Special lawyer letter issuing method and system based on CART tree

Abstract

The invention discloses a special lawyer letter issuing method and system based on a CART tree. The system comprises a client device, a distributed database device, a telescopic server, a cloud platform device, an intelligent letter sending processing device, a real-time monitoring device and a terminal visualization large screen. The method comprises the following steps: carrying out equidistant box separation on money return dates to determine the maximum concentration ratio of money return date distribution; constructing a decision-making tree model, and training the decision-making tree model through the correlation characteristics of the historical functional data; according to the output of the trained decision tree model, determining a customer group with the highest collection amount concentration degree, and matching the customer group with the maximum collection date distribution concentration degree to determine a function issuing plan; and implementing the issuing plan and carrying out real-time monitoring, and optimizing the issuing plan according to a feedback result obtained by monitoring. Therefore, the recovery rate of overdue cases is effectively improved, the goal of intelligent letter issuing plans in a big data environment is achieved, and the efficiency and accuracy of the letter issuing plans are greatly improved.

Description

Technical Field

[0001] The present invention relates to the technical field of financial big data processing, and in particular to a method and system for sending special lawyer letters based on a CART tree. Background Art

[0002] With the vigorous development of big data technology, the need for financial credit risk control to shift from traditional manual risk control to digital risk control is becoming more and more urgent. However, in the actual financial case allocation business, the traditional letter sending plan method has long relied on manual business experience to send out special lawyer letters in a timely manner after the case is placed at the beginning of the month. The manual business experience letter sending plan is not scientific and reasonable, resulting in irregular case sending and a low rate of case recovery.

[0003] The current letter sending mechanism also has many problems. Due to the differences in gender, age, province, etc., the repayment willingness and repayment ability of customers corresponding to different overdue cases are quite different, which makes their repayment date intervals different. If a random and indiscriminate letter sending plan is adopted based only on business needs, the case recovery effect will be random and poor. (Zhong Chenbiao; Lin Jianian; Guo Hongqing; Ying Haiyun; Zeng Donglin; Li Jianbin; Pan Xiaobo; Zhang Bosong, Intelligent Sending Method and System of Lawyer's Letter, 2019, 1-2).

[0004] To this end, the present invention specifically develops and constructs a special lawyer letter sending method and system based on CART tree. Through the above-mentioned intelligent letter sending method and system of the invention, the accuracy of letter sending can be significantly improved, so that the recovery rate of overdue cases can be maximized. Summary of the invention

[0005] The main purpose of the present invention is to provide a special lawyer letter sending method and system based on CART tree, aiming to solve the technical problem that the current letter sending mechanism adopts a random and indiscriminate letter sending plan, which makes the case recovery effect random and leads to poor results.

[0006] To achieve the above-mentioned purpose, the present invention proposes a special lawyer letter sending method based on CART tree. The method includes: extracting historical letter sending data from a distributed database, and performing data cleaning and feature selection on the historical letter sending data to obtain correlation features; performing equidistant binning on the collection date in the historical letter sending data to determine the maximum concentration of the collection date distribution; constructing a decision tree model, and training the decision tree model through the correlation features to adjust the hyperparameters of the decision tree model to obtain a trained decision tree model; according to the output of the trained decision tree model, determining the customer group with the highest collection amount concentration; matching the customer group with the highest collection amount concentration with the maximum collection date distribution concentration, determining the letter sending plan for the customer group with the highest collection amount concentration, and the letter sending plan is formulated based on the letter sending time of each customer group meeting the window period for maximizing the collection probability; implementing the letter sending plan and performing real-time monitoring, and optimizing the letter sending plan based on the feedback results obtained from the monitoring.

[0007] By intelligently sending special lawyer letters through the present invention, the recovery efficiency of overdue cases can be improved and the case recovery rate can be maximized.

[0008] To achieve the above purpose, the technical solution provided by the present invention is as follows:

[0009] Step 1: System construction.

[0010] To achieve the above-mentioned purpose, the present invention also proposes a special lawyer letter sending system based on CART tree, which is composed of the following key modules: client device, distributed database device, telescopic server, cloud platform device, intelligent letter sending processing device, real-time monitoring device and terminal visualization large screen; the client device is connected to the distributed database device, and the distributed database device is connected to the telescopic server, the cloud platform device, the intelligent letter sending processing device, the real-time monitoring device and the terminal visualization large screen in turn, thereby constructing an efficient and intelligent letter sending plan platform.

[0011] The client device is used to receive and process case information of special lawyer letters, wherein the special lawyer financial letters include overdue cases with different sample characteristics.

[0012] The distributed database device can store historical letter sending data on multiple physical nodes, manage data using distributed computing and storage technology, has high scalability and fault tolerance, and supports large-scale data and high-concurrency access requests.

[0013] The telescopic server and cloud platform equipment are used to provide elastic computing and storage support, while supporting dynamic expansion and processing of data.

[0014] The intelligent letter sending processing device, as a core module, includes a data extraction module, a model training module, a collection analysis module and a letter sending plan generation module; the data extraction module is used to extract historical letter sending data from a distributed database device; the model training module is used to build a decision tree model and perform model training through the historical letter sending data; the collection analysis module is used to analyze the historical letter sending data and determine the maximum concentration of the collection date distribution; the letter sending plan generation module is used to match the customer group with the highest collection amount concentration determined by the trained decision tree model with the maximum concentration of the collection date distribution to generate a letter sending plan.

[0015] The real-time monitoring device is used to track the progress of the collection of each special lawyer's letter in real time, and adjust the letter sending plan according to the collection effect, so that the allocation of cases can be completed on time, as required and reasonably.

[0016] The terminal visualization screen is used to output customer service and case allocation detailed reports, which record the processing process of each special lawyer's letter. It is used to analyze the collection effect, evaluate the effectiveness of the case allocation strategy, and provide data support for future strategy adjustments.

[0017] Step 2: Data collection and preprocessing.

[0018] The system first extracts historical letter data from the distributed database, including the province of letter issuance, gender, age, overdue days, overdue amount, new case / old case, whether it is a duplicate case, amount of payment, whether it is paid back, etc. Then the historical letter data is cleaned, including processing missing values, outliers, duplicate data, etc.

[0019] Next, perform feature construction: create new features based on existing fields (for example, segment the number of overdue days, or standardize the repayment amount).

[0020] Final feature selection: Based on correlation analysis or feature importance assessment, select the most relevant features that are most helpful to the model.

[0021] Step 3: Maximum concentration of repayment date distribution.

[0022] This process can use the equal interval binning method to bin the "receipt date" to ensure that the distribution of the collection date has the greatest concentration. This can be achieved by following the steps below:

[0023] Selecting a suitable payment collection date range from the historical letter sending data (for example, the payment collection date distribution in the past 12 months), and obtaining payment collection date data within the payment collection date range;

[0024] Use the equal-interval binning method to divide the payment date data into multiple payment date intervals at equal intervals to ensure that the payment amount concentration in each interval is maximized;

[0025] The collection amounts in the multiple collection date intervals are analyzed, and the date segment with the highest collection amount concentration is selected as the maximum collection date distribution concentration.

[0026] Step 4: Build a decision tree (Classification and Regression Tree, CART) model.

[0027] First, a decision tree model is built using a decision tree algorithm to classify customers. The goal is to explore the relationship between the concentration of the repayment customer group and the date of letter issuance.

[0028] The decision tree model can then be trained using the correlation features (such as province, gender, age, etc.) to identify the collection potential of customer groups.

[0029] Then, grid search can be used to adjust the hyperparameters of the decision tree model, including the depth of the tree, the minimum number of samples of leaf nodes, etc., to ensure the best performance of the model.

[0030] Finally, cross-validation can be used to evaluate the generalization ability of the decision tree model to avoid overfitting, and finally obtain a trained decision tree model.

[0031] Step 5: Analysis of customer groups for payment collection.

[0032] According to the output of the trained decision tree model, the customer group is divided into different types of customers, each type of customer has different collection characteristics.

[0033] Perform a collection analysis on the collection amounts of the different types of customers to identify the customer groups with the highest concentration of collection amounts (for example, customer groups with certain genders, age groups, and overdue amount ranges) to determine the customer groups with the highest concentration of collection amounts among the different types of customers.

[0034] Then, based on the concentration of the repayment amount, find the characteristic differences between the repayment customer group and the non-repayment customer group, and identify which customer groups are most likely to receive repayments.

[0035] Step 6: Generate a letter sending plan.

[0036] Match the customer group with the highest concentration of the repayment amount with the maximum concentration of the repayment date distribution to determine the date interval with the high concentration of the repayment amount, that is, in the date interval with the high concentration of the repayment amount, give priority to sending letters to those customers with a high probability of repayment;

[0037] Based on the date range, formulate a letter sending plan to ensure that the letter sending time for each customer group fits within the window period that maximizes the probability of collection.

[0038] Generate a letter sending plan based on this priority, and generate a personalized letter sending plan based on the characteristics of the customer group with the highest concentration of repayment amounts and the best repayment date. Send letters to customer groups in batches to ensure that the time interval for sending letters is aligned with the repayment characteristics of the customer groups.

[0039] Step 7: Model evaluation and adjustment.

[0040] By tracking the actual effect of the model, we can evaluate the changes in the amount of repayments and the improvement in the repayment rate. We can use indicators such as precision, recall rate, F1 score, etc. to evaluate the model effect.

[0041] Based on the evaluation results of new data and models, we continuously optimize the letter sending plan and decision tree model to ensure the maximum amount of repayments.

[0042] Step 8: Implementation and deployment.

[0043] Connect the letter sending plan with the automated letter sending system to implement a data-driven letter sending strategy.

[0044] Then, during the implementation process, we continuously monitor the collection results and adjust the letter sending strategy based on real-time data to ensure long-term maximum collection.

[0045] In this way, through machine learning modeling and optimization, the collection effect of letters can be maximized, helping to maximize the collection of overdue customers.

[0046] Step 9: Real-time monitoring and feedback.

[0047] Use real-time monitoring equipment to track the payment collection results and letter sending progress to ensure that the letter sending for each case is completed reasonably and on time. At the same time, the terminal visualization screen outputs detailed customer service and case allocation details report, recording the allocation and processing process of each case. Analyze the payment collection effect, evaluate the effectiveness of the case allocation strategy, and provide data support for future strategy adjustments.

[0048] One or more technical solutions proposed in the present invention have at least the following technical effects: a special lawyer letter sending method and system based on the CART tree constructed by the present invention effectively improves the recovery rate of overdue cases, achieves the goal of intelligent letter sending plan in a big data environment, and greatly improves the efficiency and accuracy of the letter sending plan. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.

[0050] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0051] Figure 1 A flowchart diagram of a method for sending a special lawyer's letter based on a CART tree according to the present invention;

[0052] Figure 2 A customer group distribution diagram of the decision tree algorithm provided by the present invention;

[0053] Figure 3 The present invention is a structural schematic diagram of a special lawyer letter sending system based on a CART tree.

[0054] Description of Figure Numbers:

[0055] 1. Client device; 2. Distributed database device; 3. Scalable server; 4. Intelligent letter processing device; 5. Real-time monitoring device; 6. Terminal visualization large screen.

[0056] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0057] It should be understood that the specific embodiments described herein are only used to explain the technical solutions of the present invention and are not used to limit the present invention.

[0058] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0059] It should be noted that all directional indications in the embodiments of the present invention (such as up, down, close, far...) are only used to explain the relative position relationship, movement status, etc. between the components under a certain specific posture (as shown in the accompanying drawings). If the specific posture changes, the directional indication will also change accordingly.

[0060] In addition, the descriptions of "first", "second", etc. in the present invention are only used for descriptive purposes and cannot be understood as indicating or implying their relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include at least one of the features. In addition, the technical solutions between the various embodiments can be combined with each other, but they must be based on the ability of ordinary technicians in the field to implement them. When the combination of technical solutions is contradictory or cannot be implemented, it should be deemed that such a combination of technical solutions does not exist and is not within the scope of protection required by the present invention.

[0061] The embodiment of the present invention provides a method for sending a special lawyer's letter based on a CART tree.

[0062] In this embodiment, the method for sending special lawyer letters based on the CART tree includes the steps of: extracting historical letter sending data from a distributed database, and performing data cleaning and feature selection on the historical letter sending data to obtain correlation features; equidistantly binning the collection dates in the historical letter sending data to determine the maximum concentration of the collection date distribution; constructing a decision tree model, and training the decision tree model through the correlation features to adjust the hyperparameters of the decision tree model to obtain a trained decision tree model; determining the customer group with the highest collection amount concentration according to the output of the trained decision tree model; matching the customer group with the highest collection amount concentration with the maximum collection date distribution concentration to determine a letter sending plan for the customer group with the highest collection amount concentration, the letter sending plan being formulated based on the letter sending time of each customer group being in line with the window period for maximizing the collection probability; implementing the letter sending plan and performing real-time monitoring, and optimizing the letter sending plan based on the feedback results obtained from the monitoring.

[0063] It should be noted that the distributed database may be a database system that stores case information of special lawyer letters in a dispersed manner in multiple different physical locations.

[0064] Among them, special lawyer letters can be professional legal documents prepared on behalf of clients regarding relevant facts or legal issues.

[0065] The historical letter data can be various information records related to letters sent in the past stored in the distributed database, including the sending province, gender, age, overdue days, overdue amount, new case / old case, whether it is a repeated case, the amount of payment, whether the payment has been paid, etc.

[0066] Among them, the correlation features can be the cleaning of historical letter sending data, including processing missing values, outliers, duplicate data, etc., and then creating new features based on existing fields (for example, segmenting the overdue days, or standardizing the repayment amount), and finally selecting the features that are most helpful to the subsequent model based on correlation analysis or feature importance evaluation.

[0067] It is understandable that the maximum concentration of the collection date distribution can be the time when the payment should be received in the historical letter sending data. Among the presentations at different time points, the collection amount accounts for the largest and most concentrated proportion.

[0068] It should be understood that the decision tree (Classification and Regression Tree, CART) model can be a model that analyzes historical letter sending data and splits each node according to the correlation characteristics to form a tree-like structure. It can be used to divide customer groups into different categories and mine the customer groups with the highest concentration of repayment amounts.

[0069] It is understandable that the letter sending plan can be a plan formulated based on the customer group with the highest concentration of repayment amount and the largest concentration of repayment date distribution, which can ensure that the letter sending time for each customer group is consistent with the window period with the maximum probability of repayment. The customer groups can be sent letters in batches to ensure that the time interval for sending letters is aligned with the repayment characteristics of the customer groups.

[0070] In a specific implementation, this embodiment provides a method for sending special lawyer letters based on a CART tree, which can intelligently send special lawyer letters, improve the recovery efficiency of overdue cases, and maximize the case recovery rate.

[0071] In a feasible implementation, in this embodiment, the step of equally spaced binning the collection dates in the historical letter sending data and determining the maximum concentration of the collection date distribution includes: selecting a collection date range from the historical letter sending data, and obtaining the collection date data within the collection date range; using the equally spaced binning method to divide the collection date data into multiple collection date intervals; analyzing the concentration of the collection amounts in the multiple collection date intervals to determine the maximum concentration of the collection date distribution.

[0072] It should be noted that the equidistant binning method can be a data preprocessing method, which can divide the range of historical letter sending data into several equidistant intervals to facilitate subsequent analysis and processing.

[0073] In this implementation, first select a suitable collection date range (for example, the collection date distribution in the past 12 months). Then use the equal-interval binning method to divide the "collection date" into multiple intervals (i.e., multiple collection date intervals) at equal intervals to ensure that the collection amount concentration in each interval is the largest. Finally, analyze the collection amounts in different date intervals and select the date segment with the highest collection concentration (i.e., the collection date distribution with the maximum concentration) to ensure that the collection date distribution has the greatest concentration.

[0074] In a feasible implementation manner, the steps of constructing a decision tree model and training the decision tree model through the correlation features to adjust the hyperparameters of the decision tree model to obtain a trained decision tree model in this embodiment include: constructing a decision tree model through a decision tree algorithm; training the decision tree model through the correlation features to identify the collection potential of a customer group; adjusting the hyperparameters of the decision tree model using grid search and cross-validating the generalization ability of the decision tree model to obtain a trained decision tree model, wherein the hyperparameters include the depth of the tree and the minimum number of samples of leaf nodes.

[0075] It should be noted that the hyperparameters can be the parameters set before the decision tree model is trained.

[0076] The depth of the tree can be the longest path length from the root node to the leaf node in the decision tree. A larger tree depth may make the model more complex and capture more detailed patterns, but it may also lead to overfitting; a smaller tree depth may make the model simpler and more robust, but may lose some detailed information.

[0077] The minimum number of samples for a leaf node can be the number of samples that a leaf node must contain at least. If this value is set too large, the growth of the tree will be restricted, the number of leaf nodes will be reduced, and the model will not be too complex; if it is set too small, the tree may grow too lush and overfitting may occur. By adjusting these hyperparameters, the performance and generalization ability of the decision tree model can be optimized.

[0078] It is understandable that Grid Search can be a method for optimizing the hyperparameters of a decision tree model.

[0079] By enumerating all possible combinations of a given set of hyperparameters, training the model on each of these combinations and evaluating its performance, we can finally find the hyperparameters with the best performance.

[0080] In this implementation, a decision tree model is first constructed using a decision tree algorithm and customer classification is performed, with the goal of exploring the relationship between the concentration of the customer group that is collecting payments and the date of the letter. Then, the decision tree model is trained using the collected correlation features (such as province, gender, age, etc.) to identify the payment collection potential of different customer groups. Then, the hyperparameters of the decision tree are adjusted through grid search, such as the depth of the tree, the minimum number of samples of leaf nodes, etc., to ensure the best performance of the model. Finally, cross-validation is used to evaluate the generalization ability of the model to avoid overfitting.

[0081] In a feasible implementation, the step of determining the customer group with the highest concentration of repayment amounts based on the output of the trained decision tree model described in this embodiment includes: dividing the customer group into different types of customers based on the output of the trained decision tree model; performing repayment analysis on the repayment amounts of the different types of customers, and determining the customer group with the highest concentration of repayment amounts among the different types of customers.

[0082] In this implementation, first, based on the output of the above decision tree model, the customer groups are divided into different categories, each of which has different payment collection characteristics. Then, the payment collection amount of each category is analyzed to find out the customer groups with the highest payment collection amount concentration (for example, customer groups with certain genders, age groups, and overdue amount ranges). Finally, based on the concentration of payment collection amount, the characteristic differences between the payment collection customer group and the non-payment collection customer group are found to conduct customer distribution analysis, identify which customer groups are most likely to receive payment, and determine the customer groups with the highest payment collection amount concentration among different customer categories.

[0083] In another feasible implementation, the step of matching the customer group with the highest concentration of repayment amounts with the maximum concentration of repayment date distribution to determine the letter sending plan for the customer group with the highest concentration of repayment amounts described in this embodiment includes: matching the customer group with the highest concentration of repayment amounts with the maximum concentration of repayment date distribution to determine a date range with a high concentration of repayment amounts; based on the date range, formulating a letter sending plan with priority given to the letter sending time of each customer group that meets the window period for maximizing the probability of repayment.

[0084] In this implementation, first, the customer group with the highest concentration of repayments is matched with the concentration of repayment dates. That is, in the date interval with a high concentration of repayment amounts, letters are sent first to customers with a high probability of repayment. Then a letter sending plan is formulated to ensure that the letter sending time for each customer group meets the window period for maximizing its repayment probability. A letter sending plan is generated, and a personalized letter sending plan is generated based on the characteristics of customers with a high concentration of repayments and the best repayment date. Letters are sent to customer groups in batches to ensure that the letter sending time interval is aligned with the repayment characteristics of the customer groups.

[0085] Further, to facilitate understanding of the implementation process of the above-mentioned special lawyer letter sending method based on CART tree in this embodiment, please refer to Figure 1 , Figure 1 A flowchart of a special lawyer letter sending method based on a CART tree is provided for the present invention. First, historical letter sending data is collected from a distributed database, and then data cleaning is performed, including outliers and noise data. Then the maximum concentration of repayment dates in the historical letter sending data is calculated. Then the decision tree model (CART) is constructed and trained. After the decision tree model is constructed and trained, the concentration of the repayment customer group is analyzed through the model, and a letter sending plan is generated through the model.

[0086] After generating the letter sending plan, you can evaluate and adjust the model. This process specifically includes: tracking the actual effect of the decision tree model, evaluating the changes in the collection amount and the improvement of the collection rate. You can use indicators such as precision, recall rate, F1 score, etc. to evaluate the model effect. Then, based on the new data and model evaluation results, continue to optimize the letter sending plan and decision tree model to ensure the maximum collection amount.

[0087] At the same time, after connecting the letter sending plan with the automated letter sending system to implement a data-driven letter sending strategy, you can also use real-time monitoring equipment to monitor whether the case is sent according to the letter sending progress, and continuously monitor the collection results to adjust the letter sending strategy based on real-time data to ensure long-term maximum collection. In this way, through machine learning modeling and optimization, the collection effect of letter sending can be maximized to help maximize the collection of overdue customers. Finally, the actual collection effect can be output as a detailed customer service and case allocation report to record the allocation and processing process of each case. Analyze the collection effect, evaluate the effectiveness of the case distribution strategy, and provide data support for future strategy adjustments.

[0088] In the technical solution provided in this embodiment, through the above-mentioned special lawyer letter sending method based on the CART tree, the recovery rate of overdue cases is effectively improved, the goal of intelligent letter sending plan in a big data environment is achieved, and the efficiency and accuracy of the letter sending plan are greatly improved.

[0089] For example, in order to illustrate how a special lawyer letter sending method based on a CART tree in the above embodiment generates a letter sending plan, a specific example will be used to demonstrate the calculation and analysis of the entire process. Specifically:

[0090] Step 1: Statistical period is monthly letter sending projects in terms of the concentration of number of people receiving payment and amount of payment received in each date period and the probability of payment received (the average of the percentage of the number of people receiving payment and the percentage of the amount of payment received). The details are shown in Table 1:

[0091] Table 1

[0092] Payment date The proportion of people who received payment Repayment amount ratio Probability of Repayment No. 1-9 15.15％ 16.03％ 16％ No. 10-21 63.93％ 58.54％ 61％ No. 22-31 20.92％ 25.43％ 23％

[0093] Step 2: Statistical analysis of the mail collection and delivery time for both land and air transportation, as shown in Table 2:

[0094] Table 2

[0095]

[0096] Therefore, based on actual considerations such as the length of time letters take to be transported on the road, for projects that send letters monthly, we will maximize the number of letters sent to customers with the highest concentration of repayments between the 6th and 17th, and generate a corresponding letter sending plan.

[0097] Step 3: Analyze the customer groups with the highest concentration of repayments for the monthly letter-sending projects.

[0098] Among them, the number of non-repayment customers in the project sample is 1,925,246, and the number of repayment customers is 76,322, with an overall average repayment ratio of 3.81%. By mining the customer group whose repayment ratio is higher than the average repayment ratio and sending corresponding letters on June 6-17, the priority repayment probability of this customer group can be achieved.

[0099] According to the CART tree algorithm, we can find the customer groups with high concentration of leaf node customers. Figure 2 , Figure 2 This is the customer group distribution diagram of the decision tree algorithm provided by the present invention. Figure 2 As shown:

[0100] For example, the starting node (such as period < 2.5, square error = 0.037, sample = 2001568, proportion = 0.038) can be divided into two directions of correct (such as period < 0.5, square error = 0.057, sample = 197176, proportion = 0.061) and wrong (such as range < 2.5, square error = 0.034, sample = 1804392, proportion = 0.036) nodes. Among the correct nodes, the following nodes can be divided into "range <1.5, square error = 0.069, sample = 68368, proportion = 0.074" and "period <1.5, square error = 0.05, sample = 128808, proportion = 0.053"; finally, we can get customer group 1 (such as square error = 0.064, sample = 32732, proportion = 0.068), customer group 2 (such as square error = 0.073, sample = 35636, proportion = 0.08), customer group 3 (such as square error = 0.055, sample = 51358, proportion = 0.058) and customer group 4 (square error = 0.047, sample = 77450, proportion = 0.05). Among the wrong nodes, the following nodes can be divided into "period <2.5, square error = 0.032, sample = 1174358, proportion = 0.033" and "range <2.5, square error = 0.039, sample = 630034, proportion = 0.04"; finally, other customer groups can be obtained (such as square error = 0.037, sample = 187514, proportion = 0.038; and square error = 0.031, sample = 986844, proportion = 0.032), customer group 5 (such as square error = 0.037, sample = 499147, proportion = 0.039) and customer group 6 (such as square error = 0.044, sample = 130887, proportion = 0.046).

[0101] As mentioned above, the sample number of customer group 1 is 32,732 customers, and the repayments account for 6.8% of them; the sample number of customer group 2 is 35,636 customers, and the repayments account for 8% of them; the sample number of customer group 3 is 51,358 customers, and the repayments account for 5.8% of them; the sample number of customer group 4 is 77,450 customers, and the repayments account for 5% of them; the sample number of customer group 5 is 499,147 customers, and the repayments account for 3.9% of them; the sample number of customer group 6 is 130,887 customers, and the repayments account for 4.6% of them; the sample number of other customer groups is 1,174,358 customers, and the repayments account for 3.3%.

[0102] Step 4: According to the above customer group distribution, the corresponding rules are shown in Table 3:

[0103] Table 3

[0104]

[0105] Step 5: Identify the customer groups whose repayment ratio is higher than the average repayment ratio, send corresponding letters on June 6-17, and come up with a scientific letter sending rule plan.

[0106] It should be noted that the above examples are only used to understand the present invention and do not constitute a limitation on the present invention's method for sending special lawyer letters based on the CART tree. More simple transformations based on this technical concept are all within the scope of protection of the present invention.

[0107] Based on the above-mentioned method for sending special lawyer letters based on CART tree, the present invention also provides a system for sending special lawyer letters based on CART tree, and the above-mentioned method is implemented based on this system. Figure 3 , Figure 3 The present invention is a structural schematic diagram of a special lawyer letter sending system based on a CART tree.

[0108] In this embodiment, the system includes: a client device 1, a distributed database device 2, a telescopic server 3 (e.g., a server ESC), a cloud platform device, an intelligent letter processing device 4, a real-time monitoring device 5, and a terminal visualization large screen 6;

[0109] The client device 1 is connected to the distributed database device 2, and the distributed database device 2 is connected to the telescopic server 3, the cloud platform device, the intelligent letter processing device 4, the real-time monitoring device 5 and the terminal visualization large screen 6 in sequence;

[0110] The client device 1 is used to receive and process case information of special lawyer letters;

[0111] The distributed database device 2 is used to store historical letter sending data and supports high concurrent access;

[0112] The telescopic server 3 and the cloud platform equipment are used to provide elastic computing and storage support.

[0113] It should be noted that the client device is used to receive and process case information. The distributed database device module stores data on multiple physical nodes and manages data using distributed computing and storage technology. It has high scalability and fault tolerance and supports large-scale data and high concurrent access requests. The scalable server and cloud platform devices provide elastic computing and storage capabilities and support dynamic expansion and processing of data.

[0114] In this embodiment, the entire system builds an efficient and intelligent letter sending planning platform by integrating with distributed databases, scalable servers, cloud platform equipment, intelligent case processing equipment, real-time monitoring equipment and terminal visualization large screen.

[0115] In a feasible implementation, the intelligent letter processing device 4 described in this embodiment includes a data extraction module, a model training module, a collection analysis module and a letter sending plan generation module; the data extraction module is used to extract historical letter sending data from the distributed database device 2; the model training module is used to build a decision tree model and perform model training through the historical letter sending data; the collection analysis module is used to analyze the historical letter sending data and determine the maximum concentration of the collection date distribution; the letter sending plan generation module is used to match the customer group with the highest collection amount concentration determined by the trained decision tree model with the maximum concentration of the collection date distribution to generate a letter sending plan.

[0116] In this implementation, the intelligent letter processing device is the core module of the system, responsible for extracting data, calculating different tree levels and leaf node parameters, grid search, automatic search, mining repayment customer concentration higher than the average repayment customer concentration, generating case letter sending times for different projects and different customer groups, etc.

[0117] In a feasible implementation manner, the real-time monitoring device 5 described in this embodiment is used to track the progress of the collection of each special lawyer's letter in real time, and adjust the letter sending plan according to the collection effect.

[0118] In this embodiment, the real-time monitoring device can track the collection results and the assignment of each case, so that the assignment of cases can be completed on time, as required, and reasonably.

[0119] In a feasible implementation manner, the terminal visualization large screen 6 described in this embodiment is used to output a customer service and case allocation detailed report, and the customer service and case allocation detailed report records the processing process of each special lawyer's letter.

[0120] In this implementation, the terminal visualization screen can output detailed customer service and case allocation report, record the allocation and processing process of each case, analyze the collection effect, evaluate the effectiveness of the case allocation strategy, and provide data support for future strategy adjustments.

[0121] The present invention provides a special lawyer letter sending system based on a CART tree, which adopts a special lawyer letter sending method based on a CART tree in the above embodiment, and can solve the current letter sending mechanism, which adopts a random and indiscriminate letter sending plan, making the case recovery effect random, resulting in a poor technical problem. Compared with the prior art, the beneficial effects of the special lawyer letter sending system based on a CART tree provided by the present invention are the same as the beneficial effects of the special lawyer letter sending method based on a CART tree provided in the above embodiment, which will not be repeated here.

[0122] The above descriptions are only some embodiments of the present invention, and are not intended to limit the patent scope of the present invention. All equivalent structural changes made using the contents of the present invention's specification and drawings under the technical concept of the present invention, or directly / indirectly applied in other related technical fields, are included in the patent protection scope of the present invention.

Claims

1. A special lawyer letter sending method based on CART tree, characterized in that: The method comprises: Extracting historical letter sending data from a distributed database, and performing data cleaning and feature selection on the historical letter sending data to obtain correlation features; The payment collection dates in the historical letter sending data are divided into equally spaced bins to determine the maximum concentration of the payment collection date distribution; Constructing a decision tree model, and training the decision tree model through the correlation features to adjust the hyperparameters of the decision tree model to obtain a trained decision tree model; Determine the customer group with the highest concentration of repayment amounts based on the output of the trained decision tree model; Match the customer group with the highest concentration of the repayment amount with the maximum concentration of the repayment date distribution, and determine the letter sending plan for the customer group with the highest concentration of the repayment amount, wherein the letter sending plan is formulated based on the letter sending time of each customer group conforming to the window period for maximizing the probability of repayment; The letter sending plan is implemented and monitored in real time, and the letter sending plan is optimized according to the feedback results obtained from the monitoring.

2. The method according to claim 1, characterized in that The step of equally spaced bins for the payment collection dates in the historical letter sending data and determining the maximum concentration of the payment collection date distribution includes: Select a payment collection date range from the historical letter sending data, and obtain payment collection date data within the payment collection date range; Using an equidistant binning method to divide the payment date data into multiple payment date intervals; The concentration of the payment amounts in the multiple payment date intervals is analyzed to determine the maximum concentration of the payment date distribution.

3. The method according to claim 1, characterized in that The step of constructing a decision tree model and training the decision tree model by using the correlation features to adjust the hyperparameters of the decision tree model to obtain a trained decision tree model includes: Construct a decision tree model through the decision tree algorithm; The decision tree model is trained by using the correlation features to identify the repayment potential of customer groups; The hyperparameters of the decision tree model are adjusted using grid search and the generalization ability of the decision tree model is cross-validated to obtain a trained decision tree model, wherein the hyperparameters include the depth of the tree and the minimum number of samples of leaf nodes.

4. The method according to claim 1, characterized in that The step of determining the customer group with the highest concentration of repayment amounts according to the output of the trained decision tree model comprises: According to the output of the trained decision tree model, the customer group is divided into different types of customers; Perform a collection analysis on the collection amounts of the different types of customers to determine the customer group with the highest concentration of collection amounts among the different types of customers.

5. The method according to claim 1, characterized in that The step of matching the customer group with the highest concentration of the repayment amount with the maximum concentration of the repayment date distribution to determine the letter sending plan for the customer group with the highest concentration of the repayment amount includes: Matching the customer group with the highest concentration of the repayment amount with the maximum concentration of the repayment date distribution to determine the date interval with the highest concentration of the repayment amount; Based on the date range, a letter sending plan is formulated with priority given to each customer group's letter sending time that conforms to the window period that maximizes the probability of collection.

6. A special lawyer letter sending system based on CART tree, characterized in that: The special lawyer letter sending system based on CART tree is implemented by the special lawyer letter sending method based on CART tree as claimed in claim 1, and the system includes: client device, distributed database device, telescopic server, cloud platform device, intelligent letter sending processing device, real-time monitoring device and terminal visualization large screen; The client device is connected to the distributed database device, and the distributed database device is connected to the telescopic server, the cloud platform device, the intelligent letter processing device, the real-time monitoring device and the terminal visualization large screen in sequence; The client device is used to receive and process case information of special lawyer letters; The distributed database device is used to store historical mail data and supports high concurrent access; The telescopic server and cloud platform equipment are used to provide elastic computing and storage support.

7. The system according to claim 6, characterized in that The intelligent letter-issuing processing device includes a data extraction module, a model training module, a payment collection analysis module and a letter-issuing plan generation module; The data extraction module is used to extract historical letter sending data from the distributed database device; The model training module is used to build a decision tree model and perform model training using the historical letter sending data; The payment collection analysis module is used to analyze the historical letter sending data to determine the maximum concentration of payment collection date distribution; The letter sending plan generation module is used to match the customer group with the highest concentration of repayment amounts determined by the trained decision tree model with the maximum concentration of repayment date distribution to generate a letter sending plan.

8. The system according to claim 7, characterized in that The real-time monitoring device is used to track the progress of the collection of each special lawyer's letter in real time, and adjust the letter sending plan according to the collection effect.

9. The system according to claim 8, characterized in that The terminal visualization large screen is used to output customer service and case allocation detailed reports, and the customer service and case allocation detailed reports record the processing process of each special lawyer's letter.

Images