Method and device for auditing illegal commission of agent, electronic equipment and medium

Abstract

The invention discloses a method and device for auditing illegal commission of agents, electronic equipment and a medium, and the method comprises the steps: obtaining a plurality of pieces of reward settlement data, and extracting preset data items from the reward settlement data as feature data; a training set and a verification set are obtained based on arrangement of the feature data and the mark data corresponding to all the reward settlement data; configuring a multi-layer perceptron; based on the training set, training the multilayer perceptron through forward propagation and back propagation to obtain a training model; performing performance verification on the training model based on the verification set to obtain a performance result; performing iterative training on the multi-layer perceptron until the performance result is no longer improved, and taking the training model of the last iteration as a target model; wherein the target model is used for auditing the reward settlement data of the to-be-audited agent to obtain the illegal commission condition of the to-be-audited agent. The method can efficiently and accurately realize illegal commission of auditing agents, and can be widely applied to the technical field of data processing.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to a method, device, electronic equipment and medium for auditing illegal commission arbitrage by agents. Background Art

[0002] In the communications industry, agents taking commissions refers to agents taking advantage of rules or system loopholes to obtain more commissions through improper means, which is a violation of regulations. This behavior violates the contract or agreement with the communications operator, damages the interests of the operator and other agents, and affects the brand image and reputation of the operator.

[0003] At present, the commission cost of illegal commission collection is about 1 million per month, the amount of illegal commission collection accounts for about 5% of the total commission, and the commission users account for about 6% of the total number of developed users, which directly leads to an increase in low-quality users, an increase in user churn, and a loss of operator revenue; at the same time, it indirectly leads to a decrease in high-quality users, and the interests of other honest agents are damaged, destroying the fair competition environment in the market. However, the current inspection methods for illegal commission collection are often through manual verification of settlement data, which has low processing efficiency, or verification based only on rigid indicator rules, which has great limitations and low accuracy. Summary of the invention

[0004] The present invention proposes a method, device, electronic device and medium for auditing agents' illegal commissions, aiming to solve one of the technical problems in the related art at least to a certain extent. The embodiments of the present invention can efficiently and accurately realize the illegal commissions of auditing agents.

[0005] On the one hand, an embodiment of the present invention provides a method for auditing illegal commission arbitrage by agents, comprising:

[0006] Acquire multiple remuneration settlement data, extract preset data items from the remuneration settlement data as feature data; each remuneration settlement data is marked with marking data indicating illegal commission arbitrage;

[0007] Based on the feature data and labeled data corresponding to all remuneration settlement data, a training set and a validation set are obtained;

[0008] Configure a multi-layer perceptron; based on the training set, train the multi-layer perceptron through forward propagation and back propagation to obtain a training model;

[0009] The performance of the training model is verified based on the validation set to obtain the performance result; the step of training the multilayer perceptron based on the training set through forward propagation and back propagation is returned to execute until the performance result no longer improves in a continuous preset number of iterations, and the training model of the last iteration is used as the target model;

[0010] Among them, the target model is used to audit the remuneration settlement data of the audited agents to obtain the illegal commission arbitrage of the audited agents.

[0011] Optionally, before the step of acquiring a plurality of remuneration settlement data, the method further comprises the following steps:

[0012] Pre-process the peripheral data based on the preset remuneration settlement logic, complete the remuneration settlement based on the preset remuneration policy, and obtain the remuneration settlement data;

[0013] In response to the audit result, the illegal commission arbitrage situation of each remuneration settlement data is obtained as marking data.

[0014] Optionally, extracting preset data items from the remuneration settlement data as feature data comprises the following steps:

[0015] Based on the preset commission feature data items, characteristic data is obtained from the remuneration settlement data;

[0016] Among them, the characteristic data includes multiple agent characteristic data, including the main package ordered by the user, the number of months the user has been online, the matching commission policy, the user's cumulative settled remuneration multiples, the number of settlements within 12 months of the user's corresponding ID card information, whether the user attributes have been converted, and the user's network access channel.

[0017] Optionally, obtaining a training set and a validation set based on the feature data and the labeled data corresponding to all the remuneration settlement data includes the following steps:

[0018] Sample data is obtained based on the feature data and the marked data corresponding to the same remuneration settlement data;

[0019] The training set and validation set are obtained based on the sample data corresponding to all the remuneration settlement data.

[0020] Optionally, configuring a multilayer perceptron includes the following steps:

[0021] A multilayer perceptron is obtained based on the input layer, hidden layer and output layer configuration;

[0022] Initialize the weights, biases, and activation functions of the hidden and output layers.

[0023] Optionally, the training set includes interrelated feature data and label data corresponding to a plurality of remuneration settlement data; the multilayer perceptron includes an input layer, a hidden layer, and an output layer; based on the training set, the multilayer perceptron is trained by forward propagation and back propagation to obtain a training model, including the following steps:

[0024] Input the feature data in the training set into the multilayer perceptron as input data;

[0025] Based on the input data, the prediction result is obtained through forward propagation processing in the multi-layer perceptron;

[0026] Based on the prediction results combined with the labeled data in the training set, the error results are obtained using the chain rule.

[0027] According to the error results, the network parameters of the hidden layer and the output layer are updated through back propagation to obtain the training model.

[0028] Optionally, input data is received through an input layer; based on the input data, a prediction result is obtained by forward propagation processing in a multi-layer perceptron, comprising the following steps:

[0029] Based on the input data, a hidden layer output is obtained by processing the first network parameter and the first activation function of the hidden layer;

[0030] Based on the hidden layer output, the output data is obtained by processing the second network parameters and the second activation function of the output layer;

[0031] Wherein, the first network parameter and the second network parameter both include weights and biases.

[0032] On the other hand, an embodiment of the present invention provides a device for auditing illegal commission arbitrage by agents, including:

[0033] The first module is used to obtain a plurality of remuneration settlement data, extract preset data items from the remuneration settlement data as feature data; each remuneration settlement data is marked with marking data of illegal commission arbitrage;

[0034] The second module is used to obtain a training set and a validation set based on the feature data and the marked data corresponding to all the remuneration settlement data;

[0035] The third module is used to configure the multi-layer perceptron; based on the training set, the multi-layer perceptron is trained through forward propagation and back propagation to obtain a training model;

[0036] The fourth module is used to perform performance verification on the training model based on the verification set to obtain performance results; return to execute the steps of training the multilayer perceptron based on the training set through forward propagation and back propagation in the third module until the performance results no longer improve in a continuous preset number of iterations, and the training model of the last iteration is used as the target model;

[0037] Among them, the target model is used to audit the remuneration settlement data of the audited agents to obtain the illegal commission arbitrage of the audited agents.

[0038] On the other hand, an embodiment of the present invention provides an electronic device, including: a processor and a memory; the memory is used to store programs; the processor executes the program to implement the above-mentioned method of auditing agents' illegal commission arbitrage.

[0039] On the other hand, an embodiment of the present invention provides a computer storage medium, in which a program executable by a processor is stored. When the program executable by the processor is executed by the processor, it is used to implement the above-mentioned method of auditing illegal commission arbitrage by an auditing agent.

[0040] The embodiment of the present invention obtains multiple remuneration settlement data, extracts preset data items from the remuneration settlement data as feature data; each remuneration settlement data is marked with marked data of illegal commission arbitrage; based on the feature data and marked data corresponding to all remuneration settlement data, a training set and a verification set are obtained; a multi-layer perceptron is configured; based on the training set, the multi-layer perceptron is trained by forward propagation and back propagation to obtain a training model; based on the verification set, the performance of the training model is verified to obtain a performance result; return to execute the step of training the multi-layer perceptron by forward propagation and back propagation based on the training set until the performance result is no longer improved in the continuous preset number of iterations, and the training model of the last iteration is used as the target model; wherein the target model is used to audit the remuneration settlement data of the audited agent to obtain the illegal commission arbitrage of the audited agent. The present invention uses the multi-layer perceptron model algorithm, deep learning and data analysis technology, and can accurately analyze the mapping relationship between the feature data in the remuneration settlement data and the illegal commission arbitrage situation, and then through the constructed target model, it can efficiently realize the accurate identification and prediction of the agent's remuneration arbitrage situation. The present invention can efficiently and accurately realize the audit of illegal commissions of agents. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] The accompanying drawings are used to provide a further understanding of the technical solution of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the technical solution of the present invention and do not constitute a limitation on the technical solution of the present invention.

[0042] Figure 1 A schematic diagram of an implementation environment for auditing agents' illegal commission arbitrage provided by an embodiment of the present invention;

[0043] Figure 2 A flowchart of a method for auditing illegal commissions of agents provided by an embodiment of the present invention;

[0044] Figure 3 A schematic diagram of an extended process of a method for auditing agents' illegal commission arbitrage provided by an embodiment of the present invention;

[0045] Figure 4 A schematic diagram of the expanded flow of step S200 provided in an embodiment of the present invention;

[0046] Figure 5 A schematic diagram of the process of training a multi-layer perceptron according to an embodiment of the present invention;

[0047] Figure 6 A schematic diagram of the MLP algorithm process provided by an embodiment of the present invention;

[0048] Figure 7 A schematic diagram of the audit process for auditing agents' illegal commission arbitrage provided by an embodiment of the present invention;

[0049] Figure 8 The following is a schematic diagram of the existing remuneration audit process;

[0050] Fig. 9 A schematic diagram of the module structure of a device for auditing agents' illegal commissions provided by an embodiment of the present invention;

[0051] Fig.10 A schematic diagram of the structure of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0052] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0053] It should be noted that, although the functional modules are divided in the system schematic diagram and the logical order is shown in the flowchart, in some cases, the steps shown or described may be performed in a different order than the module division in the system or the order in the flowchart. The terms "first / S100", "second / S200", etc. in the specification, claims and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence.

[0054] Reference to "embodiments" herein means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present invention. The appearance of the phrase in various places in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0055] To facilitate the understanding of the technical solution of the present invention, the technical feature terms that may appear in the embodiments of the present invention are first explained:

[0056] Perceptron: It has only input and output layers, which together form a simple neuron, i.e. a single neuron model.

[0057] Multi-Layer Perceptron (MLP): A generalization of the perceptron, its main feature is that it has multiple layers of neurons. MLP is a specific type of artificial neural network that consists of multiple neurons, usually including an input layer, one or more hidden layers, and an output layer.

[0058] Early stopping: Early stopping is a widely used method that calculates the performance of the model on the validation set during training. When the performance of the model on the validation set begins to decline, training is stopped to avoid the problem of overfitting caused by continued training.

[0059] It is understandable that the method for auditing agents' illegal commissions provided in the embodiment of the present invention can be applied to any computer device with data processing and computing capabilities, and this computer device can be various types of terminals or servers. When the computer device in the embodiment is a server, the server is an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, CDN (Content Delivery Network), and big data and artificial intelligence platforms. Optionally, the terminal is a smart phone, tablet computer, laptop computer, desktop computer, etc., but is not limited to this.

[0060] like Figure 1 FIG. 1 is a schematic diagram of an implementation environment provided by an embodiment of the present invention. Figure 1 The implementation environment includes at least one terminal 102 and a server 101. The terminal 102 and the server 101 can be connected to a network wirelessly or wired to complete data transmission and exchange.

[0061] Server 101 can be an independent physical server, or a server cluster or distributed system composed of multiple physical servers. It can also be a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, CDN (Content Delivery Network), as well as big data and artificial intelligence platforms.

[0062] In addition, the server 101 can also be a node server in the blockchain network. Blockchain is a new application model of computer technologies such as distributed data storage, peer-to-peer transmission, consensus mechanism, and encryption algorithm.

[0063] The terminal 102 may be a smart phone, a tablet computer, a laptop computer, a desktop computer, a smart speaker, a smart watch, etc., but is not limited thereto. The terminal 102 and the server 101 may be directly or indirectly connected via wired or wireless communication, which is not limited in the embodiment of the present invention.

[0064] Based on the example Figure 1 In the implementation environment shown, an embodiment of the present invention provides a method for auditing agents' illegal commissions. The following is explained by taking the method for auditing agents' illegal commissions applied in terminal 102 as an example. It can be understood that the method for auditing agents' illegal commissions can also be applied in server 101.

[0065] Reference Figure 2 , Figure 2 The flowchart of the method for checking commissions of auditing agents in violation of regulations and applied to terminals provided by the embodiment of the present invention, the execution subject of the method for checking commissions of auditing agents in violation of regulations can be any of the aforementioned computer devices (including servers or terminals). Figure 2 , the method comprises the following steps:

[0066] S100, obtaining a plurality of remuneration settlement data, and extracting preset data items from the remuneration settlement data as feature data;

[0067] Among them, each remuneration settlement data is marked with data indicating illegal commission arbitrage;

[0068] It should be noted that in some optional embodiments, such as Figure 3 As shown, before the step of obtaining multiple remuneration settlement data, the method may further include the following steps: T100, pre-processing the peripheral data based on a preset remuneration settlement logic, and completing the remuneration settlement based on a preset remuneration policy to obtain remuneration settlement data; T200, in response to the audit result, obtaining the illegal commission arbitrage situation of each remuneration settlement data as marking data.

[0069] Exemplarily, in some specific implementations, remuneration settlement: peripheral data can be pre-processed according to remuneration settlement logic, and remuneration can be settled for users according to remuneration policies to generate remuneration settlement detail data.

[0070] In some specific application scenarios, the principle logic of the T100 step can be implemented as follows:

[0071] Preset remuneration settlement logic: This logic is a set of rules and algorithms used by the system or platform to calculate remuneration. It can involve multiple factors, such as transaction volume, transaction type, transaction time, user level, etc. These factors together determine the amount of remuneration that should be obtained for each transaction or task.

[0072] Preprocessing of peripheral data: Peripheral data refers to various data related to remuneration settlement, such as transaction records, user information, product information, etc. The purpose of preprocessing this data is to ensure that they meet the requirements of remuneration settlement logic, remove invalid or erroneous data, and format or convert the data for subsequent processing.

[0073] Preset remuneration policy: The remuneration policy is a set of rules set by the platform or system to motivate users. It specifies which behaviors can be rewarded, as well as the calculation method and standard of the remuneration. The relevant policies can be adjusted and set according to actual needs. These policies can vary from platform to platform and can also change over time.

[0074] Generation of remuneration settlement data: Based on the pre-processed peripheral data and the preset remuneration policy, the system can calculate the remuneration amount for each user or transaction and generate remuneration settlement data. This data usually includes information such as user ID, remuneration amount, settlement time, etc.

[0075] The principle logic of step T200 can be implemented as follows:

[0076] Audit results: After generating the remuneration settlement data, the system or platform may audit the data to ensure the accuracy and compliance of the data. The audit may involve multiple aspects, such as the integrity, logic, and legality of the data.

[0077] Marking of illegal commissions: If the audit finds that a certain remuneration settlement data contains illegal commissions (i.e., the user obtains remuneration through improper means), the system will record this as marked data. This marked data may include the ID of the violating user, the type of violation, the time of the violation, and other information.

[0078] Additional explanation for the example:

[0079] Assume that an e-commerce platform has formulated a remuneration policy in order to encourage sellers and buyers to trade. According to the policy, sellers will receive a certain percentage of remuneration for each completed transaction, and buyers can also receive remuneration under certain specific conditions (such as purchasing designated products or participating in promotional activities).

[0080] In step T100, the platform first pre-processes the peripheral data according to the preset remuneration settlement logic. For example, it will screen out eligible transaction records and calculate the remuneration amount for each transaction. Then, according to the preset remuneration policy, the platform will generate remuneration settlement data for each seller and buyer.

[0081] In step T200, the platform will review the generated remuneration settlement data. If the review finds that a seller has obtained remuneration through fake transactions, the platform will record this situation as marked data. The marked data may include the seller's ID, the record of the fake transaction, the time of the violation, and other information.

[0082] It should be noted that, in some embodiments, extracting preset data items from the remuneration settlement data as feature data may include the following steps: obtaining feature data from the remuneration settlement data based on preset package commission feature data items; wherein the feature data includes multiple agent feature data, and the agent feature data includes the main package ordered by the user, the number of months the user has been online, the matching commission policy, the user's cumulative settled remuneration multiples, the number of settlements within 12 months of the user's corresponding ID card information, whether the user attributes have been converted, and the user's network access channel.

[0083] Exemplarily, in some specific implementations, the commission package characteristics are confirmed: 7 agent characteristic data are extracted according to the remuneration settlement data: the main package subscribed by the user, the number of months the user has been online, the matching commission policy, the user's cumulative settled remuneration multiples, the number of settlements within 12 months of the user's corresponding ID card information, whether the user attributes have been converted, and the user's network access channel.

[0084] In some specific application scenarios, the principle logic of confirming the commission arbitrage feature can be implemented as follows:

[0085] When identifying and handling agent commission arbitrage, a series of key features can be extracted through in-depth analysis of remuneration settlement data to help identify potential violations. The following is a detailed explanation of the above 7 agent feature data:

[0086] 1. User orders the main package:

[0087] Explanation: This refers to the main service package type ordered by the agent or user. Different package types may correspond to different commission policies. Therefore, analyzing the main package ordered by the user is an important aspect of identifying commission-binding behavior.

[0088] Example: An agent frequently orders packages with high commission ratios for users, but the actual usage of these packages does not meet expectations. This may be a clue of commission arbitrage.

[0089] 2. Number of months the user has been online:

[0090] Explanation: The number of months a user has been online reflects the user's continuous use and is also an important indicator for judging the authenticity and activity of users. Users who have been online for a long time are more likely to be real users, while users who register in large numbers in a short period of time and quickly obtain commissions may be suspected of commission arbitrage.

[0091] Example: If an affiliate registers a large number of new users in a short period of time, and these users quickly leave the network, this could be a sign of arbitrage behavior.

[0092] 3. Matching commission policy:

[0093] Explanation: The commission policy determines the benefits that agents and users can obtain. By analyzing the degree of match between the commission policy and agent behavior, it is possible to identify whether there is abnormal commission-earning behavior.

[0094] Example: If an agent frequently participates in high-commission campaigns, but their user activity or transaction volume does not match this, this may indicate that the agent is using some means to obtain commissions.

[0095] 4. The user's cumulative settled remuneration multiples:

[0096] Explanation: This refers to the ratio of the cumulative remuneration obtained by users or agents through the platform to their expected income. By calculating this multiple, it is possible to assess whether there is any abnormality in the efficiency of agents' remuneration acquisition.

[0097] Example: If an agent's cumulative settled remuneration multiples are much higher than the industry average and there is no reasonable explanation, this may mean that the agent is using improper means to obtain commissions.

[0098] 5. Number of settlements within 12 months corresponding to the ID card information of the user:

[0099] Explanation: This refers to the number of times a user who registered with the same ID card information participated in the remuneration settlement within 12 months. By analyzing this feature, it is possible to identify whether there is identity fraud or false registration.

[0100] For example: If an ID card information is involved in a large number of remuneration settlements within 12 months, and the user behaviors corresponding to these settlements lack consistency or rationality, this may indicate identity theft or false registration.

[0101] 6. Whether the user attributes are converted:

[0102] Explanation: User attribute conversion may involve changes in user identity, role, or status. This conversion may affect the applicability of commission policies and the calculation of commissions. By analyzing the conversion of user attributes, it is possible to identify whether there is abnormal commission earning behavior.

[0103] Example: If an agent frequently converts users from a low commission policy group to a high commission policy group, and these conversions lack reasonable business logic support, this may be a clue of commission arbitrage behavior.

[0104] 7. User access channels:

[0105] Explanation: User access channels reflect the way and means by which users enter the platform. Different access channels may correspond to different user groups and commission policies. By analyzing user access channels, it is possible to identify whether there is commission arbitrage in specific channels.

[0106] Example: If an agent signs up a large number of new users through a specific channel, and these users quickly earn high commissions, while other users from the channel do not perform in line with this, this may indicate that the agent is using the specific channel for commission arbitrage.

[0107] Additional explanation for the example:

[0108] Assume that an e-commerce platform promotes its services through agents and pays commissions based on the effectiveness of the agents’ promotion. In order to identify potential commission-binding behaviors, the above 7 agent feature data are extracted for analysis.

[0109] During the analysis, we found that the following characteristic data of a certain agent was abnormal:

[0110] The agent frequently ordered packages with high commission ratios for users;

[0111] These users have only been online for a short time and leave the platform quickly;

[0112] The agent is involved in high-commission activities that do not match the agent's user activity or trading volume;

[0113] Its cumulative settled remuneration multiples are much higher than the industry average;

[0114] The user who registered with the same ID card information participated in a large number of remuneration settlements within 12 months; the agent frequently converted users from the low commission policy group to the high commission policy group;

[0115] These users mainly enter the platform through specific channels.

[0116] Based on these abnormal characteristic data, it can be preliminarily determined that the agent has engaged in commission arbitrage, and further investigation and processing were carried out. Through in-depth analysis and investigation, it can be finally confirmed that the agent has engaged in commission arbitrage.

[0117] S200, obtaining a training set and a validation set based on the feature data and the marked data corresponding to all the remuneration settlement data;

[0118] It should be noted that in some embodiments, Figure 4As shown, obtaining a training set and a validation set based on the feature data and the marked data corresponding to all the remuneration settlement data may include the following steps: S201, obtaining sample data based on the feature data and the marked data corresponding to the same remuneration settlement data; S202, obtaining a training set and a validation set based on the sample data corresponding to all the remuneration settlement data.

[0119] For example, in some specific implementations, in the process of collating the feature data and labeled data based on the remuneration settlement data to obtain the training set and the validation set, it is necessary to ensure the accuracy, completeness and representativeness of the data so that the subsequent machine learning model can be effectively learned and verified.

[0120] S201: Sample data is obtained based on the feature data and the mark data corresponding to the same remuneration settlement data:

[0121] Explanation: This step is to integrate each remuneration settlement data (including its corresponding feature data and tag data) into a complete sample data. Feature data is the data describing the agent's behavior (such as the user's subscription to the main package, the number of months the user has been online, etc.), and tag data is the label (such as "yes" or "no") based on these feature data to determine whether the agent has engaged in commission arbitrage.

[0122] The detailed steps can be implemented as follows:

[0123] 1. Data cleaning: First, the remuneration settlement data needs to be cleaned to remove duplicate, missing or abnormal data.

[0124] 2. Feature extraction: Extract feature data related to commission-based behavior from the cleaned data, such as the type of main package ordered by the user, the number of months online, commission policy, etc.

[0125] 3. Marked data generation: Based on business rules or expert judgment, a marked data is generated for each remuneration settlement data, indicating whether the agent corresponding to the data has engaged in commission arbitrage.

[0126] 4. Data integration: Integrate feature data and label data into a complete sample data.

[0127] Example: Assume that there is a remuneration settlement data, whose characteristic data includes that the main package subscribed by the user is a "high commission package", the number of online months of the user is 2 months, the matching commission policy is a "high commission policy", etc., and according to the business rules, the agent corresponding to the data has engaged in commission arbitrage, so the marked data is "yes". Then, the sample data corresponding to this remuneration settlement data is "{characteristic data: {main package subscribed by the user: high commission package, number of online months of the user: 2, matching commission policy: high commission policy, ...}, marked data: yes}". Among them, in some preferred implementations, the agent's commission arbitrage ratio can also be marked in the marked data.

[0128] S202: Arrange the sample data corresponding to all the remuneration settlement data to obtain a training set and a validation set:

[0129] Explanation: This step is to divide all the sorted sample data into training set and validation set according to a certain ratio. The training set is used to train the machine learning model, while the validation set is used to evaluate the performance of the model.

[0130] The detailed steps can be implemented as follows:

[0131] 1. Data division: All sample data are randomly divided into training set and verification set according to a certain ratio (such as 80% training set, 20% verification set).

[0132] 2. Data balance: Ensure that the ratio of positive and negative samples in the training set and the validation set is roughly the same to avoid overfitting the model to a certain class of samples.

[0133] 3. Data preservation: Save the divided training set and validation set into corresponding files for subsequent use.

[0134] Example: Assume that there are 1,000 sample data, of which 500 are marked as "yes" (there is arbitrage behavior) and 500 are marked as "no" (there is no arbitrage behavior). Then, these data can be divided into 800 training samples and 200 validation samples according to the ratio of 80% training set and 20% validation set. During the division process, it is ensured that the ratio of positive and negative samples in the training set and validation set is roughly 1:1. Finally, save these divided data into corresponding files.

[0135] S300, configuring a multilayer perceptron; based on the training set, training the multilayer perceptron through forward propagation and back propagation to obtain a training model;

[0136] It should be noted that, in some embodiments, configuring a multilayer perceptron may include the following steps: configuring a multilayer perceptron based on an input layer, a hidden layer, and an output layer; initializing weights, biases, and activation functions of the hidden layer and the output layer.

[0137] Exemplarily, in some specific implementations, configuring a multilayer perceptron (MLP) is one of the basic steps in building a neural network model. The multilayer perceptron consists of an input layer, a hidden layer, and an output layer. By initializing weights, biases, and selecting appropriate activation functions, a neural network model with learning and prediction capabilities can be constructed. The following is a detailed explanation and supplement to the above two steps, as well as corresponding example descriptions.

[0138] 1. Based on the input layer, hidden layer and output layer configuration, a multi-layer perceptron is obtained:

[0139] 1. Input layer: The input layer is responsible for receiving input data. The dimension of the input data determines the number of neurons in the input layer. For example, if the input data is a vector containing n features, the input layer should have n neurons.

[0140] 2. Hidden layer: The hidden layer is located between the input layer and the output layer and is the key part of the neural network for feature extraction and pattern recognition. The number of hidden layers and the number of neurons in each layer can be adjusted according to the complexity of the problem and the characteristics of the data. Generally speaking, the more hidden layers and the more neurons in each layer, the stronger the network's fitting ability, but it may also lead to overfitting.

[0141] 3. The output layer is responsible for outputting the model's prediction results. The number of neurons in the output layer depends on the type of problem. For example, in a binary classification problem, the output layer usually has one neuron that outputs a probability value between 0 and 1; in a multi-classification problem, the output layer may have multiple neurons, each corresponding to the probability of a category.

[0142] Example description:

[0143] Suppose you want to build a multi-layer perceptron for a binary classification problem, the input data contains 4 features (i.e. n = 4), and select a hidden layer that contains 5 neurons. Then, the multi-layer perceptron structure can be expressed as:

[0144] Input layer: 4 neurons (corresponding to 4 input features);

[0145] Hidden layer: 5 neurons (the number can be adjusted based on experience or experiments);

[0146] Output layer: 1 neuron (outputs the probability value of binary classification);

[0147] 2. Initialize the weights, biases, and activation functions of the hidden and output layers:

[0148] 1. Weights and biases: Weights and biases are trainable parameters in a neural network, which determine the connection strength between neurons and the output threshold of neurons. Weights and biases are usually obtained by random initialization. There are many initialization methods, such as uniform distribution initialization, normal distribution initialization, He initialization, Xavier initialization, etc. Different initialization methods may affect the training effect and convergence speed of the model.

[0149] 2. Activation function: The activation function is a nonlinear transformation in a neural network, which enables the neural network to learn and represent complex nonlinear relationships. Common activation functions include ReLU (Rectified Linear Unit), Sigmoid, Tanh, etc. Different activation functions have different characteristics and applicable scenarios.

[0150] Example description:

[0151] For the multi-layer perceptron structure mentioned above, the following initialization method and activation function can be used:

[0152] Weight initialization: Use the Xavier initialization method, which automatically adjusts the initial value of the weight according to the number of neurons in the input layer and the output layer.

[0153] Bias initialization: Initialize the bias to 0 or a small random number.

[0154] Activation function: The ReLU activation function is used in the hidden layer because it has a faster convergence speed and a smaller gradient vanishing problem during training; the Sigmoid activation function is used in the output layer because it can limit the output value between 0 and 1, which is suitable for binary classification problems.

[0155] In summary, by reasonably configuring the input layer, hidden layer, and output layer, and selecting appropriate initialization methods and activation functions, a multilayer perceptron model with powerful learning and prediction capabilities can be constructed.

[0156] It should also be noted that in some embodiments, the training set includes interrelated feature data and label data corresponding to multiple remuneration settlement data; the multi-layer perceptron includes an input layer, a hidden layer and an output layer; Figure 5 As shown, based on the training set, the multilayer perceptron is trained by forward propagation and back propagation to obtain a training model, which may include the following steps: S301, inputting the feature data in the training set as input data into the multilayer perceptron; S302, based on the input data, obtaining a prediction result by forward propagation processing in the multilayer perceptron; S303, based on the prediction result combined with the labeled data in the training set, using the chain rule to obtain an error result; S304, updating the network parameters of the hidden layer and the output layer by back propagation according to the error result to obtain a training model.

[0157] For example, in some specific implementations, a process of model training is performed using a training set through a multi-layer perceptron (MLP), which includes inputting feature data into the multi-layer perceptron, performing forward propagation to obtain prediction results, calculating errors, and updating network parameters through back propagation. To facilitate the understanding of the technical solution, the above steps will be described in detail below, and corresponding examples will be given.

[0158] S301: Input the feature data in the training set as input data into the multi-layer perceptron:

[0159] Before training begins, you first need to prepare a training set. The training set contains interrelated feature data and labeled data corresponding to multiple remuneration settlement data. Feature data is the data input to the multi-layer perceptron, which contains relevant information for predicting remuneration settlement results. Labeled data is the expected output result, that is, the actual remuneration settlement result.

[0160] When inputting feature data into a multilayer perceptron, the data needs to be converted into a format suitable for network processing, such as standardizing or normalizing the data to an appropriate range, to ensure that the network can correctly process and interpret the data.

[0161] S302: Based on the input data, the prediction result is obtained through forward propagation processing in the multi-layer perceptron:

[0162] Forward propagation is the process of transferring data from the input layer to the output layer in a neural network. In this process, the input data is weighted summed and activated by neurons in each layer, and finally the prediction result of the output layer is obtained.

[0163] During the forward propagation process, the output of each layer is the input of the next layer. The neurons in the hidden layer generate outputs by calculating the weighted sum of the input data and applying an activation function. The neurons in the output layer generate the final prediction results based on the output of the hidden layer.

[0164] Example description:

[0165] Assume that the multilayer perceptron contains one hidden layer with 5 neurons and one neuron in the output layer (used to output the prediction results). During the forward propagation process, the feature data of the input layer is first passed to the hidden layer, and each neuron in the hidden layer calculates the weighted sum of its input data and applies the ReLU activation function. Then, the output of the hidden layer is passed to the output layer, and the neurons in the output layer calculate the weighted sum of their input data and apply the Sigmoid activation function (or other activation functions suitable for binary classification problems) to produce the final prediction results.

[0166] S303: Based on the prediction results and the labeled data in the training set, the error result is obtained by using the chain rule:

[0167] Once you have the predictions, you need to compare them with the labeled data in the training set to calculate the error. Error is a measure of the difference between the predictions and the actual results.

[0168] To calculate the error, a loss function is usually used (such as mean squared error, cross entropy loss, etc.). The loss function calculates the difference between the predicted result and the labeled data and returns a scalar value as the error.

[0169] In multi-layer perceptrons, the error calculation is usually performed using the chain rule (also known as the back-propagation algorithm). The chain rule allows the calculation of the gradient of the loss function with respect to each network parameter, which is used to guide the update of the network parameters.

[0170] Example description:

[0171] Assume that the cross entropy loss function is used to calculate the error between the prediction result and the labeled data. After obtaining the prediction result, it is compared with the labeled data and the cross entropy loss is calculated. Then, the gradient of the loss function with respect to each network parameter is calculated using the chain rule.

[0172] S304: Update the network parameters of the hidden layer and the output layer through back propagation according to the error results to obtain a training model:

[0173] Backpropagation is the process used in neural networks to update network parameters. During backpropagation, the weights and bias parameters in the network are updated according to the calculated gradients to reduce the error and improve the prediction performance of the network.

[0174] Updating network parameters usually uses gradient descent algorithms (such as stochastic gradient descent, mini-batch gradient descent, Adam, etc.). These algorithms update network parameters based on the calculated gradients and iteratively adjust parameter values ​​to minimize the loss function.

[0175] After updating the network parameters, forward propagation needs to be performed again to calculate the new prediction results and errors. This process is repeated multiple times (called training iterations or training rounds) until a predetermined stopping condition is reached (such as the error reaches a certain threshold, the number of iterations reaches an upper limit, etc.).

[0176] Example description:

[0177] Assume that the Adam optimization algorithm is used to update the network parameters. In each iteration, the gradient of the loss function with respect to each network parameter is first calculated (via back propagation). Then, the Adam optimization algorithm is used to update the network parameters (weights and biases) based on the gradient. After the update, the forward propagation is re-performed to calculate the new prediction results and errors. This process is repeated many times until the predetermined stopping condition is reached. The resulting network parameters constitute the trained model.

[0178] In some embodiments, input data is received through an input layer; based on the input data, a prediction result is obtained through forward propagation processing in a multilayer perceptron, which may include the following steps: based on the input data, a hidden layer output is obtained through first network parameters and a first activation function processing of the hidden layer; based on the hidden layer output, output data is obtained through second network parameters and a second activation function processing of the output layer; wherein the first network parameters and the second network parameters both include weights and biases.

[0179] For example, in some embodiments, it is assumed that there is a hidden layer with a weight of W1 and a bias of b1. The output H of the hidden layer can be calculated by the following formula: H = f1(W1X+b1) where f1 is the activation function of the hidden layer;

[0180] The weight of the output layer is W2, and the bias is b2. The output Y of the output layer can be calculated by the following formula: Y = f2(W2H+b2), where f2 is the activation function of the output layer.

[0181] In some specific application scenarios, the forward propagation process involves the processing of the hidden layer and the output layer, wherein the hidden layer and the output layer are processed using the first network parameter and the first activation function, and the second network parameter and the second activation function, respectively. This process will be described in more detail below, and corresponding examples will be given.

[0182] 1. The input layer receives input data:

[0183] The input layer is the first layer of the multilayer perceptron and is responsible for receiving raw input data. These data can be numerical, discrete, or preprocessed.

[0184] 2. Hidden layer processing:

[0185] The hidden layer is located between the input layer and the output layer, and there can be multiple layers. Each layer contains a certain number of neurons, and each neuron is connected to the neurons in the previous layer through weights.

[0186] In the hidden layer, each neuron receives input data from the previous layer (or data output by the hidden layer, for multiple hidden layers), and then applies the first activation function for nonlinear transformation.

[0187] The first network parameters (including weights and biases) are used to adjust the scale and direction of the input data to better fit the data.

[0188] The output of a hidden layer is the input to the next layer, or for the last hidden layer, the output layer.

[0189] 3. Output layer processing:

[0190] The output layer is the last layer of the multilayer perceptron and is responsible for generating predictions.

[0191] The output layer is processed similarly to the hidden layers, but usually uses a different activation function (such as softmax for classification tasks and linear for regression tasks).

[0192] The second network parameters (also including weights and biases) are used to further adjust the scale and direction of the data to obtain the final prediction results.

[0193] The role of activation function:

[0194] The activation function is used to introduce nonlinear factors so that the multilayer perceptron can fit complex nonlinear relationships.

[0195] Common activation functions include sigmoid, tanh, ReLU (Rectified Linear Unit), etc.

[0196] Update of network parameters:

[0197] During training, the network parameters (weights and biases) are updated via the back-propagation algorithm to minimize the loss function.

[0198] The back propagation algorithm calculates the gradient and updates the network parameters based on the error between the predicted result and the actual result.

[0199] S400, configure a multilayer perceptron; based on a training set, train the multilayer perceptron through forward propagation and back propagation to obtain a training model; based on a validation set, verify the performance of the training model to obtain a performance result; return to execute the step of training the multilayer perceptron through forward propagation and back propagation based on the training set until the performance result no longer improves in a continuous preset number of iterations, and use the training model of the last iteration as the target model.

[0200] Among them, the target model is used to audit the remuneration settlement data of the audited agents to obtain the illegal commission arbitrage of the audited agents;

[0201] For example, in some specific implementations, the early stopping method can be used to select the optimal parameter index of the training model to obtain the target model.

[0202] In some specific application scenarios, early stopping is a technique used to prevent neural networks from overfitting. During the training process, the performance of the model on the training set usually improves as the number of iterations increases, but the performance on the validation set may begin to decline after a certain point, indicating that the model begins to overfit the training data. Early stopping monitors the performance on the validation set and stops training when the performance no longer improves, thereby avoiding overfitting.

[0203] The early stopping method can be implemented as follows:

[0204] 1. Set monitoring indicators: Select a performance indicator on the validation set (such as accuracy, recall, F1 score, cross entropy loss, etc.) as the monitoring indicator.

[0205] 2. Record the best performance: During the training process, record the best performance indicator achieved on the validation set and its corresponding number of iterations.

[0206] 3. Set Patience: The Patience value is a preset integer that indicates the number of iterations allowed to continue training when the performance of the validation set no longer improves. If the performance still does not improve within this patience value, the training is stopped.

[0207] 4. Stop training: When the performance indicator on the validation set does not improve within a preset number of consecutive iterations (i.e., patience value), stop training and use the model obtained at the last iteration as the optimal model.

[0208] Advantages of early stopping: It can effectively prevent overfitting and improve the generalization ability of the model. It does not require additional regularization terms or parameter adjustments, which simplifies the model training process.

[0209] Example of early stopping method application, example scenario:

[0210] Suppose you are training a multi-layer perceptron model to audit the remuneration settlement data of agents to detect illegal commission arbitrage. The training set contains a large amount of historical remuneration settlement data and its corresponding labels (whether it is illegal commission arbitrage), and the validation set is used to monitor the performance of the model.

[0211] The implementation steps can be achieved as follows:

[0212] 1. Configure a multi-layer perceptron: set the input layer, hidden layer (can have multiple layers), output layer, and corresponding activation function, loss function, and optimization algorithm.

[0213] 2. Training model: Use the training set to train the multilayer perceptron through forward propagation and back propagation.

[0214] 3. Performance verification: At the end of each iteration cycle, the validation set is used to calculate the performance indicators (such as accuracy) of the model.

[0215] 4. Record the best performance: During the training process, record the highest accuracy achieved on the validation set and its corresponding number of iterations.

[0216] 5. Set the patience value: Assume that the patience value is set to 10, that is, if the accuracy on the validation set does not improve within 10 consecutive iterations, stop training.

[0217] 6. Stop training and save the model: When the accuracy on the validation set does not improve within 10 consecutive iterations, stop training and save the model obtained in the last iteration as the optimal model (i.e., the target model).

[0218] Example results: By using the early stopping method, we can obtain a multi-layer perceptron model that performs well on the validation set. This model can effectively audit the agent's remuneration settlement data and accurately detect illegal commission arbitrage. At the same time, due to the use of the early stopping method, the problem of overfitting the model to the training data is avoided, and the generalization ability of the model is improved.

[0219] In order to explain the principle of the technical solution of the present invention in detail, the overall process of the present invention is described below in combination with some specific embodiments. It is easy to understand that the following is an explanation of the technical principle of the present invention and cannot be regarded as a limitation of the present invention.

[0220] First of all, it should be noted that in response to the problem of existing agents illegally taking commissions, the present invention provides a method for auditing agents' illegal commissions. This method is based on the multi-layer perceptron (MLP) model algorithm. Through the algorithm, the suspected illegal commission users developed by the agent in the current month are audited, and an audit work order is automatically generated and sent to the district, county, and grid leaders for manual verification, forming a closed-loop process of auditing, warning, verification, tracing, and closing orders for illegal commissions.

[0221] In some embodiments, the present invention provides a method for avoiding agent remuneration and commission-based compensation, which uses a multi-layer perceptron (MLP) model algorithm, through deep learning and data analysis technology, to accurately identify and predict the agent remuneration and commission-based compensation situation of the month, and then comprehensively monitor and warn the channel package situation of agents throughout the province. The specific process is as follows:

[0222] The remuneration settlement data is deposited into the data warehouse and commission arbitrage features are extracted.

[0223] 1. Remuneration settlement: pre-process the peripheral data according to the remuneration settlement logic, settle remuneration for users according to the remuneration policy, and generate remuneration settlement detailed data;

[0224] 2. Confirm the commission package characteristics: Extract 7 agent characteristic data according to the remuneration settlement data: the main package ordered by the user, the number of months the user has been online, the matching commission policy, the cumulative multiples of the user's settled remuneration, the number of settlements within 12 months of the user's corresponding ID card information, whether the user's attributes have been converted, and the user's network access channel;

[0225] 3. Use the multi-layer perceptron algorithm model to predict the agent commission situation, such as Figure 6 As shown, the MLP algorithm process can be implemented as follows:

[0226] 3.1 The input layer receives the (T-1, T-6) monthly remuneration settlement original feature data X, where X is a matrix including N samples, and each sample includes M features;

[0227] 3.2 Assume there is a hidden layer with weight W1 and bias b1. The output H of the hidden layer can be calculated by the following formula: H = f1(W1X+b1) where f1 is the activation function of the hidden layer;

[0228] 3.3 The weight of the output layer is W2, and the bias is b2. The output Y of the output layer can be calculated by the following formula: Y =

[0229] f2(W2H+b2), where f2 is the activation function of the output layer;

[0230] 3.4 The error degree between the predicted result and the actual result of the output layer result data is calculated using the loss function;

[0231] 3.5 Use the chain rule to calculate the gradient of the loss function to the weights and biases of the output layer, and use the gradient descent optimization algorithm to update the weights and biases;

[0232] 3.6 According to the process from 3.1 to 3.5, the result data can be trained according to the 7 feature data in 2, forward propagation is performed to calculate the predicted value, the loss function is calculated, back propagation is performed to calculate the gradient, and the weights and biases are updated; the overall training is implemented as follows:

[0233] 3.6.1 Data preparation and input layer reception:

[0234] Data preparation: First, you need to prepare training data, which includes 7 agent feature data extracted from the remuneration settlement data (user subscription main package, number of user online months, matching commission policy, user cumulative settled remuneration multiples, number of settlements within 12 months corresponding to the user's ID card information, whether the user attribute is converted, and user network access channel). These data will be used as the original feature data received by the input layer.

[0235] Input layer receives: The input layer receives the original feature data X of the remuneration settlement for the month (T-1, T-6), where X is a matrix containing N samples, each sample containing M features. Here T represents the current month, and the month (T-1, T-6) represents the data from the past month to the past six months.

[0236] 3.6.2 Forward propagation calculates the predicted value:

[0237] Hidden layer calculation: Assume there is a hidden layer with weight W1 and bias b1. The output H of the hidden layer can be calculated by the formula H = f1(W1X+b1), where f1 is the activation function of the hidden layer (such as ReLU, Sigmoid or Tanh, etc.). The role of this activation function is to convert the result of the linear combination into a nonlinear output to increase the expressive power of the model.

[0238] Output layer calculation: The weight of the output layer is W2, and the bias is b2. The output Y of the output layer can be calculated by the formula Y = f2 (W2H + b2), where f2 is the activation function of the output layer. In the scenario of predicting the commission of agents, the output layer usually uses activation functions such as Sigmoid or Softmax to output one or more probability values ​​as the prediction results.

[0239] 3.6.3 Calculate the loss function:

[0240] Loss function selection: According to the specific requirements of the prediction task (such as binary classification, multi-classification or regression, etc.), select a suitable loss function to evaluate the error between the predicted result and the true result. For example, in a binary classification task, you can use the cross entropy loss function; in a regression task, you can use loss functions such as mean square error (MSE) or mean absolute error (MAE).

[0241] Loss calculation: Using the selected loss function, calculate the error between the predicted result Y and the true result. This error value will be used for subsequent back propagation and weight update.

[0242] 3.6.4 Back propagation gradient calculation:

[0243] Chain rule: Use the chain rule to calculate the gradient of the loss function with respect to weights and biases. These gradients represent the rate of change of the loss function with respect to weights and biases, and are key to subsequent weight updates.

[0244] Gradient calculation: Based on the hidden layer output H and output layer output Y obtained during the forward propagation process, as well as the gradient of the loss function to the output layer (calculated by the loss function), the gradient of the loss function to the weights and bias of the hidden layer and input layer can be obtained by reverse calculation layer by layer.

[0245] 3.6.5 Update weights and biases:

[0246] Gradient descent optimization algorithm: Use gradient descent (or its variants such as Adam, RMSprop, etc.) optimization algorithm to update weights and biases. These algorithms adjust the values ​​of weights and biases according to the calculated gradient values ​​and a certain learning rate (LearningRate) to reduce the value of the loss function.

[0247] Weight update: According to the gradient descent algorithm, the weights and biases of the hidden layer and the output layer are updated. The updated weights and biases will be used for the next forward propagation calculation until the model performance no longer improves on the validation set (i.e., the condition for stopping training is met).

[0248] 3.6.6 Validation set monitoring and model stopping conditions:

[0249] Validation set monitoring: During the training process, the performance of the model is monitored using a validation set. The validation set is a dataset independent of the training set and is used to evaluate the performance of the model on unseen data.

[0250] Model stopping condition: When the performance of the model on the validation set no longer improves in multiple consecutive iterations, stop training to prevent overfitting. This usually means that the model has learned enough information, and continuing training may cause the model to overfit on the training set and reduce generalization ability.

[0251] 3.6.7 Output the optimal agent commission data results:

[0252] Final output: After training is completed, the model will output the optimal agent commission data results. These results are the predicted values ​​calculated based on the input feature data (i.e. 7 agent feature data) and the trained model.

[0253] Application of results: These prediction results can be used to guide agents in actual business scenarios such as remuneration settlement and commission policy formulation.

[0254] 3.7 Use the validation set to monitor the performance of the model. When the performance no longer improves in multiple consecutive iterations, stop training to prevent overfitting and finally output the optimal agent commission data results;

[0255] 4. The method of selecting the optimal indicator using the early stopping method in step 3.7 is as follows:

[0256] 4.1 For a training subset A, establish a user data feature set Q = q(O T ,M T ,F T ,B T ,C T ,S T ,L T), the main package ordered by the channel user O, the number of online months of the user M, the matching commission policy F, the cumulative settled remuneration multiples of the user B, the number of settlements within 12 months of the ID card information corresponding to the user C, whether the user attribute is converted S, the user's network access channel L, q is the weight of a certain feature group indicator;

[0257] 4.2 The original settlement data set with feature set is divided into two parts during the training process, namely training set n1, n1 = q(O T ,M T ,F T ,B T ,C T ,S T ,L T ) and test set n2, n2 = q(O T ,M T ,F T ,B T ,C T ,S T ,L T ), using U as the error function of the training algorithm, then Utr(n1) is the error on the training data, Ute(n2) is the error on the test set, and the validation set error is used to estimate it;

[0258] 4.3 Define a k cycle, the training algorithm is as follows:

[0259]

[0260] P is the probability that the training subset A meets the reward feature Q.

[0261] 4.4 Calculate the Gini coefficient of each feature of each split node, select the feature with the smallest Gini coefficient as the optimal feature indicator, and finally obtain the data of the agent's commission and return it to the settlement system.

[0262] Embodiment 1:

[0263] like Figure 7 As shown, the whole process of auditing illegal commissions using the method of the present invention is as follows:

[0264] 1. Collect source data from different systems and pre-process them;

[0265] 2. The system settles remuneration and outputs remuneration settlement details;

[0266] 3. Use algorithms to audit agents who violate regulations on commission arbitrage;

[0267] 4. The audit results will be automatically generated into work orders and sent to the work numbers of the responsible persons in each district, county and grid;

[0268] 5. The prefecture-level city shall manually check the audit work order, feedback the results, and conduct system tracing if the situation is true;

[0269] 6. Check and label the users and agents who engage in illegal commission arbitrage, and re-enter the manually verified and labeled data into the algorithm to retrain the model and optimize the parameters.

[0270] Table 1

[0271]

[0272] Table 1 above is the result of using the present invention to audit the illegal commissions of agents. It can be seen from the table that after using the method of the present invention, the proportion of illegal commissions in the province's total commissions is 1.2%.

[0273] Table 2

[0274]

[0275] Table 2 above shows the duration of auditing agents' illegal commission arbitrage using the present invention. It can be seen from the table that the average duration of auditing using the method of the present invention is 3.5 days.

[0276] Comparative Example 1:

[0277] Use the existing model to conduct audits on agents’ illegal commissions, such as Figure 8 As shown, the specific process is as follows:

[0278] 1. Collect source data from different systems and pre-process them;

[0279] 2. The system settles remuneration and outputs remuneration settlement details;

[0280] 3. The provincial company's business management center audits the remuneration settlement data according to fixed rules and distributes the audit results to the prefecture-level cities;

[0281] 4. The local and municipal administrators review the audit data;

[0282] 5. The city administrator will report the users who violated the rules to the provincial company supervisor, and the users who are found to have violated the rules will be manually adjusted;

[0283] Table 3

[0284]

[0285] Table 3 shows the proportion of illegal commissions audited using the existing method. It can be seen from the table that the current proportion of illegal commissions in the province is about 5.3%. The proportion of illegal commissions in the method of the present invention is about 1.2%, which is a decrease of 4.1%, saving commission costs of about 800,000 yuan, indicating that the present invention can improve the accuracy of audits and reduce the risk of illegal commissions by agents.

[0286] Table 4

[0287]

[0288] Table 4 shows the total audit time using the existing method. It can be seen from the table that the average total audit time is 7.5 days. The audit time of the method of the present invention is 3.5 days, saving 4 days and improving efficiency by nearly 57%, indicating that the present invention can improve audit efficiency and save labor costs.

[0289] In summary, by applying the technical solution of the present invention, the suspected illegal commission users developed by the agents in the current month were audited, and the proportion of illegal commissions in the total commission amount was reduced from 5.3% to 1.2%, the commission cost achieved value maximization, improved user quality, and encouraged agents to operate in good faith and compete fairly; compared with traditional auditing methods, the use of the multi-layer perceptron algorithm has improved the comprehensiveness and accuracy of the audit, greatly shortened the audit time in cities and counties, and reduced the average monthly audit time from 7.5 days to 3.5 days; it promotes agents to operate in good faith, develop high-quality users, reduce user churn, create revenue with quality, and help enterprises with digital transformation.

[0290] Compared with the prior art, the present invention has at least the following beneficial effects:

[0291] 1. In the process of auditing agents' illegal commissions, the identification of illegal behaviors often involves multiple variables and complex interactive relationships. The multi-layer perceptron algorithm can handle complex nonlinear relationships by introducing hidden layers and nonlinear activation functions, effectively capture nonlinear features such as users and policies, and improve the accuracy and efficiency of audits.

[0292] 2. The multi-layer perceptron algorithm used in the present invention can continuously learn new violation patterns and features. With the accumulation of settlement data and optimization in actual operation, the accuracy and efficiency of auditing will continue to improve. At the same time, auditing and model updating can be carried out in real time to adapt to new auditing needs and scenarios and expand to more complex auditing tasks.

[0293] 3. The present invention audits users who illegally use commissions as agents based on the multi-layer perceptron algorithm. Compared with traditional auditing methods, it improves the comprehensiveness and accuracy of auditing and greatly shortens the auditing time in cities and counties. It maximizes the value of commission costs, improves user quality, reduces user churn rate, creates revenue with quality, and helps companies with their digital transformation.

[0294] On the other hand, Fig. 9 As shown, an embodiment of the present invention provides a device 900 for auditing illegal commission arbitrage by agents, comprising:

[0295] The first module 901 is used to obtain a plurality of remuneration settlement data, extract preset data items from the remuneration settlement data as feature data; each remuneration settlement data is marked with marking data indicating illegal commission arbitrage;

[0296] The second module 902 is used to obtain a training set and a validation set based on the feature data and the marked data corresponding to all the remuneration settlement data;

[0297] The third module 903 is used to configure a multi-layer perceptron; based on the training set, the multi-layer perceptron is trained through forward propagation and back propagation to obtain a training model;

[0298] The fourth module 904 is used to perform performance verification on the training model based on the verification set to obtain a performance result; return to execute the step of training the multilayer perceptron based on the training set by forward propagation and back propagation in the third module until the performance result no longer improves in a continuous preset number of iterations, and the training model of the last iteration is used as the target model;

[0299] Among them, the target model is used to audit the remuneration settlement data of the audited agents to obtain the illegal commission arbitrage of the audited agents.

[0300] The contents of the method embodiments of the present invention are all applicable to the device embodiments. The functions specifically implemented by the device embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0301] On the other hand, an embodiment of the present invention further provides an electronic device, the electronic device includes a memory and a processor, the memory stores a computer program, and the processor implements the above sensitive information method when executing the computer program. The electronic device can be any intelligent terminal including a tablet computer, a car computer, etc.

[0302] It can be understood that the contents of the above method embodiments are all applicable to the present device embodiments, the functions specifically implemented by the present device embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0303] like Fig.10 As shown, Fig.10 A specific example of the hardware structure of an electronic device 1000 according to an embodiment is shown. The electronic device 1000 includes:

[0304] The processor 1001 may be implemented by a general-purpose CPU (Central Processing Unit), a microprocessor, an application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of the present invention.

[0305] The memory 1002 can be implemented in the form of a read-only memory (ROM), a static storage device, a dynamic storage device, or a random access memory (RAM). The memory 1002 can store an operating system and other application programs. When the technical solution provided in the embodiment of this specification is implemented by software or firmware, the relevant program code is stored in the memory 1002, and the processor 1001 calls and executes the network node population optimization method of the embodiment of the present invention;

[0306] Input / output interface 1003, used to implement information input and output;

[0307] The communication interface 1004 is used to realize the communication interaction between the device and other devices. The communication can be realized through a wired manner (such as USB, network cable, etc.) or a wireless manner (such as mobile network, WIFI, Bluetooth, etc.);

[0308] A bus 1005 , which transmits information between various components of the device (e.g., the processor 1001 , the memory 1002 , the input / output interface 1003 , and the communication interface 1004 );

[0309] The processor 1001 , the memory 1002 , the input / output interface 1003 and the communication interface 1004 are connected to each other in communication within the device via the bus 1005 .

[0310] The electronic device embodiments described above are merely illustrative, and the units described as separate components may or may not be physically separated, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0311] The contents of the method embodiments of the present invention are all applicable to the electronic device embodiments. The functions specifically implemented by the electronic device embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0312] Another aspect of an embodiment of the present invention further provides a computer-readable storage medium, wherein the storage medium stores a program, and the program is executed by a processor to implement the above method.

[0313] It should be noted that the computer-readable medium shown in the embodiment of the present invention may be a computer-readable signal medium or a computer-readable storage medium or any combination of the above two. The computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more 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), a flash memory, an optical fiber, a portable compact disk read-only memory (Compact Disc Read to Only Memory, CD to ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present invention, a computer-readable storage medium may be any tangible medium containing or storing a program, which may be used by or in combination with an instruction execution system, device or device. In the present invention, a computer-readable signal medium may include a data signal propagated in a baseband or as part of a carrier wave, which carries a computer-readable program code. Such propagated data signals may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. Computer readable signal media may also be any computer readable medium other than computer readable storage media, which may send, propagate, or transmit programs for use by or in conjunction with an instruction execution system, apparatus, or device. The program code contained on the computer readable medium may be transmitted using any suitable medium, including but not limited to: wireless, wired, etc., or any suitable combination of the above.

[0314] The contents of the method embodiments of the present invention are all applicable to the computer-readable storage medium embodiments. The functions specifically implemented by the computer-readable storage medium embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0315] The embodiment of the present invention also discloses a computer program product or a computer program, which includes a computer instruction stored in a computer-readable storage medium. A processor of a computer device can read the computer instruction from the computer-readable storage medium, and the processor executes the computer instruction, so that the computer device executes the above method.

[0316] The flow chart and block diagram in the accompanying drawings illustrate the possible architecture, function and operation of the system, method and computer program product according to various embodiments of the present invention. In this regard, each box in the flow chart or block diagram can represent a module, a program segment, or a part of a code, and the above-mentioned module, program segment, or a part of a code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram or flow chart, and the combination of the boxes in the block diagram or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0317] It should be noted that, although several modules of the device for action execution are mentioned in the above detailed description, this division is not mandatory. In fact, according to an embodiment of the present invention, the features and functions of two or more modules or units described above can be embodied in one module or unit. On the contrary, the features and functions of one module or unit described above can be further divided into being embodied by multiple modules or units.

[0318] Through the description of the above implementation, it is easy for those skilled in the art to understand that the example implementation described here can be implemented by software, or by software combined with necessary hardware. Therefore, the technical solution according to the implementation of the present invention can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD to ROM, a USB flash drive, a mobile hard disk, etc.) or on a network, including several instructions to enable a computing device (which can be a personal computer, a server, a touch terminal, or a network device, etc.) to execute the method according to the implementation of the present invention.

[0319] In some selectable embodiments, the function / operation mentioned in the block diagram may not occur in the order mentioned in the operation diagram. For example, depending on the function / operation involved, the two boxes shown in succession can actually be executed substantially simultaneously or the boxes can sometimes be executed in reverse order. In addition, the embodiment presented and described in the flow chart of the present invention is provided by way of example, for the purpose of providing a more comprehensive understanding of technology. The disclosed method is not limited to the operation and logic flow presented herein. Selectable embodiments are expected, wherein the order of various operations is changed and the sub-operation of a part for which is described as a larger operation is performed independently.

[0320] In addition, although the present invention is described in the context of functional modules, it should be understood that, unless otherwise specified, one or more of the functions and / or features can be integrated into a single physical device and / or software module, or one or more functions and / or features can be implemented in a separate physical device or software module. It is also understood that a detailed discussion of the actual implementation of each module is unnecessary for understanding the present invention. More specifically, in view of the properties, functions and internal relationships of the various functional modules in the device disclosed herein, the actual implementation of the module will be understood within the conventional skills of the engineer. Therefore, those skilled in the art can implement the present invention set forth in the claims without excessive experimentation using ordinary techniques. It is also understood that the specific concepts disclosed are merely illustrative and are not intended to limit the scope of the present invention, which is determined by the full scope of the appended claims and their equivalents.

[0321] If the function is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the methods of each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc. Various media that can store program codes.

[0322] The logic and / or steps represented in the flowchart or otherwise described herein, for example, may be considered as an ordered list of executable instructions for implementing logical functions, and may be embodied in any computer-readable medium for use by an instruction execution device, apparatus or device (such as a computer-based device, a device including a processor, or other device that can fetch instructions from an instruction execution device, apparatus or device and execute the instructions), or in conjunction with such instruction execution device, apparatus or device. For purposes of this specification, "computer-readable medium" may be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution device, apparatus or device, or in conjunction with such instruction execution device, apparatus or device.

[0323] More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection with one or more wires (electronic device), a portable computer disk case (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), a fiber optic device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be a paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, deciphering or, if necessary, processing in another suitable manner, and then stored in a computer memory.

[0324] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, a plurality of steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution device. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0325] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "examples", "specific examples", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner.

[0326] Although the embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the claims and their equivalents.

[0327] The above is a specific description of the preferred implementation of the present invention, but the present invention is not limited to the embodiments. Those skilled in the art can make various equivalent modifications or substitutions without violating the spirit of the present invention. These equivalent modifications or substitutions are all included in the scope defined by the claims of the present invention.

Claims

1. A method for auditing agents' illegal commission arbitrage, characterized in that: The following steps are involved: Acquire multiple remuneration settlement data, extract preset data items from the remuneration settlement data as feature data; each remuneration settlement data is marked with marking data indicating illegal commission arbitrage; Based on the feature data and the marked data corresponding to all the remuneration settlement data, a training set and a validation set are obtained; Configure a multi-layer perceptron; Based on the training set, the multilayer perceptron is trained by forward propagation and back propagation to obtain a training model; Performing performance verification on the training model based on the verification set to obtain a performance result; Returning to the step of training the multilayer perceptron based on the training set by forward propagation and back propagation until the performance result no longer improves in a continuous preset number of iterations, and using the training model of the last iteration as the target model; The target model is used to audit the remuneration settlement data of the audited agent to obtain the illegal commission situation of the audited agent.

2. The method for auditing illegal commissions of agents according to claim 1, characterized in that: Before the step of obtaining a plurality of remuneration settlement data, the method further comprises the following steps: Preprocessing the peripheral data based on the preset remuneration settlement logic, and completing the remuneration settlement based on the preset remuneration policy to obtain the remuneration settlement data; In response to the audit result, the illegal commission arbitrage situation of each remuneration settlement data is obtained as the marking data.

3. The method for auditing illegal commissions of agents according to claim 1, characterized in that: The step of extracting preset data items from the remuneration settlement data as feature data comprises the following steps: Based on the preset commission characteristic data items, the characteristic data is obtained from the remuneration settlement data; Among them, the characteristic data includes multiple agent characteristic data, including the main package ordered by the user, the number of months the user has been online, the matching commission policy, the user's cumulative settled remuneration multiples, the number of settlements within 12 months of the user's corresponding ID card information, whether the user attributes have been converted and the user's network access channel.

4. The method for auditing illegal commissions of agents according to claim 1, characterized in that: The step of arranging the feature data and the marked data corresponding to all the remuneration settlement data to obtain a training set and a validation set comprises the following steps: The sample data is obtained based on the feature data and the mark data corresponding to the same remuneration settlement data; A training set and a validation set are obtained by sorting out the sample data corresponding to all the remuneration settlement data.

5. The method for auditing illegal commissions of agents according to claim 1, characterized in that: The configuration of the multi-layer perceptron includes the following steps: The multilayer perceptron is obtained based on the input layer, the hidden layer and the output layer configuration; Initialize the weights, biases, and activation functions of the hidden layer and the output layer.

6. The method for auditing illegal commissions of agents according to claim 1, characterized in that: The training set includes the mutually related feature data and the label data corresponding to the plurality of remuneration settlement data; the multilayer perceptron includes an input layer, a hidden layer and an output layer; and the multilayer perceptron is trained based on the training set by forward propagation and back propagation to obtain a training model, including the following steps: Inputting the feature data in the training set as input data into the multilayer perceptron; Based on the input data, a prediction result is obtained by forward propagation processing in the multi-layer perceptron; Based on the prediction result and the labeled data in the training set, the error result is obtained by using the chain rule; The network parameters of the hidden layer and the output layer are updated through back propagation according to the error result to obtain the training model.

7. The method for checking illegal commissions of agents according to claim 6, characterized in that: The input data is received through the input layer; and based on the input data, a prediction result is obtained through forward propagation processing in the multi-layer perceptron, comprising the following steps: Based on the input data, obtaining a hidden layer output by processing the first network parameter and the first activation function of the hidden layer; Based on the hidden layer output, obtaining output data through the second network parameter and the second activation function of the output layer; The first network parameter and the second network parameter both include weights and biases.

8. A device for auditing agents' illegal commission arbitrage, characterized in that: include: The first module is used to obtain a plurality of remuneration settlement data and extract preset data items from the remuneration settlement data as feature data; Each of the remuneration settlement data is marked with marking data indicating illegal commission arbitrage; The second module is used to obtain a training set and a validation set based on the feature data and the marked data corresponding to all the remuneration settlement data; The third module is used to configure the multi-layer perceptron; Based on the training set, the multilayer perceptron is trained by forward propagation and back propagation to obtain a training model; The fourth module is used to perform performance verification on the training model based on the verification set to obtain performance results; Return to executing the step of training the multilayer perceptron based on the training set by forward propagation and back propagation in the third module until the performance result no longer improves in a continuous preset number of iterations, and use the training model of the last iteration as the target model; The target model is used to audit the remuneration settlement data of the audited agent to obtain the illegal commission situation of the audited agent.

9. An electronic device, characterized in that: including a processor and a memory; The memory is used to store programs; The processor executes the program to implement the method according to any one of claims 1 to 7.

10. A computer storage medium storing a program executable by a processor, characterized in that: The program executable by the processor is used to implement the method according to any one of claims 1 to 7 when executed by the processor.

Images