A user credit evaluation method and system based on an improved migration genetic algorithm with dynamic weights

Abstract

The application discloses a kind of user credit evaluation methods of dynamic weight based on improved immigration genetic algorithm, comprising: selecting whether current bad account, ARPU, network duration, DOU, whether credit card automatic deduction, whether fusion, whether there is vice card, whether one card multi-number, whether black and white list as multi-scene general evaluation factor, form global variable;Global variable weight is initialized by traditional experience scoring method, the objective function and constraint condition of multi-objective optimization are established, the optimal weight is solved by using improved immigration genetic algorithm, and the user basic credit level is calculated.The application decouples user credit system, credit level and application scene, builds a unified user credit evaluation system through global variables, outputs user credit level to different application scenes, combines its own business conditions (i.e.application local variables), serves the corresponding scene, and provides useful feedback data to the user credit system, thereby affecting the fluctuation of rating, forming a closed-loop control.

Description

Technical Field

[0001] This invention relates to the field of billing and accounting technology, specifically to a user credit assessment method and system based on dynamic weights of an improved immigrant genetic algorithm. Background Technology

[0002] Credit is a capability built on trust—the ability to obtain funds, goods, services, etc., without immediate payment. Telecommunication operators can effectively reduce operational risks, improve user satisfaction, and promote sustainable corporate development by implementing credit control management for users.

[0003] The traditional credit control management process is as follows: 1. Credit control processing is triggered when a user's cumulative account data, account balance data, or credit limit data changes.

[0004] 2. Calculate the user's real-time charges based on overdue payment data.

[0005] 3. Credit monitoring, based on user status and in accordance with credit control rules, will handle payment reminders and service suspension / reactivation: Credit monitoring determines whether a user's account is suspended. If so, it checks whether the sum of the user's account balance, credit limit, current month's real-time call charges, and accumulated outstanding call charges exceeds the activation threshold. If the conditions are met, the user's account is activated; otherwise, the operation on that user is terminated.

[0006] If the user's status is not suspended, determine whether the user's account balance + user credit limit - user's real-time monthly call charges - user's accumulated arrears have reached the arrears warning threshold. If so, generate collection data and send it to the CRM user contact management; otherwise, end the operation for the user.

[0007] If the user's status is not suspended, check whether the user's account balance + user credit limit - user's current month's real-time call charges - user's accumulated arrears have reached the suspension threshold. If so, generate suspension data and send the data to CRM order processing; otherwise, end the operation for the user.

[0008] 4. Record detailed credit monitoring logs; 5. The credit monitoring process has ended.

[0009] The aforementioned traditional credit control management methods mainly have the following problems: Inaccurate user credit assessment: The credit control system mainly assesses a user's credit based on their historical arrears records and other relevant information. However, in reality, a user's credit status may change due to various factors, so this assessment method may be inaccurate.

[0010] Credit control rules are too simplistic: Control rules are often too simplistic, only considering the user's outstanding payments and credit rating, while ignoring other factors such as the user's behavior and spending habits. Therefore, they may not be able to comprehensively assess the user's credit status.

[0011] Lack of personalized services: The output of the credit control system can usually only be applied to two standardized services: user service suspension and restart, and user overdue payment reminders. It cannot be applied to personalized service scenarios based on the needs and characteristics of different users, which will affect user satisfaction and loyalty. Summary of the Invention

[0012] To address the shortcomings of existing technologies, this invention provides a user credit assessment method based on dynamic weights using an improved immigrant genetic algorithm.

[0013] This invention discloses a user credit assessment method based on dynamic weights using an improved immigrant genetic algorithm, comprising the following steps: Step 1: Construct global variables: The following factors are selected as general evaluation factors for multiple scenarios: whether there is bad debt, ARPU, network access duration, DOU, whether there is automatic credit card deduction, whether it is integrated, whether there are supplementary cards, whether one card has multiple numbers, and whether it is on a blacklist or whitelist, forming a global variable; Step 2, Unified Credit Assessment: The weights of global variables are initialized using the traditional empirical scoring method. The objective function and constraints of multi-objective optimization are established. An improved immigrant genetic algorithm is used to solve for the optimal weights and calculate the user's basic credit rating.

[0014] As a further improvement to the present invention, it also includes: Step 3, Local Variable Fusion: Local variables are selected for specific application scenarios, and global and local variables are combined. The weights are then optimized again through an improved migration genetic algorithm to generate scenario-based credit ratings. Step 4, Closed-loop optimization: Collect feedback data from application scenarios, adjust the weights of global variables accordingly, and achieve continuous optimization.

[0015] As a further improvement of the present invention, in step 2, the weights of the global variables are initialized as follows: current bad debt 0.25, ARPU 0.2, network access duration 0.15, DOU 0.1, whether credit card automatic deduction is enabled 0.1, whether it is integrated 0.05, whether there is a supplementary card 0.05, whether it has multiple numbers on one card 0.05, and whether it is on a blacklist or whitelist 0.05.

[0016] As a further improvement of the present invention, in step 2, the precocious convergence judgment rule of the improved immigrant genetic algorithm is implemented based on the population mean square error, and the mean square error formula is: In the formula: For the first t The mean square error in the population It is the i-th individual in the population. It is the average fitness value in the current population, and N is the population size.

[0017] As a further improvement of the present invention, in step 2, the immigration admission rule of the improved immigration genetic algorithm is as follows: In the formula, Mod() is the interval algebra for performing immigration operations, and it is the floor function. t Let be the current generation number; this expression represents every 100 generations. To conduct an immigration process on your behalf.

[0018] As a further improvement of the present invention, in step 2, the formula for calculating the number of immigrants in the improved immigrant genetic algorithm is as follows: In the formula, N This represents the number of immigrants currently required. `Integer[]` refers to rounding the parameter to the nearest integer. a The parameter for the immigration curve can be set to 5. For the maximum number of generations, The preset number of immigrants.

[0019] As a further improvement of the present invention, in step 2, the improved migration genetic algorithm's migration quality control method is as follows: 10% of excellent individuals and 10% of poor individuals are each selected for mutation to generate migrants with fitness higher than the population average; that is, 10% of individuals, representing 10% of the total population, are each selected from both excellent and poor individuals for mutation until a migration quality is generated. A fitness value greater than the current population average fitness value The individual, and replace the worst one in the current group. Individual.

[0020] As a further improvement of the present invention, in step 3, the specific application scenarios include overdue payment reminders, user account suspension, payment collection, and free data allowance. The local variables are selected according to the characteristics of the specific application scenario, such as contract expiration duration, consumption saturation, whether there is an overdue payment, consumption amount, and consumption frequency.

[0021] As a further improvement of the present invention, in step 4, the feedback data includes user consumption behavior, overdue payment status, and service usage records, which are used to dynamically adjust the weights of global variables.

[0022] As a further improvement of the present invention, the credit rating is divided into six levels: A, AA, AAA, AAAA, AAAAA, and blacklist, which are approximately normally distributed.

[0023] This invention also discloses a user credit assessment system that implements the above-described user credit assessment method, comprising: The data acquisition module is used to obtain accounting data and customer information data from the accounting management system and CRM system, and supports variable construction; The global variable construction module is used to filter common evaluation factors for multiple scenarios and form global variables; The weight optimization module is used to dynamically optimize the weights of global variables and "global + local" fusion variables by employing an improved migration genetic algorithm. The credit assessment module is used to calculate the basic credit rating and the scenario-based credit rating based on the optimal weights. The feedback loop module is used to collect scenario application data, adjust the weights in reverse, and optimize the evaluation system.

[0024] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention, based on traditional credit control and management methods, decouples the user credit system, credit rating, and application scenarios. It constructs a unified user credit assessment system through global variables, outputting user credit ratings to different application scenarios. Combined with its own business situation (i.e., application-specific local variables), it serves the corresponding scenarios and provides useful feedback data to the user credit system, thereby influencing rating fluctuations and forming a closed-loop control. Furthermore, the implementation employs a factor-weighted user credit rating calculation method and designs an improved genetic algorithm to solve the problem of "multi-objective optimization where the objective function is non-convex," enabling intelligent assessment and dynamic adjustment of the weight ratios of each dimension of the credit assessment model. Attached Figure Description

[0025] Figure 1 This is a framework diagram of the user credit assessment system disclosed in this invention; Figure 2 This is a flowchart of the user credit assessment method disclosed in this invention; Figure 3 A flowchart of the traditional genetic algorithm solution process; Figure 4 This diagram illustrates the number of iterations and convergence when adjusting weights using a traditional genetic algorithm. Figure 5 A diagram showing the convergence of the algorithm before and after optimization with the number of iterations; Figure 6 This is a flowchart of the combination of global and local variable factors disclosed in this invention. Detailed Implementation

[0026] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0027] The present invention will now be described in further detail with reference to the accompanying drawings: like Figure 1 As shown, the present invention provides a user credit assessment system, comprising: The data acquisition module includes an accounting management system and a CRM system. The accounting management system provides accounting-side business data to the user credit assessment module for business tagging and model training. The CRM system provides user and user information business data to the user credit assessment module for business tagging and model training. The user credit assessment unit provides a credit rating query API or data file for use in scenarios such as overdue payment collection, user account suspension, and free data usage. It includes a global variable construction module, a weight optimization module, a credit assessment module, and a feedback loop module. The global variable construction module is used to filter common evaluation factors for multiple scenarios and form global variables, which serve as the common basis for each characteristic scenario. The weight optimization module is used to dynamically optimize the weights of global variables and "global + local" fusion variables by employing an improved migration genetic algorithm. The credit assessment module is used to calculate basic credit rating and scenario-based credit rating based on optimal weights, including five main steps: importance ranking, weight allocation, credit score calculation, user classification and stratification, and credit rating determination. The feedback loop module is used to collect scenario application data, adjust the weights in reverse, and optimize the evaluation system.

[0028] like Figure 2 As shown, this invention provides a user credit assessment method based on dynamic weights using an improved immigrant genetic algorithm, comprising the following steps: S1. Construct global variables: This step primarily involves selecting common credit assessment criteria across multiple scenarios. The aim is to establish a foundational system beforehand to conduct a preliminary assessment of user credit, rapidly build a characteristic assessment model during the process, and output specific assessment results for particular scenarios, thus achieving rapid and scientific credit assessment for specific scenarios. To this end, this invention selects factors such as current bad debt status, ARPU, network access duration, DOU, whether automatic credit card deductions are used, whether the account is integrated, whether there are supplementary cards, whether there are multiple phone numbers on one card, and whether the account is on a blacklist or whitelist as common assessment factors for multiple scenarios, forming global variables. S2. Unified Credit Assessment: First, initialize the weights of global variables based on the traditional experience-based scoring method. Then, set the objective function and constraints, use the optimized genetic algorithm to obtain the optimal weight values, and calculate the final user credit score and user basic credit rating based on these weight values. 1) Using the subjective weighting method with the traditional experience scoring method, weights are set for the global variables. The specific weights are shown in Table 1.

[0029] Table 1 The result is then calculated using the Score formula and converted to a percentage system, resulting in a weighted user credit rating score: Score=(0.25 [Current bad debt status] +0.15 [Network access duration] +0.2 [ARPU]+ 0.1 [DOU]+0.1 [Whether to automatically deduct from credit card] +0.05 [Whether to merge]+ 0.05 [Do you have a secondary card?] +0.05 [Whether it's a SIM card with multiple phone numbers]+ 0.05 [Whether it is a blacklist or whitelist]) 100 Note: The above "whether it is currently a bad debt" and "whether it is on the blacklist or whitelist" are negative variables. When calculating the score, for "whether it is currently a bad debt", if the value is bad debt, the value is -1, and if the value is not bad debt, the value is 1; for "whether it is on the blacklist or whitelist", if the value is blacklist, the value is -1, and if the value is whitelist, the value is 1.

[0030] Traditional experience-based scoring methods have shortcomings and defects, such as being prone to omissions and errors in complex quantitative relationships, insufficient precision in quantitative analysis, and the generation of subjective bias. In this scheme, by establishing an objective function and constraints, a multi-objective optimization algorithm is used to reasonably calculate the weights. The weight calculation process of the optimization algorithm is as follows.

[0031] 2) Establish the objective function and constraints Let S represent the number of consumers and Y represent the number of bad debtors. ①The current formula for calculating credit rating scores is: Score=(0.25 [Current bad debt status] +0.15 [Network access duration] +0.2 [ARPU]+ 0.1 [DOU]+0.1 [Whether to automatically deduct from credit card] +0.05 [Whether to merge]+ 0.05 [Do you have a secondary card?] +0.05 [Whether it's a SIM card with multiple phone numbers]+ 0.05 [Whether it is a blacklist or whitelist]) 100= The matrix expression is: Note: Among them For the weight vector, Feature matrix.

[0032] ② Set the objective function as follows ③ Set the constraints as follows As can be seen, the above is a multi-objective optimization problem with a non-convex objective function, and the optimal solution can be obtained using a multi-objective optimization algorithm.

[0033] 3) Solving for optimal weights using multi-objective optimization algorithms An optimized genetic algorithm is used to solve for the global optimum W, where the initialization of the population (feasible solutions) includes weight values ​​obtained using a traditional empirical scoring method. The initial W is shown in Table 1.

[0034] like Figure 3 The solution process of the traditional genetic algorithm shown is to repeat steps 2 to 6 until the stopping criteria are met, such as reaching the maximum number of iterations or the objective function converging.

[0035] Step 1: Initialize the population: Based on the characteristics and constraints of the problem, randomly generate a set of initial solutions as the population. Each solution is usually represented as a chromosome, and chromosomes are composed of genes.

[0036] Step 2: Calculate fitness: For each chromosome, calculate its fitness value, which measures how well that individual performs in the optimization problem. The fitness function is designed according to the characteristics of the problem and can be a metric for the objective function value.

[0037] Step 3, Selection Operation: Drawing inspiration from the concept of "survival of the fittest," a selection operation is performed to choose individuals with higher fitness from the current population as parents for subsequent crossover and mutation operations. Selection operations are typically implemented using methods such as roulette wheel selection or tournament selection.

[0038] Step 4, Crossover: Simulating the biological crossover (mating) process, a pair of individuals is selected from the parents, and a crossover operation is performed on their chromosomes with a certain probability. The crossover operation generates a new individual by exchanging a portion of the genetic information of the chromosomes.

[0039] Step 5, Mutation Operation: Simulating the biological mutation process, mutation operations are performed on the chromosomes of new individuals with a certain probability, i.e., random gene changes occur on the chromosomes. Mutation operations increase the diversity of the population and help to escape local optima.

[0040] Step 6: Generate the next generation population: Update the current population based on the new individuals obtained from selection, crossover, and mutation operations.

[0041] Step 7: Repeat steps 2 to 6 until the stopping criteria are met, such as reaching the maximum number of iterations or the objective function converging.

[0042] Taking the consumer proxy scenario as an example, the number of iterations and convergence when adjusting weights using traditional genetic algorithms are as follows: Figure 4 As shown.

[0043] Genetic algorithms still have many areas for improvement in both theory and application methods. A prominent issue is the trade-off between their global search performance and convergence speed. Figure 4 As shown. To address this, this invention proposes a novel improved genetic algorithm, starting with the overall algorithm design and addressing aspects such as crossover strategy, crossover probability, mutation probability, and migration method, enabling the algorithm to balance global search performance and convergence speed.

[0044] This invention addresses the shortcomings of genetic algorithms by introducing an immigration method for optimization. Immigration is an effective means of suppressing premature convergence and maintaining population diversity, effectively compensating for the search limitations of crossover and mutation operators. However, determining when to introduce immigrants, how many to introduce, and what quality of immigrants to introduce is often difficult. Therefore, considering both computational complexity and algorithm performance, this paper adopts the following approach: The standard mean square deviation of the population is used to measure the degree of premature convergence, and the immigration admission rules are determined based on the number of generations. Once premature convergence is detected or a certain number of generations have passed, the required number of high-quality immigrants are generated according to certain rules to replace inferior individuals in the population, based on the evolutionary situation.

[0045] The precocious puberty criterion is based on the mean square deviation, which is: In the formula: Let be the mean squared error in the t-th generation population. It is the i-th individual in the population. It is the average fitness value in the current population; N is the population size.

[0046] The drawbacks of relying solely on this method for judging premature convergence are obvious—because the mean squared error is similar and doesn't change much when the population converges in the later stages of the calculation. If immigration is introduced even when the algorithm has converged, it will not only increase the computational burden unnecessarily, but also affect the stability of the algorithm's convergence in the later stages, and may even cause the algorithm to fail to converge.

[0047] To this end, this paper designs immigration admission rules related to the number of generations, ensuring that the population can be replenished with a certain number of superior individuals at least every certain number of generations. In the formula, Mod() is the integer division operator for the interval algebra used in immigration operations; this expression means every [number of algebraic intervals]. This is to conduct an immigration process on behalf of someone else. The formula for calculating the number of immigrants is: In the formula, N is the number of immigrants required; Integer[] refers to rounding the parameter to the nearest integer; a is the immigration curve parameter, which can be 5. Maximum number of generations The preset number of immigrants.

[0048] The quality of immigrants also has a significant impact on algorithm improvement. In this invention, the generation of new immigrants involves selecting 10% of the total population from both high-quality and low-quality individuals and performing mutation operations until a new immigrant is generated. A fitness value greater than the current population average fitness value The individual, and replace the worst one in the current group. individual The improved migration operation derived from the above three formulas is as follows: If the premature convergence judgment rule or the migration calculation condition is met, then the migration number is calculated, and the aforementioned migration operation is performed accordingly. As the evolution progresses, the number of generated migrations will gradually decrease to protect the stability of the algorithm.

[0049] Taking a consumer proxy scenario as an example, the convergence of the algorithm before and after optimization with the number of iterations is compared to... Figure 5 As shown, it is evident that the optimized model converges faster, and the final weight changes are less different, indicating that the optimization method proposed in this scheme improves the convergence speed and the performance of the global optimal solution.

[0050] S3, Local Variable Fusion: Local variables are selected for specific application scenarios, and global and local variables are combined. The weights are then optimized again through an improved migration genetic algorithm to generate scenario-based credit ratings. like Figure 6 As shown, the combination of global and local variable factors is based on the unified user credit assessment using global variable factors. However, according to the specific use case, characteristic local variables of that scenario are introduced, and the multi-objective optimization algorithm used in the proposal is reused to calculate the final weights. These weights are then used to calculate the user's credit rating under that specific scenario. The specific steps are as follows: Step 1: Select local variable factors based on specific business scenarios; Step 2: Merge the global variable factors and local variable factors, and reinitialize the weights of each factor; Step 3: Calculate the optimal weights of each factor based on the migration genetic algorithm used in this invention; Step 4: Calculate the final user credit assessment for the specific scenario using the optimal weights; the unified user credit system can be divided into 6 levels: A, AA, AAA, AAAA, AAAAA, and blacklist. The distribution of the 6 levels is reasonable, uniform, and approximately normal.

[0051] S4, Closed-loop optimization: Collect feedback data from application scenarios, adjust the weights of global variables accordingly, and achieve continuous optimization.

[0052] Application effect In addition to its application in traditional scenarios such as service suspension / reactivation and overdue payment collection, the credit assessment system is also applied to scenarios involving bill payment and free data usage. Taking one city as an example, after implementing this solution in the bill payment scenario, user spending increased by 1.566 million yuan; in the free data usage scenario, data vouchers were appropriately given to approximately 420,000 users with high credit ratings to enhance user care, effectively improving user satisfaction and user stickiness.

[0053] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A user credit assessment method based on dynamic weights using an improved immigrant genetic algorithm, characterized in that, Includes the following steps: Step 1: Select the following as general evaluation factors for multiple scenarios: whether there is bad debt, ARPU, network access duration, DOU, whether credit card deduction is automatic, whether the account is integrated, whether there are supplementary cards, whether one card has multiple numbers, and whether it is on a blacklist or whitelist, to form a global variable; Step 2: Initialize the weights of global variables using the traditional empirical scoring method, establish the objective function and constraints for multi-objective optimization, and use the improved immigrant genetic algorithm to solve for the optimal weights and calculate the user's basic credit rating.

2. The user credit assessment method according to claim 1, characterized in that, Also includes: Step 3: Select local variables for specific application scenarios, integrate global and local variables, and re-optimize the weights through an improved immigrant genetic algorithm to generate scenario-based credit ratings; Step 4: Collect scenario application feedback data and adjust the weights of global variables accordingly.

3. The user credit assessment method according to claim 1 or 2, characterized in that, In step 2, Initialize the weights of the global variables as follows: current bad debt status 0.25, ARPU 0.2, network access duration 0.15, DOU 0.1, whether credit card automatic deduction is enabled 0.1, whether it is integrated 0.05, whether there is a supplementary card 0.05, whether it has multiple numbers on one card 0.05, whether it is on a blacklist or whitelist 0.05; The precocious convergence judgment rule of the improved immigrant genetic algorithm is based on the population mean square error, and the mean square error formula is: ； In the formula: For the first t The mean square error in the population It is the i-th individual in the population. It is the average fitness value in the current population, and N is the population size.

4. The user credit assessment method according to claim 1 or 2, characterized in that, In step 2, the immigration admission rules of the improved immigrant genetic algorithm are as follows: ； In the formula, Mod() is the interval algebra for performing immigration operations, and it is the floor function. t This represents the current generation.

5. The user credit assessment method according to claim 1 or 2, characterized in that, In step 2, the formula for calculating the number of immigrants using the improved immigrant genetic algorithm is as follows: ； In the formula, N This represents the number of immigrants currently required. `Integer[]` refers to rounding the parameter to the nearest integer. a These are parameters for the immigration curve. For the maximum number of generations, The preset number of immigrants.

6. The user credit assessment method according to claim 1 or 2, characterized in that, In step 2, the improved migration genetic algorithm's migration quality control method is as follows: 10% of excellent individuals and 10% of poor individuals are each mutated to generate migrants with fitness higher than the population average.

7. The user credit assessment method according to claim 2, characterized in that, In step 3, the specific application scenarios include overdue payment reminders, user account suspension, payment collection on behalf of others, and free data allowance. Local variables are selected based on the characteristics of the specific application scenario, such as contract expiration duration, consumption saturation, whether there is an overdue payment, consumption amount, and consumption frequency.

8. The user credit assessment method according to claim 2, characterized in that, In step 4, the feedback data includes user consumption behavior, outstanding payments, and service usage records, which are used to dynamically adjust the weights of global variables.

9. The user credit assessment method according to claim 1 or 2, characterized in that, The credit rating is divided into six levels: A, AA, AAA, AAAA, AAAAA, and blacklist, which are approximately normally distributed.

10. A user credit assessment system that implements the user credit assessment method according to any one of claims 1 to 9, characterized in that, include: The data acquisition module is used to obtain accounting data and customer information data from the accounting management system and CRM system, and supports variable construction; The global variable construction module is used to filter common evaluation factors for multiple scenarios and form global variables; The weight optimization module is used to dynamically optimize the weights of global variables and "global + local" fusion variables by employing an improved migration genetic algorithm. The credit assessment module is used to calculate the basic credit rating and the scenario-based credit rating based on the optimal weights. The feedback loop module is used to collect scenario application data, adjust the weights in reverse, and optimize the evaluation system.

Images