Credit quota evaluation method and device, electronic equipment and storage medium

Abstract

The embodiment of the invention provides a credit line evaluation method and device, electronic equipment and a storage medium, and belongs to the technical field of artificial intelligence. The method comprises the following steps: acquiring historical medicine order data of a medicine circulation customer, performing time sequence feature extraction to obtain an order time sequence feature vector, and calculating a customer repayment risk coefficient of the order time sequence feature vector based on the historical medicine order data; based on the historical medicine order data, the customer portrait data and the order time sequence feature vector, performing customer clustering on the medicine circulation customers to obtain a plurality of customer clusters; performing exclusive model construction on each customer group cluster to obtain a plurality of customer order quota prediction models; based on the customer order limit prediction model group, predicting customer order limit prediction data of the medicine circulation customer; and based on the customer order quota prediction data and the customer repayment risk coefficient, calculating the customer credit quota of the medicine circulation customer. According to the embodiment of the invention, the accuracy of credit line evaluation can be improved.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and in particular to a model screening method and apparatus, electronic device and storage medium. Background Technology

[0002] Credit limit assessment is used to evaluate the maximum amount of products a customer can purchase on credit within a certain period based on their historical order data. For example, in the pharmaceutical distribution industry, by assessing the credit limit of pharmaceutical distribution customers, it is possible to avoid operational abnormalities caused by excessively low credit limits, and also to reduce economic losses caused by bad debts due to excessively high credit limits.

[0003] Currently, in the pharmaceutical distribution sector, order amounts from pharmaceutical distribution customers often follow an exponential distribution, while machine learning models are only suitable for normally distributed data. This makes it difficult for machine learning models to predict future order amounts based on existing order data. Consequently, common credit limit assessment methods often involve sales personnel manually evaluating customers' historical pharmaceutical order data to determine each customer's credit limit. However, this method is prone to subjective biases due to the inability to guarantee the professionalism and experience of sales personnel, potentially leading to either excessively high or low credit limits for pharmaceutical distribution customers. Therefore, improving the accuracy of credit limit assessment has become a pressing technical challenge. Summary of the Invention

[0004] The main objective of this application is to provide a credit limit assessment method, apparatus, electronic device, and storage medium, which aims to improve the accuracy of credit limit assessment.

[0005] To achieve the above objectives, a first aspect of this application proposes a credit limit assessment method, the method comprising: Obtain customer profile data and historical pharmaceutical order data from pharmaceutical distribution clients; The historical pharmaceutical order data is subjected to time-series feature extraction to obtain the order time-series feature vector; Based on the historical pharmaceutical order data, the repayment risk of the pharmaceutical distribution customers is calculated to obtain the customer repayment risk coefficient. Based on the historical pharmaceutical order data, the customer profile data, and the order time-series feature vector, the pharmaceutical distribution customers are clustered to obtain a customer group cluster list, wherein the customer group cluster list includes multiple customer group clusters; Based on the multiple customer group clusters, the customer profile data, the historical pharmaceutical order data, and the order time series feature vector, the pre-constructed initial customer order amount prediction model is trained to obtain a customer order amount prediction model group, wherein the customer order amount prediction model group includes multiple customer order amount prediction models, and each customer order amount prediction model corresponds one-to-one with the customer group cluster. Based on the aforementioned customer order amount prediction model group, the order amount of the pharmaceutical distribution customers is predicted to obtain customer order amount prediction data. Based on the customer order limit prediction data and the customer repayment risk coefficient, the credit limit of the pharmaceutical distribution customer is calculated to obtain the customer credit limit.

[0006] In some embodiments, the process of clustering pharmaceutical distribution customers based on the historical pharmaceutical order data, the customer profile data, and the order time-series feature vectors to obtain a customer group cluster list includes: Based on the historical pharmaceutical order data, the activity level of the pharmaceutical distribution customers is assessed to obtain customer activity level; Based on the customer activity level, the pharmaceutical distribution customers are classified into active order customers and dormant order customers. Based on the historical pharmaceutical order data, a mixture Gaussian clustering method was used to identify the active customers who placed orders, resulting in an active customer clustering scheme. Based on the active customer clustering scheme, the customer profile data, and the order time-series feature vector, the pre-constructed initial customer classification model is trained to obtain the target customer classification model; Based on the target customer classification model, the customer profile data, and the order time-series feature vector, the dormant customers who have placed orders are classified to obtain a dormant customer clustering scheme; Based on the active customer clustering scheme and the dormant customer clustering scheme, the customer group cluster list of the pharmaceutical distribution customers is determined.

[0007] In some embodiments, the step of performing mixture Gaussian clustering on the active customers based on the historical pharmaceutical order data to obtain an active customer clustering scheme includes: Based on the active customers who placed orders, the historical pharmaceutical order data was filtered to obtain active pharmaceutical order data. Perform normal distribution hybrid clustering on the active pharmaceutical order data to obtain a normal distribution clustering scheme for the target data; Based on the normal distribution clustering scheme of the target data, the active customers who placed orders are classified to obtain the clustering scheme of active customers.

[0008] In some embodiments, performing normal distribution hybrid clustering on the active pharmaceutical order data to obtain a normal distribution clustering scheme for the target data includes: Based on a preset list of clustering categories, the active medical order data is clustered to obtain a list of data clustering schemes. The list of clustering categories includes multiple clustering categories, and the list of data clustering schemes includes multiple normal distribution clustering schemes. The normal distribution clustering schemes correspond one-to-one with the number of clustering categories. The clustering quality of the normally distributed clustering scheme is evaluated to obtain clustering quality evaluation data; Based on the clustering quality assessment data, the list of data clustering schemes is screened to obtain the target data normal distribution clustering scheme.

[0009] In some embodiments, the method for training a pre-constructed initial customer order amount prediction model based on the multiple customer group clusters, the customer profile data, the historical medical order data, and the order time-series feature vector to obtain a customer order amount prediction model group includes: The historical pharmaceutical order data is classified to obtain pharmaceutical order training data and pharmaceutical order test data; Based on the customer group cluster, the customer profile data, the pharmaceutical order training data, and the order time series feature vector, the parameters of the initial customer order amount prediction model are adjusted to obtain the first customer order amount prediction model. The training data for the medical orders is transformed to obtain transformed order data; Based on the customer group cluster, the customer profile data, the conversion order data, and the order time series feature vector, the parameters of the initial customer order amount prediction model are adjusted to obtain the second customer order amount prediction model. Based on the pharmaceutical order test data, the performance of the first customer order amount prediction model and the second customer order amount prediction model is compared to obtain model performance comparison information. Based on the model performance comparison information, a model selection is performed between the first customer order amount prediction model and the second customer order amount prediction model to obtain the customer order amount prediction model.

[0010] In some embodiments, the calculation of repayment risk for pharmaceutical distribution customers based on the historical pharmaceutical order data to obtain a customer repayment risk coefficient includes: Based on pre-constructed risk indicators and the historical pharmaceutical order data, a risk assessment is conducted on the pharmaceutical distribution customers to obtain a repayment risk score; The repayment risk score is mapped to a coefficient to obtain the customer's repayment risk coefficient.

[0011] In some embodiments, the step of predicting the order amount of the pharmaceutical distribution customer based on the customer order amount prediction model group to obtain customer order amount prediction data includes: Based on the customer group cluster list, the target customer group cluster of the pharmaceutical distribution customer is determined; Based on the pharmaceutical distribution customers, the customer profile data is used to locate the target customer profile, and the historical pharmaceutical order data is used to locate the target pharmaceutical order data. Based on the target customer group cluster, the customer order amount prediction model group is screened to obtain the customer order amount prediction model; Based on the customer order amount prediction model, the target customer profile and the target pharmaceutical order data are used to output predictions to obtain the customer order amount prediction data.

[0012] To achieve the above objectives, a second aspect of this application provides a credit limit assessment device, the device comprising: The data acquisition module is used to acquire customer profile data and historical pharmaceutical order data of pharmaceutical distribution customers; The feature extraction module is used to extract time-series features from the historical medical order data to obtain order time-series feature vectors; The risk calculation module is used to calculate the repayment risk of the pharmaceutical distribution customers based on the historical pharmaceutical order data, and obtain the customer repayment risk coefficient. The customer clustering module is used to perform customer clustering on the pharmaceutical distribution customers based on the historical pharmaceutical order data, the customer profile data, and the order time-series feature vector, to obtain a customer group cluster list, wherein the customer group cluster list includes multiple customer group clusters; The model training module is used to train a pre-constructed initial customer order amount prediction model based on the multiple customer group clusters, the customer profile data, the historical medical order data, and the order time series feature vector, to obtain a customer order amount prediction model group, wherein the customer order amount prediction model group includes multiple customer order amount prediction models, and each customer order amount prediction model corresponds one-to-one with the customer group cluster. The order limit prediction module is used to predict the order limits of the pharmaceutical distribution customers based on the customer order limit prediction model group, and obtain customer order limit prediction data. The credit limit calculation module is used to calculate the credit limit of the pharmaceutical distribution customer based on the customer order limit prediction data and the customer repayment risk coefficient, and obtain the customer credit limit.

[0013] To achieve the above objectives, a third aspect of the present application provides an electronic device, the electronic device including a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the method described in the first aspect.

[0014] To achieve the above objectives, a fourth aspect of the present application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the method described in the first aspect.

[0015] This application embodiment acquires customer profile data and historical pharmaceutical order data of pharmaceutical distribution customers, extracts time-series features from the historical pharmaceutical order data to obtain order time-series feature vectors, and calculates repayment risk for pharmaceutical distribution customers based on historical pharmaceutical order data to obtain customer repayment risk coefficients. This clarifies the transaction patterns and repayment capabilities of pharmaceutical distribution customers. Furthermore, based on historical pharmaceutical order data, customer profile data, and order time-series feature vectors, pharmaceutical distribution customers are clustered to obtain a customer group cluster list. This customer group cluster list includes multiple customer group clusters, which solves the problem that traditional machine learning models cannot handle pharmaceutical distribution customers. In addition, by using multiple feature data to cluster pharmaceutical distribution customers, the accuracy of the customer group cluster list is improved. Furthermore, based on multiple customer group clusters, customer profile data, and historical pharmaceutical orders... Using data and order time-series feature vectors, a pre-constructed initial customer order limit prediction model is trained to obtain a customer order limit prediction model group. This group comprises multiple customer order limit prediction models, each corresponding one-to-one with a customer group cluster. This improves the adaptability of the customer order limit prediction model to the customer group cluster, thereby enhancing the prediction accuracy. Finally, based on this customer order limit prediction model group, order limits are predicted for pharmaceutical distribution customers, yielding customer order limit prediction data. Based on this data and the customer repayment risk coefficient, credit limits are calculated for pharmaceutical distribution customers, resulting in their creditworthiness. This not only increases the assessment dimensions for credit limit evaluation of pharmaceutical distribution customers but also reduces the involvement of manual assessment, thus improving the accuracy of credit limit evaluation. Attached Figure Description

[0016] Figure 1 This is a flowchart of the credit limit assessment method provided in the embodiments of this application; Figure 2 yes Figure 1 The flowchart of step S103 in the process; Figure 3 yes Figure 1 The flowchart of step S104 in the process; Figure 4 yes Figure 3 The flowchart of step S303 in the process; Figure 5 yes Figure 4 The flowchart of step S402 in the document; Figure 6 yes Figure 1 The flowchart of step S105 in the process; Figure 7 yes Figure 1 The flowchart of step S106 in the process; Figure 8 This is a schematic diagram of the structure of the credit limit assessment device provided in the embodiments of this application; Figure 9 This is a schematic diagram of the hardware structure of the electronic device provided in the embodiments of this application. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0018] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing embodiments of this application only and is not intended to limit this application.

[0019] First, let's analyze some of the terms used in this application: Exponential distribution: The exponential distribution is a continuous probability distribution mainly used to describe the time intervals between independent random events, such as the time it takes for equipment to operate without failure, the time between two customer orders, and the time it takes for a service window to process a single request. The core of the exponential distribution is determined by the rate parameter λ, which represents the average number of times an event occurs per unit time. The larger λ is, the more frequently events occur, and the shorter the corresponding time interval; conversely, the smaller λ is, the longer the interval.

[0020] Gaussian Mixture Model (GM): A GM is a probabilistic model formed by a weighted combination of multiple normal distributions. It falls under the category of unsupervised learning and is primarily used to fit complex probability distributions of data or to perform clustering tasks. A GM contains three key parameters: the mean, variance / covariance of each normal distribution, and the weights of each normal distribution.

[0021] Artificial intelligence (AI) is a new branch of computer science that studies, develops, and applies theories, methods, technologies, and systems to simulate, extend, and expand human intelligence. It aims to understand the essence of intelligence and produce intelligent machines that can react in a way similar to human intelligence. Research in this field includes robotics, speech recognition, image recognition, natural language processing, and expert systems. AI can simulate human consciousness and thought processes. Furthermore, AI utilizes digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceiving the environment, acquiring knowledge, and using that knowledge to achieve optimal results.

[0022] Credit limit assessment is used to evaluate the maximum amount of products a customer can purchase on credit within a certain period based on their historical order data. For example, in the pharmaceutical distribution industry, by assessing the credit limit of pharmaceutical distribution customers, it is possible to avoid operational abnormalities caused by excessively low credit limits, and also to reduce economic losses caused by bad debts due to excessively high credit limits.

[0023] Currently, in the pharmaceutical distribution sector, order amounts from pharmaceutical distribution customers often follow an exponential distribution, while machine learning models are only suitable for normally distributed data. This makes it difficult for machine learning models to predict future order amounts based on existing order data. Consequently, common credit limit assessment methods often involve sales personnel manually evaluating customers' historical pharmaceutical order data to determine each customer's credit limit. However, this method is prone to subjective biases due to the inability to guarantee the professionalism and experience of sales personnel, potentially leading to either excessively high or low credit limits for pharmaceutical distribution customers. Therefore, improving the accuracy of credit limit assessment has become a pressing technical challenge.

[0024] Based on this, embodiments of this application provide a credit limit assessment method and apparatus, electronic device and storage medium, aiming to improve the accuracy of credit limit assessment.

[0025] The credit limit assessment method, apparatus, electronic device, and storage medium provided in this application are specifically described through the following embodiments. First, the credit limit assessment method in this application is described.

[0026] The embodiments of this application can acquire and process relevant data based on artificial intelligence technology. Artificial intelligence (AI) refers to the theories, methods, technologies, and application systems that use digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to obtain optimal results.

[0027] The credit limit assessment method provided in this application relates to the field of artificial intelligence technology. This method can be applied to a terminal, a server, or software running on either a terminal or a server. In some embodiments, the terminal can be a smartphone, tablet, laptop, desktop computer, etc.; the server can be configured as an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms; the software can be an application implementing the credit limit assessment method, but is not limited to the above forms.

[0028] This application can be used in a wide variety of general-purpose or special-purpose computer system environments or configurations. Examples include: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics, network PCs, minicomputers, mainframe computers, and distributed computing environments including any of the above systems or devices. This application can be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc., that perform specific tasks or implement specific abstract data types. This application can also be practiced in distributed computing environments where tasks are performed by remote processing devices connected via a communication network. In distributed computing environments, program modules can reside in local and remote computer storage media, including storage devices.

[0029] Figure 1 This is an optional flowchart of the model screening method provided in the embodiments of this application. Figure 1 The method may include, but is not limited to, steps S101 to S107.

[0030] Step S101: Obtain customer profile data and historical pharmaceutical order data of pharmaceutical distribution customers; Step S102: Extract time-series features from historical pharmaceutical order data to obtain order time-series feature vectors; Step S103: Based on historical pharmaceutical order data, calculate the repayment risk of pharmaceutical distribution customers to obtain the customer repayment risk coefficient; Step S104: Based on historical pharmaceutical order data, customer profile data, and order time-series feature vectors, pharmaceutical distribution customers are clustered to obtain a list of customer groups, which includes multiple customer groups. Step S105: Based on multiple customer group clusters, customer profile data, historical pharmaceutical order data and order time series feature vectors, train the pre-constructed initial customer order amount prediction model to obtain a customer order amount prediction model group. The customer order amount prediction model group includes multiple customer order amount prediction models, and each customer order amount prediction model corresponds one-to-one with a customer group cluster. Step S106: Based on the customer order amount prediction model group, predict the order amount of pharmaceutical distribution customers to obtain customer order amount prediction data; Step S107: Based on the customer order limit prediction data and the customer repayment risk coefficient, calculate the credit limit of the pharmaceutical distribution customer to obtain the customer credit limit.

[0031] Steps S101 to S107 of this embodiment involve acquiring customer profile data and historical pharmaceutical order data of pharmaceutical distribution customers, extracting time-series features from the historical pharmaceutical order data to obtain order time-series feature vectors, and calculating repayment risk for pharmaceutical distribution customers based on the historical pharmaceutical order data to obtain customer repayment risk coefficients. This clarifies the transaction patterns and repayment capabilities of pharmaceutical distribution customers. Furthermore, based on historical pharmaceutical order data, customer profile data, and order time-series feature vectors, pharmaceutical distribution customers are clustered to obtain a customer group cluster list. This customer group cluster list includes multiple customer group clusters, which solves the problem that traditional machine learning models cannot handle pharmaceutical distribution customers. In addition, by using multiple feature data to cluster pharmaceutical distribution customers, the accuracy of the customer group cluster list is improved. Furthermore, based on multiple customer group clusters and customer profiles... Using data, historical pharmaceutical order data, and order time-series feature vectors, a pre-constructed initial customer order limit prediction model is trained to obtain a customer order limit prediction model group. This group comprises multiple customer order limit prediction models, each corresponding one-to-one with a customer group cluster. This improves the adaptability of the model to the customer group, thereby enhancing the prediction accuracy. Finally, based on this model group, order limits are predicted for pharmaceutical distribution customers, yielding predicted order limit data. Then, based on this data and the customer repayment risk coefficient, credit limits are calculated for these customers. This not only increases the assessment dimensions for credit limit evaluation of pharmaceutical distribution customers but also reduces the involvement of manual assessment, thus improving the accuracy of credit limit evaluation.

[0032] In step S101 of some embodiments, pharmaceutical distribution customers refer to downstream customers in the pharmaceutical field who have business cooperation with pharmaceutical distribution companies. Among them, pharmaceutical distribution companies may be companies that manufacture drugs or medical devices, and downstream customers may be a municipal first people's hospital, a chain of private medical examination hospitals, a commercial company engaged in pharmaceutical wholesale, or a community convenience pharmacy.

[0033] Customer profile data refers to a set of data used to characterize the status of pharmaceutical distribution customers. This set of data typically includes information such as customer type, customer configuration, cooperation duration, lifecycle status, and historical cooperation scale. Among them, lifecycle status can be divided into new customers, active customers, old customers, and churned customers. For example, in the customer management scenario of a pharmaceutical distribution company, the customer profile data of a community pharmacy could be: Customer type: retail pharmacy; Cooperation duration: 2 years; Lifecycle status: active customer; Historical annual purchase amount: 800,000 yuan.

[0034] Historical pharmaceutical order data refers to various data related to orders generated by pharmaceutical distribution customers in their past cooperation with pharmaceutical distribution companies, such as the number of orders, order frequency, order time, order number, order amount, payment received, return amount, and outstanding amount.

[0035] In this embodiment, by searching for characteristics such as the number of patient beds, surgical drug needs, departmental structure, registered capital, qualification level, specialty department configuration, specialty category, drug procurement participation, and medical insurance designation of pharmaceutical distribution clients, the basic information of pharmaceutical distribution clients in multiple dimensions can be clarified. Then, customer profile data can be generated based on this multi-dimensional basic information. For example, for a tertiary hospital A, the customer profile data may include the number of hospital beds, number of surgeries, type and number of hospital departments, registered capital, tertiary hospital level, specialty departments, and whether the hospital is a non-specialty hospital. In addition to the centralized procurement of pharmaceutical products and the use of medical insurance information, this application embodiment can also calculate the outstanding amount of pharmaceutical distribution customers in each time window based on the order amount, payment amount, return amount, order time, payment time, and return time in the transaction records. The length of each time window can be a length commonly used in the pharmaceutical distribution field, such as a natural month, week, or quarter. Furthermore, by calculating the outstanding amount of pharmaceutical distribution customers, the order data of pharmaceutical distribution products of pharmaceutical distribution customers is enriched, providing more reference features for customer clustering and future order amount prediction of pharmaceutical distribution customers, thereby improving the accuracy of customer clustering and future order amount prediction.

[0036] In step S102 of some embodiments, the order time-series feature vector refers to the numerical vector representation of the trend characteristics, periodic characteristics, etc., exhibited by historical pharmaceutical order data when it exists in the form of time-series data. For example, if the historical pharmaceutical order data of pharmaceutical distribution customer a shows that a placed 1 order for pharmaceutical products in January, 2 orders for pharmaceutical products in February, and 3 orders for pharmaceutical products in March, then the order time-series feature vector of this historical pharmaceutical order data can be a numerical vector representation of the gradually increasing trend of the number of orders placed by pharmaceutical distribution customer a.

[0037] In this embodiment of the application, the data of each dimension in the historical medical order data can be converted into time series data based on the time tags such as the order date and payment date carried by the historical medical order data. For example, the discrete number of customer orders can be converted into a data combination arranged by the order date, i.e., the time series data of the number of orders. The discrete customer payment amount can also be converted into a data combination arranged by the payment date, i.e., the time series data of the payment amount. Furthermore, the data trend and data periodicity of the time series data of each dimension in the historical medical order data can be represented in the form of a numerical vector to obtain the order time series feature vector.

[0038] In step S103 of some embodiments, the customer repayment risk coefficient refers to the quantitative value of the pharmaceutical distribution customer's ability to repay the outstanding amount in a repayment scenario. It usually takes a value range of 0-1. When the customer repayment risk coefficient is closer to 1, it means that the pharmaceutical distribution customer has a stronger ability to repay the outstanding amount and a lower risk. When the customer repayment risk coefficient is closer to 0, it means that the pharmaceutical distribution customer has a weaker ability to repay the outstanding amount and a higher risk.

[0039] This application embodiment can estimate the repayment ability of each pharmaceutical distribution customer based on the historical pharmaceutical order data of each customer according to the pre-set risk indicators. Furthermore, the repayment ability of each pharmaceutical distribution customer is mapped to the risk coefficient space to determine the customer repayment risk coefficient of each pharmaceutical distribution customer.

[0040] For details, please refer to Figure 2 In some embodiments, step S103 may include, but is not limited to, steps S201 to S202: Step S201: Based on pre-built risk indicators and historical pharmaceutical order data, conduct risk assessment on pharmaceutical distribution customers to obtain repayment risk scores; Step S202: Map the repayment risk score to obtain the customer's repayment risk coefficient.

[0041] In step S201 of some embodiments, the risk indicator refers to the core quantitative indicator used to measure the repayment risk of pharmaceutical distribution customers, such as overdue amount, overdue days, customer size, etc.

[0042] The repayment risk score refers to the quantitative value obtained by pharmaceutical distribution customers during the repayment risk assessment process. Therefore, the repayment risk score can directly reflect the degree of repayment risk of pharmaceutical distribution customers.

[0043] In this embodiment, the risk indicators can be customer repayment risk assessment indicators set by experts in the pharmaceutical distribution industry based on their experience. In addition, to ensure the comprehensiveness of the risk indicators, this application can also add assessment indicators given by the risk model on the basis of the customer repayment risk assessment indicators specified by the experts, thereby forming a comprehensive risk indicator. For example, the customer repayment risk assessment indicators set by experts based on their own experience include indicators such as overdue amount and overdue days. The risk model can clarify the customer scale, customer repayment speed and other dimensions of pharmaceutical distribution customers through historical pharmaceutical order data, which are also important indicators for assessing the repayment ability of pharmaceutical distribution customers. At this time, the risk indicator can be a combination of indicators such as overdue amount and overdue days and indicators such as customer scale and customer repayment speed.

[0044] It's important to understand that the risk model is a scorecard model. The risk model takes whether a pharmaceutical distribution customer is overdue as the model output target and uses logistic regression to classify pharmaceutical distribution customers. This allows the trained risk model to obtain a risk score indicating the probability of a customer being overdue. The risk score ranges from 0 to 1, representing the level of overdue risk for pharmaceutical distribution customers. Specifically, the closer the risk score is to 1, the higher the overdue risk for pharmaceutical distribution customers.

[0045] It is important to understand that after determining the risk indicators, this application can set corresponding risk assessment rules based on different risk indicators. For example, when the risk indicator is the number of overdue days, the risk assessment rule could be: overdue days between one day and one month, risk score of 1; overdue days between one month and one quarter, risk score of 2; overdue days of one quarter or more, risk score of 3. Furthermore, by using the risk assessment rules to evaluate the historical pharmaceutical order data of pharmaceutical distribution customers, the risk score for each risk indicator can be obtained. Finally, for... The risk scores for each risk indicator are weighted and summed or averaged. For example, if pharmaceutical distribution customer C has a risk score of 2 for overdue amount and a risk score of 3 for overdue days, then based on the pre-set weight of 0.8 for overdue amount and 0.2 for overdue days, the repayment risk score for pharmaceutical distribution customer C can be calculated as 2×0.8+3×0.2=2.2. Alternatively, the repayment risk score for pharmaceutical distribution customer C can be obtained by averaging the risk scores of the two risk indicators, namely 2×0.5+3×0.5=2.5.

[0046] In step S202 of some embodiments, the repayment risk score of pharmaceutical distribution customers can be converted into the customer repayment risk coefficient of pharmaceutical distribution customers by a pre-constructed mapping table or mapping function of risk coefficient and risk score. The mapping table of risk coefficient and risk score can be a table containing repayment risk score and repayment risk coefficient, and each repayment risk score corresponds to a repayment risk coefficient. The mapping function of risk coefficient and risk score can be a univariate polynomial equation with repayment risk score as independent variable and repayment risk coefficient as dependent variable.

[0047] Steps S201 to S202, as illustrated in this embodiment, provide a standardized framework for assessing the repayment risk of pharmaceutical distribution customers by pre-constructing risk indicators. Furthermore, by combining historical pharmaceutical order data of pharmaceutical distribution customers, risk assessment is conducted, ensuring that the obtained repayment risk score is supported by objective data and avoiding the bias of relying on subjective experience. In addition, the accuracy of risk assessment is improved. Finally, the repayment risk score is mapped to obtain the customer's repayment risk coefficient, unifying the output format of the repayment risk assessment results and enabling it to be directly applied to the business scenario of credit limit calculation, thereby improving the efficiency of credit limit assessment.

[0048] In step S104 of some embodiments, the customer group cluster list refers to the set of all customer group clusters formed after pharmaceutical distribution customers are clustered. Here, a customer group cluster refers to a set of pharmaceutical distribution customers with the same or similar characteristics. For example, when pharmaceutical distribution customer a and pharmaceutical distribution customer b have similar characteristics in terms of transaction volume, repayment performance, etc., pharmaceutical distribution customer a and pharmaceutical distribution customer b can be pharmaceutical distribution customers in the same customer group cluster.

[0049] This application embodiment can assess the activity level of pharmaceutical distribution customers based on historical pharmaceutical order data, and classify them into active order customers and dormant order customers. Active order customers are those who have placed orders for pharmaceutical products within a specified time period, while dormant order customers are those who have not placed orders for pharmaceutical products within a specified time period. It should be noted that the specified time period can be a month, a quarter, or half a year, etc. Further, a Gaussian mixture clustering scheme is applied to active order customers to obtain an active customer clustering scheme. Then, based on the active customer clustering scheme, customer profile data of active order customers, and order time-series feature vectors, a target customer classification model is trained. This target customer classification model is then used to classify dormant order customers to obtain a dormant customer clustering scheme. Finally, by combining the active customer clustering scheme and the dormant customer clustering scheme, the customer group cluster to which each pharmaceutical distribution customer belongs can be determined, thus forming a customer group cluster list.

[0050] For details, please refer to Figure 3 In some embodiments, step S104 may include, but is not limited to, steps S301 to S306: Step S301: Based on historical pharmaceutical order data, assess the activity level of pharmaceutical distribution customers to obtain customer activity level; Step S302: Based on customer activity, classify pharmaceutical distribution customers to obtain active order customers and dormant order customers; Step S303: Based on historical pharmaceutical order data, perform mixed Gaussian clustering on active customers to obtain an active customer clustering scheme; Step S304: Based on the active customer clustering scheme, customer profile data and order time-series feature vectors, train the pre-built initial customer classification model to obtain the target customer classification model; Step S305: Based on the target customer classification model, customer profile data and order time-series feature vectors, classify dormant customers who have placed orders to obtain a dormant customer clustering scheme; Step S306: Based on the active customer clustering scheme and the dormant customer clustering scheme, determine the customer group cluster list of pharmaceutical distribution customers.

[0051] In step S301 of some embodiments, customer activity refers to the quantitative result of the activity assessment of pharmaceutical distribution customers. Customer activity can be presented in the form of high level, low level or specific score of 0-100. It should be noted that customer activity can intuitively reflect the frequency and scale of transactions between pharmaceutical distribution customers and pharmaceutical distribution companies in the near future.

[0052] This application embodiment can extract order activity data related to activity from historical pharmaceutical order data. For example, the number of orders placed in the past 30 / 90 / 180 days, the total order amount in the past 30 / 90 / 180 days, and the interval between consecutive orders are all from historical pharmaceutical order data. Furthermore, based on pre-set activity indicators and indicator evaluation rules, for example, for the activity indicator "number of orders," the evaluation rule could be that each additional order increases the activity score by 1 point. By evaluating the extracted order activity data, a quantitative result of the order activity data's activity evaluation can be obtained, i.e., customer activity level.

[0053] In step S302 of some embodiments, an active customer who places an order refers to a pharmaceutical distribution customer whose customer activity level exceeds an activity threshold, wherein the activity threshold is the minimum standard used to assess whether a pharmaceutical distribution customer is active.

[0054] Dormant customers refer to pharmaceutical distribution customers whose activity level has not exceeded the activity threshold.

[0055] In this embodiment, the customer activity level of each pharmaceutical distribution customer is compared with a pre-set activity threshold. If the customer activity level is less than or equal to the activity threshold, the customer is determined to be a dormant customer. If the customer activity level is greater than the activity threshold, the customer is determined to be an active customer. In one embodiment, if the activity threshold is 0, the customer will only be determined to be a dormant customer when the customer activity level is 0, and only when the customer activity level is greater than 0 will the customer be determined to be an active customer. For example, the historical medical records of pharmaceutical distribution customer D... The pharmaceutical order data shows that pharmaceutical distribution customer D has placed 0 orders in the past 180 days, has a total order amount of 0 in the past 180 days, and has a continuous order interval of 1 year. Based on the activity level calculation, pharmaceutical distribution customer D's customer activity level is 0. At this time, pharmaceutical distribution customer D will be judged as a dormant customer. The historical pharmaceutical order data of pharmaceutical distribution customer E shows that pharmaceutical distribution customer E has placed 1 order in the past 180 days, has a total order amount of 120,000 in the past 180 days, and has a continuous order interval of 70 days. Based on the activity level calculation, pharmaceutical distribution customer E's customer activity level is 5. At this time, pharmaceutical distribution customer E will be judged as an active customer.

[0056] In step S303 of some embodiments, the active customer clustering scheme refers to the set of all active customer groups formed after clustering active customers who have placed orders. The active customer clustering scheme includes the composition information of active customers who have placed orders for each active customer group and the cluster number information of the active customer group.

[0057] This application embodiment can filter active pharmaceutical order data from historical pharmaceutical order data based on active customers who have placed orders. Furthermore, it can perform normal distribution mixed clustering on the active pharmaceutical order data to obtain a target data clustering scheme, and then classify active customers based on this to obtain an active customer clustering scheme.

[0058] For details, please refer to Figure 4 In some embodiments, step S303 may include, but is not limited to, steps S401 to S403: Step S401: Based on active customers who have placed orders, filter historical pharmaceutical order data to obtain active pharmaceutical order data; Step S402: Perform normal distribution mixed clustering on the active pharmaceutical order data to obtain the normal distribution clustering scheme for the target data; Step S403: Based on the normal distribution clustering scheme of the target data, classify active customers who have placed orders to obtain an active customer clustering scheme.

[0059] In step S401 of some embodiments, active pharmaceutical order data refers to historical pharmaceutical order data of active customers.

[0060] In this embodiment of the application, all order records of active customers can be matched from the historical pharmaceutical order data, i.e., active pharmaceutical order data, based on the matching relationship between active customers and historical pharmaceutical order data.

[0061] It's important to know that active customers and historical pharmaceutical order data can be matched using unique identifiers such as customer IDs, making it easier to filter active pharmaceutical order data from historical pharmaceutical order data.

[0062] In step S402 of some embodiments, the target data normal distribution clustering scheme refers to the clustering scheme in which the active medical order data exists in the form of a normal distribution. It should be noted that the target data normal distribution clustering scheme includes the number of clusters and the active medical order data corresponding to each cluster.

[0063] This application embodiment can perform multiple clusterings on active pharmaceutical order data based on a preset list of clustering categories containing multiple clustering categories, to obtain a data normal distribution clustering scheme corresponding to the number of clustering categories. Furthermore, the clustering quality of the data normal distribution clustering scheme is evaluated, and a target data normal distribution clustering scheme is selected from the data normal distribution clustering schemes based on the clustering quality.

[0064] For details, please refer to Figure 5 In some embodiments, step S402 may include, but is not limited to, steps S501 to S503: Step S501: Based on the preset list of clustering categories, cluster the active pharmaceutical order data to obtain a list of data clustering schemes. The list of clustering categories includes multiple clustering categories, and the list of data clustering schemes includes multiple normal distribution clustering schemes. The normal distribution clustering schemes correspond one-to-one with the number of clustering categories. Step S502: Evaluate the clustering quality of the normal distribution clustering scheme to obtain clustering quality evaluation data; Step S503: Based on the clustering quality assessment data, the list of data clustering schemes is screened to obtain the target data normal distribution clustering scheme.

[0065] In step S501 of some embodiments, the cluster category number list refers to a set containing multiple possible cluster category numbers. The cluster category number list is used to provide different cluster number selections for clustering active medical order data. For example, when the cluster category number list is (2, 3, 4, 5), it means that the number of clusters for active medical order data can be 2, 3, 4 or 5.

[0066] The number of cluster categories refers to the number of individual clusters in the list of cluster categories, that is, the number of groups that the active pharmaceutical order data is to be divided into during clustering.

[0067] A normal distribution clustering scheme refers to a scheme obtained by clustering active pharmaceutical order data for a specific number of cluster categories. It is important to know that each number of cluster categories corresponds to an independent normal distribution clustering scheme, and the active pharmaceutical order data within each cluster approximately follows a normal distribution.

[0068] In this embodiment, active pharmaceutical order data can be clustered multiple times using a Gaussian mixture model based on a pre-defined list of clustering categories containing multiple clustering categories, in order to form multiple data normal distribution clustering schemes that correspond one-to-one with the number of clustering categories.

[0069] In step S502 of some embodiments, clustering quality assessment data refers to a quantitative value that measures the quality of a normal distribution clustering scheme. Specifically, the larger the clustering quality assessment data, the better the quality of the normal distribution clustering scheme, that is, the more correct the normal distribution clustering scheme.

[0070] In this embodiment, by calculating the silhouette coefficient, CH index, and DB index, which represent the clustering quality of each data normal distribution clustering scheme, the silhouette coefficient score, CH index score, and DB index score of each data normal distribution clustering scheme can be obtained. Furthermore, any one or more of the silhouette coefficient score, CH index score, and DB index score can be selected and combined as the clustering quality evaluation rule for the data normal distribution clustering scheme. In one embodiment, when the clustering quality evaluation rule is to use the silhouette coefficient as the clustering quality evaluation index for the data normal distribution clustering scheme... When evaluating clustering quality, the silhouette coefficient scores of each normally distributed clustering scheme can be used as clustering quality assessment data. For example, when the silhouette coefficient score of normally distributed clustering scheme F is 0.3, the CH index score is 0.32, and the DB index score is 0.31, the clustering quality assessment data for normally distributed clustering scheme F can be a silhouette coefficient score of 0.3. When the clustering quality assessment rule uses the silhouette coefficient, CH index, and DB index as clustering quality assessment indicators for normally distributed clustering schemes, the silhouette coefficient score, CH index score, and DB index score of each normally distributed clustering scheme can be evaluated. We perform a weighted summation to obtain the clustering quality assessment data for each data normal distribution clustering scheme. For example, when the silhouette coefficient score of the data normal distribution clustering scheme F is 0.3, the CH index score is 0.32, and the DB index score is 0.31, and the weights of the silhouette coefficient score, the CH index score, and the DB index score in the clustering quality assessment data are 0.2, 0.3, and 0.5 respectively, the clustering quality assessment data for the data normal distribution clustering scheme F can be 0.3×0.2+0.32×0.3+0.31×0.5=0. 311. Alternatively, the average of the silhouette coefficient score, CH index score, and DB index score of each normal distribution clustering scheme can be calculated to obtain the clustering quality assessment data of each normal distribution clustering scheme. For example, when the silhouette coefficient score of the normal distribution clustering scheme F is 0.30, the CH index score is 0.32, and the DB index score is 0.31, and the silhouette coefficient score, CH index score, and DB index score have the same weight in the clustering quality assessment data, the clustering quality assessment data of the normal distribution clustering scheme F can be (0.3 + 0.32 + 0.32) ÷ 3 = 0.31.

[0071] In step S503 of some embodiments, when clustering quality assessment data is obtained, and the larger the clustering quality assessment data is, the better the clustering quality of the data normal distribution clustering scheme is, the data normal distribution clustering scheme with the largest clustering quality assessment data can be selected from the list of data clustering schemes as the target data normal distribution clustering scheme based on the size of the clustering quality assessment data. When there are multiple data normal distribution clustering schemes with the largest clustering quality assessment data, the selection of the target data normal distribution clustering scheme can be carried out manually.

[0072] Steps S501 to S503 of this embodiment involve clustering active pharmaceutical order data using a pre-defined list of clustering categories with varying numbers of clustering types. This yields multiple sets of normally distributed clustering schemes, avoiding the bias and deviation that can result from clustering with a single cluster, thus improving the clustering accuracy of active pharmaceutical order data. Secondly, the normally distributed clustering schemes are evaluated for clustering quality, resulting in clustering quality evaluation data that replaces the user's subjective experience, improving the reliability and accuracy of the evaluation data. Finally, based on the clustering quality evaluation data, the list of clustering schemes is filtered to obtain a target normally distributed clustering scheme. This ensures that the target normally distributed clustering scheme achieves both high similarity and compact distribution within clusters, as well as significant and clear differences between clusters, accurately adapting to the actual distribution characteristics of active pharmaceutical order data and improving the clustering accuracy of active pharmaceutical order data.

[0073] It is also important to know that active pharmaceutical order data contains order data from multiple dimensions. The dimensions used for normal distribution mixed clustering can be, but are not limited to, order count, order frequency, order amount, etc. Therefore, by integrating the multidimensional active pharmaceutical order data of each pharmaceutical distribution customer into the form of a feature vector, and then clustering the active pharmaceutical order data in the form of the feature vector of each pharmaceutical distribution customer, a normal distribution clustering scheme for the target data can be obtained.

[0074] In step S403 of some embodiments, after obtaining the target data normal distribution clustering scheme of active pharmaceutical order data, the active customers can also be assigned to different customer groups according to the matching relationship between active pharmaceutical order data and active customers who place orders, thereby forming an active customer clustering scheme.

[0075] Steps S401 to S403 of this embodiment involve filtering historical pharmaceutical order data based on active customers to obtain active pharmaceutical order data. This reduces the impact of historical pharmaceutical order data from dormant customers on clustering, thereby improving the accuracy of clustering active customers. Secondly, a normal distribution hybrid clustering method is applied to the active pharmaceutical order data to obtain a normal distribution clustering scheme for the target data. This makes the distribution of the clustered active pharmaceutical order data approximate a normal distribution, making the clustered active pharmaceutical order data more suitable for traditional machine learning models, thus improving the model's accuracy. Finally, based on the normal distribution clustering scheme for the target data, active customers are classified to obtain an active customer clustering scheme, ensuring the accuracy of the active customer clustering scheme.

[0076] In step S304 of some embodiments, the initial customer classification model refers to a model that can only classify customers based on a single characteristic of pharmaceutical distribution customers. For example, the initial customer classification model can classify pharmaceutical distribution customers based on the order amount of pharmaceutical products placed by the pharmaceutical distribution customers, but it cannot classify pharmaceutical distribution customers by combining the order amount of pharmaceutical products placed by the pharmaceutical distribution customers with profile features such as the scale of patient beds, the demand for medication related to surgical business, and the department configuration structure. It should be noted that the initial customer classification model can be an XGBoost or MLP model.

[0077] The target customer classification model refers to a model that can cluster active customers based on customer profile data of pharmaceutical distribution customers, such as patient bed size, surgical business-related drug needs, and department configuration structure, as well as order time-series feature vectors such as order amount of pharmaceutical products, and output active customer clustering scheme.

[0078] In this embodiment, active customer profiles can be selected from customer profile data based on active customers who have placed orders. At the same time, active order feature vectors corresponding to active customers can be selected from order time-series feature vectors. Furthermore, the composition and number of active customers in each active customer cluster included in the active customer clustering scheme, as well as information such as active customer profiles and active order feature vectors, are used as model inputs to conduct supervised training on the pre-constructed initial customer classification model, thereby obtaining a target customer classification model suitable for pharmaceutical distribution customers.

[0079] It is important to know that, in order to improve the training efficiency of the target customer classification model, it is better to use a subset of active customers who have placed orders and their corresponding active customer profiles and active order feature vectors for model training, rather than using all active customers who have placed orders and their corresponding active customer profiles and active order feature vectors for model training.

[0080] In step S305 of some embodiments, the dormant customer clustering scheme refers to the set of all dormant customer groups formed after clustering dormant customers who have placed orders. The dormant customer clustering scheme includes information such as the composition of dormant customers who have placed orders and the number of clusters in each dormant customer group.

[0081] In this embodiment of the application, after obtaining the target customer classification model applicable to pharmaceutical distribution customers, the customer profile data and order time-series feature vectors corresponding to dormant customers who have placed orders can be input into the target customer classification model. This allows the target customer classification model to output information such as the dormant customer clusters corresponding to each dormant customer who has placed orders, as well as the number of clusters in each dormant customer cluster. Finally, by integrating the information from the outputs of these target customer classification models, a dormant customer clustering scheme can be obtained.

[0082] In step S306 of some embodiments, after obtaining the clustering schemes for active customers and dormant customers, it is possible to check whether there are duplicates or overlaps in the clustering schemes for active customers and dormant customers to ensure that each pharmaceutical distribution customer belongs to only one customer group cluster. Finally, the information of each pharmaceutical distribution customer and its customer group cluster is integrated into a list to obtain the customer group cluster list.

[0083] Steps S301 to S306 of this embodiment assess the activity of pharmaceutical distribution customers based on historical pharmaceutical order data, obtaining customer activity levels. Then, based on customer activity levels, pharmaceutical distribution customers are categorized into active order customers and inactive order customers. This transforms the subjective assessment of customer activity into a measurable indicator, avoiding the bias of relying on experience-based judgments and thus improving the accuracy of customer classification. Secondly, based on historical pharmaceutical order data, a mixture of Gaussian clustering is performed on active order customers to obtain an active customer clustering scheme. This solves the problem that order data from active order customers in pharmaceutical distribution cannot be predicted using traditional machine learning models. Finally, based on the activity... By using customer clustering schemes, customer profile data, and order time-series feature vectors, a pre-constructed initial customer classification model is trained to obtain a target customer classification model. Then, based on the target customer classification model, customer profile data, and order time-series feature vectors, dormant customers who have placed orders are classified to obtain a dormant customer clustering scheme. Based on the active customer clustering scheme and the dormant customer clustering scheme, a list of customer groups for pharmaceutical distribution customers is determined. This solves the problem that order data of all pharmaceutical distribution customers cannot be used for model training of traditional machine learning models, making it impossible for traditional machine learning models to predict the future order amount of pharmaceutical distribution customers, thereby improving the accuracy of predicting the future order amount of pharmaceutical distribution customers.

[0084] In step S105 of some embodiments, the initial customer order amount prediction model refers to a model such as XGBoost (eXtreme GradientBoosting) or MLP (Multi-Layer Perceptron) that has basic order amount prediction function but has not been adapted and optimized for the characteristics of different customer groups.

[0085] The customer order amount prediction model group refers to the set of models formed by training a dedicated customer order amount prediction model for each customer group in the customer group cluster list. Each customer order amount prediction model corresponds one-to-one with a customer group cluster.

[0086] This application embodiment can divide historical pharmaceutical order data into training data and test data, adjust the pre-constructed initial customer order amount prediction model based on the training data and the transformed training data, and combine information such as customer group clusters to obtain the first and second customer order amount prediction models, then compare the model performance of the first and second customer order amount prediction models with the test data, and finally select the customer order amount prediction model from the first and second customer order amount prediction models according to the model performance.

[0087] For details, please refer to Figure 6 In some embodiments, step S105 may include, but is not limited to, steps S601 to S606: Step S601: Classify the historical pharmaceutical order data to obtain pharmaceutical order training data and pharmaceutical order test data; Step S602: Based on customer group clusters, customer profile data, pharmaceutical order training data and order time series feature vectors, adjust the parameters of the initial customer order amount prediction model to obtain the first customer order amount prediction model. Step S603: Perform data transformation on the pharmaceutical order training data to obtain transformed order data; Step S604: Based on customer group clusters, customer profile data, conversion order data and order time series feature vectors, adjust the parameters of the initial customer order amount prediction model to obtain the second customer order amount prediction model; Step S605: Based on the pharmaceutical order test data, compare the model performance of the first customer order amount prediction model and the second customer order amount prediction model to obtain model performance comparison information; Step S606: Based on the model performance comparison information, select the first customer order amount prediction model and the second customer order amount prediction model to obtain the customer order amount prediction model.

[0088] In step S601 of some embodiments, the pharmaceutical order training data refers to a portion of the data that is divided from the historical pharmaceutical order data and used for adjusting the model parameters. It should be noted that the pharmaceutical order training data usually accounts for 70%-80% of the historical pharmaceutical order data.

[0089] Pharmaceutical order test data refers to a portion of historical pharmaceutical order data used to evaluate model performance. Pharmaceutical order test data typically accounts for 20%-30% of historical pharmaceutical order data.

[0090] It is important to know that the training data for pharmaceutical orders must not overlap with the test data for pharmaceutical orders, so as to ensure the independence of model training.

[0091] This application embodiment can classify historical pharmaceutical order data using time division or random sampling methods to obtain pharmaceutical order training data and pharmaceutical order test data. For example, when dividing historical pharmaceutical order data by time, historical pharmaceutical order data from 2020 to 2022 can be used as pharmaceutical order training data, and historical pharmaceutical order data from 2023 can be used as pharmaceutical order test data. If historical pharmaceutical order data is classified using random sampling, it is only necessary to ensure that the ratio of randomly sampled pharmaceutical order training data to pharmaceutical order test data meets a pre-set ratio.

[0092] In some embodiments, steps S602 to S604 refer to an initial customer order amount prediction model that has basic order amount prediction functions but is not adapted to the characteristics of the customer group cluster.

[0093] The first customer order amount prediction model refers to an optimized model obtained by adjusting the parameters of the initial model based on training data of unconverted pharmaceutical orders.

[0094] Transformed order data refers to pharmaceutical order training data after data transformation. For example, logarithmic data obtained by performing a logarithmic transformation on pharmaceutical order training data can be transformed order data from pharmaceutical order training data.

[0095] The second customer order amount prediction model refers to the optimized model obtained by adjusting the parameters of the initial model based on the converted order data.

[0096] In this embodiment, after obtaining the training data for pharmaceutical orders, the training data is backed up to obtain backup training data. Then, a portion of the backup training data undergoes data transformation processing, such as logarithmic transformation, to obtain transformed order data. Further, using the untransformed data in the backup training data, along with the aforementioned customer group clusters, customer profile data, and order time-series feature vectors, a pre-constructed initial customer order amount prediction model is trained under model supervision to obtain a first customer order amount prediction model. Simultaneously, using the transformed order data in the backup training data, along with the aforementioned customer group clusters, customer profile data, and order time-series feature vectors, a pre-constructed initial customer order amount prediction model is trained under model supervision to obtain a second customer order amount prediction model.

[0097] It is important to note that if the order data is converted to logarithmic transformed data, then when performing supervised training on the pre-built initial customer order amount prediction model, the loss function of the initial customer order amount prediction model needs to be adjusted in order to successfully train the second customer order amount prediction model, since the logarithmic data may be negative. Specifically, the loss function of the initial customer order amount prediction model can be adjusted to a negative log-likelihood loss function.

[0098] In steps S605 and S606 of some embodiments, the model performance comparison information refers to the information obtained after model performance comparison, wherein model performance can be prediction accuracy, prediction speed, etc.

[0099] This application embodiment can generate model performance comparison information by inputting medical order test data into a first customer order amount prediction model and a second customer order amount prediction model, and comparing the model performance such as output speed and output accuracy of the first customer order amount prediction model and the second customer order amount prediction model. Furthermore, based on the comparison results of various model performance aspects in the model performance comparison information, the model with the best model performance can be selected from the first customer order amount prediction model and the second customer order amount prediction model as the customer order amount prediction model.

[0100] Steps S601 to S606 of this embodiment involve dividing historical pharmaceutical order data into pharmaceutical order training data and pharmaceutical order test data. Based on customer group clusters, customer profile data, pharmaceutical order training data, and order time-series feature vectors, the parameters of the initial customer order amount prediction model are adjusted to obtain a first customer order amount prediction model. Simultaneously, by transforming the data format of the aforementioned customer group clusters, customer profile data, and pharmaceutical order training data to obtain converted order data and order time-series feature vectors, the parameters of the initial customer order amount prediction model are adjusted to obtain a second customer order amount prediction model. This avoids model adaptation bias caused by a single data processing method. Furthermore, based on pharmaceutical order test data, the performance of the first and second customer order amount prediction models is compared to obtain model performance comparison information. Based on this information, a model selection is performed between the first and second customer order amount prediction models to obtain a final customer order amount prediction model. This ensures the model performance of the customer order amount prediction model, thereby improving the accuracy of customer order amount prediction.

[0101] In step S106 of some embodiments, the customer order amount prediction data refers to the order amount of a pharmaceutical distribution customer in a pharmaceutical distribution enterprise in the future, as predicted by the customer order amount prediction model. The prediction time length of the customer order amount prediction model is related to the model parameters of the customer order amount prediction model.

[0102] In this embodiment, the customer group of the pharmaceutical distribution customer to be predicted for order amount can be determined based on the customer group cluster list. Then, the customer profile data of the pharmaceutical distribution customer is located to obtain the target customer profile, and the historical pharmaceutical order data of the pharmaceutical distribution customer is located to obtain the target pharmaceutical order data. Subsequently, the customer order amount prediction model is selected from the customer order amount prediction model group based on the customer group of the pharmaceutical distribution customer. Finally, the customer order amount prediction model is used to predict the target customer profile and the target pharmaceutical order data to obtain the customer order amount prediction data of the pharmaceutical distribution customer.

[0103] For details, please refer to Figure 7 In some embodiments, step S106 may include, but is not limited to, steps S701 to S704: Step S701: Based on the customer group cluster list, determine the target customer group clusters for pharmaceutical distribution customers; Step S702: Based on pharmaceutical distribution customers, perform data positioning on customer profile data to obtain target customer profiles, and perform data positioning on historical pharmaceutical order data to obtain target pharmaceutical order data; Step S703: Based on the target customer group cluster, perform model screening on the customer order amount prediction model group to obtain the customer order amount prediction model; Step S704: Based on the customer order amount prediction model, output predictions are made on the target customer profile and target pharmaceutical order data to obtain customer order amount prediction data.

[0104] In some embodiments, steps S701 to S703 refer to the target customer group cluster determined from the customer group cluster list for pharmaceutical distribution customers to whom order amount prediction is required.

[0105] Target customer profile refers to the customer profile data of the aforementioned pharmaceutical distribution customers obtained through data positioning.

[0106] Targeted pharmaceutical order data refers to the historical pharmaceutical order data of the aforementioned pharmaceutical distribution customers obtained through data positioning.

[0107] A customer order amount prediction model is a model that is adapted to predict customer order amounts for a target customer group.

[0108] In this embodiment of the application, the customer group to which the pharmaceutical distribution customer belongs, i.e., the target customer group, can be determined based on the clustering data of each pharmaceutical distribution customer in the aforementioned customer group cluster list. Furthermore, the target customer profile that matches the pharmaceutical distribution customer can be found from the customer profile data based on the unique identifier of the pharmaceutical distribution customer, and the target pharmaceutical order data that matches the pharmaceutical order data can be found from the historical pharmaceutical order data. At the same time, after determining the target customer group, the customer order amount prediction model that matches the target customer group can also be found from the customer order amount prediction model group.

[0109] In step S704 of some embodiments, after clarifying the customer order amount prediction model, the target customer profile, and the target pharmaceutical order data, the target customer profile and the target pharmaceutical order data can be input into the customer order amount prediction model to obtain the customer order amount prediction data of the aforementioned pharmaceutical distribution customer output by the customer order amount prediction model.

[0110] Steps S701 to S704, as illustrated in this embodiment, involve determining the target customer group of pharmaceutical distribution customers through a customer group cluster list, locating the customer profile data of the pharmaceutical distribution customer to obtain the target customer profile, locating the historical pharmaceutical order data of the pharmaceutical distribution customer to obtain the target pharmaceutical order data, and simultaneously filtering the customer order amount prediction model group based on the target customer group cluster to obtain the customer order amount prediction model. Furthermore, based on the customer order amount prediction model, output predictions are made on the target customer profile and the target pharmaceutical order data to obtain the customer order amount prediction data. This can improve the fit between the target customer profile, the target pharmaceutical order data, and the customer order amount prediction model, thereby ensuring that the generated customer order amount prediction data is more accurate.

[0111] In step S107 of some embodiments, the customer credit limit refers to the maximum amount that a pharmaceutical distribution company allows its customers to purchase pharmaceutical products on credit within a certain period. For example, in the scenario where the credit limit is implemented, the customer credit limit of a certain tertiary hospital can be RMB 5.6 million for six months, which means that the tertiary hospital can purchase pharmaceutical products on credit for a cumulative period of no more than RMB 5.6 million in the next six months.

[0112] In this embodiment of the application, given the clear customer order limit forecast data and customer repayment risk coefficient of the pharmaceutical distribution customer, the maximum amount of pharmaceutical products that the pharmaceutical distribution customer can purchase on credit within a certain period can be determined by calculating the product of the customer order limit forecast data and the customer repayment risk coefficient, i.e., the customer credit limit. It should be noted that the period in the customer credit limit is the same as the forecast time length of the customer order limit forecast data. For example, if the customer order limit forecast data represents the total order amount of the pharmaceutical distribution customer within the next 6 months, then the period of the customer credit limit is also 6 months.

[0113] This application obtains customer profile data and historical pharmaceutical order data of pharmaceutical distribution customers, extracts time-series features from the historical pharmaceutical order data to obtain order time-series feature vectors, and calculates repayment risk coefficients for pharmaceutical distribution customers based on historical pharmaceutical order data. This clarifies the transaction patterns and repayment capabilities of pharmaceutical distribution customers. Furthermore, based on historical pharmaceutical order data, customer profile data, and order time-series feature vectors, pharmaceutical distribution customers are clustered to obtain a customer group cluster list. This customer group cluster list includes multiple customer group clusters, which solves the problem that traditional machine learning models cannot handle pharmaceutical distribution customers. In addition, by using multiple feature data to cluster pharmaceutical distribution customers, the accuracy of the customer group cluster list is improved. Furthermore, based on multiple customer group clusters, customer profile data, and historical pharmaceutical order data... Based on the order time-series feature vector, the pre-constructed initial customer order limit prediction model is trained to obtain a customer order limit prediction model group. This group includes multiple customer order limit prediction models, each corresponding one-to-one with a customer group cluster. This improves the adaptability of the customer order limit prediction model to the customer group cluster, thereby increasing the prediction accuracy. Finally, based on the customer order limit prediction model group, order limits are predicted for pharmaceutical distribution customers, yielding customer order limit prediction data. Based on this data and the customer repayment risk coefficient, credit limits are calculated for pharmaceutical distribution customers, resulting in their creditworthiness. This not only increases the assessment dimensions for credit limit evaluation of pharmaceutical distribution customers but also reduces the involvement of manual assessment, thus improving the accuracy of credit limit evaluation.

[0114] Please see Figure 8 This application also provides a credit limit assessment device that can implement the above-mentioned credit limit assessment method. The device includes: Data acquisition module 801 is used to acquire customer profile data and historical pharmaceutical order data of pharmaceutical distribution customers; Feature extraction module 802 is used to extract time-series features from historical pharmaceutical order data to obtain order time-series feature vectors; The risk calculation module 803 is used to calculate the repayment risk of pharmaceutical distribution customers based on historical pharmaceutical order data, and obtain the customer repayment risk coefficient. The customer clustering module 804 is used to cluster pharmaceutical distribution customers based on historical pharmaceutical order data, customer profile data and order time-series feature vectors, and obtain a list of customer group clusters, which includes multiple customer group clusters. The model training module 805 is used to train a pre-built initial customer order amount prediction model based on multiple customer group clusters, customer profile data, historical medical order data and order time series feature vectors, to obtain a customer order amount prediction model group. The customer order amount prediction model group includes multiple customer order amount prediction models, and each customer order amount prediction model corresponds one-to-one with a customer group cluster. The order amount prediction module 806 is used to predict the order amount of pharmaceutical distribution customers based on a group of customer order amount prediction models, and obtain customer order amount prediction data. The credit limit calculation module 807 is used to calculate the credit limit of pharmaceutical distribution customers based on customer order limit prediction data and customer repayment risk coefficient, and obtain the customer's credit limit.

[0115] The specific implementation method of this credit limit assessment device is basically the same as the specific implementation method of the above-mentioned credit limit assessment method, and will not be described again here.

[0116] This application also provides an electronic device, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the aforementioned credit limit assessment method. This electronic device can be any smart terminal, including tablet computers, in-vehicle computers, etc.

[0117] Please see Figure 9 , Figure 9 The hardware structure of an electronic device according to another embodiment is illustrated. The electronic device includes: The processor 901 can be implemented using a general-purpose CPU (Central Processing Unit), microprocessor, application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of this application. The memory 902 can be implemented as a read-only memory (ROM), a static storage device, a dynamic storage device, or a random access memory (RAM). The memory 902 can store the operating system and other application programs. When the technical solutions provided in the embodiments of this specification are implemented through software or firmware, the relevant program code is stored in the memory 902 and is called and executed by the processor 901 using the credit limit assessment method of the embodiments of this application. The input / output interface 903 is used to implement information input and output; The communication interface 904 is used to enable communication and interaction between this device and other devices. Communication can be achieved through wired means (such as USB, Ethernet cable, etc.) or wireless means (such as mobile network, WIFI, Bluetooth, etc.). Bus 905 transmits information between various components of the device (e.g., processor 901, memory 902, input / output interface 903, and communication interface 904); The processor 901, memory 902, input / output interface 903, and communication interface 904 are connected to each other within the device via bus 905.

[0118] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described credit limit assessment method.

[0119] Memory, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs and non-transitory computer-executable programs. Furthermore, memory may include high-speed random access memory, and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some embodiments, memory may optionally include memory remotely located relative to the processor, and these remote memories can be connected to the processor via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0120] The credit limit assessment method, credit limit assessment device, electronic device, and storage medium provided in this application embodiment acquire customer profile data and historical pharmaceutical order data of pharmaceutical distribution customers. They extract time-series features from the historical pharmaceutical order data to obtain an order time-series feature vector. Based on the historical pharmaceutical order data, they calculate the repayment risk of pharmaceutical distribution customers to obtain a customer repayment risk coefficient. Based on the historical pharmaceutical order data, customer profile data, and order time-series feature vector, they cluster pharmaceutical distribution customers to obtain a customer group cluster list. This customer group cluster list includes multiple customer group clusters, based on multiple customer... Using customer clusters, customer profile data, historical pharmaceutical order data, and order time-series feature vectors, a pre-constructed initial customer order limit prediction model is trained to obtain a customer order limit prediction model group. This group includes multiple customer order limit prediction models, each corresponding one-to-one with a customer cluster. Based on this model group, order limits are predicted for pharmaceutical distribution customers to obtain customer order limit prediction data. Based on this data and the customer repayment risk coefficient, the credit limit of each pharmaceutical distribution customer is calculated to obtain their creditworthiness.

[0121] The embodiments described in this application are for the purpose of more clearly illustrating the technical solutions of the embodiments of this application, and do not constitute a limitation on the technical solutions provided by the embodiments of this application. As those skilled in the art will know, with the evolution of technology and the emergence of new application scenarios, the technical solutions provided by the embodiments of this application are also applicable to similar technical problems.

[0122] Those skilled in the art will understand that the technical solutions shown in the figures do not constitute a limitation on the embodiments of this application, and may include more or fewer steps than shown, or combine certain steps, or different steps.

[0123] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0124] Those skilled in the art will understand that all or some of the steps in the methods disclosed above, as well as the functional modules / units in the systems and devices, can be implemented as software, firmware, hardware, or suitable combinations thereof.

[0125] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms “comprising” and “having,” and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0126] It should be understood that in this application, "at least one (item)" means one or more, and "more than" means two or more. "And / or" is used to describe the relationship between related objects, indicating that three relationships can exist. For example, "A and / or B" can represent three cases: only A exists, only B exists, and both A and B exist simultaneously, where A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. "At least one (item) of the following" or similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one (item) of a, b, or c can represent: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, and c can be single or multiple.

[0127] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of the units described above is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. The coupling or direct coupling or communication connection between the shown or discussed units may be through some interfaces, or indirect coupling or communication connection between the apparatus or units, and may be electrical, mechanical, or other forms.

[0128] The units described above as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0129] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0130] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes multiple instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned storage medium includes various media capable of storing programs, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0131] The preferred embodiments of the present application have been described above with reference to the accompanying drawings, but this does not limit the scope of the claims of the present application. Any modifications, equivalent substitutions, and improvements made by those skilled in the art without departing from the scope and substance of the embodiments of the present application shall be within the scope of the claims of the present application.

Claims

1. A method for assessing credit limits, characterized in that, The method includes: Obtain customer profile data and historical pharmaceutical order data from pharmaceutical distribution clients; The historical pharmaceutical order data is subjected to time-series feature extraction to obtain the order time-series feature vector; Based on the historical pharmaceutical order data, the repayment risk of the pharmaceutical distribution customers is calculated to obtain the customer repayment risk coefficient. Based on the historical pharmaceutical order data, the customer profile data, and the order time-series feature vector, the pharmaceutical distribution customers are clustered to obtain a customer group cluster list, wherein the customer group cluster list includes multiple customer group clusters; Based on the multiple customer group clusters, the customer profile data, the historical pharmaceutical order data, and the order time series feature vector, the pre-constructed initial customer order amount prediction model is trained to obtain a customer order amount prediction model group, wherein the customer order amount prediction model group includes multiple customer order amount prediction models, and each customer order amount prediction model corresponds one-to-one with the customer group cluster. Based on the aforementioned customer order amount prediction model group, the order amount of the pharmaceutical distribution customers is predicted to obtain customer order amount prediction data. Based on the customer order limit prediction data and the customer repayment risk coefficient, the credit limit of the pharmaceutical distribution customer is calculated to obtain the customer credit limit.

2. The method according to claim 1, characterized in that, The process involves clustering pharmaceutical distribution customers based on the historical pharmaceutical order data, the customer profile data, and the order time-series feature vectors to obtain a customer group cluster list, including: Based on the historical pharmaceutical order data, the activity level of the pharmaceutical distribution customers is assessed to obtain customer activity level; Based on the customer activity level, the pharmaceutical distribution customers are classified into active order customers and dormant order customers. Based on the historical pharmaceutical order data, a mixture Gaussian clustering method was used to identify the active customers who placed orders, resulting in an active customer clustering scheme. Based on the active customer clustering scheme, the customer profile data, and the order time-series feature vector, the pre-constructed initial customer classification model is trained to obtain the target customer classification model; Based on the target customer classification model, the customer profile data, and the order time-series feature vector, the dormant customers who have placed orders are classified to obtain a dormant customer clustering scheme; Based on the active customer clustering scheme and the dormant customer clustering scheme, the customer group cluster list of the pharmaceutical distribution customers is determined.

3. The method according to claim 2, characterized in that, The step of performing a mixture Gaussian clustering method on the active customers based on the historical pharmaceutical order data to obtain an active customer clustering scheme includes: Based on the active customers who placed orders, the historical pharmaceutical order data was filtered to obtain active pharmaceutical order data. Perform normal distribution hybrid clustering on the active pharmaceutical order data to obtain a normal distribution clustering scheme for the target data; Based on the normal distribution clustering scheme of the target data, the active customers who placed orders are classified to obtain the clustering scheme of active customers.

4. The method according to claim 3, characterized in that, The step of performing normal distribution mixed clustering on the active pharmaceutical order data to obtain a normal distribution clustering scheme for the target data includes: Based on a preset list of clustering categories, the active medical order data is clustered to obtain a list of data clustering schemes. The list of clustering categories includes multiple clustering categories, and the list of data clustering schemes includes multiple normal distribution clustering schemes. The normal distribution clustering schemes correspond one-to-one with the number of clustering categories. The clustering quality of the normally distributed clustering scheme is evaluated to obtain clustering quality evaluation data; Based on the clustering quality assessment data, the list of data clustering schemes is screened to obtain the target data normal distribution clustering scheme.

5. The method according to claim 1, characterized in that, The method for training a pre-constructed initial customer order amount prediction model based on the multiple customer group clusters, the customer profile data, the historical medical order data, and the order time-series feature vector, to obtain a customer order amount prediction model group, includes: The historical pharmaceutical order data is classified to obtain pharmaceutical order training data and pharmaceutical order test data; Based on the customer group cluster, the customer profile data, the pharmaceutical order training data, and the order time series feature vector, the parameters of the initial customer order amount prediction model are adjusted to obtain the first customer order amount prediction model. The training data for the medical orders is transformed to obtain transformed order data; Based on the customer group cluster, the customer profile data, the conversion order data, and the order time series feature vector, the parameters of the initial customer order amount prediction model are adjusted to obtain the second customer order amount prediction model. Based on the pharmaceutical order test data, the performance of the first customer order amount prediction model and the second customer order amount prediction model is compared to obtain model performance comparison information. Based on the model performance comparison information, a model selection is performed between the first customer order amount prediction model and the second customer order amount prediction model to obtain the customer order amount prediction model.

6. The method according to any one of claims 1 to 5, characterized in that, The process of calculating the repayment risk of pharmaceutical distribution customers based on the historical pharmaceutical order data to obtain a customer repayment risk coefficient includes: Based on pre-constructed risk indicators and the historical pharmaceutical order data, a risk assessment is conducted on the pharmaceutical distribution customers to obtain a repayment risk score; The repayment risk score is mapped to a coefficient to obtain the customer's repayment risk coefficient.

7. The method according to any one of claims 1 to 5, characterized in that, The method of predicting order amounts for pharmaceutical distribution customers based on the customer order amount prediction model group, to obtain customer order amount prediction data, includes: Based on the customer group cluster list, the target customer group cluster of the pharmaceutical distribution customer is determined; Based on the pharmaceutical distribution customers, the customer profile data is used to locate the target customer profile, and the historical pharmaceutical order data is used to locate the target pharmaceutical order data. Based on the target customer group cluster, the customer order amount prediction model group is screened to obtain the customer order amount prediction model; Based on the customer order amount prediction model, the target customer profile and the target pharmaceutical order data are used to output predictions to obtain the customer order amount prediction data.

8. A credit limit assessment device, characterized in that, The device includes: The data acquisition module is used to acquire customer profile data and historical pharmaceutical order data of pharmaceutical distribution customers; The feature extraction module is used to extract time-series features from the historical medical order data to obtain order time-series feature vectors; The risk calculation module is used to calculate the repayment risk of the pharmaceutical distribution customers based on the historical pharmaceutical order data, and obtain the customer repayment risk coefficient. The customer clustering module is used to perform customer clustering on the pharmaceutical distribution customers based on the historical pharmaceutical order data, the customer profile data, and the order time-series feature vector, to obtain a customer group cluster list, wherein the customer group cluster list includes multiple customer group clusters; The model training module is used to train a pre-constructed initial customer order amount prediction model based on the multiple customer group clusters, the customer profile data, the historical medical order data, and the order time series feature vector, to obtain a customer order amount prediction model group, wherein the customer order amount prediction model group includes multiple customer order amount prediction models, and each customer order amount prediction model corresponds one-to-one with the customer group cluster. The order limit prediction module is used to predict the order limits of the pharmaceutical distribution customers based on the customer order limit prediction model group, and obtain customer order limit prediction data. The credit limit calculation module is used to calculate the credit limit of the pharmaceutical distribution customer based on the customer order limit prediction data and the customer repayment risk coefficient, and obtain the customer credit limit.

9. An electronic device, characterized in that, The electronic device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the credit limit assessment method according to any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the credit limit assessment method according to any one of claims 1 to 7.

Images