Risk fund identification method and device
By identifying the risks of logistics companies' accounts receivable through machine learning models, the problems of low efficiency and insufficient accuracy in existing technologies are solved, efficient and accurate screening of risky accounts is achieved, the risk of overdue collection is reduced, and the operating efficiency of logistics companies is improved.
Patent Information
- Application Number
- CN202410303477.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-03-15
- Publication Date
- 2025-09-16
AI Technical Summary
In existing technologies, logistics companies have low efficiency and insufficient accuracy in identifying the risks of accounts receivable, resulting in a high risk of overdue collection and affecting operating efficiency.
A machine learning model is used to train a classification model based on pre-labeled historical payment feature information. By calculating the overdue probability of current payments, payments with high overdue probability and large amounts are screened out, and the number of risky payments is determined based on resource data for efficient and accurate identification.
It has achieved efficient and accurate identification of risky funds, reduced the risk of overdue repayments, increased the business repayment rate, and improved the operating efficiency of logistics companies.
Smart Images
Figure CN120654019A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a method and device for identifying risky payments. Background Art
[0002] Cash flow is a core factor in the development of logistics companies, and accounts receivable are a key indicator of their liquidity. Therefore, prioritizing the management of logistics accounts receivable is crucial for improving operational efficiency. In practice, logistics businesses often suffer significant losses due to overdue payments from logistics customers. Therefore, it is essential to identify and proactively address receivables with anticipated risks. Existing technology typically relies on financial personnel to identify risky accounts by screening accounts with near-due dates or customers with a history of overdue payments. This is inefficient, and determining risky accounts based solely on account age and historical overdue behavior is inaccurate. Summary of the Invention
[0003] In view of this, an embodiment of the present invention provides a method and device for identifying risky payments, which realize efficient and accurate identification of risky payments through a machine learning model.
[0004] To achieve the above objectives, according to one aspect of the present invention, a method for identifying risky payments is provided.
[0005] The method for identifying risky payments in an embodiment of the present invention includes: training a preset classification model based on pre-labeled historical payments; wherein, the characteristic information of the historical payments is used as training input data, and the labeling information of the historical payments is used as training label data, and the labeling information includes overdue repayments or on-time repayments; inputting the characteristic information of the payments to be identified in the current period into the trained classification model to obtain the overdue probability of the payments to be identified, and determining the payments to be identified whose overdue probability is greater than a preset first threshold as preliminary payments; determining the planned number of risky payments for the current period based on the payment processing resource data of the current period; arranging the preliminary payments into a queue in descending order according to the amount, and selecting the preliminary payments whose overdue probability is greater than the preset dynamic threshold and whose number is equal to the planned number of risky payments in order from the head to the end of the queue, and determining the selected preliminary payments as risky payments.
[0006] Optionally, the payment processing resource data of the current cycle includes: the total amount of payment processing resources of the current cycle, and the average amount of payment processing resources consumed by a single payment in historical cycles; and determining the number of risk payment plans for the current cycle based on the payment processing resource data of the current cycle includes: dividing the total amount of payment processing resources of the current cycle by the average amount of payment processing resources consumed by a single payment to obtain the number of risk payment plans for the current cycle.
[0007] Optionally, the method of selecting the preliminary funds whose overdue probability is greater than a preset dynamic threshold and whose quantity is equal to the planned quantity of risk funds in sequence from the head to the tail of the queue includes: determining the maximum value and the step reduction amount of the dynamic threshold, constructing an arithmetic decreasing sequence with the maximum value as the first value and the step reduction amount as the tolerance; starting from the maximum value, selecting the values in the arithmetic decreasing sequence in sequence as the dynamic threshold in sequence according to the numerical order of the arithmetic decreasing sequence until the current dynamic threshold meets the preset judgment condition; selecting the preliminary funds whose overdue probability is greater than the current dynamic threshold in sequence from the head to the tail of the queue. The preliminary selected funds with a dynamic threshold are used as the risk funds; the discrimination conditions are: the preliminary selected funds with an overdue probability greater than the current dynamic threshold are selected in sequence from the head to the tail of the queue, and the number of the selected initial options is determined as the first number; the preliminary selected funds with an overdue probability greater than the next dynamic threshold are selected in sequence from the head to the tail of the queue, and the number of the selected initial options is determined as the second number; the first number is less than or equal to the planned number of risk funds, and the second number is greater than or equal to the planned number of risk funds; wherein, the next dynamic preset is the next value of the current dynamic threshold in the arithmetic decreasing sequence.
[0008] Optionally, the characteristic information includes at least one of the following information of the payer: basic information, transaction information, industry information, external environment information, abnormal business information, waybill information, feedback information, and order information; the dynamic threshold is greater than or equal to the first threshold.
[0009] To achieve the above objectives, according to another aspect of the present invention, a device for identifying risky payments is provided.
[0010] The risk payment identification device of an embodiment of the present invention includes: a training unit, which is used to train a preset classification model based on pre-labeled historical payments; wherein the characteristic information of the historical payments is used as training input data, and the labeling information of the historical payments is used as training label data, and the labeling information includes overdue repayments or on-time repayments; a preliminary selection unit, which is used to input the characteristic information of the payments to be identified in the current period into the trained classification model to obtain the overdue probability of the payments to be identified, and determine the payments to be identified whose overdue probability is greater than a preset first threshold as preliminary selected payments; a planned quantity calculation unit, which is used to determine the planned quantity of risk payments for the current period based on the payment processing resource data of the current period; a selection unit, which is used to arrange the preliminary selected payments into a queue in descending order according to the amount, and select the preliminary selected payments whose overdue probability is greater than the preset dynamic threshold and whose quantity is equal to the planned quantity of risk payments in order from the head to the end of the queue, and determine the selected preliminary selected payments as risk payments.
[0011] Optionally, the payment processing resource data for the current cycle includes: the total amount of payment processing resources for the current cycle, and the average amount of payment processing resources consumed by a single payment in historical cycles; and the planned quantity calculation unit is further used to: divide the total amount of payment processing resources for the current cycle by the average amount of payment processing resources consumed by a single payment to obtain the planned quantity of risk payments for the current cycle.
[0012] Optionally, the selection unit is further used to: determine the maximum value and the step reduction amount of the dynamic threshold, construct an arithmetic decreasing sequence with the maximum value as the first value and the step reduction amount as the tolerance; starting from the maximum value, select the values in the arithmetic decreasing sequence in sequence according to the numerical order of the arithmetic decreasing sequence as the dynamic threshold until the current dynamic threshold meets the preset judgment condition; select the preliminary funds with the overdue probability greater than the current dynamic threshold as the risk funds in sequence from the head to the end of the queue; the judgment condition is : Select the preliminary amounts whose overdue probability is greater than the current dynamic threshold in order from the head to the end of the queue, and determine the number of the selected initial options as the first number; select the preliminary amounts whose overdue probability is greater than the next dynamic threshold in order from the head to the end of the queue, and determine the number of the selected initial options as the second number; the first number is less than or equal to the number of risk amount plans, and the second number is greater than or equal to the number of risk amount plans; wherein, the next dynamic preset is the next value of the current dynamic threshold in the arithmetic decreasing sequence.
[0013] Optionally, the characteristic information includes at least one of the following information of the payer: basic information, transaction information, industry information, external environment information, abnormal business information, waybill information, feedback information, and order information; the dynamic threshold is greater than or equal to the first threshold.
[0014] To achieve the above objective, according to another aspect of the present invention, an electronic device is provided.
[0015] An electronic device of the present invention includes: one or more processors; a storage device for storing one or more programs, and when the one or more programs are executed by the one or more processors, the one or more processors implement the risk payment identification method provided by the present invention.
[0016] To achieve the above objective, according to another aspect of the present invention, a computer-readable storage medium is provided.
[0017] A computer-readable storage medium of the present invention stores a computer program, which, when executed by a processor, implements the risk payment identification method provided by the present invention.
[0018] According to the technical solution of the present invention, the embodiments of the above invention have the following advantages or beneficial effects:
[0019] First, a pre-trained classification model is used to calculate the overdue probability of each to-be-identified payment in the current cycle, and the to-be-identified payments with an overdue probability greater than the first threshold are determined as preliminary payments. The number of risk payment plans for the current cycle is determined based on the payment processing resource data of the current cycle. Afterwards, the preliminary payments are arranged into queues in descending order according to the amount, and preliminary payments with an overdue probability greater than the dynamic threshold and a number equal to the number of risk payment plans are selected in order from the head to the tail of the queue. Finally, the selected preliminary payments are determined as risk payments. Through the above steps, efficient and accurate identification of risk payments is achieved based on the machine learning model for early processing, avoiding the generation of payers with overdue risks, reducing the occurrence of collection risks, and improving business collection rates. At the same time, the above identification method takes into account factors such as amount, overdue probability, and the current status of payment processing resources in the current cycle. An algorithm is designed to screen out payments that match the resources of the current cycle and guarantee the maximum total amount and a large overdue probability as risk payments, which is highly practical.
[0020] The further effects of the above-mentioned non-conventional optional manner will be described below in conjunction with specific embodiments. BRIEF DESCRIPTION OF THE DRAWINGS
[0021] The accompanying drawings are provided for a better understanding of the present invention and are not intended to limit the present invention.
[0022] Figure 1 This is a schematic diagram of the main steps of the method for identifying risky funds in an embodiment of the present invention;
[0023] Figure 2 This is a schematic diagram of specific execution steps of the method for identifying risky funds in an embodiment of the present invention;
[0024] Figure 3 This is a schematic diagram of the components of the risk payment identification device in an embodiment of the present invention;
[0025] Figure 4 is an exemplary system architecture diagram in which embodiments of the present invention may be applied;
[0026] Figure 5 This is a schematic diagram of the structure of an electronic device used to implement the risk payment identification method in an embodiment of the present invention. DETAILED DESCRIPTION
[0027] The following description of exemplary embodiments of the present invention is made in conjunction with the accompanying drawings, in which various details of the embodiments of the present invention are included to facilitate understanding. These details should be considered as merely exemplary. Therefore, it should be appreciated by those skilled in the art that various changes and modifications may be made to the embodiments described herein without departing from the scope and spirit of the present invention. Similarly, for the sake of clarity and conciseness, descriptions of well-known functions and structures are omitted in the following description.
[0028] It should be pointed out that, in the absence of conflict, the embodiments of the present invention and the technical features therein may be combined with each other.
[0029] Figure 1 2 is a schematic diagram of the main steps of the method for identifying risky payments according to an embodiment of the present invention.
[0030] like Figure 1 As shown, the risky payment identification method according to the embodiment of the present invention can be specifically performed according to the following steps:
[0031] Step S101: training a preset classification model based on pre-labeled historical payments.
[0032] During the model training process in this step, the characteristic information of the historical payments is used as training input data, and the annotated information of the historical payments is used as training label data. The annotated information may include overdue repayments or on-time repayments. Exemplarily, the classification model may be Xgboost (eXtreme Gradient Boosting). In practical applications, the above characteristic information may include at least one of the following information about the payer: basic information, transaction information, industry information, external environment information, abnormal business information, waybill information, feedback information, and order information.
[0033] In one embodiment, basic information represents the payer's basic profile attributes, including registration time, legal person information, financial information, and tax status. Transaction information represents the payer's transaction-related attributes, including the payer's first cooperation date, active cooperation markets, recent transaction amounts, and historical overdue payments. Industry information represents the payer's industry, including the payer's main business, upstream and downstream industries, and industry volatility. External environment information represents the constraints of the payer's external environment, including health-related information in the payer's city. Abnormal transaction information represents available abnormal information about the payer, including claims data, business adjustment data, and prepayment data. Waybill information represents the payer's overall waybill data, including waybill fluctuation data, average waybill revenue data, and the number of cities covered by waybill data. Feedback information primarily includes customer complaint information, including average daily customer complaint data and customer complaint loss data. Order information represents overall online order data, including daily order amounts, wallet balances, and selected products. It can be understood that the various features mentioned above are directly related to whether the payer can repay on time, so these features are selected as input to train the subsequent classification model. In this way, by selecting the above comprehensive features that have a significant impact on the risk payment identification task to conduct data modeling in the subsequent prediction process and applying these features to the subsequent classification model, the classification model can learn the deep differences between overdue and on-time repayments in the risk payment identification task through the model training process, thereby improving the classification model's prediction accuracy for unknown data.
[0034] Step S102: Input the characteristic information of the to-be-identified accounts in the current period into the trained classification model to obtain the overdue probability of the to-be-identified accounts, and determine the to-be-identified accounts whose overdue probability is greater than a preset first threshold as preliminary selected accounts.
[0035] In this step, the trained classification model is used to calculate the overdue probability of each pending payment. Payments with an overdue probability greater than a first threshold (e.g., 0.5) are identified as preliminary payments. Subsequent risky payments are screened from the preliminary selection. For example, the current period is the current month, and the pending payments in the current period are those due that month.
[0036] Step S103: determining the planned amount of risk payments for the current period according to the payment processing resource data for the current period.
[0037] In this step, the payment processing resource data for the current cycle may include: the total amount of payment processing resources for the current cycle (e.g., the total number of manual hours required to collect overdue payments) and the average amount of processing resources consumed per payment in historical cycles (e.g., the average number of manual hours required to collect a single payment). Accordingly, the total amount of payment processing resources for the current cycle may be divided by the average amount of processing resources consumed per payment to obtain the planned number of risk payments for the current cycle.
[0038] Step S104: Arrange the preliminary selected funds into a queue in descending order according to the amount, and select the preliminary selected funds whose overdue probability is greater than the preset dynamic threshold and whose quantity is equal to the planned quantity of risk funds in order from the head to the tail of the queue, and determine the selected preliminary selected funds as risk funds.
[0039] As a preferred solution, this step can be performed in the following manner. First, determine the maximum value and the step reduction amount of the dynamic threshold, and construct an arithmetic decreasing sequence with the maximum value as the first value and the step reduction amount as the tolerance. For example, with 0.9 as the maximum value of the dynamic threshold and 0.05 as the tolerance, construct the following arithmetic decreasing sequence: 0.9, 0.85, 0.8, 0.75, 0.7, 0.65, 0.6... Then, starting from the maximum value, select the values in the arithmetic decreasing sequence in the numerical order of the arithmetic decreasing sequence as the dynamic threshold until the current dynamic threshold meets the preset discrimination condition. Continuing with the above example, starting from 0.9, select a value in the sequence as the current dynamic threshold, and judge whether the current dynamic threshold meets the following discrimination condition. If so, stop selecting the dynamic threshold, otherwise continue to select the next value in the sequence as the dynamic threshold and continue to judge the discrimination condition.
[0040] The discrimination conditions are as follows: First, the preliminary selections of funds whose overdue probability is greater than the current dynamic threshold are selected in order from the head of the queue to the tail, and the number of the selected initial options is determined as the first number; then, the preliminary selections of funds whose overdue probability is greater than the next dynamic threshold are selected in order from the head of the queue to the tail, and the number of the selected initial options is determined as the second number; wherein the next dynamic preset is the next value of the current dynamic threshold in the arithmetic descending sequence. The conditions that need to be met are: the first number is less than or equal to the planned number of risk funds, and the second number is greater than or equal to the planned number of risk funds.
[0041] Finally, the preliminary funds whose overdue probability is greater than the current dynamic threshold (i.e., the dynamic threshold that meets the judgment condition) are selected in order from the head to the end of the queue as the risk funds.
[0042] Continuing with the above example, if the number of risky payment plans is 10, the number of preliminary selected payments with an overdue probability greater than the current dynamic threshold is selected from the queue in the above manner as follows: the number of preliminary selected payments is 1 under the dynamic threshold of 0.9, the number of preliminary selected payments is 2 under the dynamic threshold of 0.8, the number of preliminary selected payments is 8 under the dynamic threshold of 0.75, and the number of preliminary selected payments is 11 under the dynamic threshold of 0.7. The dynamic threshold of 0.75 satisfies the above judgment conditions (the number of preliminary selected payments of 8 for this dynamic threshold < the number of risky payment plans of 10 < the number of preliminary selected payments of 11 for the next dynamic threshold). Therefore, 8 preliminary selected payments can be obtained as risky payments based on the dynamic threshold of 0.75. Thereafter, the two adjacent preliminary selected payments selected using the next dynamic threshold (0.7) will also be used as risky payments, so that the number of risky payments is equal to the number of risky payment plans. It can be understood that the above dynamic threshold is greater than or equal to the first threshold.
[0043] Figure 2 This is a schematic diagram of specific execution steps of the risk payment identification method according to an embodiment of the present invention. Figure 2 In step S201, various attribute-related feature information is added to each receivable, and the receivables due in the past and the receivables due in the current month are distinguished; in step S202, the overdue status of the receivables due in the past is marked; in step S203, the classification model is trained using the marked data; in step S204, the overdue probability of each to-be-identified payment is calculated using the trained classification model; in step S205, the to-be-identified payment with an overdue probability greater than the first threshold is determined as the preliminary payment; in step S206, the planned number of risky payments for the current month is calculated; in step S208, the number of risky payments for the current month is calculated; in step S209, the number of risky payments for the current month is calculated; in step S210, the number of risky payments for the current month is calculated. In step S207, an arithmetic decreasing series of dynamic thresholds is generated; in step S208, the current dynamic threshold is selected from the series; in step S209, multiple preliminary selected items are selected from the preliminary selected items arranged in descending order of amount according to the current dynamic threshold; in step S210, it is determined whether the discrimination condition is met; if so, step S211 is executed; if not, step S212 is executed; in step S211, risky items are selected according to the dynamic threshold that meets the discrimination condition; in step S212, the next value in the series is selected as the dynamic threshold and step S208 is re-executed.
[0044] In the technical solution of the embodiment of the present invention, a pre-trained classification model is first used to calculate the overdue probability of each to-be-identified payment in the current cycle, and the to-be-identified payment with an overdue probability greater than a first threshold is determined as a preliminary payment, and the number of risk payment plans for the current cycle is determined based on the payment processing resource data of the current cycle. Thereafter, the preliminary payments are arranged into queues in descending order according to the amount, and preliminary payments with an overdue probability greater than a dynamic threshold and a number equal to the number of risk payment plans are selected in order from the head to the tail of the queue, and finally the selected preliminary payments are determined as risk payments. Through the above steps, efficient and accurate identification of risk payments is achieved based on the machine learning model for early processing, avoiding the generation of payers with overdue risks, reducing the occurrence of collection risks, and improving business collection rates; at the same time, the above identification method takes into account factors such as amount, overdue probability, and the current payment processing resource status of the current cycle, and designs an algorithm to screen out payments that match the current cycle resources and guarantee the maximum total amount and a large overdue probability as risk payments, which has high practicality.
[0045] A specific embodiment of the present invention is described below.
[0046] Cash flow is one of the core issues concerning the development of logistics companies. Accounts receivable is a key indicator of the liquidity of logistics companies. Paying attention to the management of logistics accounts receivable is of great significance to improving the operating efficiency of logistics companies. Logistics business often leads to huge losses due to overdue payments from logistics customers.
[0047] To identify customers with collection risks and implement corresponding strategic management to avoid overdue payments and improve the collection rate, the existing technology is to use manual reconciliation to screen out accounts with aging that are about to expire or customers with a history of overdue payments and mark them as high-risk.
[0048] The financial staff reconciles accounts regularly, checking accounts with customers once every month or quarter. At the same time, the financial staff conducts an aging analysis of the accounts receivable for each collection and formulates a collection plan schedule. The sales department designates a person to be responsible for the collection of collections, which is done once every certain period.
[0049] The existing technology has the following disadvantages:
[0050] 1. Manual reconciliation is inefficient for assessing collection risk. With hundreds of thousands of diverse customers, a complex product structure, and varying product collection deadlines, or even varying deadlines for the same product across different customers, each finance department requires a large number of financial personnel to perform aging analysis on accounts receivable for each customer across different products. This analysis is prone to errors and delays, resulting in delayed collection of receivables.
[0051] 2. Because the factors contributing to collection risk are complex, simply judging risk based on account age and past due dates is insufficiently accurate. Consequently, collection plans can be irrational. Collection plans fail to accurately distinguish between customers who will likely not pay due when their accounts mature, and those who will pay within the aging period without reminders. This results in a lack of prioritization for collection actions by sales representatives, delays in timely payment collection for high-risk customers, and a low overall collection rate.
[0052] The finance department has invested a large number of personnel in analyzing the aging of accounts receivable. The customer repayment collection plan formulated has inaccurate judgments, resulting in no priority for collection actions and overdue repayments. In order to solve the problem of not relying on human factors to determine the possibility of customer repayment overdue, this embodiment adopts machine learning methods to accurately identify overdue customers, and adopts pre-identification of business risks in combination with business scenarios. From the perspective of overdue customer identification, three superposition models are used to accurately predict and identify customers who will be overdue and customers who will not be overdue among existing customers and newly signed customers within the account period. These models are used for sales execution and collection in a timely manner in the early stage of the account period to avoid the emergence of customers with overdue risks, reduce the occurrence of collection risks, and improve the business collection rate.
[0053] (1) Consider more factors as features of machine learning models.
[0054] (2) Learn the relationship between the features in (1) and the collection risk by training a supervised machine learning model.
[0055] (3) Use the trained machine learning model to predict the probability of collection risk. For accounts receivable with a collection risk probability exceeding the threshold, trigger sales to promptly perform collection collection in the early stages of the payment period.
[0056] First, the technical problem to be solved by this invention is modeled. The problem is modeled as an optimization problem of improving the monthly collection rate of logistics business under the limitation of collection labor cost. That is, the labor time C h <θ, find a set of collections to be collected Belongs to S, which maximizes the monthly logistics customer receivables collection rate R.
[0057] Where R is the monthly logistics customer receivables collection rate;
[0058] R = (actual amount of payments received from logistics customers) / (monthly amount of payments due from logistics customers)
[0059] C h The manual time spent on collecting payments for logistics settlement personnel;
[0060] (C i The manual time required to split the payment to a single high-risk customer);
[0061] C i : The manual time required to collect payments from a single high-risk customer;
[0062] N high : The number of receivables for collection;
[0063] S: the quantity of all payable receivables;
[0064] θ: The upper limit of manual time required for payment collection.
[0065] The overall flow chart of the method adopted by the present invention is as follows:
[0066] (1) Supplement attributes for each receivable
[0067] Each receivable is supplemented with attribute data such as basic customer information, transaction information, industry information, sales behavior, external environment information, abnormal transaction information, waybill information, customer complaint information, consumer information, and sales feedback. Basic information includes registration date, legal person information, financial report information, and tax status. Transaction information includes the first time the logistics customer cooperates, active cooperative markets, recent transaction amounts, and historical overdue payments. Industry information includes core business, upstream and downstream industries, and industry volatility. Sales behavior includes sales personnel's payment collection status, sales personnel's customer visits, and sales personnel's feedback on risk warnings. External environment information includes health and wellness information in the customer's city. Abnormal transaction information includes claims data, business adjustment data, and prepayment data. Waybill information includes waybill fluctuation data, average waybill revenue data, and the number of cities covered by waybill. Customer complaint information includes daily average customer complaint data and customer complaint loss data. Consumer information includes daily order amount, wallet balance, and selected logistics products. Sales feedback includes daily feedback data from sales personnel to customers. The attribute data supplemented in this step will serve as the input of the classification model in the subsequent steps.
[0068] (2) Mark each historical due receivable as overdue
[0069] Obtain customers with a history of overdue payment repayments and mark the data in step (1);
[0070] Overdue data for logistics customers is marked based on past historical samples of overdue customers. Based on whether a customer's payment is overdue, the following factors can also be used to determine:
[0071] a. Overdue Days: refers to the number of days a customer is overdue, usually based on the most recent payment date.
[0072] b. Number of overdue payments: refers to the number of times a customer has overdue payments within a certain period of time.
[0073] c. Overdue Amount: refers to the total amount overdue by the customer.
[0074] d. Repayment Status: This refers to the customer's current repayment status, such as paid or outstanding. This data helps to more accurately assess the customer's overdue status and repayment ability.
[0075] (3) Use the data obtained in the previous step to train the Xgboost machine learning model M
[0076] After calling step (2), the labeled samples are put into the Xgboost model for training. The model inputs the label of whether the historical customer samples are overdue. It can be understood that Xgboost is a machine learning model integrated by multiple decision trees.
[0077] For each sample i, calculate its predicted value for the current model: For each historical sample of customers who should be paid back, K represents the number of decision trees in the model, f k (x i ) represents the predicted value of sample i by the kth decision tree in the model.
[0078] After that, we need to calculate the loss function under the current model state. Represents the loss function of sample i, Ω(f k ) represents the regularization term under the kth decision tree; after each iteration, the value of the model's loss function is obtained, it is necessary to update the value of each decision tree to the leaf node to minimize the loss function value; when the iterative training round reaches the preset loss function value threshold, the Xgboost model M trained based on the historical overdue customer sample is completed.
[0079] (4) Obtain the accounts receivable for the current month with the supplementary completion attribute and use model M to predict the overdue probability
[0080] Take the accounts receivable from customers that are due in the current month, supplement them with the current customer settlement information, customer complaint information, order quantity information and other daily fluctuation attributes, use the model M trained in step (3) to predict and output the probability value of whether the customer has the risk of overdue payment.
[0081] a. For each logistics payment risk sample, its prediction score needs to be calculated. The prediction score in Xgboost is obtained by summing the output values of multiple trees. The specific calculation method is as follows:
[0082] score=sum(w i *f i (x))+b
[0083] Among them, w i is the weight of the i-th tree, fi (x) is the output value of the i-th tree for the risk sample x, and b is the bias term. In the model, the output value of each tree is obtained by the weight sum of its leaf nodes, which can be expressed as:
[0084] f i (x) = sum(w j ,if x in R j ; else 0)
[0085] Among them, R j is the sample area corresponding to the jth leaf node, w j is the weight of the j-th leaf node.
[0086] b. Convert the predicted score to a probability. Because the binary classification problem only requires a probability value (i.e., the probability that the sample belongs to the positive class), the Sigmoid function can be used to convert the predicted score to a probability. The specific calculation method is: p = 1 / (1 + exp(-score)), where p is the probability that the sample belongs to the positive class.
[0087] c. Convert probabilities to classification labels. The customer's payment risk probability can be converted into a classification label based on a threshold. For example, if the threshold is set to 0.5, then when the probability of a sample belonging to the positive class is greater than 0.5, it is predicted as a positive class; otherwise, it is predicted as a negative class.
[0088] (5) Calculate the amount of money that can be collected based on the collection manpower limit for that month
[0089] Based on the monthly collection labor cost limit, the number of high-risk payments that can be collected is obtained, which is generally determined by manual experience: N high =θ / c, where the labor cost is limited to θ.
[0090] (6) Generate a collection of high-risk collection amounts
[0091] a. Set the initial threshold of overdue probability σ′ = 0.8.
[0092] b. Sort the customers in descending order according to their payable amounts, and take the high-risk customer payments where σ>σ′ until N are reached. high High-risk customer funds;
[0093] Assume there are n samples, and the amount of the i-th sample is a i , the estimated probability is p i . Define the decision variable x i Indicates whether to select the i-th sample, x i =1 means selection, x i =0 means no selection. The goal is to select some samples so that the total amount is maximized while satisfying the estimated probability of not less than 0.8, that is:
[0094]
[0095]
[0096] c. If the quantity does not reach N high , then repeat step b until the required amount of risky funds is reached or the condition σ′ < 0.01 is satisfied. Here, σ′ is a variable representing the limit on the estimated probability. If the required amount cannot be reached, σ′ is gradually adjusted downward until a solution is found. Once all risky customer samples that meet the conditions are found, the model can be derived into an online table.
[0097] (7) Issue a collection of high-risk collection amounts
[0098] The generated data is sent to sales personnel through the logistics online system for delegated collection work.
[0099] Compared to the manual processing methods of existing technologies, this embodiment adds attribute features to accounts receivable, supporting over 200 sets of attribute features, including basic information, transaction data, industry data, sales behavior, real-time order volume fluctuations, and customer complaint data. This reduces the workload of manual data analysis through machine learning. By predicting the probability of collection, labor costs are concentrated on accounts receivable with low collection probabilities and large amounts. This helps to timely execute sales collection in the early stages of the account period, avoid the generation of customers with high-risk overdue payments, reduce collection risks, and improve business collection rates.
[0100] It should be noted that the collection, collection, updating, analysis, processing, use, transmission, and storage of user personal information that may be involved in the technical solution of the present invention all comply with the provisions of relevant laws and regulations, are used for legitimate purposes, and do not violate public order and good morals. Necessary measures are taken with respect to user personal information to prevent unauthorized access to user personal information data and to maintain the security of user personal information, network security, and national security.
[0101] For ease of description, the aforementioned method embodiments are described as a series of actions. However, those skilled in the art should be aware that the present invention is not limited to the order of the actions described, and certain steps can actually be performed in other orders or simultaneously. In addition, those skilled in the art should also be aware that the embodiments described in this specification are preferred embodiments, and the actions and modules involved are not necessarily required to implement the present invention.
[0102] In order to better implement the above solutions of the embodiments of the present invention, relevant devices for implementing the above solutions are also provided below.
[0103] See also Figure 3As shown, the risk payment identification device 300 provided by the embodiment of the present invention may include: a training unit 301, a preliminary selection unit 302, a planned quantity calculation unit 303 and a selection unit 304.
[0104] Among them, the training unit 301 is used to train a preset classification model based on pre-labeled historical payments; wherein, the characteristic information of the historical payments is used as training input data, and the labeling information of the historical payments is used as training label data, and the labeling information includes overdue repayments or on-time repayments; the preliminary selection unit 302 is used to input the characteristic information of the payments to be identified in the current period into the trained classification model to obtain the overdue probability of the payments to be identified, and determine the payments to be identified whose overdue probability is greater than a preset first threshold as preliminary selected payments; the planned quantity calculation unit 303 is used to determine the planned number of risk payments for the current period based on the payment processing resource data of the current period; the selection unit 304 is used to arrange the preliminary selected payments into a queue in descending order according to the amount, and select the preliminary selected payments whose overdue probability is greater than the preset dynamic threshold and whose quantity is equal to the planned number of risk payments in order from the head to the end of the queue, and determine the selected preliminary selected payments as risk payments.
[0105] As a preferred solution, the payment processing resource data of the current cycle includes: the total amount of payment processing resources of the current cycle, and the average amount of payment processing resources consumed by a single payment in the historical cycle; and the planned quantity calculation unit 303 can be further used to: divide the total amount of payment processing resources of the current cycle by the average amount of payment processing resources consumed by a single payment to obtain the planned quantity of risk payments for the current cycle.
[0106] Preferably, the selection unit 304 can be further used to: determine the maximum value and the step reduction amount of the dynamic threshold, construct an arithmetic decreasing sequence with the maximum value as the first value and the step reduction amount as the tolerance; starting from the maximum value, select the values in the arithmetic decreasing sequence in sequence according to the numerical order of the arithmetic decreasing sequence as the dynamic threshold until the current dynamic threshold meets the preset judgment condition; select the preliminary funds with the overdue probability greater than the current dynamic threshold as the risk funds in sequence from the head to the tail of the queue; the judgment condition The components are: selecting the preliminary amounts with an overdue probability greater than the current dynamic threshold in order from the head to the tail of the queue, and determining the number of the selected initial options as the first number; selecting the preliminary amounts with an overdue probability greater than the next dynamic threshold in order from the head to the tail of the queue, and determining the number of the selected initial options as the second number; the first number is less than or equal to the number of risk amount plans, and the second number is greater than or equal to the number of risk amount plans; wherein, the next dynamic preset is the next value of the current dynamic threshold in the arithmetic decreasing sequence.
[0107] In addition, in an embodiment of the present invention, the characteristic information includes at least one of the following information of the payer: basic information, transaction information, industry information, external environment information, abnormal business information, waybill information, feedback information, and order information; the dynamic threshold is greater than or equal to the first threshold.
[0108] According to the technical solution of the embodiment of the present invention, first, a pre-trained classification model is used to calculate the overdue probability of each to-be-identified payment in the current cycle, and the to-be-identified payment with an overdue probability greater than the first threshold is determined as a preliminary payment, and the number of risk payment plans for the current cycle is determined based on the payment processing resource data of the current cycle. Thereafter, the preliminary payments are arranged into queues in descending order according to the amount, and preliminary payments with an overdue probability greater than the dynamic threshold and a number equal to the number of risk payment plans are selected in order from the head to the tail of the queue. Finally, the selected preliminary payments are determined as risk payments. Through the above steps, efficient and accurate identification of risk payments is achieved based on the machine learning model for early processing, avoiding the generation of payers with overdue risks, reducing the occurrence of collection risks, and improving business collection rates. At the same time, the above identification method takes into account factors such as amount, overdue probability, and the current status of payment processing resources in the current cycle, and designs an algorithm to screen out payments that match the current cycle resources and guarantee the maximum total amount and a large overdue probability as risk payments, which has high practicality.
[0109] Figure 4 An exemplary system architecture 400 is shown to which the risky payment identification method or risky payment identification device according to an embodiment of the present invention can be applied.
[0110] like Figure 4 As shown, system architecture 400 may include terminal devices 401, 402, and 403, a network 404, and a server 405 (this architecture is merely an example, and the components included in the specific architecture may be adjusted based on the specific application). Network 404 is used to provide a medium for communication links between terminal devices 401, 402, and 403 and server 405. Network 404 may include various connection types, such as wired or wireless communication links or fiber optic cables.
[0111] Users can use terminal devices 401, 402, 403 to interact with server 405 via network 404 to receive or send messages, etc. Various communication client applications can be installed on terminal devices 401, 402, 403, such as risk payment identification application (only as an example).
[0112] The terminal devices 401 , 402 , and 403 may be various electronic devices having a display screen and supporting web browsing, including but not limited to smart phones, tablet computers, laptop computers, and desktop computers.
[0113] Server 405 may be a server that provides various services, such as a backend server that supports risky payment identification applications operated by users using terminal devices 401, 402, and 403 (for example only). The backend server may process received risky payment identification requests and provide feedback (for example, the probability of payment being overdue—for example only) to terminal devices 401, 402, and 403.
[0114] It should be noted that the risky payment identification method provided in the embodiment of the present invention is generally executed by the server 405 , and accordingly, the risky payment identification device is generally set in the server 405 .
[0115] It should be understood that Figure 4 The number of terminal devices, networks and servers in the embodiment is merely illustrative. Any number of terminal devices, networks and servers may be provided as required.
[0116] The present invention also provides an electronic device. The electronic device in an embodiment of the present invention includes: one or more processors; and a storage device configured to store one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the risky payment identification method provided by the present invention.
[0117] Reference below Figure 5 , which shows a schematic structural diagram of a computer system 500 of an electronic device suitable for implementing an embodiment of the present invention. Figure 5 The electronic device shown is only an example and should not limit the functions and scope of use of the embodiments of the present invention.
[0118] like Figure 5 As shown, the computer system 500 includes a central processing unit (CPU) 501, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 502 or a program loaded from a storage unit 508 into a random access memory (RAM) 503. Various programs and data required for the operation of the computer system 500 are also stored in the RAM 503. The CPU 501, the ROM 502, and the RAM 503 are connected to each other via a bus 504. An input / output (I / O) interface 505 is also connected to the bus 504.
[0119] The following components are connected to the I / O interface 505: an input section 506 including a keyboard, a mouse, and the like; an output section 507 including devices such as a cathode ray tube (CRT), a liquid crystal display (LCD), and a speaker; a storage section 508 including a hard disk; and a communication section 509 including a network interface card such as a LAN card or a modem. The communication section 509 performs communication processing via a network such as the Internet. A drive 510 is also connected to the I / O interface 505 as needed. A removable medium 511, such as a magnetic disk, an optical disk, a magneto-optical disk, or a semiconductor memory, is installed in the drive 510 as needed, so that computer programs read therefrom can be installed in the storage section 508 as needed.
[0120] In particular, according to embodiments disclosed herein, the processes described in the main step diagrams above can be implemented as computer software programs. For example, embodiments of the present invention include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for executing the methods shown in the main step diagrams. In the above embodiments, the computer program can be downloaded and installed from a network via the communication section 509 and / or installed from removable media 511. When the computer program is executed by the central processing unit 501, the above-described functions defined in the system of the present invention are performed.
[0121] It should be noted that the computer-readable medium described in the present invention may be a computer-readable signal medium or a computer-readable storage medium, or any combination thereof. A computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or component, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to, an electrical connection having one or more conductors, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In the present invention, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, device, or component. In the present invention, a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. This propagated data signal may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium may also be any computer-readable medium other than a computer-readable storage medium that can transmit, propagate, or transport a program for use by or in conjunction with an instruction execution system, apparatus, or device. Program code embodied on a computer-readable medium may be transmitted using any suitable medium, including but not limited to wireless, wireline, optical cable, RF, or any suitable combination thereof.
[0122] The flowcharts and block diagrams in the accompanying drawings illustrate the possible implementation architecture, functions and operations of the systems, methods and computer program products according to various embodiments of the present invention. In this regard, each box in the flowchart or block diagram can represent a module, program segment, or a part of code, and the above-mentioned module, program segment, or a part of code contains one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in an order different from that 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 flowchart, and the combination of boxes in the block diagram or flowchart, can be implemented with a dedicated hardware-based system that performs the specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.
[0123] The units involved in the embodiments of the present invention may be implemented in software or hardware. The units described may also be provided in a processor. For example, they may be described as follows: a processor comprising: a training unit, a preliminary selection unit, a planned quantity calculation unit, and a selection unit. The names of these units do not, in some cases, limit the units themselves. For example, the training unit may also be described as a "unit that provides a trained classification model to the preliminary selection unit."
[0124] As another aspect, the present invention further provides a computer-readable medium, which may be included in the device described in the above embodiment; or may exist independently and not be assembled into the device. The computer-readable medium carries one or more programs, and when the one or more programs are executed by the device, the device performs the following steps: training a preset classification model based on pre-labeled historical payments; wherein the characteristic information of the historical payments is used as training input data, and the labeling information of the historical payments is used as training label data, and the labeling information includes overdue repayments or on-time repayments; inputting the characteristic information of the payments to be identified in the current period into the trained classification model to obtain the overdue probability of the payments to be identified, and determining the payments to be identified whose overdue probability is greater than a preset first threshold as preliminary selected payments; determining the number of risk payment plans for the current period based on the payment processing resource data of the current period; arranging the preliminary selected payments into a queue in descending order of amount, selecting preliminary selected payments whose overdue probability is greater than a preset dynamic threshold and whose number is equal to the number of risk payment plans in order from the head of the queue to the tail of the queue, and determining the selected preliminary selected payments as risk payments.
[0125] In the technical solution of the embodiment of the present invention, a pre-trained classification model is first used to calculate the overdue probability of each to-be-identified payment in the current cycle, and the to-be-identified payment with an overdue probability greater than a first threshold is determined as a preliminary payment, and the number of risk payment plans for the current cycle is determined based on the payment processing resource data of the current cycle. Thereafter, the preliminary payments are arranged into queues in descending order according to the amount, and preliminary payments with an overdue probability greater than a dynamic threshold and a number equal to the number of risk payment plans are selected in order from the head to the tail of the queue, and finally the selected preliminary payments are determined as risk payments. Through the above steps, efficient and accurate identification of risk payments is achieved based on the machine learning model for early processing, avoiding the generation of payers with overdue risks, reducing the occurrence of collection risks, and improving business collection rates; at the same time, the above identification method takes into account factors such as amount, overdue probability, and the current payment processing resource status of the current cycle, and designs an algorithm to screen out payments that match the current cycle resources and guarantee the maximum total amount and a large overdue probability as risk payments, which has high practicality.
[0126] The above specific embodiments do not limit the scope of protection of the present invention. Those skilled in the art will appreciate that various modifications, combinations, sub-combinations, and substitutions may occur depending on design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention are intended to be included within the scope of protection of the present invention.
Claims
1. A method for identifying risky funds, characterized in that: include: Training a preset classification model based on pre-labeled historical payments; wherein the characteristic information of the historical payments serves as training input data, and the labeling information of the historical payments serves as training label data, wherein the labeling information includes overdue repayments or on-time repayments; Inputting the characteristic information of the to-be-identified accounts in the current period into the trained classification model to obtain the overdue probability of the to-be-identified accounts, and determining the to-be-identified accounts whose overdue probability is greater than a preset first threshold as preliminary selected accounts; Determining the planned amount of risk payments for the current period based on the payment processing resource data for the current period; The preliminary selected funds are arranged into a queue in descending order according to the amount, and the preliminary selected funds whose overdue probability is greater than the preset dynamic threshold and whose quantity is equal to the planned quantity of risk funds are selected in sequence from the head to the tail of the queue, and the selected preliminary selected funds are determined as risk funds.
2. The method according to claim 1, characterized in that The payment processing resource data for the current cycle includes: the total amount of payment processing resources for the current cycle and the average amount of processing resources consumed by a single payment in historical cycles; and determining the number of risk payment plans for the current cycle based on the payment processing resource data for the current cycle includes: The total amount of payment processing resources in the current cycle is divided by the average amount of processing resources consumed for a single payment to obtain the planned number of risk payments in the current cycle.
3. The method according to claim 1, characterized in that The step of selecting, in order from the head to the tail of the queue, the preliminary selected funds whose overdue probability is greater than a preset dynamic threshold and whose quantity is equal to the planned quantity of risky funds includes: Determine the maximum value and the step reduction amount of the dynamic threshold, and construct an arithmetic decreasing sequence with the maximum value as the first value and the step reduction amount as the tolerance; Starting from the maximum value, values in the arithmetic decreasing sequence are sequentially selected in the numerical order of the arithmetic decreasing sequence as the dynamic threshold value until the current dynamic threshold value meets the preset judgment condition; Selecting, in order from the head of the queue to the tail of the queue, the preliminarily selected funds with overdue probabilities greater than the current dynamic threshold as the risky funds; The discrimination conditions are: selecting the preliminary amounts whose overdue probability is greater than the current dynamic threshold in order from the head to the tail of the queue, and determining the number of the selected initial options as the first number; selecting the preliminary amounts whose overdue probability is greater than the next dynamic threshold in order from the head to the tail of the queue, and determining the number of the selected initial options as the second number; the first number is less than or equal to the number of risk amount plans, and the second number is greater than or equal to the number of risk amount plans; wherein, the next dynamic preset is the next value of the current dynamic threshold in the arithmetic decreasing sequence.
4. The method according to claim 1, wherein The characteristic information includes at least one of the following information of the payer: basic information, transaction information, industry information, external environment information, abnormal business information, waybill information, feedback information, and order information; The dynamic threshold is greater than or equal to the first threshold.
5. A risky payment identification device, characterized in that: include: A training unit, configured to train a preset classification model based on pre-labeled historical payments; wherein the characteristic information of the historical payments serves as training input data, and the labeling information of the historical payments serves as training label data, wherein the labeling information includes whether the payment is overdue or on time; A preliminary selection unit is configured to input feature information of the to-be-identified accounts in the current period into the trained classification model to obtain an overdue probability of the to-be-identified accounts, and to determine the to-be-identified accounts whose overdue probability is greater than a preset first threshold as preliminary selected accounts; a planned quantity calculation unit, configured to determine the planned quantity of risk payments for the current period according to the payment processing resource data for the current period; A selection unit is used to arrange the preliminary selected funds into a queue in descending order according to the amount, and select the preliminary selected funds whose overdue probability is greater than a preset dynamic threshold and whose number is equal to the planned number of risk funds in order from the head to the end of the queue, and determine the selected preliminary selected funds as risk funds.
6. The device according to claim 5, characterized in that The payment processing resource data of the current cycle includes: the total amount of payment processing resources of the current cycle and the average amount of payment processing resources consumed by a single payment in the historical cycle; and the planned quantity calculation unit is further used to: The total amount of payment processing resources in the current cycle is divided by the average amount of processing resources consumed for a single payment to obtain the planned number of risk payments in the current cycle.
7. The device according to claim 5, characterized in that The selection unit is further configured to: Determine the maximum value and the step reduction amount of the dynamic threshold, construct an arithmetic decreasing sequence with the maximum value as the first value and the step reduction amount as the tolerance; starting from the maximum value, select values in the arithmetic decreasing sequence in order of the values in the arithmetic decreasing sequence as the dynamic threshold until the current dynamic threshold satisfies the preset judgment condition; select the preliminary selected funds with an overdue probability greater than the current dynamic threshold as the risk funds in order from the head of the queue to the tail of the queue; The discrimination conditions are: selecting the preliminary amounts whose overdue probability is greater than the current dynamic threshold in order from the head to the tail of the queue, and determining the number of the selected initial options as the first number; selecting the preliminary amounts whose overdue probability is greater than the next dynamic threshold in order from the head to the tail of the queue, and determining the number of the selected initial options as the second number; the first number is less than or equal to the number of risk amount plans, and the second number is greater than or equal to the number of risk amount plans; wherein, the next dynamic preset is the next value of the current dynamic threshold in the arithmetic decreasing sequence.
8. The device according to claim 5, characterized in that The characteristic information includes at least one of the following information of the payer: basic information, transaction information, industry information, external environment information, abnormal business information, waybill information, feedback information, and order information; The dynamic threshold is greater than or equal to the first threshold.
9. An electronic device, characterized in that: include: one or more processors; a storage device for storing one or more programs, When the one or more programs are executed by the one or more processors, the one or more processors implement the method according to any one of claims 1 to 4.
10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the method according to any one of claims 1 to 4 is implemented.