Customer full-duty investigation task allocation method, storage medium and electronic equipment

Abstract

The invention discloses a customer full-duty investigation task allocation method, a storage medium and electronic equipment, and the method comprises the steps: carrying out the customer feature extraction through calling industrial and commercial information by a system after a new customer is introduced into a channel, and matching a customer manager set according to the customer account opening full-duty investigation demand, customer feature information and historical task execution time and empirical value information of customer managers are used as input parameters, task duration is automatically predicted through a constructed random forest model, an optimal customer manager is selected for task allocation, and after tasks are completed, empirical values of the customer managers are adjusted in combination with actual task execution conditions, so that the task allocation efficiency is improved. And task allocation and execution are realized. According to the method, the enterprise customer features are extracted, and multiple factors such as industries, personnel, assets, business scales and stock equity modes are combined, so that task prediction and allocation are performed by adopting an improved random forest model algorithm, and optimal allocation of the full-duty investigation tasks and dynamic self-adaption of the system are ensured.

Description

Technical Field

[0001] The present invention relates to a method, storage medium and electronic device for allocating customer due diligence tasks, belonging to the field of financial services, and in particular to a method and system for automatically allocating customer due diligence tasks based on an improved random forest algorithm. Background Art

[0002] Before a company opens an account, the account manager of a financial institution must conscientiously perform on-site due diligence. This proposal intends to use an improved random forest algorithm to realize the automatic online allocation of due diligence tasks for industrial and commercial account opening companies, thereby improving the quality and efficiency of account opening for corporate customers of financial institutions.

[0003] At present, the acquisition and allocation of two-person on-site due diligence tasks in the financial industry are mainly done offline. There is no digital task allocation plan and means for due diligence scenarios. Expanding to automatic task allocation, there are many allocation methods after retrieval, such as multi-drone task allocation for artificial bee colony algorithms, complaint task allocation methods based on keyword matching, etc. However, none of them can meet the task allocation method affected by multiple factors in due diligence business scenarios.

[0004] The present invention is to maintain a feature value library of account managers by the system background, which contains the experience values ​​of account managers in terms of industry, scale, enterprise location, account type and other features. When there is a willingness to open an account, the system extracts features from the customer according to the industry, scale and holding nature of the customer, combines the geographical location and free time of the account manager, and uses a rule-based matching algorithm to match the appropriate account manager set. The improved random forest algorithm is used to predict the time required for the account manager to complete the due diligence task to achieve automatic task allocation. After the account is opened, the account opening efficiency and experience value are dynamically adjusted and post-evaluated. Summary of the invention

[0005] The present invention aims at the problems in the prior art and provides a method, storage medium and electronic device for assigning customer due diligence tasks. Starting from the introduction of corporate customers, the complexity of the task, the experience value of the account manager, the geographical location distance, the expected completion time of the task, and the actual completion time of the task are trained to predict the current task completion time. The most suitable account manager is matched in combination with the account manager's geographical location, the time to execute the task, the account manager's experience value, and the current load level of the account manager, and the account manager's experience value and account opening risk value are adjusted in combination with the account opening results and account usage, so as to achieve the purpose of improving the efficiency and quality of corporate customer account opening.

[0006] In order to achieve the above object, the technical solution adopted by the present invention is:

[0007] A method for assigning customer due diligence tasks, characterized in that after a new customer is introduced into a channel, the system extracts customer features by calling industrial and commercial information, matches a customer manager set according to the customer account opening due diligence requirements (location, time), takes customer feature information and the historical task execution time and experience value information of the customer manager as input parameters, automatically predicts the task duration through a constructed random forest model, selects the optimal customer manager for task assignment, and after the task is completed, adjusts the experience value of the customer manager in combination with the actual task execution situation to achieve task assignment and execution; the specific steps are as follows:

[0008] S1. Extract customer features to obtain a customer feature value library;

[0009] S2. Calculate the matching account manager set using a simple rule matching algorithm based on the customer feature value database;

[0010] S3, predicting task duration through the constructed random forest model;

[0011] S4. Perform task allocation and optimization based on the obtained duration prediction;

[0012] S5. After the task is completed, conduct a post-evaluation based on the actual task completion time.

[0013] Furthermore, in step S1, the customer characteristics, including the industry, holding type, asset size, employee size, etc., are extracted by calling the business registration information of the enterprise, and used as input characteristics for the subsequent account manager matching and task allocation process. The system needs to continuously maintain a feature value library of account managers and determine the manager allocation based on these characteristics.

[0014] Furthermore, in step S2, the customer feature value library includes industry categories, scale features, and other industry-related features, wherein industry categories include IT, finance, manufacturing, etc., scale features are classified according to the number of employees, annual income, or business scale of the enterprise, and other industry-related features include the geographical location of the enterprise, the type of account opening requirements, etc. These feature values ​​will be matched with the experience and current workload of each account manager to help select the most suitable account manager.

[0015] Furthermore, in step S2, the simple rule matching algorithm calculates the geographical distance between the customer and the manager and ensures that it is within a given radius; when matching the account manager, priority is given to screening based on the customer's geographical location and due diligence time requirements; during the preliminary screening, the matched managers meet the following two conditions: geographical proximity: calculate the distance between the customer's geographical location and the location of the account manager's branch, and give priority to matching managers with a closer distance; time availability: based on the account manager's work schedule and current task load, screen out account managers who have time to handle new tasks.

[0016] Furthermore, in step S3, the improved random forest regression model is used to input historical task features (such as industry type, scale, account manager experience, etc.) and known task completion time for training; through training, the model can automatically learn the relationship between features and task duration. The following steps are included:

[0017] (1) Input feature definition: In the improved random forest regression model, each task has multiple features as input. These features form a feature vector, which is used to describe the relevant information of the task. The model predicts the duration of the task based on this information. The features include: industry (such as software development services, finance, etc.), which is used to describe the customer industry served by the task; scale attributes, such as initial scale, medium scale, large scale, etc., which are used to characterize the size of the task; sales revenue: used to identify the business operation of the enterprise; account manager experience value, which is used to quantify the experience level of the account manager for such characteristic customers; account manager available time, which is used to indicate the time window that the account manager can allocate to the task;

[0018] (2) Data preprocessing: First, the missing values ​​of the original historical task data are filled, and the format of the number of employees, asset scale, and sales revenue is converted to extract feature values ​​to improve the accuracy of model training;

[0019] (3) Prepare training and test sets; divide the preprocessed data into training and test sets, where the training set is used to train the parameters and weights of the model and to train the model so that the model can learn the relationship between features and task duration; the test set is used to evaluate the performance and generalization ability of the model, and the validation set is used to adjust the hyperparameters of the model and perform model selection; at the same time, in order to avoid the situation where the data set has category imbalance, the data is divided into layers;

[0020] Input feature set: various feature values ​​of historical tasks, such as industry type, enterprise size, account manager experience, etc. defined above.

[0021] Output target value: The actual duration of each task, which serves as the target value of the regression model.

[0022] (4) Model construction: It consists of multiple decision trees (usually dozens to hundreds). The specific process is as follows:

[0023] Bootstrap: When training each decision tree, a sample subset is randomly selected from the original data set for training. Through the method of sampling with replacement, the training data of each tree is different, thus introducing the diversity of the model.

[0024] Random feature selection: When splitting the nodes of each decision tree, a subset of features is randomly selected instead of using all features for splitting, which is used to reduce the correlation between trees and further improve the generalization ability of the model;

[0025] (5) Decision tree training: During the decision tree training process, the K-fold cross-validation method is used, that is, the original training data set is divided into K subsets, and K-1 subsets are used as training sets each time, and the remaining subset is used as the validation set. This is repeated K times and the average result is taken. The training set is used to adjust the hyperparameters of the model and perform model selection. This method can prevent the risk of overfitting of a single tree and improve the generalization ability of the model; Node splitting: Each node of the tree is binary split according to a certain characteristic value (such as enterprise size or industry type). Each split attempts to maximize the purity of the data in the node so that the prediction result is more accurate; Leaf node generation: When no effective split can be performed, a leaf node is generated. The leaf node contains samples under the same characteristic conditions, and the value of the leaf node is the average duration of these samples.

[0026] The formula for subset partitioning of the K-fold cross-validation method is as follows:

[0027] D=D1∪D2∪D3∪…∪D K

[0028]

[0029] Among them, D is the original training data set, D1D2…D K It is obtained by stratified sampling from D. Each time, the union of K-1 subsets is used as the training set, and the remaining subsets are used as the test set.

[0030] The error calculation formula of the K-fold cross validation method is as follows:

[0031]

[0032] Among them, CV (K) It identifies the cross-validation error, K represents the number of data samples, y_test represents the actual observed value, and y_pred identifies the model predicted value.

[0033] The cross-validation formula is to obtain the performance index of the model by calculating the average error between the actual observed value y_test and the model predicted value y_pred.

[0034] (6) Improved random forest model generation: The decision tree is cross-validated by K-fold and CV (K) Sort and take the first S CVs (K) The decision trees corresponding to the values ​​constitute the improved random forest model.

[0035] (7) Task duration prediction: When a new task arrives, its features are input into the trained random forest regression model. The model outputs the predicted task duration, which helps to reasonably allocate tasks and avoid manager workload overload or task delays.

[0036] Furthermore, in step S4, priority is sorted according to the duration of the task, the workload of the account manager, and the characteristic weight of the task enterprise. Tasks with longer duration or higher priority will be preferentially assigned to account managers who meet the enterprise's characteristic attributes. If there is no account manager who fully meets the enterprise's characteristic attributes, the task and the account manager are matched according to the weight of the characteristic attributes. In addition, according to the current load of each account manager (the number of assigned tasks and the estimated working time), the most suitable manager is selected to undertake the new task to ensure balanced workload distribution.

[0037] Furthermore, in step S5, a post-evaluation is performed based on the actual task completion time, and the account manager's experience value is adjusted; in this process, the experience value will be updated based on the following factors: task completion efficiency, if a certain account manager completes a task faster or slower than expected, his experience value will be adjusted according to the actual situation, affecting his matching degree in future task allocation; accumulation of specific industry experience, the more tasks a manager performs in a certain industry, the more experience the manager has in that industry. This step of dynamic adjustment helps to improve the accuracy and rationality of the system in long-term task allocation.

[0038] Furthermore, it also includes visualization and feedback processes. The results of account manager task allocation will be visualized through maps to facilitate subsequent analysis and manual intervention; the locations of customers and account managers are displayed on the map, marking the process of task allocation and providing feedback on the rationality of task allocation. This process design can effectively cope with complex task allocation scenarios, especially considering multiple factors such as geographical location, time arrangement, industry characteristics and account manager experience, to ensure the optimization of task allocation and dynamic adaptation of the system.

[0039] The present invention provides a computer storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps of the customer due diligence task allocation method are implemented.

[0040] The present invention also provides an electronic device, including a memory and one or more processors, wherein the memory is used to store one or more programs; when the one or more programs are executed by the one or more processors, the customer due diligence task allocation method is implemented.

[0041] Beneficial effects of the present invention:

[0042] The present invention extracts the characteristics of corporate customers, combines multiple factors such as industry, personnel, assets and business scale, equity structure, etc., and adopts an improved random forest model algorithm to predict and allocate tasks, thereby ensuring the optimal allocation of due diligence tasks and dynamic self-adaptation of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] 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 work.

[0044] Figure 1 It is a flow chart of the method of the present invention.

[0045] Figure 2 It is a flow chart of random forest regression model training in the present invention.

[0046] Figure 3 It is an example diagram of historical data in the present invention.

[0047] Figure 4 It is a visualization effect diagram of task allocation in the present invention. DETAILED DESCRIPTION

[0048] 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.

[0049] like Figure 1 As shown, the present invention provides a method for assigning customer due diligence tasks. After a new customer is introduced into a channel, the system extracts customer features by calling industrial and commercial information, matches a customer manager set according to the customer account opening due diligence requirements (location, time), takes customer feature information and the historical task execution time and experience value information of the customer manager as input parameters, automatically predicts the task duration through the constructed random forest model, selects the optimal customer manager for task assignment, and adjusts the experience value of the customer manager in combination with the actual task execution situation after the task is completed, so as to realize task assignment and execution; the specific steps are as follows:

[0050] S1. Extract customer features to obtain a customer feature value library;

[0051] S2. Calculate the matching account manager set using a simple rule matching algorithm based on the customer feature value database;

[0052] S3, predicting task duration through the constructed random forest model;

[0053] S4. Perform task allocation and optimization based on the obtained duration prediction;

[0054] S5. After the task is completed, conduct a post-evaluation based on the actual task completion time.

[0055] In this embodiment, in step S1, the customer characteristics, including the industry, holding type, asset size, employee size, etc., are extracted by calling the business registration information of the enterprise, and used as input characteristics for the subsequent account manager matching and task allocation process. The system needs to continuously maintain a feature value library of account managers and determine the allocation of managers based on these characteristics.

[0056] In this embodiment, in step S2, the customer feature value library includes industry categories, scale features, and other industry-related features, wherein industry categories include IT, finance, manufacturing, etc., scale features are classified according to the number of employees, annual income, or business scale of the enterprise, and other industry-related features include the geographical location of the enterprise, the type of account opening requirements, etc. These feature values ​​will be matched with the experience and current workload of each account manager to help select the most suitable account manager.

[0057] In this embodiment, in step S2, the simple rule matching algorithm calculates the geographical distance between the customer and the manager and ensures that it is within a given radius; when matching the customer manager, priority is given to screening based on the customer's geographical location and due diligence time requirements; during the preliminary screening, the matched managers meet the following two conditions: geographical proximity: calculate the distance between the customer's geographical location and the location of the customer manager's branch, and give priority to matching managers with a closer distance; time availability: based on the customer manager's work schedule and current task load, screen out customer managers who have time to handle new tasks.

[0058] In this embodiment, in step S3, an improved random forest regression model is used to input historical task features (such as industry type, scale, account manager experience, etc.) and known task completion time for training; through training, the model can automatically learn the relationship between features and task duration, such as Figure 2 As shown. The following steps are included:

[0059] (1) Input feature definition: In the improved random forest regression model, each task has multiple features as input. These features form a feature vector that is used to describe the relevant information of the task. The model predicts the duration of the task based on this information.

[0060] The features include: industry (such as software development services, finance, etc.), which is used to describe the industry of the customers served by the task; scale attributes, such as initial scale, medium scale, large scale, etc., which are used to characterize the size of the task; sales revenue: used to identify the business operation of the enterprise; account manager experience value, which is used to quantify the account manager's experience level with such characteristic customers; account manager's available time, which is used to indicate the time window that the account manager can allocate to the task;

[0061] (2) Data preprocessing: First, the missing values ​​of the original historical task data are filled, and the format of the number of employees, asset scale, and sales revenue is converted to extract feature values ​​to improve the accuracy of model training, such as Figure 3 As shown;

[0062] (3) Prepare training and test sets; divide the preprocessed data into training and test sets, where the training set is used to train the parameters and weights of the model and to train the model so that the model can learn the relationship between features and task duration; the test set is used to evaluate the performance and generalization ability of the model, and the validation set is used to adjust the hyperparameters of the model and perform model selection; at the same time, in order to avoid the situation where the data set has category imbalance, the data is divided into layers;

[0063] Input feature set: various feature values ​​of historical tasks, such as industry type, enterprise size, account manager experience, etc. defined above.

[0064] Output target value: The actual duration of each task, which serves as the target value of the regression model.

[0065] (4) Model construction: It consists of multiple decision trees (usually dozens to hundreds). The specific process is as follows:

[0066] Bootstrap: When training each decision tree, a sample subset is randomly selected from the original data set for training. Through the method of sampling with replacement, the training data of each tree is different, thus introducing the diversity of the model.

[0067] Random feature selection: When splitting the nodes of each decision tree, a subset of features is randomly selected instead of using all features for splitting, which is used to reduce the correlation between trees and further improve the generalization ability of the model;

[0068] (5) Decision tree training: During the decision tree training process, the K-fold cross-validation method is used, that is, the original training data set is divided into K subsets, and K-1 subsets are used as training sets each time, and the remaining subset is used as the validation set. This is repeated K times and the average result is taken. The training set is used to adjust the model's hyperparameters and perform model selection. This method can prevent the risk of overfitting of a single tree and improve the generalization ability of the model. Node splitting: Each node of the tree is binary split according to a certain feature value (such as enterprise size or industry type). Each split attempts to maximize the purity of the data in the node so that the prediction result is more accurate; Leaf node generation: When no effective split can be performed, a leaf node is generated. The leaf node contains samples under the same feature conditions, and the value of the leaf node is the average duration of these samples.

[0069] K-fold cross validation method subset partitioning:

[0070] D=D1∪D2∪D3∪…∪D K

[0071]

[0072] Among them, D is the original training data set, D1D2…D K It is obtained by stratified sampling from D. Each time, the union of K-1 subsets is used as the training set, and the remaining subsets are used as the test set.

[0073] Cross validation error calculation:

[0074]

[0075] Among them, CV (K) It identifies the cross-validation error, K represents the number of data samples, y_test represents the actual observed value, and y_pred identifies the model predicted value.

[0076] The cross-validation formula is to obtain the performance index of the model by calculating the average error between the actual observed value y_test and the model predicted value y_pred.

[0077] (6) Improved random forest model generation: The decision tree is cross-validated by K-fold and CV (K) Sort and take the first S CVs (K) The decision trees corresponding to the values ​​constitute the improved random forest model.

[0078] (7) Task duration prediction: When a new task arrives, its features are input into the trained random forest regression model. The model outputs the predicted task duration, which helps to reasonably allocate tasks and avoid manager workload overload or task delays.

[0079] In this embodiment, in step S4, priority is sorted according to the duration of the task, the workload of the account manager, and the characteristic weight of the enterprise to which the task belongs. Tasks with longer duration or higher priority will be preferentially assigned to account managers who meet the enterprise's characteristic attributes. If there is no account manager who fully meets the enterprise's characteristic attributes, the task and the account manager are matched according to the weight of the characteristic attributes. In addition, according to the current load of each account manager (the number of assigned tasks and the estimated working time), the most suitable manager is selected to undertake the new task to ensure balanced workload distribution.

[0080] In this embodiment, in step S5, a post-evaluation is performed based on the actual task completion time, and the account manager's experience value is adjusted; in this process, the experience value will be updated based on the following factors: task completion efficiency, if a certain account manager completes a task faster or slower than expected, his experience value will be adjusted according to the actual situation, affecting his matching degree in future task allocation; accumulation of specific industry experience, the more tasks a manager performs in a certain industry, the more experience the manager has in that industry. This step of dynamic adjustment helps to improve the accuracy and rationality of the system in long-term task allocation.

[0081] In this embodiment, visualization and feedback processes are also included. The task assignment results of the account manager will be visualized through a map to facilitate subsequent analysis and manual intervention. Figure 4 As shown in the figure, the location of the customer and the account manager are shown on the map, the process of task allocation is marked, and feedback on the rationality of task allocation is given. This process design can effectively cope with complex task allocation scenarios, especially considering multiple factors such as geographical location, time arrangement, industry characteristics and account manager experience, ensuring the optimization of task allocation and dynamic adaptation of the system.

[0082] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above examples. The above examples and descriptions are only for explaining the principles of the present invention. The present invention may be subject to various changes and improvements without departing from the spirit and scope of the present invention. These changes and improvements fall within the scope of the present invention. The scope of protection of the present invention is defined by the attached claims and their equivalents.

Claims

1. A method for allocating customer due diligence tasks, characterized in that: After a new customer is introduced into the channel, the system extracts customer features by calling the industrial and commercial information, matches the account manager set according to the customer account opening due diligence requirements, takes the customer feature information and the account manager's historical task execution time and experience value information as input parameters, automatically predicts the task duration through the constructed random forest model, selects the best account manager for task assignment, and adjusts the account manager's experience value based on the actual task execution situation after the task is completed, so as to realize task assignment and execution; the specific steps are as follows: S1. Extract customer features to obtain a customer feature value library; S2. Calculate the matching account manager set using a simple rule matching algorithm based on the customer feature value database; S3, predicting task duration through the constructed random forest model; S4. Perform task allocation and optimization based on the obtained duration prediction; S5. After the task is completed, conduct a post-evaluation based on the actual task completion time.

2. A customer due diligence task allocation method according to claim 1, characterized in that: In step S1, customer characteristics are extracted by calling the business registration information of the enterprise, including the industry, holding type, asset size, and employee size, and used as input characteristics for the subsequent account manager matching and task allocation process.

3. A customer due diligence task allocation method according to claim 2, characterized in that: In step S2, the customer feature value library includes industry categories, scale characteristics, and other industry-related characteristics, wherein the industry categories include IT, finance, and manufacturing; the scale characteristics are classified according to the number of employees, annual income, or business scale; and other industry-related characteristics include the geographical location of the enterprise and the type of account opening requirements.

4. A method for allocating customer due diligence tasks according to claim 3, characterized in that: In step S2, the simple rule matching algorithm calculates the geographical distance between the customer and the manager and ensures that it is within a given radius; when matching the account manager, priority is given to screening based on the customer's geographical location and due diligence time requirements; during the initial screening, the matched managers meet the following two conditions: geographical proximity: calculate the distance between the customer's geographical location and the location of the account manager's branch, and give priority to matching managers with a closer distance; time availability: based on the account manager's work schedule and current task load, screen out account managers who have time to handle new tasks.

5. A method for allocating customer due diligence tasks according to claim 4, characterized in that: In step S3, the improved random forest regression model is used to input historical task features and known task completion time for training; the following steps are included: (1) Input feature definition: In the improved random forest regression model, each task has multiple features as input. These features form a feature vector that is used to describe the relevant information of the task. The model predicts the duration of the task based on this information. The features include: industry, which is used to describe the industry of the customers served by the task; scale attribute, which is used to characterize the size of the task; sales revenue, which is used to identify the business operation of the enterprise; account manager experience value, which is used to quantify the account manager's experience level with such characteristic customers; account manager's available time, which is used to indicate the time window that the account manager can allocate to the task; (2) Data preprocessing: First, missing values ​​are filled in the historical task data, and the number of employees, asset size, and sales revenue are converted to a different format to extract feature values ​​to improve the accuracy of model training. (3) Prepare training and test sets; divide the preprocessed data into training and test sets, where the training set is used to train the parameters and weights of the model and to train the model so that the model can learn the relationship between features and task duration; the test set is used to evaluate the performance and generalization ability of the model, and the validation set is used to adjust the hyperparameters of the model and perform model selection; at the same time, in order to avoid the situation where the data set has category imbalance, the data is divided into layers; (4) Model construction: It consists of multiple decision trees. The specific process is as follows: Random sampling: When training each decision tree, a sample subset is randomly selected from the original data set for training; through the method of sampling with replacement, the training data of each tree is different, thus introducing model diversity; Random feature selection: When each decision tree node is split, a subset of features is randomly selected to reduce the correlation between trees and further improve the generalization ability of the model; (5) Decision tree training; During the decision tree training process, the K-fold cross-validation method is used, that is, the training data set is divided into K subsets, and K-1 subsets are used as training sets each time, and the remaining subset is used as validation sets. This is repeated K times and the average result is taken. The training set is used to adjust the model's hyperparameters and perform model selection to improve the training effect of the decision tree. Each node of the tree is binary split according to a certain feature value, and each split attempts to maximize the purity of the data in the node so that the prediction result is more accurate. When no effective split can be performed, a leaf node is generated. The leaf node contains samples under the same feature conditions, and the value of the leaf node is the average duration of these samples. The formula for subset partitioning of the K-fold cross-validation method is as follows: D=D1∪D2∪D3∪…∪D K Among them, D is the original training data set, D1D2…D K Obtained by stratified sampling from D; each time the union of K-1 subsets is used as the training set, and the remaining subsets are used as the test set; The error calculation formula of the K-fold cross validation method is as follows: Among them, CV (K) Identifies the cross-validation error, K represents the number of data samples, y_test represents the actual observation value, and y_pred identifies the model prediction value; the performance index of the model is obtained by calculating the average error between the actual observation value y_test and the model prediction value y_pred; (6) Improved random forest model generation: The decision tree is cross-validated by K-fold and CV (K) Sort and take the first S CVs (K) The decision trees corresponding to the values ​​constitute the improved random forest model; (7) Task duration prediction: When a new task arrives, its features are input into the trained random forest regression model, and the model outputs the predicted task duration, which is used to reasonably allocate tasks and avoid manager workload overload or task delays.

6. A method for allocating customer due diligence tasks according to claim 5, characterized in that: In step S4, priority is sorted according to the duration of the task, the workload of the account manager, and the characteristic weight of the task enterprise. Tasks with longer duration or higher priority will be preferentially assigned to account managers who meet the enterprise characteristic attributes. If there is no account manager who fully meets the enterprise's characteristic attributes, the task and account manager are matched according to the weight of the characteristic attributes; in addition, based on the current load of each account manager, the most suitable manager is selected to take on the new task to ensure a balanced distribution of workload.

7. A method for allocating customer due diligence tasks according to claim 6, characterized in that: In step S5, a post-evaluation is performed based on the actual task completion time, and the account manager's experience value is adjusted; in this process, the experience value is updated based on the following factors: task completion efficiency. If a certain account manager completes a task faster or slower than expected, his experience value will be adjusted according to the actual situation, affecting his matching degree in future task allocation; accumulation of specific industry experience. The more tasks a manager performs in a certain industry, the more experience value the manager will automatically increase in that industry.

8. A method for allocating customer due diligence tasks according to claim 7, characterized in that: It also includes visualization and feedback processes. The results of account manager task assignments are visualized through maps to facilitate subsequent analysis and manual intervention. The map shows the locations of customers and account managers respectively, marks the task assignment process, and gives feedback on the rationality of task assignments.

9. A computer storage medium, characterized in that A computer program is stored thereon, and when the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 8 are implemented.

10. An electronic device, characterized in that: The method comprises a memory and one or more processors, wherein the memory is used to store one or more programs; when the one or more programs are executed by the one or more processors, the method according to any one of claims 1 to 8 is implemented.

Images