Account receivable management method and system based on digital system

Abstract

The invention provides an account receivable management method and system based on a digital system, electronic equipment and a storage medium, and aims to solve the problems that account receivable data is unclear, information is unknown, early warning is not timely, and general service and ICT (information and communication technology) under-clearance targets are uncontrollable. Performing data processing on the client data; performing customer importance level classification on the processed customer data through a neural network model, and classifying customers into types with different importance levels by using a clustering optimization importance level classification result; and according to the type of the customer, a strategy gradient algorithm is used, and an optimal arrearage payment reminding strategy is matched according to historical user arrearage records and payment reminding behaviors. The method can standardize the collection process, reduces the risk, increases the management means of account receivable of government and enterprise, and promotes the pressure drop of account receivable.

Description

Technical Field

[0001] The present disclosure relates to the field of artificial intelligence technology, and in particular to an accounts receivable management method based on a digital system, an accounts receivable management system based on a digital system, an electronic device, and a computer-readable storage medium. Background Art

[0002] In today's social context of economic development, communication operators are facing increasingly fierce market competition. It is particularly important to continuously improve the internal management system and continuously improve the management level of enterprises. Accounts receivable management is an important part of enterprise management, which helps to maintain the stability of corporate cash flow and improve the efficiency of capital utilization. However, in the actual operation and management of communication operators, there are many problems in the management of accounts receivable. Although the accounts receivable of communication service companies account for a relatively low proportion of revenue, due to their large revenue base, the absolute amount is large. In addition, due to the large number of government and enterprise customers of communication operators, the complex industries involved, and the diverse payment channels, it is difficult to effectively control the problem of users' arrears. At the same time, the ARPU (Average Revenue Per User) value of a single user is not high. When considering the recovery plan, it is necessary to balance the cost and benefit. As a result, it is difficult for communication operators to recover accounts receivable, and the management of accounts receivable has also become one of the management difficulties of major communication operators.

[0003] At present, the accounts receivable of telecommunications operators are facing a severe situation. The effectiveness of reducing accounts receivable has become the main determinant affecting the completion of the company's annual revenue budget target. At this stage, operators lack effective management and control measures for accounts receivable of government and enterprise customers. The usual method is for account managers to manually record and urge them to handle the accounts receivable, which cannot ensure that the accounts receivable are recovered on time and in quantity. In the process of collecting accounts receivable from government and enterprise customers of telecommunications operators, there are often problems such as unclear collection requirements, untimely and irregular collection, which leads to the inability to recover large amounts of debts. There are three key problems in the management process: "unclear data", that is, information is scattered everywhere, account managers do not understand the arrears of their customers, cannot fully grasp the detailed information such as customers, amounts, payment periods, etc. related to the arrears, and cannot carry out timely debt collection; "untimely" warnings, that is, there is a lack of arrears warnings and abnormal customer information related to government and enterprise accounts receivable, and the warnings cannot be issued to the corresponding account managers in a timely and accurate manner, resulting in delays in debt collection; "uncontrollable" processes, that is, there is no top-down task supervision means and statistical analysis capabilities, and there is no systematic debt collection work process, making the debt collection goals of Tongfu and ICT (information and communications technology) uncontrollable. Summary of the invention

[0004] In order to at least solve the problems of unclear accounts receivable data, unclear information, untimely warning, and uncontrollable communication service and ICT debt collection targets in the prior art. The present disclosure provides an accounts receivable management method based on a digital system, an accounts receivable management system based on a digital system, an electronic device, and a computer-readable storage medium. Based on the digital system support, the accounts receivable customers of government and enterprise customers of communication operators are graded and classified, and collection strategies are formulated according to different customer levels, which can further standardize the collection process, reduce risks, increase the means of government and enterprise accounts receivable management, and promote the reduction of accounts receivable.

[0005] In a first aspect, the present disclosure provides a method for managing accounts receivable based on a digital system, the method comprising:

[0006] Obtain data on customers who are in arrears;

[0007] Carry out data processing on customer data;

[0008] The processed customer data is classified into customer importance levels through a neural network model, and clustering is used to optimize the importance level classification results to divide customers into types with different importance levels;

[0009] According to the type of customer, the policy gradient algorithm is used to match the best overdue payment collection strategy based on the historical user's overdue payment records and collection behavior.

[0010] Furthermore, the obtaining of the customer data of the arrears includes:

[0011] A list of customers in arrears is obtained through the enterprise ledger to form a data table of customers in arrears, wherein key fields of the data table of customers in arrears include customer name, customer address, place of origin, corresponding account manager, business scale and amount of arrears.

[0012] Furthermore, the processing of customer data includes:

[0013] Process missing data values. If the missing rate of a variable is greater than the first threshold, delete the variable directly. If the missing rate is lower than the second threshold, fill it with the mode or median. If the missing rate of a variable is between the first and second thresholds, retain the original variable, introduce a derived variable, assign the missing field a value of 0 directly, and fill in the remaining values ​​as they are.

[0014] Check whether the data is duplicated and delete the duplicate data directly;

[0015] Check whether the data has outliers and treat them as missing values;

[0016] Convert non-numeric fields to numeric values;

[0017] Min-Max normalization is used to scale all data to the interval [0, 1]. The Min-Max representation is:

[0018]

[0019] Among them, x′ is the normalized data value, and x is the original data value.

[0020] Furthermore, the processing of the customer data by using a neural network model to classify the customer importance level includes:

[0021] The neural network model uses the TextCNN model, and a tested model is obtained after passing the training set training and the test set testing;

[0022] The processed customer data is converted into a multi-dimensional vector and input into the tested model. The model outputs the probabilities of multiple categories and classifies the customers into the category with the highest probability.

[0023] Furthermore, the use of clustering to optimize the importance level classification results includes:

[0024] For each category after model classification, the k-means clustering algorithm is used to optimize the important level classification results, including:

[0025] After completing the cluster center selection, for the newly added customer information, calculate its Euclidean distance with each cluster center, classify it into the cluster with the nearest center, and obtain the classification result. Determine the final importance level classification result of the user based on the model training result and the classification result, and add the customer level field to the customer data.

[0026] Furthermore, the method further comprises:

[0027] After obtaining the type of customer importance level, accounts receivable warning information is pushed to sales units and customer managers according to the corresponding account managers and customer importance levels, in accordance with business orientation and hierarchical classification warning rules.

[0028] Furthermore, the method of using a policy gradient algorithm according to the type of customer to match the best overdue payment collection strategy according to the historical user overdue payment records and collection behaviors includes:

[0029] For each type of customer, a policy gradient algorithm is used to determine the overdue payment collection strategy based on the historical user's overdue payment records and collection behavior, including:

[0030] Determine user status: the duration of arrears; determine actions: different collection actions;

[0031] Define strategy: Use convolutional neural network to approximate the strategy function, with the input being the current user status and the output being the probability of debt recovery for each action;

[0032] Randomly initialize the weights and biases of the neural network as the parameters of the strategy, let the user run it multiple times in different states, choose actions based on the current strategy, and record the state, action, and reward of each run;

[0033] Use the recorded data to calculate the policy gradient, use Monte Carlo methods to calculate the long-term return of each action in a given state, and use these returns to update the policy parameters;

[0034] Repeat the above steps until the strategy converges or reaches the preset number of iterations, matching the best collection behavior for different arrears durations.

[0035] In a second aspect, the present disclosure provides an accounts receivable management system based on a digital system, the system comprising:

[0036] An acquisition module configured to acquire data of customers in arrears;

[0037] A data processing module configured to perform data processing on customer data;

[0038] A classification module is configured to classify the processed customer data into customer importance levels through a neural network model, and use clustering to optimize the importance level classification results to classify customers into types with different importance levels;

[0039] The collection processing module is configured to use a policy gradient algorithm according to the type of customer and match the best collection strategy based on historical user arrears records and collection behaviors.

[0040] In a third aspect, the present disclosure provides an electronic device comprising a memory and a processor, wherein the memory stores a computer program, and when the processor runs the computer program stored in the memory, the processor executes the accounts receivable management method based on a digital system as described in any one of the first aspects.

[0041] In a fourth aspect, the present disclosure provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the digital system-based accounts receivable management method described in any one of the first aspects above is implemented.

[0042] Beneficial effects:

[0043] The accounts receivable management method based on a digital system, the accounts receivable management system based on a digital system, the electronic device and the storage medium provided by the present invention are aimed at the in-service process management of the customers in arrears of communication operators. Based on the support of a digital system, importance is graded according to the overall customer portrait, and different types of arrears collection plans are formulated based on a comprehensive assessment of the importance levels of existing customers and the amount of arrears, so as to improve the ability to control the accounts receivable process and improve the level of clearing arrears. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 A flowchart of an accounts receivable management method based on a digital system provided in the first embodiment of the present disclosure;

[0045] Figure 2 A flowchart of a method for managing and controlling accounts receivable based on a digital system provided in the second embodiment of the present disclosure;

[0046] Figure 3 A schematic diagram of the structure of an accounts receivable management and control system based on a digital system provided in the second embodiment of the present disclosure;

[0047] Figure 4 This is an architecture diagram of an accounts receivable management and control system based on a digital system provided in Embodiment 3 of the present disclosure;

[0048] Figure 5 This is an architecture diagram of an electronic device provided in Embodiment 4 of the present disclosure. DETAILED DESCRIPTION

[0049] In order to enable those skilled in the art to better understand the technical solution of the present disclosure, the present disclosure is further described in detail below in conjunction with the drawings and embodiments. It should be understood that the specific embodiments and drawings described herein are only used to explain the present invention, rather than to limit the present invention.

[0050] It should be noted that the terms "first", "second", etc. in the specification and claims of the present disclosure and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence; and, in the absence of conflict, the embodiments in the present disclosure and the features in the embodiments can be arbitrarily combined with each other.

[0051] The terms used in the embodiments of the present disclosure are only for the purpose of describing specific embodiments and are not intended to limit the present disclosure. The singular forms of "a", "said" and "the" used in the embodiments of the present disclosure and the appended claims are also intended to include plural forms unless the context clearly indicates other meanings.

[0052] In the subsequent description, the suffixes such as "module", "component" or "unit" used to represent elements are only used to facilitate the description of the present disclosure, and have no specific meanings. Therefore, "module", "component" or "unit" can be used in a mixed manner.

[0053] The technical solution of the present invention and how the technical solution of the present invention solves the above-mentioned problems existing in the prior art are described in detail below with specific embodiments. It is understandable that in the embodiments of the present application, the execution subject can perform some or all of the steps in the embodiments of the present application, and these steps or operations are only examples. The embodiments of the present application can also perform other operations or deformations of various operations. In addition, each step can be performed in a different order presented in the embodiments of the present application, and it is possible not to perform all the operations in the embodiments of the present application. And, the following several specific embodiments can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments.

[0054] Figure 1 A flowchart of a method for managing accounts receivable based on a digital system is provided in the first embodiment of the present disclosure, such as Figure 1 As shown, the method includes:

[0055] Step S101: Acquire the customer data of the customer in arrears;

[0056] Step S102: Processing customer data;

[0057] Step S103: classifying the processed customer data by customer importance level through a neural network model, and using clustering to optimize the importance level classification results to classify customers into types with different importance levels;

[0058] Step S104: According to the type of customer, use the policy gradient algorithm to match the best overdue payment collection strategy based on the historical user's overdue payment records and collection behavior.

[0059] In view of the pain points of unclear data, untimely warning and uncontrollable targets in the accounts receivable of the communication service + ICT project for government and enterprise customers, the present disclosed embodiment obtains the customer data of arrears through the enterprise ledger, including customer information data and related arrears data, such as customer name, customer address, place of origin, customer manager, business scale, arrears amount, arrears duration, historical arrears records and other specific information, and forms machine-readable and processable numerical data by processing the data, which is convenient for the model to classify customers, and trains and tests the neural network model through part of the customer data and part of the actual data that have been marked, and classifies the processed customer data by customer importance through the trained model, and uses clustering to optimize the importance level classification results for more accurate positioning, and divides customers into types with different importance levels. For customers of different levels, the policy gradient algorithm is used to match the best arrears collection strategy according to historical user arrears records and collection behaviors. For example, by regularly pushing information, collecting by phone, and making actual visits, we provide customers with different repayment plans, corresponding preferential activities, and necessary punishment measures, such as suspending services, collecting arrears through legal channels, etc.; thus, we handle different arrears situations in different ways, achieve arrears early warning, collection and supervision process control, and reduce arrears for government and enterprise customers. This will standardize the collection process, reduce risks, increase government and enterprise accounts receivable management methods, and promote the reduction of accounts receivable.

[0060] Furthermore, the obtaining of the customer data of the arrears includes:

[0061] A list of customers in arrears is obtained through the enterprise ledger to form a data table of customers in arrears, wherein key fields of the data table of customers in arrears include customer name, customer address, place of origin, corresponding account manager, business scale and amount of arrears.

[0062] After obtaining the customer's data, the data table of customers in arrears is formed through various key fields, as shown in Table 1 below

[0063] Table 1: Data information of customers in arrears

[0064] Customer Name Customer Address Place of Origin account Manager Business scale Amount of debt

[0065] By forming a data table of the list of customers in arrears, the status of customers in arrears corresponding to each account manager can be obtained, and scattered information can be summarized so that the account manager can understand the arrears of his own customers and fully grasp the detailed information such as customers, amounts, and billing periods related to arrears. It is also convenient to search for data.

[0066] Furthermore, the processing of customer data includes:

[0067] Process missing data values. If the missing rate of a variable is greater than the first threshold, delete the variable directly. If the missing rate is lower than the second threshold, fill it with the mode or median. If the missing rate of a variable is between the first and second thresholds, retain the original variable, introduce a derived variable, assign the missing field a value of 0 directly, and fill in the remaining values ​​as they are.

[0068] Check whether the data is duplicated and delete the duplicate data directly;

[0069] Check whether the data has outliers and treat them as missing values;

[0070] Convert non-numeric fields to numeric values;

[0071] Min-Max normalization is used to scale all data to the interval [0, 1]. The Min-Max representation is:

[0072]

[0073] Among them, x′ is the normalized data value, and x is the original data value.

[0074] During data processing, the first threshold can be set to any value between 85% and 95%, preferably 90%, and the second threshold can be set to 5% to 15%, preferably 10%. When converting non-numeric fields to numeric values, for binary variables, if the value is "yes", it is changed to "1", and if the value is "no", it is changed to "0". For ordered multi-classification variables, the values ​​are assigned to "1", "2", ..., "n" according to the order. For multi-category unordered categorical variables, they are split into multiple binary variables, and then processed according to the processing method of binary variables.

[0075] Furthermore, the processing of the customer data by using a neural network model to classify the customer importance level includes:

[0076] The neural network model uses the TextCNN model, and a tested model is obtained after passing the training set training and the test set testing;

[0077] The processed customer data is converted into a multi-dimensional vector and input into the tested model. The model outputs the probabilities of multiple categories and classifies the customers into the category with the highest probability.

[0078] The TextCNN model is used to classify customer importance levels. The specific method is as follows: there are m customer data, each of which is converted into an n-dimensional vector, and the m×n matrix is ​​input into the model embedding layer; the embedding layer is an n×128-dimensional matrix, which embeds the m×n-dimensional sparse matrix into a new real vector space to become an m×128-dimensional matrix, and the initial parameters of the matrix are determined by random generation; two convolution layers are constructed, the number of sequential convolution kernels is 128 and 98 respectively, the size of the sequential convolution kernel is 4, and the activation function is the hyperbolic tangent function; two convolution layers are constructed, the number of sequential convolution kernels is 128 and 98 respectively, the size of the sequential convolution kernel is 4, and the activation function is the hyperbolic tangent function; two Pooling layer, the two pooling layers select average pooling and square root pooling respectively; input the results of the two pooling layers into the fully connected layer, the activation function is SoftMax, the loss function selects cross entropy, and the optimizer selects adaptive gradient descent, outputs the probabilities of the four categories A, B, C, and D, and classifies the customer into the category with the highest probability; uses accuracy to measure classification accuracy; after using the training set to build the model, use the test set data to test, and output the average loss value and accuracy; after a certain number of iterations, determine the parameters of the embedding layer and convolution layer based on the performance on the test set. All pre-prepared customer information is randomly divided into training and test sets in a ratio of 8:2. After obtaining the trained model, the processed customer data can be converted into a multi-dimensional vector and input into the tested model. The model outputs the probabilities of multiple categories and classifies the customer into the category with the highest probability.

[0079] Furthermore, the use of clustering to optimize the importance level classification results includes:

[0080] For each category after model classification, the k-means clustering algorithm is used to optimize the important level classification results, including:

[0081] After completing the cluster center selection, for the newly added customer information, calculate its Euclidean distance with each cluster center, classify it into the cluster with the nearest center, and obtain the classification result. Determine the final importance level classification result of the user based on the model training result and the classification result, and add the customer level field to the customer data.

[0082] Use k-means clustering to optimize the hierarchical classification results. First, determine the number of cluster centers. The specific steps are: randomly select n (n = 3, 4, 5, 6, 7, 8, 9) samples as the initial cluster centers; calculate the Euclidean distance from each sample in the data set to the n cluster centers, and assign it to the cluster corresponding to the cluster center with the smallest distance; for each cluster, recalculate its cluster center position; repeat the above operation of changing the cluster center to calculate the distance until a certain termination condition is reached (the number of iterations reaches the threshold or the cluster center position remains unchanged); calculate the overall clustering silhouette coefficient; change the value of n, calculate the clustering silhouette coefficient under different cluster numbers, select the n value corresponding to the maximum silhouette coefficient, and use its clustering result as the final clustering result. After clustering, for a newly added customer information, after model training, calculate its Euclidean distance from each cluster center and classify it into the cluster with the closest center.

[0083] In one implementation, clustering is performed after classification, and the model recognition results can be accurately verified and refined more accurately. For example, after the clustering results, customers of category A may be clustered into the central cluster of category B, and the data of these customers need to be analyzed to finally determine the classification results of the customers.

[0084] In another embodiment, after the model is classified, the k-means clustering algorithm can be optimized for each category of the classification, and the corresponding center can be determined in each category. The new data can be classified by calculating the distance from each center in the category. The accuracy of clustering can be improved by the method of classification before clustering: by first classifying, the data set can be divided into more specific categories, and then these categories can be clustered, which can improve the accuracy and efficiency of clustering. Better understanding of data: through classification, the structure and pattern of the data can be better understood, which helps to discover the potential relationships and laws in the data. Improve the quality and efficiency of clustering.

[0085] After the model training obtains 4 categories of results, the importance level classification results are optimized through the clustering algorithm. According to the customer importance level classification results, the level of customers with outstanding payments is marked and divided into multiple categories (with decreasing importance priority). After the customer classification is completed, the "Customer Level" field is added to the customer database in Table 1 to form the customer classification data table information as the basic data.

[0086] Furthermore, the method further comprises:

[0087] After obtaining the type of customer importance level, accounts receivable warning information is pushed to sales units and customer managers according to the corresponding account managers and customer importance levels, in accordance with business orientation and hierarchical classification warning rules.

[0088] The acquired customer classification data will be sorted out, and the basic data of key customers will be distinguished between government and enterprise channel service accounts receivable and emerging ICT project accounts receivable. The data levels are divided into five levels: province, city, district and county, grid, and customer manager level, and listed and displayed in multiple dimensions including balance, percentage of revenue, and recovery amount.

[0089] After the customer classification is completed, the early warning phase begins. The formulation of early warning rules follows the principle of "business-oriented, classification and classification". Early warning information is regularly pushed to each sales unit and account manager through office platforms, text messages, etc. for key businesses and key scenarios every month to advance the collection action and reduce the risk of arrears.

[0090] Furthermore, the method of using a policy gradient algorithm according to the type of customer to match the best overdue payment collection strategy according to the historical user overdue payment records and collection behaviors includes:

[0091] For each type of customer, a policy gradient algorithm is used to determine the overdue payment collection strategy based on the historical user's overdue payment records and collection behavior, including:

[0092] Determine user status: the duration of arrears; determine actions: different collection actions;

[0093] Define strategy: Use convolutional neural network to approximate the strategy function, with the input being the current user status and the output being the probability of debt recovery for each action;

[0094] Randomly initialize the weights and biases of the neural network as the parameters of the strategy, let the user run it multiple times in different states, choose actions based on the current strategy, and record the state, action, and reward of each run;

[0095] Use the recorded data to calculate the policy gradient, use Monte Carlo methods to calculate the long-term return of each action in a given state, and use these returns to update the policy parameters;

[0096] Repeat the above steps until the strategy converges or reaches the preset number of iterations, matching the best collection behavior for different arrears durations.

[0097] After the warning is pushed, the collection execution phase begins. In order to normalize the collection operation, the account manager uses the received arrears data information to conduct closed-loop control of arrears data viewing, collection record filling, and collection data statistics. At the same time, by setting the time configuration for the unread reminder of the pushed warning, the warning will be pushed again after the time is exceeded. In order to improve the collection effect, the policy gradient algorithm (Policy Gradient Methods) is used to match the collection behavior of different customer levels.

[0098] The policy gradient algorithm is used to determine the overdue payment collection strategy based on the historical user's overdue payment records and collection behaviors. The specific steps are as follows: determine the state: the duration of overdue payment; determine the action: different collection behaviors; define the strategy: use a convolutional neural network (CNN) to approximate the policy function, the input is the current user state (duration of overdue payment), and the output is the probability of debt recovery for each action; randomly initialize the weights and biases of the neural network as the parameters of the strategy, let the user run multiple times in different states, select actions according to the current strategy, and record the state, action and reward of each run (the reward is defined as 1 when the debt recovery is completed, and 0 in other cases); use the collected data to calculate the policy gradient, use the Monte Carlo method to calculate the long-term return of each action in a given state, and use these returns to update the policy parameters; repeat the above steps until the strategy converges or reaches the preset number of iterations. During the iteration process, the performance of debt collection will gradually improve, and the best collection behavior can be matched for different overdue payment durations; according to the collection strategy, the account manager can obtain pending warnings and overdue payment details to formulate a collection strategy.

[0099] The disclosed embodiment is based on digital system support, and grades and classifies the accounts receivable customers of the communication operator's government and enterprise customers, and formulates collection strategies according to different customer levels. This can further standardize the collection process, reduce risks, increase the means of managing government and enterprise accounts receivable, and promote the reduction of accounts receivable.

[0100] By analyzing the acquired data of customers in arrears, and comprehensively using data empowerment, tool empowerment, and application scenario empowerment, the automation of daily collection, early warning, supervision, statistical analysis, and other tasks of government and enterprise customer communication service + lCT income can be realized, making debt collection more timely. In addition, it can also achieve: visualization, based on the headquarters communication service and ICT data, processing and processing, timely display of arrears information, customer information and other data for processing and display; early warning, early warning and collection of arrears information and abnormal customers, and related to-do and SMS reminders; tracking, feedback of arrears collection and supervision information, and leaving operation tracks and logs to form data analysis and provide management decisions; supervision, forming corresponding supervision tasks for arrears information and abnormal customer information, and penetrating and supervising from provinces to cities, from cities to districts and counties to business services to improve the efforts of debt collection.

[0101] like Figure 2 As shown, the second embodiment of the present disclosure also provides an accounts receivable management method based on a digital system, which includes seven steps, namely, obtaining a list of customers in arrears, grading the importance of customers in arrears, generating early warning information, pushing early warning information, generating collection tasks, executing collection tasks, and assessment and evaluation.

[0102] Through Figure 3The system implementation shown is used to assist and support each step in data processing and retention, including a collection processing module, a pending warning module and a data collection module. The data collection module is used to automatically obtain the required original data from each data source on time, and classify and summarize it, and process it. The rules are preset by the business. The module contains eight aspects in three categories: arrears list, arrears details, a pending list, a supervision list, a project list, customer details, and warning data; the pending warning module is mainly used for collection, supervision and warning information push; the collection processing module is used for the operation and processing of daily tasks for actual users and management personnel, including collection task execution, recording, supervision processing, communication service warning setting, ICT warning setting, and custom configuration.

[0103] The specific functions of each module are as follows:

[0104] 1. Data acquisition module

[0105] 1. Data Collection

[0106] The list of customers in arrears is obtained through the enterprise ledger, which includes five key fields: customer name, customer address, location, account manager, business scale, and amount of arrears, to form a data table of customers in arrears, as shown in Table 1 below.

[0107] Table 1: Data information of customers in arrears

[0108] Customer Name Customer Address Place of Origin account Manager Business scale Amount of debt

[0109] 2. Data processing

[0110] (1) Check whether the data has missing values. If the missing rate of the variable is high (greater than 90%), delete the variable directly; if the missing rate is low (less than 10%), fill it with the mode or median; if the missing rate of the variable is between 10% and 90%, retain the original variable and introduce a derived variable, that is, directly assign the missing field to 0 and fill the remaining values ​​as they are.

[0111] (2) Check whether there are duplicate values ​​in the data and delete them directly.

[0112] (3) Check whether there are outliers in the data and treat them as missing values.

[0113] (4) Convert non-numeric fields to numeric values. For binary variables, if the value is "yes", change it to "1", and if the value is "no", change it to "0". For ordered multi-category variables, assign the values ​​to "1", "2", ..., "n" according to the order. For multi-category unordered categorical variables, split them into multiple binary variables and then process them according to the processing method of binary variables.

[0114] (5) Min-Max normalization is used to scale all data to the interval [0,1]. The Min-Max representation method is:

[0115]

[0116] Among them, x′ is the normalized data value, and x is the original data value.

[0117] 2. To-do warning module

[0118] 1. Classification

[0119] First, the TextCNN model is used to classify customer importance levels. The specific method is as follows: there are m customer data, each of which is converted into an n-dimensional vector, and the m×n matrix is ​​input into the model embedding layer; the embedding layer is an n×128-dimensional matrix, which embeds the m×n-dimensional sparse matrix into a new real vector space to become an m×128-dimensional matrix, and the initial parameters of the matrix are determined by random generation; two convolutional layers are constructed, the number of sequential convolution kernels is 128 and 98 respectively, the size of the sequential convolution kernel is 4, and the activation function is the hyperbolic tangent function; 2 The two pooling layers are average pooling and square root pooling respectively; the results of the two pooling layers are input into the fully connected layer, the activation function is SoftMax, the loss function is cross entropy, the optimizer is adaptive gradient descent, and the probabilities of the four categories A, B, C, and D are output, and the customer is classified into the category with the highest probability; the accuracy is used to measure the classification accuracy; after using the training set to build the model, the test set data is used for testing, and the average loss value and accuracy are output; after a certain number of iterations, the parameters of the embedding layer and the convolution layer are determined according to the performance on the test set. All sample customer information is randomly divided into training set and test set in a ratio of 8:2.

[0120] Subsequently, k-means clustering is used to optimize the hierarchical classification results. The specific steps are as follows: randomly select n (n = 3, 4, 5, 6, 7, 8, 9) samples as the initial cluster centers; for each sample in the data set, calculate its Euclidean distance to the n cluster centers, and assign it to the cluster corresponding to the cluster center with the smallest distance; for each cluster, recalculate its cluster center position; repeat the above operation of changing the cluster center to calculate the distance until a certain termination condition is reached (the number of iterations reaches the threshold or the cluster center position remains unchanged); calculate the overall clustering silhouette coefficient; change the value of n, calculate the clustering silhouette coefficient under different cluster numbers, select the n value corresponding to the maximum silhouette coefficient, and use its clustering result as the final clustering result. After clustering is completed, for a newly added customer information after training, calculate its Euclidean distance to each cluster center and classify it into the cluster with the closest center.

[0121] According to the classification results of customer importance, the levels of customers with outstanding payments are classified into A, B, C, and D (with decreasing importance priority). After the customer classification is completed, the "Customer Level" field is added to the customer database in Table 1 to form the customer classification data table information as the basic data.

[0122] 2. Customer data classification

[0123] The acquired customer classification data will be sorted out, and the basic data of key customers will be distinguished between government and enterprise channel service accounts receivable and emerging ICT project accounts receivable. The data levels are divided into five levels: province, city, district and county, grid, and customer manager level, and listed and displayed in multiple dimensions including balance, percentage of revenue, and recovery amount.

[0124] 3. Customer accounts receivable warning

[0125] After the customer classification is completed, the early warning phase begins. The formulation of early warning rules follows the principle of "business-oriented, classification and classification". Early warning information is regularly pushed to each sales unit and account manager through office platforms, text messages, etc. for key businesses and key scenarios every month to advance the collection action and reduce the risk of arrears.

[0126] 3. Collection Processing Module

[0127] 1. Request payment from customers in arrears

[0128] After the early warning is pushed, the collection execution stage begins. The collection operation is standardized and normalized. The account manager uses the received arrears data information to conduct closed-loop control of arrears data viewing, collection record filling, and collection data statistics. At the same time, by setting the time configuration for the unread reminder of the pushed early warning, the early warning will be pushed again after the time exceeds the limit.

[0129] Collection strategy process:

[0130] The Policy Gradient Methods algorithm is used to determine the overdue payment collection strategy based on the historical user's overdue payment records and collection behaviors. The specific steps are as follows: determine the state: the duration of overdue payment; determine the action: different collection behaviors; define the strategy: use a convolutional neural network (CNN) to approximate the policy function, the input is the current user state (duration of overdue payment), and the output is the probability of debt collection for each action; randomly initialize the weights and biases of the neural network as the parameters of the strategy, let the user run multiple times in different states, select actions according to the current strategy, and record the state, action and reward of each run (the reward is defined as 1 when the debt collection is completed, and 0 in other cases); use the collected data to calculate the policy gradient, use the Monte Carlo method to calculate the long-term return of each action in a given state, and use these returns to update the policy parameters; repeat the above steps until the strategy converges or reaches the preset number of iterations. During the iteration process, the performance of debt collection will gradually improve, and the best collection behavior can be matched for different overdue payment durations; according to the collection strategy, the account manager can obtain pending warnings and debt details and formulate a collection strategy.

[0131] The third embodiment of the present disclosure also provides an accounts receivable management system based on a digital system, such as Figure 4 As shown, the system comprises:

[0132] An acquisition module 11 is configured to acquire data of customers in arrears;

[0133] A data processing module 12, which is configured to process customer data;

[0134] A classification module 13, which is configured to classify the processed customer data by customer importance level through a neural network model, and use clustering to optimize the importance level classification results to classify customers into types with different importance levels;

[0135] The collection processing module 14 is configured to use a policy gradient algorithm according to the type of customer and match the best collection strategy based on the historical user's arrears records and collection behaviors.

[0136] Furthermore, the acquisition module 11 is specifically configured as follows:

[0137] A list of customers in arrears is obtained through the enterprise ledger to form a data table of customers in arrears, wherein key fields of the data table of customers in arrears include customer name, customer address, place of origin, corresponding account manager, business scale and amount of arrears.

[0138] Furthermore, the data processing module 12 is specifically configured as follows:

[0139] Process missing data values. If the missing rate of a variable is greater than the first threshold, delete the variable directly. If the missing rate is lower than the second threshold, fill it with the mode or median. If the missing rate of a variable is between the first and second thresholds, retain the original variable, introduce a derived variable, assign the missing field a value of 0 directly, and fill in the remaining values ​​as they are.

[0140] Check whether the data is duplicated and delete the duplicate data directly;

[0141] Check whether the data has outliers and treat them as missing values;

[0142] Convert non-numeric fields to numeric values;

[0143] Min-Max normalization is used to scale all data to the interval [0, 1]. The Min-Max representation is:

[0144]

[0145] Among them, x′ is the normalized data value, and x is the original data value.

[0146] Furthermore, the classification module 13 is specifically configured as follows:

[0147] The neural network model uses the TextCNN model, and a tested model is obtained after passing the training set training and the test set testing;

[0148] The processed customer data is converted into a multi-dimensional vector and input into the tested model. The model outputs the probabilities of multiple categories and classifies the customers into the category with the highest probability.

[0149] Furthermore, the classification module 13 is further configured to:

[0150] For each category after model classification, the k-means clustering algorithm is used to optimize the important level classification results, including:

[0151] After completing the cluster center selection, for the newly added customer information, calculate its Euclidean distance with each cluster center, classify it into the cluster with the nearest center, and obtain the classification result. Determine the final importance level classification result of the user based on the model training result and the classification result, and add the customer level field to the customer data.

[0152] Furthermore, the system also includes an early warning module 15;

[0153] The warning module 15 is configured to push accounts receivable warning information to the sales unit and the account manager according to the business orientation and classification warning rules based on the corresponding account manager and the customer importance level after the classification module 13 obtains the type of customer importance level.

[0154] Furthermore, the collection processing module 14 is specifically configured as follows:

[0155] For each type of customer, a policy gradient algorithm is used to determine the overdue payment collection strategy based on the historical user's overdue payment records and collection behavior, including:

[0156] Determine user status: the duration of arrears; determine actions: different collection actions;

[0157] Define strategy: Use convolutional neural network to approximate the strategy function, with the input being the current user status and the output being the probability of debt recovery for each action;

[0158] Randomly initialize the weights and biases of the neural network as the parameters of the strategy, let the user run it multiple times in different states, choose actions based on the current strategy, and record the state, action, and reward of each run;

[0159] Use the recorded data to calculate the policy gradient, use Monte Carlo methods to calculate the long-term return of each action in a given state, and use these returns to update the policy parameters;

[0160] Repeat the above steps until the strategy converges or reaches the preset number of iterations, matching the best collection behavior for different arrears durations.

[0161] The accounts receivable management system based on a digital system in the disclosed embodiment is used to implement the accounts receivable management system method based on a digital system in method embodiment 1 and embodiment 2, so the description is relatively simple. For details, please refer to the relevant description in the previous method embodiment, which will not be repeated here.

[0162] In addition, if Figure 5 As shown, the fourth embodiment of the present disclosure further provides an electronic device, including a memory 100 and a processor 200, wherein the memory 100 stores a computer program, and when the processor 200 runs the computer program stored in the memory 100, the processor 200 executes the above-mentioned various possible methods.

[0163] The memory 100 is connected to the processor 200. The memory 100 may be a flash memory, a read-only memory or other memory, and the processor 200 may be a central processing unit or a single-chip microcomputer.

[0164] In addition, an embodiment of the present disclosure further provides a computer-readable storage medium, on which a computer program is stored, and the computer program is executed by a processor to perform the above-mentioned various possible methods.

[0165] The computer-readable storage medium includes volatile or non-volatile, removable or non-removable media implemented in any method or technology for storing information (such as computer-readable instructions, data structures, computer program modules or other data). Computer-readable storage media include, but are not limited to, RAM (Random Access Memory), ROM (Read-Only Memory), EEPROM (Electrically Erasable Programmable read only memory), flash memory or other memory technology, CD-ROM (Compact Disc Read-Only Memory), Digital Video Disc (DVD) or other optical disk storage, magnetic cassettes, magnetic tapes, magnetic disk storage or other magnetic storage devices, or any other medium that can be used to store the desired information and can be accessed by a computer.

[0166] It is to be understood that the above embodiments are merely exemplary embodiments used to illustrate the principles of the present disclosure, but the present disclosure is not limited thereto. For those of ordinary skill in the art, various modifications and improvements can be made without departing from the spirit and substance of the present disclosure, and these modifications and improvements are also considered to be within the scope of protection of the present disclosure.

Claims

1. A method for managing accounts receivable based on a digital system, characterized in that: The method comprises: Obtain data on customers who are in arrears; Carry out data processing on customer data; The processed customer data is classified into customer importance levels through a neural network model, and clustering is used to optimize the importance level classification results to divide customers into types with different importance levels; According to the type of customer, the policy gradient algorithm is used to match the best overdue payment collection strategy based on the historical user's overdue payment records and collection behavior.

2. The method according to claim 1, characterized in that The acquisition of the customer data of the arrears includes: A list of customers in arrears is obtained through the enterprise ledger to form a data table of customers in arrears, wherein key fields of the data table of customers in arrears include customer name, customer address, place of origin, corresponding account manager, business scale and amount of arrears.

3. The method according to claim 2, characterized in that The data processing of customer data includes: Process missing data values. If the missing rate of a variable is greater than the first threshold, delete the variable directly. If the missing rate is lower than the second threshold, fill it with the mode or median. If the missing rate of a variable is between the first and second thresholds, retain the original variable, introduce a derived variable, assign the missing field a value of 0 directly, and fill in the remaining values ​​as they are. Check whether the data is duplicated and delete the duplicate data directly; Check whether the data has outliers and treat them as missing values; Convert non-numeric fields to numeric values; Min-Max normalization is used to scale all data to the interval [0, 1]. The Min-Max representation is: Among them, x′ is the normalized data value, and x is the original data value.

4. The method according to claim 1, characterized in that: The processing of the customer data by using a neural network model to classify the customer importance level includes: The neural network model uses the TextCNN model, and a tested model is obtained after passing the training set training and the test set testing; The processed customer data is converted into a multi-dimensional vector and input into the tested model. The model outputs the probabilities of multiple categories and classifies the customers into the category with the highest probability.

5. The method according to claim 4, characterized in that The classification results of importance level optimization using clustering include: For each category after model classification, the k-means clustering algorithm is used to optimize the important level classification results, including: After completing the cluster center selection, for the newly added customer information, calculate its Euclidean distance with each cluster center, classify it into the cluster with the nearest center, and obtain the classification result. Determine the final importance level classification result of the user based on the model training result and the classification result, and add the customer level field to the customer data.

6. The method according to claim 5, characterized in that The method further comprises: After obtaining the type of customer importance level, accounts receivable warning information is pushed to sales units and customer managers according to the corresponding account managers and customer importance levels, in accordance with business orientation and hierarchical classification warning rules.

7. The method according to claim 1, characterized in that According to the type of customer, the policy gradient algorithm is used to match the best overdue payment collection strategy according to the historical user's overdue payment records and collection behavior, including: For each type of customer, a policy gradient algorithm is used to determine the overdue payment collection strategy based on the historical user's overdue payment records and collection behavior, including: Determine user status: the duration of arrears; determine actions: different collection actions; Define strategy: Use convolutional neural network to approximate the strategy function, with the input being the current user status and the output being the probability of debt recovery for each action; Randomly initialize the weights and biases of the neural network as the parameters of the strategy, let the user run it multiple times in different states, choose actions based on the current strategy, and record the state, action, and reward of each run; Use the recorded data to calculate the policy gradient, use Monte Carlo methods to calculate the long-term return of each action in a given state, and use these returns to update the policy parameters; Repeat the above steps until the strategy converges or reaches the preset number of iterations, matching the best collection behavior for different arrears durations.

8. An accounts receivable management system based on a digital system, characterized in that: The system comprises: An acquisition module configured to acquire data of customers in arrears; A data processing module configured to perform data processing on customer data; A classification module is configured to classify the processed customer data into customer importance levels through a neural network model, and use clustering to optimize the importance level classification results to classify customers into types with different importance levels; The collection processing module is configured to use a policy gradient algorithm according to the type of customer and match the best collection strategy based on historical user arrears records and collection behaviors.

9. An electronic device, characterized in that: It includes a memory and a processor, wherein the memory stores a computer program, and when the processor runs the computer program stored in the memory, the processor executes the accounts receivable management method based on a digital system as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method for managing accounts receivable based on a digital system as described in any one of claims 1 to 7 is implemented.

Images