risk transmission path between enterprises, evaluate the risk status of the enterprise to be evaluated as a whole, and based on the enterprise evaluation index, the concentrated bill intermediary group can be found, and the identification ability of the hidden risk can be improved.

[0070] Step S03, score calibration is performed on the probability score to generate a target risk score;

[0071] It should be noted that the target risk score is the final score obtained after score calibration, and is used to comprehensively evaluate the risk level of the enterprise to be evaluated.

[0072] Exemplarily, considering that the probability score output by the bill intermediary identification model may have deviation, it is necessary to calibrate it. The calibration process can adjust the probability score output by the model in combination with the preset business rules and the original business data. After score calibration, the target risk score is generated, which more accurately reflects the actual risk level of the enterprise to be evaluated.

[0073] It can be understood that, since the probability score output by the bill intermediary identification model may be affected by factors such as model training data distribution and feature selection, there is a certain deviation, which leads to misjudgment or omission of risk enterprises, and affects the identification effect, so step S03 is performed to adjust the probability score by score calibration, so that it is more consistent with the real risk distribution, thereby generating a more accurate target risk score, thereby providing a more reliable basis for subsequent risk assessment and decision-making, and reducing the risk caused by model deviation.

[0074] At step S04, the enterprise risk portrait is generated based on the target risk score, the network graph, the enterprise evaluation index, and the original business data, and the evaluation result of the enterprise to be evaluated is generated based on the enterprise risk portrait, where the evaluation result is used to indicate whether the enterprise to be evaluated is a legal bill intermediary.

[0075] It should be noted that the enterprise risk portrait integrates the multi-aspect information of the enterprise to be evaluated, and can comprehensively describe and depict the risk characteristics of the enterprise to be evaluated. The evaluation result is determined based on the enterprise risk portrait and the preset evaluation rule, and is the result of whether the enterprise to be evaluated is a legal bill intermediary. The legal bill intermediary refers to an enterprise that complies with laws and regulations and relevant provisions in bill business, normally carries out bill business, and does not have irregular operations and fraudulent behaviors.

[0076] Exemplarily, the enterprise risk portrait is generated by integrating the target risk score, the network graph, the enterprise evaluation index, and other related information (such as enterprise qualification and transaction record) in the original business data. The portrait comprehensively describes the risk characteristics of the enterprise to be evaluated, including risk level, correlation, and potential risk points. Based on the generated enterprise risk portrait, the preset judgment rule and threshold are combined to judge whether the enterprise to be evaluated is a legal bill intermediary. If the enterprise to be evaluated has high risk characteristics (such as high target risk score and abnormal enterprise evaluation index), it is judged to be a risk bill intermediary; otherwise, it is judged to be a legal bill intermediary.

[0077] In addition, it should be noted that the system can integrate an explainability tool such as SHAP or LIME. The explainability tool is used to help understand and explain the decision-making process of the machine learning model. SHAP (SHapley Additive exPlanations) measures the contribution of each feature to the model prediction by calculating the Shapley value of each feature, which can provide global and local explanations and help understand the behavior of the model. LIME (Local Interpretable Model-agnostic Explanations) generates perturbed samples around the target sample, calculates the prediction values of these samples, and then fits a simple model (such as linear regression or decision tree) on these perturbed samples to explain the prediction result of the target sample. After generating the enterprise risk portrait, an explanation report is generated based on the explainability tool. The explanation report is used to explain the specific features that contribute most to the determination of the risk bill intermediary, such as “overlap with the directors and supervisors of blacklisted enterprises”, “blacklist proportion of the sub-network is higher than the corresponding threshold”, and “sudden increase in discount frequency in the preset period”.

[0078] It can be understood that, since a single risk indicator is difficult to comprehensively and accurately describe the risk status of an enterprise, step S04 is performed to generate an enterprise risk portrait by integrating the target risk score, the network diagram, the enterprise evaluation indicators and other relevant information in the original business data, so as to comprehensively and integrally evaluate the enterprise from multiple perspectives, and the judgment result based on the risk portrait is more scientific and reliable, which can effectively solve the problem of poor recognition effect in the current bill intermediary recognition. In addition, the enterprise risk portrait can also provide detailed risk information for the financial institutions, which is helpful for them to make more reasonable risk prevention and control strategies and decisions.

[0079] In a feasible implementation, in step S01, before the step of inputting the original business data into the bill intermediary recognition model, steps A01-A03 are further included:

[0080] Step A01, the pre-stored enterprise historical data is taken as a training set, and the training set is input into a preset gradient decision tree model for iterative training;

[0081] It should be noted that the system has previously established data acquisition channels with each enterprise sample. These enterprise samples are objects for bill intermediary recognition in the historical period of the system. The enterprise historical data of each enterprise sample is obtained by data acquisition through the data acquisition channels, including data related to bill business, transactions and industrial and commercial affairs generated by the enterprise sample in the historical period, such as bill endorsement records, transaction amounts, industrial and commercial registration information in the past year, etc. The enterprise historical data is pre-processed to obtain a training set for model training. The preset gradient decision tree model is a decision tree model based on the gradient boosting framework, which is used to identify bill intermediaries and can be an XGboost algorithm model in machine learning. The training set is input into the preset gradient decision tree model to start the iterative training process. In each iteration, a decision tree is constructed according to the current data characteristics to attempt to predict whether the enterprise is a bill intermediary.

[0082] Step A02, the loss function between the prediction value generated in each round of training and the corresponding true label is calculated, and a new decision tree is constructed based on the gradient direction of the loss function, wherein the new decision tree is used to fit the residual between the prediction value and the corresponding true label;

[0083] It should be noted that the predicted value refers to the result of the preset gradient decision tree model predicting whether the enterprise corresponding to the training set is a bill intermediary in the training process. The true label refers to the annotation result of whether the enterprise corresponding to the training set is actually a bill intermediary, which has been pre-annotated. The loss function is a function for measuring the difference between the predicted value and the true label. The gradient direction refers to the change direction of the loss function in the parameter space, which is used to guide the construction of a new decision tree. The new decision tree is a decision tree constructed according to the gradient direction of the loss function to fit the residual error between the previous predicted value and the true label. The residual error is used to represent the gap between the predicted value and the corresponding true label, for example, the true label of a certain enterprise is 1, and the model predicted value is 0.7, then the residual error is 0.3.

[0084] Exemplarily, the XGboost algorithm model is taken as the preset gradient decision tree model, and the XGBoost (eXtreme Gradient Boosting) is a high-efficiency machine learning algorithm based on gradient boosting decision tree (GBDT), which is superior to the traditional GBDT in accuracy, robustness and computational efficiency by introducing L1 / L2 regularization, column / row sampling, high-efficiency histogram algorithm and other technologies. XGBoost constructs multiple decision trees through iteration and gradually corrects the residual error of the previous model to form a strong learner. The core objective of XGBoost is to minimize the combination of weighted loss function and model complexity, and the objective function is to optimize the performance of the model by minimizing the loss function, which specifically includes two parts: the loss function and the regularization term, and the objective function can be represented as:

[0085]

[0086] wherein, is the objective function, n is the number of samples, y i is the true label of the i-th sample, is the predicted value of the i-th sample, is the loss function, which is used to measure the residual error between the predicted value and the true label, K is the number of weak learners, and the weak learner is usually a shallow decision tree, which is called a regression tree. Each decision tree is a simple model that can learn local features in data, f k is the k-th weak learner, and Ω(f k ) is the regularization term.

[0087] In addition, it should be noted that the regularization term (also known as the complexity penalty term) is an important part of the objective function, which is used to prevent overfitting of the model. The regularization term limits the complexity of the model, so that the model performs well on the training data, and also maintains good generalization ability on new data. It usually includes two parts: the number of leaf nodes of the tree and the sum of the squares of the weights of the leaf nodes, and the regularization term can be represented as:

[0088]

[0089] Wherein, Ω(f k ) represents the regularization term, T is the number of leaf nodes in the tree, and ω is the regularization term. j is the weight of the j-th leaf node, and γ and λ are regularization parameters. γ is used to control the penalty for the number of leaf nodes, and λ is used to control the penalty for the weight of the leaf nodes.

[0090] Additionally, it's important to note that during training, feature importance is quantified using metrics such as Gain (split gain), Cover (sample coverage), and Split (number of times a feature is selected as a split node) to help identify key risk features. Gain measures the average gain a feature gains when used as a split point across all trees, reflecting its contribution to reducing the model's loss function; gain refers to the reduction in the objective function at each split node. Cover measures the number of samples a feature covers when used as a split point across all trees, reflecting the number of samples the feature is involved in during training. Split measures the number of times a feature is selected as a split node across all trees, reflecting the frequency of feature usage in the model.

[0091] Step A03: Terminate training when the change in residual is below a preset threshold, and obtain a bill intermediary recognition model integrated from several gradient boosting decision trees.

[0092] It should be noted that the preset threshold is a pre-set threshold used to determine whether the change in residuals is small enough to terminate training. When the change in residuals falls below the preset threshold, it indicates that the model has basically converged, and further iterations will no longer significantly correct the residuals; at this point, training is terminated. The final invoice intermediary identification model is an ensemble of several gradient boosting decision trees. These decision trees work together to make comprehensive predictions based on the input enterprise data.

[0093] Additionally, it should be noted that in the application of the bill intermediary identification model, for binary classification tasks, the model employs a binary:logistic objective function, combined with a sigmoid function, to map the output value of the regression tree to probability scores between 0 and 1, used to determine whether a sample belongs to a bill intermediary client. Here, binary:logistic is an objective function used for binary classification problems; it predicts the probability of a sample belonging to the positive class through logistic regression. The mathematical form of logistic regression is as follows:

[0094]

[0095] where f i is the probability that the ith sample belongs to the positive class, z i is the output of the linear combination, usually denoted as z i = ω · x i + b, where ω is the weight vector, x i is the feature vector of the ith sample, and b is the bias term.

[0096] In the embodiment, by inputting the enterprise historical data as a training set into the preset gradient decision tree model for iterative training, the problem of low efficiency of manual data collection and processing is solved, the data utilization efficiency is improved, by calculating the loss function and constructing a new decision tree based on the gradient to fit the residual error, the problem that the traditional method is difficult to capture the complex nonlinear relationship of data is solved, the feature mining capability is improved, the training is terminated when the residual error change is less than a threshold, overfitting is avoided, the model is more accurate, and the current problem of poor recognition effect of bill intermediaries is effectively solved, the recognition accuracy and efficiency are improved.

[0097] In a feasible implementation, in step S02, the step of constructing a network graph representing the structure of the enterprise association network based on the original business data includes steps A11-A13:

[0098] Step A11, extracting each enterprise entity, each associated person, and each network address in the original business data;

[0099] It should be noted that the enterprise entity refers to an organization with independent legal or operating entity qualifications in bill business, such as a company, an enterprise, etc. The associated person is an individual who has a specific association relationship with the enterprise entity, such as the legal representative, shareholder, senior manager, etc. of the enterprise. The network address is used to identify the address information of devices or resources in the network, including IP address, domain name, etc.

[0100] Illustratively, the enterprise entity information is accurately located and extracted from the original business data by using a data analysis algorithm and a preset rule engine; the associated person information is identified and extracted from the text data by using natural language processing technology; the network address is extracted from data such as logs, transaction records, etc. by using regular expressions and specific data pattern matching methods.

[0101] Step A12, determining the first relationship between each enterprise entity, the second relationship between each enterprise entity and each associated person, and the third relationship between each enterprise entity and each network address based on the original business data;

[0102] It should be noted that the first relationship is the relationship between each enterprise entity, such as upstream and downstream cooperation relationship, mutual transfer, mutual endorsement, etc. The second relationship is the relationship between each enterprise entity and each associated person, such as employment relationship, stock holding relationship, etc. The third relationship is the relationship between each enterprise entity and each network address, such as the network address used by the enterprise and the association between the network address and the business activities of the enterprise.

[0103] Exemplarily, by analyzing the transaction records, cooperation agreements and other data in the original business data, the business relationship between the enterprise entities is determined, and the first relationship between the enterprise entities is obtained. According to the business registration information, employment records and other information in the original business data, the employment, stock holding and other relationships of the associated persons in the enterprise entities are clarified, and the second relationship between the enterprise entities and the associated persons is obtained. By analyzing the network access logs, electronic bill transmission records and other data in the original business data, the network addresses used by the enterprise entities and the association between the network addresses and the business activities of the enterprise are determined, and the third relationship between the enterprise entities and each network address is obtained.

[0104] Step A13, each enterprise entity, each associated person and each network address is defined as a network node, and the first relationship, the second relationship and the third relationship are defined as the connection edges between the network nodes, to obtain a network graph representing the structure of the enterprise association network.

[0105] Exemplarily, each enterprise entity, each associated person and each network address extracted in step A11 are abstracted as network nodes, each node has a unique identifier and attribute information, the first relationship, the second relationship and the third relationship determined in step A12 are abstracted as connection edges between the network nodes, the direction of the edge can be defined according to the type of the relationship, and a visual network graph representing the structure of the enterprise association network is constructed, which displays the complex relationship between the nodes. The network graph is updated regularly, such as once a week.

[0106] Exemplarily, in order to help understand the technical concept or technical principle of the present application, please refer to Figure 2 , Figure 2A business association network structure diagram is provided, which includes business subject a, business subject b, associated person a, associated person b, associated person c, network address 1, network address 2 and network address 3, wherein the business subject a and the business subject b represent actual business entities, and the two are in a relationship of mutual transfer and parent-subsidiary company, the associated person a is the actual controller of the business subject a, the associated person b has a payment contact relationship with the business subject a and is also the actual controller of the business subject b, and the associated person c is the beneficial owner and the actual controller of the business subject b, each business subject is associated with a specific network address through MAC or IP, specifically, the business subject a and the business subject b are associated with the network address 1 through MAC, the business subject a and the business subject b are associated with the network address 3 through IP, and the business subject b is associated with the network address 2 through MAC.

[0107] In addition, it should be noted that after obtaining the network diagram, in order to further capture the dynamic evolution characteristics of the business association relationship, a plurality of network snapshots divided according to time sequences can be constructed based on the timestamp information in the original business data, and a dynamic network embedding algorithm is used to process the network snapshots to obtain a dynamic embedding vector capable of representing the time sequence evolution mode of the entity relationship. Specifically, the timestamp information in the original business data is extracted, the network diagram is divided based on the timestamp information and a preset time interval, and each network snapshot arranged in time sequence is generated; each network snapshot is input into a preset dynamic network embedding model for time sequence correlation feature learning, and a dynamic embedding vector of each network node in the network diagram at each time node is obtained. The dynamic network embedding model is a model for processing network structures that change over time, and is designed to embed nodes in a dynamic network into a low-dimensional vector space while preserving the structure and time information of the network. The dynamic embedding vector is used as a time sequence feature and is supplemented to the calculation process of the enterprise evaluation index to enhance the measurement of the evolution stability of the target sub-network structure and the time sequence abnormality of the association behavior of the enterprise to be evaluated.

[0108] In this embodiment, by extracting the enterprise entities, associated persons and network addresses in the original business data, the problem of scattered data being difficult to focus on key elements is solved, the analysis object can be accurately positioned by determining various relationships, the shortcomings of traditional identification ignoring association relationships are made up, potential risk clues are mined, the network diagram representing the enterprise association network structure is constructed by converting the elements and relationships into network nodes and connection edges, the overall picture of enterprise association is intuitively presented, the information silos are broken, the complex associations in the bill business can be comprehensively captured, the problem of poor recognition effect of the current bill intermediary is effectively solved, and the recognition accuracy and comprehensiveness are improved.

[0109] Based on the first embodiment of the present application, in the second embodiment of the present application, the same or similar contents as the above first embodiment can be referred to the above introduction, and will not be described hereinafter. On this basis, please refer to Figure 3In step S02, the enterprise evaluation indexes include sub-network dimension indexes and enterprise dimension indexes, and the step of calculating the enterprise evaluation indexes of the enterprise to be evaluated based on the network graph includes steps S11-S13:

[0110] In step S11, the network graph is divided into sub-networks, and a target sub-network to which the enterprise to be evaluated belongs is determined.

[0111] It should be noted that a sub-network is a group of network nodes such as enterprise entities, associated persons and physical addresses that are closely connected in a network graph. For example, a group of enterprises and associated persons that have frequent transaction and stock control relationships among each other in a network graph can be regarded as a sub-network. The target sub-network refers to the sub-network to which the enterprise to be evaluated belongs.

[0112] In step S12, sub-network dimension indexes of the target sub-network are calculated based on topological information of the target sub-network.

[0113] It should be noted that the topological information is information used to describe the distribution and characteristics of nodes and connection edges in a sub-network in a network graph, including the number of nodes, the number of connection edges, the degree of a node (the number of edges connected to the node), the clustering coefficient (reflecting the closeness between neighbors around the node), etc. The sub-network dimension indexes are quantitative indexes used to measure the closeness of the target sub-network, the structural aggregation characteristics and the risk association status, and can include sub-network density, average clustering coefficient, abnormal association density, blacklisted customer proportion and blacklisted node internal density.

[0114] In addition, it should be noted that the sub-network density refers to the ratio of the actual number of edges in the sub-network to the maximum possible number of edges, and is used to reflect the closeness of the internal association of the sub-network. The calculation formula of the sub-network density is as follows:

[0115]

[0116] wherein Density is the sub-network density, N real is the actual number of edges, indicating the number of connection edges between all network nodes in the target sub-network, is the number of enterprise entity-enterprise entity connection edges, N ent *N per is the number of enterprise entity-associated person connection edges, N ent *N addr is the number of enterprise entity-network address connection edges, is the maximum possible number of edges, indicating the connection edges between the enterprise to be evaluated and other network nodes in the target sub-network. If the maximum possible number of edges is 0, the sub-network density is set to 0.

[0117] In addition, it should be noted that the average clustering coefficient is used to measure the closeness of the edges formed between each network node and the neighbor nodes in the target sub-network, and can reflect the structural aggregation characteristics of the target sub-network, and the value range is [0, 1], and the average clustering coefficient calculation formula is:

[0118]

[0119] Wherein, Clustering Coefficient is the average clustering coefficient, k is the number of neighbor nodes of the network node, E v is the number of actual existing connection edges between the network node and the neighbor nodes, is the maximum possible number of edges between the network node and the neighbor nodes, if k<2,

[0120] In addition, it should be noted that the abnormal association concentration refers to the proportion of the blacklist node related connection edge number in the total connection edge number of the target sub-network, and reflects the potential risk, and the abnormal association concentration calculation formula is:

[0121]

[0122] Wherein, Abnormal Concentration is the abnormal association concentration, N Black is the blacklist node related connection edge number, N real is the actual edge number, indicating the connection edge number between all network nodes in the target sub-network. The blacklist node related connection edge number can be determined by judging whether the network node satisfies the blacklist edge condition or not, and the blacklist edge condition refers to: expanding from the blacklist node, if the blacklist node and a certain enterprise entity have direct edge association, the connection edge is calculated into the blacklist node related connection edge number; if the blacklist node and a certain network node have a shared network address node or an associated person node, then the two connection edges of the shared network address node and the associated person node are calculated into the blacklist node related connection edge number.

[0123] In addition, it should be noted that the blacklist customer ratio refers to the proportion of the blacklist node in the target sub-network, and reflects the overall risk level of the target sub-network, and the blacklist customer ratio calculation formula is:

[0124]

[0125] Wherein, Black Customer Ratio is the blacklist customer ratio, P Blcak is the number of blacklist nodes, and Z Total is the total number of nodes in the target sub-network.

[0126] In addition, it should be noted that the internal density of the blacklist node is similar to the subnetwork concentration, but the range is narrowed to the blacklist node itself and the network address nodes and associated person nodes associated with it. The internal density of the blacklist node refers to the ratio of the number of edges between blacklist nodes to the maximum possible number of edges, reflecting the closeness of the blacklist. The formula for calculating the internal density of the blacklist node is:

[0127]

[0128] where Nlack Density is the internal density of the blacklist node, N B_real represents the number of actual connection edges between all network nodes related to the blacklist node (representing a blacklist enterprise), represents the number of connection edges between blacklist enterprises, B_ent *N per represents the number of connection edges between blacklist enterprises and associated persons, B_ent *N addr represents the number of connection edges between blacklist enterprises and network addresses, represents the number of possible connection edges between the blacklist node and all network nodes in the target subnetwork.

[0129] In addition, it should be noted that the subnetwork dimension indicator can also include a comprehensive vector similarity. The dynamic embedding vectors of the network nodes in the target subnetwork at each time node are aggregated (e.g., using the average, sum, or other appropriate aggregation function) to obtain a comprehensive vector representing the overall characteristics of the target subnetwork at the corresponding time node. The similarity between the corresponding comprehensive vectors of adjacent time nodes of the target subnetwork (i.e., comprehensive vector similarity) is calculated, and the comprehensive vector similarity is used to evaluate the evolutionary stability of the structure of the target subnetwork. For example, the higher the similarity, the smaller the change in the structure of the subnetwork at adjacent time points, and the more stable the evolution; on the contrary, it indicates that the structure of the subnetwork has changed greatly, and the evolution is unstable.

[0130] Step S13, based on the network nodes and connection edges associated with the node corresponding to the enterprise to be evaluated in the network graph, calculating the enterprise dimension indicator of the enterprise to be evaluated.

[0131] It should be noted that the enterprise dimension indicator is a quantitative indicator for representing the external association characteristics and risk exposure level of the enterprise to be evaluated, which can include the number of shared nodes, the number of blacklist associations, and whether it is a blacklist. Whether it is a blacklist is used to identify a blacklist enterprise, which is pre-marked.

[0132] Additionally, it should be noted that the number of shared nodes is the number of nodes that share the network address (IP, MAC address) and the associated person with the enterprise to be evaluated. It can be determined by calculating whether the enterprise to be evaluated and other network nodes have shared IP, MAC address, and associated person, for example, customer A is associated with a certain IP / MAC address / associated person, and at the same time, the IP / MAC address / associated person is also associated with customer B. Then the number of shared nodes + 1.

[0133] Additionally, it should be noted that the number of blacklisted associations is the number of nodes associated with the blacklisted enterprise. It can be determined by calculating whether the enterprise to be evaluated and the blacklisted enterprise have direct association (direct edge relationship between enterprise entities) or indirect association (shared IP / MAC address / associated person). If the enterprise to be evaluated and the blacklisted enterprise have direct association or indirect association, then the number of blacklisted associations + 1. If the enterprise to be evaluated and the same blacklisted enterprise have multiple associations, it is counted as 1, that is, the number of blacklisted associations is only + 1.

[0134] Additionally, it should be noted that the enterprise dimension indicator can also include the number of abnormal time nodes. The dynamic embedding vector of the enterprise to be evaluated at each time node is extracted to form a vector sequence. Based on time series analysis methods, such as statistical methods (such as mean, variance analysis), machine learning methods (such as isolation forest, one-class support vector machine), etc., the vector sequence is analyzed to identify abnormal time nodes that do not conform to the normal association behavior pattern of the enterprise to be evaluated. The association behavior corresponding to these abnormal time nodes may be abnormal. The number of abnormal time nodes (i.e., the number of abnormal time nodes) is counted. The number of abnormal time nodes is used to measure the time sequence abnormality of the enterprise association behavior.

[0135] In this embodiment, by dividing the network graph into sub-networks and determining the target sub-network to which the enterprise to be evaluated belongs, the problem of difficulty in focusing on analyzing the association range of the enterprise is solved, making the analysis more targeted. The sub-network dimension indicators are calculated based on the topology information of the target sub-network, which can reveal the internal association tightness of the sub-network, discover potential risk aggregation areas, and present the external association and risk exposure of the enterprise to be evaluated by calculating the enterprise dimension indicators. From the two dimensions of sub-network and enterprise, a comprehensive evaluation is made to mine the complex association and risk characteristics of the bill intermediary, effectively solving the problem of poor recognition effect of the current bill intermediary.

[0136] In a feasible implementation, before the step of dividing the network graph into sub-networks in step S11, steps B01-B03 are further included:

[0137] Step B01, initialize the sub-network label of each network node in the enterprise association network structure of the network graph;

[0138] It should be noted that the sub-network label is a label used to identify the sub-network to which the network node belongs. In the sub-network division process of the enterprise association network structure, each network node is assigned a sub-network label, and nodes with the same label belong to the same sub-network. Among them, for the marked network node, the sub-network label is known and does not need to be assigned again. For the unmarked network node, the corresponding sub-network label is empty.

[0139] Exemplarily, the network graph can be represented in a graph structure G=(V, E), where V is a set of network nodes, and E is a set of connection edges. The adjacency matrix W of the network graph G is calculated, and the element W ij in the adjacency matrix W represents the similarity or weight between the network node i and the network node j. The label matrix Y is initialized, where the sub-network label of the marked network node is a known value, and the sub-network label of the unmarked network node is empty.

[0140] Step B02, performing multiple rounds of label propagation iterations, in each round of iteration, traversing all target network nodes with empty sub-network labels, determining the high-frequency label with the highest occurrence frequency among the adjacent nodes of the target network node, and updating the sub-network label of the target network node to the high-frequency label;

[0141] It should be noted that the target network node refers to the network node that needs to update the sub-network label in the current round of label propagation iteration. It usually refers to the node whose sub-network label is empty or whose final sub-network affiliation has not been determined. For example, in the initial iteration stage, all network nodes can be regarded as target network nodes. As the iteration proceeds, part of the nodes have determined the sub-network label, and the remaining nodes with undetermined labels are the target network nodes in the subsequent iterations. The adjacent node is a network node directly connected to the target network node, and the adjacent node and the target node have a direct association relationship, such as the transaction relationship between enterprises, the position relationship of associated natural persons, etc. The high-frequency label refers to the sub-network label with the highest occurrence frequency among the adjacent nodes of the target network node, which reflects the most concentrated sub-network affiliation information of the adjacent nodes of the target node, and is used to update the sub-network label of the target node.

[0142] Additionally, it should be noted that after starting the multi-round label propagation iteration, for each round of iteration, the system traverses all target nodes with empty subnetwork labels in the enterprise association network structure, which are nodes that have not yet determined the final subnetwork affiliation, and for each target node, the system checks the subnetwork labels of its adjacent nodes (i.e., nodes directly connected to the node), counts the frequency of occurrence of each label, finds the high-frequency label with the highest frequency of occurrence, and then updates the subnetwork label of the target node to this high-frequency label. For example, among the adjacent nodes of node A, there are 3 labels of "1" and 2 labels of "2", so the high-frequency label of node A is "1", and the system will update the subnetwork label of node A to "1".

[0143] Exemplarily, in each round of iteration, the label value of each target node is updated to the weighted average of the label values of the adjacent nodes, and the update formula is:

[0144] F (t+1) =αSF (t) +(1-α)Y

[0145] where F (t) is the label matrix of the tth iteration, F (t+1) is the label matrix of the (t+1)th iteration, S is the normalized adjacent matrix, which is usually calculated by S=D -1 W or S=d -1 / 2 WD -1 / 2 , D is the degree matrix, which is a diagonal matrix representing the degree of each node in the graph. For a graph G=(V, E), where V is the set of nodes and E is the set of edges, the degree matrix D is an n×n diagonal matrix, where n is the number of nodes, and the diagonal elements of the degree matrix represent the degree of node i, i.e., the number of edges connected to node i. ii D ii =∑ j W ij , α is the propagation coefficient, usually taking a value between (0, 1), representing the strength of label propagation, and Y is the initial label matrix.

[0146] Step B03, terminate the iteration when the subnetwork label of each network node in the network graph no longer changes, and collect the network nodes with the same subnetwork label in the network graph into the same subnetwork.

[0147] It should be noted that the system continues to perform label propagation iteration until the subnetwork label of each network node in the enterprise association network structure no longer changes (or the number of iterations reaches the preset maximum number of iterations), i.e., it reaches a state of convergence, at which point the algorithm considers that the subnetwork division has stabilized and no further iteration is needed. The network nodes with the same subnetwork label are collected into the same subnetwork, and the subnetwork division process is completed.

[0148] In this embodiment, the initialization of the enterprise association network node subnetwork label lays the foundation for subsequent subnetwork division. Through multiple rounds of label propagation iteration, the subnetwork label is updated based on the high-frequency label in the corresponding adjacent nodes of the target node, the node ownership is determined, and the iteration is terminated when the node label is stable and the subnetwork is collected. By subnetwork division, potential associations are mined, and the bill intermediary related enterprise group can be more accurately identified. Without explicit model training, the calculation efficiency is high, and the problem of poor recognition effect caused by insufficient data association mining in current bill intermediary recognition is effectively solved.

[0149] In a feasible implementation, in step S03, the step of generating a target risk score by score calibration on the probability score includes steps B11-B13:

[0150] Step B11, based on the preset bonus item rule, determine the bonus item rule satisfied by the enterprise to be evaluated in combination with the original business data, and increase the corresponding score value for the probability score based on the bonus item rule;

[0151] It should be noted that the bonus item rule is a series of rules preset for increasing the risk score of the enterprise, which are targeted at the typical behaviors and risk characteristics of bill intermediaries. Once the enterprise meets these rules, it means that it has a higher risk of being a bill intermediary. For example, the bonus item rule can be: consistent with the Mac address, IP address, director, supervisor, senior management, payment contact, actual controller, and account handler of the blacklisted enterprise, and has mutual large amount of transfer and mutual endorsement with the blacklisted enterprise, each hit adds 10 points.

[0152] Step B12, based on the preset deduction item rule, determine the deduction item rule satisfied by the enterprise to be evaluated in combination with the original business data, and reduce the corresponding score value for the probability score based on the deduction item rule;

[0153] It should be noted that the deduction item rule is a series of rules preset for reducing the risk score of the enterprise, which reflect the positive performance and compliance characteristics of the enterprise in bill business, indicating that the enterprise is unlikely to be a bill intermediary. For example, the deduction item rule can be: having a credit line, being a listed enterprise or a state-owned enterprise, having a qualification label (such as a high-tech enterprise, a small giant enterprise, etc.), or having more than 5 patents or works, each hit reduces 10 points.

[0154] Step B13, the adjusted probability score is subjected to upper and lower limit range constraints, and the target risk score is output.

[0155] It should be noted that after the adjustment of the bonus item and the deduction item, the probability score may exceed a reasonable range. In order to avoid extreme values of the score, the system sets an upper and lower limit range, for example, the range of the target risk score is set between 0-100 points.

[0156] In the embodiment, by presetting the bonus item rule and combining the original business data to add points to the probability score, the risk behaviors of enterprises such as violations can be considered, and the risk identification is strengthened. By presetting the deduction item rule to deduct points, the factors that are beneficial to reducing the suspicion of bill intermediaries, such as the active compliance of enterprises, can be quantitatively reflected, which makes up for the deficiency of relying only on the probability of the basic model. The score after adjustment is constrained to ensure the reasonableness and consistency of the score, and solves the problem that the current bill intermediary identification does not fully consider the positive and negative factors of enterprises, the score lacks standardization, and the identification effect is poor, which improves the identification accuracy and reliability.

[0157] In a feasible implementation, in step S04, the step of generating the evaluation result of the enterprise to be evaluated based on the enterprise risk portrait includes steps B21-B22:

[0158] Step B21, extracting the target risk score in the enterprise risk portrait, and comparing the target risk score with a preset risk threshold;

[0159] It should be noted that the system extracts the target risk score of the enterprise to be evaluated from the constructed enterprise risk portrait, and the target risk score comprehensively reflects the risk degree of the enterprise to be evaluated in the bill business. The system compares the extracted target risk score with a preset risk threshold (for example, 85 points).

[0160] Step B22, in the case where the target risk score is higher than the preset risk threshold, extracting the enterprise evaluation index in the enterprise risk portrait, comparing the enterprise evaluation index with the corresponding index threshold, determining the exceeding condition of the enterprise evaluation index and the corresponding index threshold, and generating the evaluation result of the enterprise to be evaluated based on the exceeding condition.

[0161] It should be noted that when it is judged in step B21 that the target risk score is higher than the preset risk threshold, the system will further extract the enterprise evaluation index from the enterprise risk portrait. The enterprise evaluation index reflects the characteristics of the enterprise to be evaluated in the bill business sub-network. The system compares these enterprise evaluation indexes with the corresponding index thresholds set in advance, determines the exceeding condition of the enterprise evaluation index and the corresponding index threshold, and obtains a comprehensive evaluation result according to the exceeding condition. The evaluation result clearly indicates whether the enterprise to be evaluated is a legal bill intermediary.

[0162] Exemplarily, in the case that the enterprise evaluation index selection sub-network dimension index is evaluated, if more than half of the indexes of the sub-network concentration, the average clustering coefficient, the abnormal association concentration, the blacklisted customer proportion, and the blacklisted node internal density are higher than the corresponding index threshold, the evaluation result is that the enterprise to be evaluated is not a legal bill intermediary. The user can also customize other judgment rules, which will not be described here.

[0163] In the embodiment, by extracting the target risk score and comparing it with the preset threshold, high-risk enterprises are quickly screened out, avoiding complex analysis of all enterprises, solving the problem of low identification efficiency caused by time-consuming and labor-consuming comprehensive evaluation, and further comparing with the enterprise evaluation index when the score is high, in-depth analysis of the characteristics of the enterprise association network, solving the problem that it is difficult to accurately distinguish legal and illegal bill intermediaries only by the score, determining the nature of the enterprise based on the comparison result, and improving the judgment accuracy by combining the score and the sub-network index. The current poor identification effect problem in bill intermediary identification is effectively solved, and the identification accuracy and reliability are improved.

[0164] Exemplarily, in order to help understand the technical concept or technical principle of the present application, please refer to Figure 4 , Figure 4 The overall flowchart of the bill intermediary identification method is provided, which includes three stages of data processing, data processing and information integration. In the data processing stage, various types of data are collected, including enterprise basic data, enterprise business data, bill behavior data, transaction behavior data, equipment login data and enterprise associated person data. These data are the basis for subsequent analysis, covering multi-dimensional information such as basic information, business behavior, transaction characteristics, login situation and associated person situation of the enterprise. In the data processing stage, on the one hand, the collected data (enterprise basic data, enterprise business data, bill behavior data, transaction behavior data) are used for XGboost model training, and the target risk score is obtained through the model, which is used to evaluate the possibility of the existence of bill intermediary risk of the enterprise. On the other hand, the associated sub-network division is carried out, the enterprise is divided into different sub-networks, the association relationship between enterprises and the contact between enterprises and associated persons are clearly presented, the network graph representing the enterprise association network structure is generated by using the collected data (bill behavior data, transaction behavior data, equipment login data and enterprise associated person data), and the enterprise evaluation index is calculated. In the information integration stage, the target risk score, the network graph, the enterprise evaluation index and the original business data are integrated, and finally the enterprise risk portrait is generated, which fully shows the risk status of the enterprise, providing strong support for accurately identifying bill intermediaries.

[0165] It should be noted that the above examples are only used to understand the present application and do not constitute a limitation on the bill intermediary identification method of the present application. More forms of simple transformation based on this technical concept are within the protection scope of the present application.

[0166] The application also provides a bill intermediary identification system, please refer to Figure 5 The bill intermediary identification system comprises:

[0167] A model evaluation module 10 is configured to obtain original business data of an enterprise to be evaluated, input the original business data into a bill intermediary identification model, and output a probability score of the enterprise to be evaluated being a risk bill intermediary;

[0168] An index calculation module 20 is configured to construct a network graph representing a network structure of the enterprise based on the original business data, and calculate an enterprise evaluation index of the enterprise to be evaluated based on the network graph;

[0169] A score calibration module 30 is configured to perform score calibration on the probability score to generate a target risk score;

[0170] A result generation module 40 is configured to generate an enterprise risk portrait based on the target risk score, the network graph, the enterprise evaluation index, and the original business data, and generate an evaluation result of the enterprise to be evaluated based on the enterprise risk portrait, wherein the evaluation result is used to indicate whether the enterprise to be evaluated is a legal bill intermediary.

[0171] Optionally, the model evaluation module 10 is further configured to:

[0172] use pre-stored enterprise historical data as a training set, and input the training set into a preset gradient decision tree model for iterative training;

[0173] calculate a loss function between a predicted value generated in each round of training and a corresponding true label, and construct a new decision tree based on a gradient direction of the loss function, wherein the new decision tree is used to fit a residual error between the predicted value and the corresponding true label;

[0174] terminate the training when a change amount of the residual error is lower than a preset threshold, and obtain a bill intermediary identification model integrated by a plurality of gradient boosting decision trees.

[0175] Optionally, the index calculation module 20 is further configured to:

[0176] extract each enterprise entity, each associated person, and each network address in the original business data;

[0177] determine a first relationship between each enterprise entity, a second relationship between each enterprise entity and each associated person, and a third relationship between each enterprise entity and each network address based on the original business data;

[0178] define each enterprise entity, each associated person, and each network address as a network node, and define the first relationship, the second relationship, and the third relationship as a connection edge between the network nodes, to obtain a network graph representing the network structure of the enterprise.

[0179] Optionally, the enterprise evaluation index includes a sub-network dimension index and an enterprise dimension index, and the index calculation module 20 is further configured to:

[0180] divide the network graph into sub-networks, and determine a target sub-network to which the enterprise to be evaluated belongs;

[0181] calculate the sub-network dimension index of the target sub-network based on the topology information of the target sub-network;

[0182] calculate the enterprise dimension index of the enterprise to be evaluated based on the network nodes and connection edges associated with the node corresponding to the enterprise to be evaluated in the network graph.

[0183] Optionally, the index calculation module 20 is further configured to:

[0184] initialize the sub-network label of each network node in the enterprise associated network structure in the network graph;

[0185] perform multiple rounds of label propagation iterations, in each round of iteration, iterate through all target network nodes with an empty sub-network label, determine the high-frequency label with the highest occurrence frequency among the adjacent nodes of the target network node, and update the sub-network label of the target network node to the high-frequency label;

[0186] terminate the iteration when the sub-network label of each network node in the network graph no longer changes, and collect the network nodes with the same sub-network label in the network graph into the same sub-network.

[0187] Optionally, the score calibration module 30 is further configured to:

[0188] determine the bonus item rule satisfied by the enterprise to be evaluated based on the preset bonus item rule and the original business data, and increase the corresponding score value for the probability score based on the bonus item rule;

[0189] determine the penalty item rule satisfied by the enterprise to be evaluated based on the preset penalty item rule and the original business data, and decrease the corresponding score value for the probability score based on the penalty item rule;

[0190] perform upper and lower limit range constraints on the adjusted probability score, and output the target risk score.

[0191] Optionally, the result generation module 40 is further configured to:

[0192] extract the target risk score in the enterprise risk portrait, and compare the target risk score with a preset risk threshold;

[0193] In a case where the target risk score is higher than the preset risk threshold, an enterprise evaluation index in the enterprise risk portrait is extracted, the enterprise evaluation index is compared with a corresponding index threshold, an exceeding condition of the enterprise evaluation index and the corresponding index threshold is determined, and an evaluation result of the enterprise to be evaluated is generated based on the exceeding condition.

[0194] The bill intermediary recognition system provided in the application adopts the bill intermediary recognition method in the above embodiments, and can solve the technical problem of poor recognition effect in the current recognition of bill intermediaries. Compared with the prior art, the bill intermediary recognition system provided in the application has the same beneficial effects as the bill intermediary recognition method provided in the above embodiments, and other technical features in the bill intermediary recognition system are the same as the features disclosed in the above embodiments, and thus will not be described here.

[0195] The application provides an electronic device, which comprises at least one processor and a memory connected with the at least one processor; the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the bill intermediary recognition method in the above embodiment one.

[0196] Reference will be made to the accompanying drawings Figure 6 which shows a structural schematic diagram of an electronic device suitable for implementing the embodiments of the application. The electronic device in the embodiments of the application can include but is not limited to mobile terminals such as mobile phones, notebook computers, PADs (Portable Application Description: tablet computers) and the like, and fixed terminals such as digital TVs, desktop computers and the like. Figure 6 The electronic device shown is only an example, and should not bring any limitation to the functions and use range of the embodiments of the application.

[0197] As Figure 6As shown, the electronic device can include a processing device 1001 (e.g., a central processor, a graphics processor, etc.) that can perform various appropriate actions and processes according to programs stored in a read-only memory 1002 or loaded from a storage device 1003 into a random access memory 1004. Various programs and data required for operation of the electronic device are also stored in the random access memory 1004. The processing device 1001, the read-only memory 1002, and the random access memory 1004 are connected to each other through a bus 1005. An input / output interface 1006 is also connected to the bus. Generally, the following systems can be connected to the input / output interface 1006: input devices 1007 including, for example, a touch screen, a touch pad, a keyboard, a mouse, an image sensor, a microphone, an accelerometer, a gyroscope, etc.; output devices 1008 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, etc.; the storage device 1003 including, for example, a magnetic tape, a hard disk, etc.; and a communication device 1009. The communication device 1009 can allow the electronic device to communicate wirelessly or wired with other devices to exchange data. Although the electronic device having various systems is shown in the figure, it should be understood that all of the shown systems are not required to be implemented or possessed. More or less systems can be alternatively implemented or possessed.

[0198] In particular, according to the embodiments disclosed in the present application, the processes described above with reference to the flowcharts can be implemented as a computer software program. For example, the embodiments disclosed in the present application include a computer program product comprising a computer program carried on a computer readable medium, the computer program containing program codes for executing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network through the communication device, or installed from the storage device 1003, or installed from the read-only memory 1002. When the computer program is executed by the processing device 1001, the above-mentioned functions defined in the methods of the embodiments disclosed in the present application are performed.

[0199] The electronic device provided by the present application adopts the bill intermediary identification method in the above-mentioned embodiments, and can solve the technical problem of poor identification effect in the current bill intermediary identification. Compared with the prior art, the electronic device provided by the present application has the same beneficial effects as the bill intermediary identification method provided by the above-mentioned embodiments, and other technical features in the electronic device are the same as the features disclosed in the previous embodiment method, which will not be repeated here.

[0200] It should be understood that parts of the present application can be realized by hardware, software, firmware or a combination thereof. In the description of the above-mentioned embodiments, specific features, structures, materials or characteristics can be combined in any one or more embodiments or examples in a suitable manner.

[0201] The above merely provides the specific implementation of the present application, but the protection scope of the present application is not limited to this. Any person skilled in the art can easily think of the changes or replacements within the technical range disclosed by the present application, which should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

[0202] The present application provides a computer readable storage medium having computer readable program instructions (i.e. computer programs) stored thereon, which are used to execute the bill intermediary identification method in the above embodiments.

[0203] The computer readable storage medium provided by the present application may, for example, be a U disk, but is not limited to an electric, magnetic, optical, electromagnetic, infrared, or semiconductor system or device, or any combination of the above. More specific examples of the computer readable storage medium can include, but are not limited to, an electric connection with one or more conductive wires, a portable computer disk, a hard disk, a random access memory (RAM), a read only memory (ROM), an erasable programmable read only memory (EPROM or flash memory), an optical fiber, a portable compact disk read only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present embodiment, the computer readable storage medium can be any tangible medium containing or storing a program, which can be used by or in combination with an instruction execution system or device. The program code contained on the computer readable storage medium can be transmitted by any suitable medium, including but not limited to an electric wire, an optical cable, an RF (Radio Frequency), etc., or any suitable combination of the above.

[0204] The above computer readable storage medium can be contained in an electronic device; or can exist separately without being assembled into an electronic device.

[0205] The computer readable storage medium described above carries one or more programs, when the one or more programs are executed by the electronic device, the bill intermediary identification device: acquires original business data of an enterprise to be evaluated, inputs the original business data into a bill intermediary identification model, and outputs a probability score of the enterprise to be evaluated as a risk bill intermediary; constructs a network graph representing a network structure of the enterprise based on the original business data, and calculates an enterprise evaluation index of the enterprise to be evaluated based on the network graph; performs score calibration on the probability score to generate a target risk score; generates an enterprise risk portrait based on the target risk score, the network graph, the enterprise evaluation index, and the original business data, and generates an evaluation result of the enterprise to be evaluated based on the enterprise risk portrait, wherein the evaluation result is used to represent whether the enterprise to be evaluated is a legal bill intermediary.

[0206] Computer program code for carrying out operations of the present application can be written in any combination of one or more programming languages, including an object oriented programming language such as Java, Smalltalk, C++ or the like and conventional procedural programming languages, such as the "C" programming language or similar programming languages. The program code can execute entirely on the user's computer, partly on the user's computer, as a stand-alone software package, partly on the user's computer and partly on a remote computer or entirely on the remote computer or server. In the latter scenario, the remote computer can be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or the connection can be made to an external computer (for example, through the Internet using an Internet Service Provider).

[0207] The flow diagrams and the block diagrams in the drawings are illustrations of architectures, functionalities, and operations of possible implementations of systems, methods, and computer program products according to various embodiments of present application. In this regard, each block in the flow diagrams or block diagrams can represent a module, a procedure, or a portion of code, which comprises one or more executable instructions for implementing the specified logical functions. It should also be noted that in some alternative implementations, the functions noted in the blocks can occur out of the order noted in the figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently or the blocks may

[0208] The modules described in the embodiments of the present application can be implemented in the form of software or in the form of hardware. In some cases, the names of the modules do not constitute a limitation on the modules themselves.

[0209] The readable storage medium provided by the present application is a computer readable storage medium, which stores computer readable program instructions (i.e., a computer program) for executing the above-mentioned bill intermediary identification method, and can solve the technical problem of poor identification effect in the current identification of bill intermediaries. Compared with the prior art, the computer readable storage medium provided by the present application has the same beneficial effects as the bill intermediary identification method provided by the above-mentioned embodiments, and will not be described here.

[0210] The present application also provides a computer program product comprising a computer program, which, when executed by a processor, implements the steps of the bill intermediary identification method as described above.

[0211] The computer program product provided by the present application can solve the technical problem of poor identification effect in the current identification of bill intermediaries. Compared with the prior art, the computer program product provided by the present application has the same beneficial effects as the bill intermediary identification method provided by the above-mentioned embodiments, and will not be described here.

[0212] The above only describes some embodiments of the present application, and does not limit the patent scope of the present application. Any equivalent structural transformation, direct / indirect application in other related technical fields based on the technical concept of the present application, and the contents of the present application and the accompanying drawings are included in the patent protection scope of the present application.

Claims

1. A method for identifying intermediaries in negotiable instruments, characterized in that, The method for identifying intermediaries in bills includes: Obtain the original business data of the enterprise to be evaluated, input the original business data into the bill intermediary identification model, and output the probability score of the enterprise to be evaluated as a risky bill intermediary. Based on the original business data, a network graph representing the enterprise's interconnected network structure is constructed, and enterprise evaluation indicators for the enterprise to be evaluated are calculated based on the network graph. The probability scores are calibrated to generate a target risk score; A risk profile of an enterprise is generated based on the target risk score, the network diagram, the enterprise evaluation indicators, and the original business data. An evaluation result of the enterprise to be evaluated is generated based on the risk profile of the enterprise. The evaluation result is used to indicate whether the enterprise to be evaluated is a legitimate bill intermediary.

2. The bill intermediary identification method as described in claim 1, characterized in that, Before the step of inputting the original business data into the bill intermediary identification model, the following steps are also included: The pre-stored historical data of the enterprise is used as the training set, and the training set is input into the preset gradient decision tree model for iterative training. Calculate the loss function between the predicted value generated in each training round and the corresponding true label, and construct a new decision tree based on the gradient direction of the loss function, wherein the new decision tree is used to fit the residual between the predicted value and the corresponding true label; Training is terminated when the change in the residual is below a preset threshold, resulting in a bill intermediary recognition model integrated from several gradient boosting decision trees.

3. The bill intermediary identification method as described in claim 1, characterized in that, The steps of constructing a network graph representing the enterprise's interconnected network structure based on the original business data include: Extract each enterprise entity, each associated person, and each network address from the original business data; Based on the original business data, determine the first relationship between each enterprise entity, the second relationship between each enterprise entity and each associated person, and the third relationship between each enterprise entity and each network address; Each enterprise entity, each associated person, and each network address is defined as a network node, and the first relationship, the second relationship, and the third relationship are defined as connection edges between the network nodes, thus obtaining a network graph representing the enterprise association network structure.

4. The bill intermediary identification method as described in claim 1, characterized in that, The enterprise evaluation indicators include sub-network dimension indicators and enterprise dimension indicators. The step of calculating the enterprise evaluation indicators of the enterprise to be evaluated based on the network graph includes: The network diagram is divided into sub-networks, and the target sub-network to which the enterprise to be evaluated belongs is determined; Based on the topology information of the target sub-network, calculate the sub-network dimension index of the target sub-network; Based on the network nodes and connecting edges associated with the corresponding nodes of the enterprise to be evaluated in the network graph, the enterprise dimension indicators of the enterprise to be evaluated are calculated.

5. The bill intermediary identification method as described in claim 4, characterized in that, The step of dividing the network graph into sub-networks includes: Initialize the sub-network label of each network node under the enterprise-related network structure in the network diagram; Perform multiple rounds of label propagation iterations. In each round of iteration, traverse all target network nodes with empty sub-network labels, determine the high-frequency label that appears most frequently among the neighboring nodes of the target network node, and update the sub-network label of the target network node with the high-frequency label. The iteration terminates when the sub-network label of each network node in the network graph no longer changes, and network nodes with the same sub-network label in the network graph are grouped into the same sub-network.

6. The bill intermediary identification method as described in claim 1, characterized in that, The step of calibrating the probability scores to generate a target risk score includes: Based on the preset bonus rules, the bonus rules that the enterprise to be evaluated meets are determined in combination with the original business data, and the corresponding score is added to the probability score based on the bonus rules. Based on the preset deduction rules, the deduction rules that the enterprise to be evaluated meets are determined in combination with the original business data, and the corresponding score is reduced based on the deduction rules. The adjusted probability scores are constrained by upper and lower limits, and the target risk score is output.

7. The bill intermediary identification method as described in claim 1, characterized in that, The step of generating the assessment result for the enterprise to be assessed based on the enterprise risk profile includes: Extract the target risk score from the enterprise risk profile and compare the target risk score with a preset risk threshold; If the target risk score is higher than the preset risk threshold, the enterprise assessment indicators in the enterprise risk profile are extracted, the enterprise assessment indicators are compared with the corresponding indicator thresholds, the excess of the enterprise assessment indicators and the corresponding indicator thresholds is determined, and the assessment result of the enterprise to be assessed is generated based on the excess.

8. An electronic device, characterized in that, The device includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, the computer program being configured to implement the steps of the bill intermediary identification method as described in any one of claims 1 to 7.

9. A storage medium, characterized in that, The storage medium is a computer-readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by a processor, it implements the steps of the bill intermediary identification method as described in any one of claims 1 to 7.

10. A computer program product, characterized in that, The computer program product includes a computer program that, when executed by a processor, implements the steps of the bill intermediary identification method as described in any one of claims 1 to 7.

Images