A collection action recommendation method, system, electronic device, and storage medium

Abstract

The present invention discloses a collection action recommendation method, system, electronic device and storage medium, which relate to the field of collection. The method includes obtaining overdue customer information; calculating a customer risk level according to the overdue customer information by using a risk rule; the risk rule is determined by using a genetic programming algorithm according to historical overdue customer information; and recommending collection actions according to the customer risk level to obtain collection actions corresponding to different customer risk levels. The present invention can automatically recommend the best collection actions and improve the collection effect.

Description

Technical Field

[0001] The present invention relates to the field of debt collection, and particularly to a method, a system, an electronic device and a storage medium for recommending debt collection actions. Background Art

[0002] Currently, debt collection work faces many challenges, such as limited number of debt collectors and lack of pertinence in debt collection strategies. Traditional debt collection methods mainly rely on manual experience and fixed debt collection processes, with low efficiency and prone to misjudgment. Therefore, an intelligent debt collection action recommendation method is needed, which can automatically recommend the best debt collection actions according to customer characteristics and historical repayment situations to improve the debt collection effect. Summary of the Invention

[0003] The purpose of the present invention is to provide a method, a system, an electronic device and a storage medium for recommending debt collection actions to automatically recommend the best debt collection actions and improve the debt collection effect.

[0004] To achieve the above purpose, the present invention provides the following solutions:

[0005] A method for recommending debt collection actions includes:

[0006] Obtaining overdue customer information;

[0007] Calculating the customer risk degree according to the overdue customer information by using risk rules; the risk rules are determined by using a genetic programming algorithm according to historical overdue customer information;

[0008] Recommending debt collection actions according to the customer risk degree to obtain debt collection actions corresponding to different customer risk degrees.

[0009] Optionally, after obtaining the overdue customer information, it further includes:

[0010] Cleaning the data of the overdue customer information; the overdue customer information includes basic information, loan information, behavior information, overdue information and previous debt collection information; the data cleaning includes outlier deletion and missing value filling.

[0011] Optionally, the determination process of the risk rules specifically includes:

[0012] Generating rule subtrees according to historical overdue customer information;

[0013] Generating initial risk rules according to the rule subtrees and calculating the fitness corresponding to the initial risk rules;

[0014] Taking the initial risk rules as the initial population of the genetic programming algorithm for iteration to obtain risk rules and the fitness corresponding to each risk rule.

[0015] Optionally, the expression of the customer risk level is as follows:

[0016]

[0017] where P i is the risk level of the i-th customer, w j is the weight of the j-th risk rule, δ ij indicates whether the i-th customer hits the j-th risk rule, and n is the number of risk rules.

[0018] Optionally, collection action recommendations are made according to the customer risk level to obtain collection actions corresponding to different customer risk levels, specifically including:

[0019] Sort and classify the customer risk levels to obtain multiple customer risk grades;

[0020] Determine collection actions corresponding to different customer risk levels according to the multiple customer risk grades; the collection actions include legal litigation, telephone collection, intelligent outbound call collection, SMS collection, and no collection for the time being.

[0021] The present invention also provides a collection action recommendation system, including:

[0022] A data acquisition module for acquiring overdue customer information;

[0023] A collection action recommendation module for calculating the customer risk level according to the overdue customer information using risk rules; the risk rules are determined by using a genetic programming algorithm according to historical overdue customer information; collection action recommendations are made according to the customer risk level to obtain collection actions corresponding to different customer risk levels.

[0024] Optionally, the collection action recommendation system further includes:

[0025] A data cleaning module for cleaning the overdue customer information; the overdue customer information includes basic information, loan information, behavior information, overdue information, and previous collection information; the data cleaning includes outlier deletion and missing value filling.

[0026] The present invention also provides an electronic device, including:

[0027] One or more processors;

[0028] A storage device on which one or more programs are stored;

[0029] When the one or more programs are executed by the one or more processors, the one or more processors are caused to implement the method as described.

[0030] The present invention also provides a computer storage medium, on which a computer program is stored, wherein the computer program, when executed by a processor, implements the method as described above.

[0031] According to the specific embodiments provided by the present invention, the following technical effects are disclosed:

[0032] The present invention obtains overdue customer information; calculates the customer risk degree according to the overdue customer information by using a risk rule; the risk rule is determined by using a genetic programming algorithm according to historical overdue customer information; and recommends collection actions according to the customer risk degree to obtain collection actions corresponding to different customer risk degrees. By performing different collection actions on customers according to different customer risk degrees, the best collection actions can be automatically recommended, and the collection effect can be improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required to be used in the embodiments. Obviously, the drawings in the following description are only some embodiments of the present invention, and those of ordinary skill in the art can also obtain other drawings based on these drawings without creative efforts.

[0034] Figure 1 It is a flowchart of the collection action recommendation method provided by the present invention;

[0035] Figure 2 It is a schematic diagram of the collection action recommendation system provided by the present invention;

[0036] Figure 3 It is a binary tree schematic diagram of a risk rule;

[0037] Figure 4 It is a schematic diagram of the risk rule crossing process;

[0038] Figure 5 It is a schematic diagram of generating a risk rule by a rule sub; DETAILED DESCRIPTION OF THE EMBODIMENTS

[0039] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, rather than all embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts belong to the scope of protection of the present invention.

[0040] The object of the present invention is to provide a collection action recommendation method, system, electronic device and storage medium to automatically recommend the best collection actions and improve the collection effect.

[0041] To make the above objects, features, and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0042] As Figure 1 shown, a collection action recommendation method provided by the present invention includes:

[0043] Step 101: Obtain overdue customer information. The data acquisition module automatically obtains overdue customer information from the big data platform, including: basic information (age, gender, marital status, occupation, education type, working years, etc.), loan information (loan amount, down payment ratio, monthly repayment amount, loan term, etc.), behavior information (number of periods repaid, proportion of periods repaid, remaining periods, proportion of remaining periods, loan balance, etc.), overdue information (number of overdue days, overdue amount, etc.), and previous collection information (number of contact times, whether lost contact, etc.).

[0044] After obtaining the overdue customer information, it further includes: cleaning the overdue customer information; the overdue customer information includes basic information, loan information, behavior information, overdue information, and previous collection information; the data cleaning includes outlier deletion and missing value filling. Perform one-hot transformation on the categorical features (such as: education type) in the basic information.

[0045] Step 102: Calculate the customer risk degree according to the overdue customer information using risk rules; the risk rules are determined by using the genetic programming algorithm based on historical overdue customer information. Among them, the risk rules can also be determined by methods such as machine learning and xgboost.

[0046] The specific process of determining the risk rules includes: generating rule subtrees according to historical overdue customer information; generating initial risk rules according to the rule subtrees and calculating the fitness corresponding to the initial risk rules; iterating the initial risk rules as the initial population of the genetic programming algorithm to obtain risk rules and the fitness corresponding to each risk rule.

[0047] The algorithm processing module generates risk rules and their weights through the genetic programming algorithm. The specific process is as follows:

[0048] Genetic Programming (GP) is an evolutionary computing technology that simulates the genetic mechanism and natural selection process in biological evolution and is used to solve optimization and machine learning problems.

[0049] The basic idea of the genetic programming algorithm is to find the optimal solution or approximate optimal solution to a problem by simulating the process of evolution. It starts the execution of the algorithm by constructing a set of initial solutions (called the population), and each solution is a candidate solution to the problem. Then, through a series of evolutionary operations (such as selection, crossover, and mutation), the solutions in the population evolve generation by generation, gradually tending towards better solutions.

[0050] The basic concepts and operations in the genetic programming algorithm are as follows:

[0051] Representation of solutions: The genetic programming algorithm usually uses a tree structure to represent solutions, where each node represents a function or termination condition, such as addition, subtraction, multiplication, etc. The root node of the tree represents the entire solution.

[0052] Initialization of the population: At the beginning of the algorithm, a set of initial solutions (trees) are randomly generated as the population.

[0053] Fitness evaluation: For each solution, its quality is evaluated through a fitness function. The fitness function is defined according to the specific requirements of the problem and can be the objective function of the problem or other measurement indicators.

[0054] Selection: Based on the fitness of the solutions, a part of the solutions are selected from the current population as the parents of the next generation. Selection strategies such as roulette wheel selection or tournament selection are usually used.

[0055] Crossover: Two solutions are selected from the parents, and offspring are generated through the crossover operation. The crossover operation can be a partial exchange or recombination of the tree structure.

[0056] Mutation: The offspring are mutated to introduce new genetic information. Mutation can be randomly changing the nodes or connections in the tree.

[0057] Replacement: A part of the solutions in the current population are replaced with the offspring to form a new population.

[0058] Termination condition: The algorithm determines whether to stop execution according to a predefined termination condition (such as reaching the maximum number of iterations or finding a solution that meets the requirements).

[0059] By continuously iterating the above steps, the genetic programming algorithm can search the solution space and gradually find better solutions. It has a wide range of applications in solving complex optimization problems, function approximation, symbolic regression, and machine learning and other fields.

[0060] Obtain the cleaned data from the data cleaning module.

[0061] The algorithm initialization unit performs the initialization operation of the algorithm as follows:

[0062] (1) Generate rule seeds: The initialization unit sends operators, features, and eigenvalues to the rule seed pool unit; the rule seed pool unit randomly selects a certain feature, operator, and eigenvalue. The operators include >, <, ≥, ≤, =. It generates rule seeds, and repeats this operation to generate a large number of rule seeds, forming a "rule seed" pool, as shown in Table 1.

[0063] Table 1 Rule Seed Examples

[0064] Feature Operator Eigenvalue Description Age ＞ 30 Age > 30 Gender ＝ Male Gender = Male Number of Repaid Periods ≤ 3 Number of Repaid Periods ≤ 3 ... ... ... ...

[0065] (2) Initialize the population: Generate n risk rules as the initial population P(t) = {x 1, , x 2 , x 3 ... x n}. The generation process of a single risk rule is as follows:

[0066] Randomly select several rule seeds from the rule seed pool unit to form a rule seed sequence, as shown in Table 2.

[0067] Table 2 Rule Seed Sequence Examples

[0068] Rule Sub - 1 Rule Sub - 2 Rule Sub - 3 Rule Sub - 4 Rule Sub - 5 ...

[0069] Randomly add rule seed operators and the calculation priorities of the operators in the rule seed sequence. The operators are ∩ (and), ∪ (union) to form a risk rule, as Figure 5 shown, and represented as a binary tree shape (risk rule tree) as Figure 3 shown.

[0070] The meaning of the risk rule is:

[0071] According to the rule seeds, screen the data sets that hit each rule seed, and perform the ∩ (and) or ∪ (union) operation on the data sets according to the operators and operator priorities between the rule seeds to obtain the final data set S that hits the risk rule i .

[0072] The fitness calculation unit calculates the fitness of the risk rule, calculates the fitness of n risk rules, and the fitness s(i,t) formula is as follows:

[0073] s(i,t) = ω 1 c′ i + ω 2 g′ i + ω 3 z′ i

[0074] c' i is the normalized risk rule complexity term, and the calculation method is:

[0075]

[0076]

[0077] Among them, c i is the original complexity of the i-th risk rule, and l i is the length of the risk subsequence corresponding to the i-th risk rule.

[0078] g' i is the normalized utility term, and the calculation method is:

[0079]

[0080]

[0081] Among them, g i is the original utility of the i-th risk rule, is the proportion of customers who finally did not repay in the dataset S i corresponding to the i-th risk rule, and p S is the proportion of customers who finally did not repay in the original dataset S.

[0082] z' i is the normalized integrity term, and the calculation method is:

[0083]

[0084]

[0085] Among them, z i is the original integrity of the i-th risk rule.

[0086] ω 1 and ω 2 and ω 3 are weight factors, and:

[0087] ω 1 + ω 2 + ω 3 = 1

[0088] Among them, the risk rule: for example, (overdue days > 30) ∩ (overdue amount > 10000). "Overdue days" and "overdue amount" are the overdue customer information.

[0089] The population evolution unit sorts the individuals in the t-th generation population according to their fitness from large to small, retains the largest n' individuals, and performs crossover and mutation operations on the retained individuals:

[0090] Crossover: The n' individuals perform crossover pairwise with probability P cPerform crossover with a certain probability and breed to form new individuals. The breeding process randomly selects non-leaf nodes in the risk rule trees of the parent and mother individuals for crossover operations, as Figure 4 shown.

[0091] Mutation: For n ‘ individuals, each individual mutates with a probability of P v to form new individuals. Individual mutation includes:

[0092] ① Deletion: Randomly delete one or several rule subsequences and their related operators in the rule subsequence.

[0093] ② Insertion: Randomly insert one or more new rule subsequences, operators, and priorities at a certain position in the rule subsequence.

[0094] ③ Variation: Randomly replace a certain rule subsequence with a new rule subsequence in the rule subsequence, or randomly change the operator priority, or randomly change the operator.

[0095] Merge the n ‘ individuals retained in the t-th generation population with the m new individuals generated through the crossover and mutation processes to form the (t + 1)-th generation population.

[0096] The algorithm termination condition judgment unit judges the termination conditions. When the termination conditions are met, the algorithm terminates and outputs the finally retained risk rule set, the fitness of each risk rule, and uses the fitness as the risk rule weight. Otherwise, the two processes of the fitness calculation unit and the population evolution unit are iteratively executed in a loop. The algorithm termination conditions are to meet the maximum number of iterations or to meet:

[0097]

[0098] where S t is the individual retained after t generations of iteration, S t-1 is the individual retained after (t - 1) generations of iteration, and T is the termination threshold.

[0099] Step 103: Recommend collection actions according to the customer risk degree to obtain collection actions corresponding to different customer risk degrees.

[0100] Step 103 specifically includes: sorting and grading the customer risk degree to obtain multiple customer risk levels; determining collection actions corresponding to different customer risk degrees according to the multiple customer risk levels; the collection actions include legal litigation, phone collection, intelligent outbound call collection, SMS collection, and no collection for the time being.

[0101] Specifically, the collection action recommendation module obtains the cleaned data from the data cleaning module, obtains the risk rules and weights from the algorithm processing module, recommends collection actions, and outputs the collection actions corresponding to the overdue cases. The process is as follows:

[0102] 1. From the algorithm processing module, generate risk rules through the genetic programming algorithm, such as generating n risk rules in total.

[0103] 2. The risk degree calculation unit calculates the customer risk degree and performs standard deviation normalization. If there are m overdue customers in total, the calculation formula for the risk degree of the i-th customer is as follows:

[0104]

[0105] Among them, n is the number of risk rules, P i is the risk degree of the i-th customer, w j is the weight of the j-th risk rule, and δ ij indicates whether the i-th customer hits the j-th risk rule:

[0106]

[0107] Risk degree standard deviation normalization:

[0108]

[0109]

[0110]

[0111] Among them, P' i is the normalized risk degree of the i-th customer, P mean is the average customer risk degree, and P std is the standard deviation of the customer risk degree.

[0112] 3. The collection action matching unit matches collection actions for each case. The process is as follows:

[0113] Sort the risk degrees of m customers from high to low, as shown in Table 3.

[0114] Table 3 Example of customer risk degree ranking

[0115]

[0116]

[0117] Divide the m customers into different risk levels in proportion. There are 5 levels from high to low. The 5th level has the highest risk, and the 1st level has the lowest risk, as shown in Table 4.

[0118] Table 4 Example of customer risk level

[0119] Customer Number Risk Level 1 5 2 4 3 3 ... ... 234 2 235 1

[0120] Overdue customers with different risk levels are assigned to corresponding collection actions, as shown in Table 5. The so-called collection actions are the collection methods in Table 5.

[0121] Table 5 Collection methods corresponding to risk levels

[0122] Risk Level Collection Method 5 Legal Litigation 4 Phone Collection 3 Intelligent Outbound Call Collection 2 SMS Collection 1 Temporarily Do Not Collect

[0123] The collection action execution module obtains the overdue cases and their corresponding collection actions from the collection action recommendation module, assigns them to the units corresponding to each collection action, and executes the specific collection actions.

[0124] As Figure 2 shown, the present invention also provides a collection action recommendation system, including:

[0125] A data acquisition module for acquiring overdue customer information.

[0126] A collection action recommendation module for calculating the customer risk degree according to the overdue customer information using risk rules; the risk rules are determined using a genetic programming algorithm based on historical overdue customer information; and collection actions are recommended according to the customer risk degree to obtain collection actions corresponding to different customer risk degrees.

[0127] As an optional implementation manner, the collection action recommendation system further includes:

[0128] A data cleaning module for cleaning the overdue customer information; the overdue customer information includes basic information, loan information, behavior information, overdue information, and previous collection information; the data cleaning includes outlier deletion and missing value filling.

[0129] The present invention also provides an electronic device, including: one or more processors; a storage device storing one or more programs thereon; when the one or more programs are executed by the one or more processors, the one or more processors are caused to implement the method as described.

[0130] The present invention also provides a computer storage medium storing a computer program thereon, wherein the computer program, when executed by a processor, implements the method as described.

[0131] The present invention generates risk rules through a genetic programming algorithm, calculates the customer risk degree through the risk rules, divides customers into different risk levels according to the risk degree, and matches the corresponding collection actions according to the customer risk levels. The higher the customer risk level, it means that, judged from historical performance, it is more difficult to succeed in collection. Therefore, more severe collection means need to be used in a timely manner. In addition, in the case of limited collection personnel, customers with a relatively high expected risk level and a relatively low collection success rate are promptly subjected to legal proceedings, avoiding waste of manpower and improving the collection efficiency.

[0132] The present invention has the following advantages:

[0133] The output risk rules have strong generalization ability. When defining the fitness, the rule complexity, utility degree, and integrity are considered simultaneously. The risk rule mining efficiency is high. By using genetic programming, a large number of risk rules can be generated. The entire collection action recommendation process is completely based on historical data and is automatic without the participation of manpower, with high efficiency.

[0134] In this specification, each embodiment is described in a progressive manner. The key point of each embodiment is to illustrate the differences from other embodiments. For the same or similar parts among the embodiments, reference can be made to each other. For the system disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and reference can be made to the description in the method part for related parts.

[0135] Specific examples are used in this article to elaborate on the principle and implementation manner of the present invention. The description of the above embodiments is only used to help understand the method and its core idea of the present invention. At the same time, for those of ordinary skill in the art, there will be changes in the specific implementation manner and application scope according to the idea of the present invention. In summary, the content of this specification should not be construed as a limitation to the present invention.

Claims

1. A method for recommending collection actions, characterized in that, it includes: Obtain overdue customer information; Perform data cleaning on the overdue customer information; the overdue customer information includes basic information, loan information, behavior information, overdue information, and previous collection information; the data cleaning includes outlier deletion and missing value filling; Calculate the customer risk level according to the overdue customer information using risk rules; The risk rules are determined using a genetic programming algorithm based on historical overdue customer information; The determination process of the risk rules specifically includes: Generate rule seeds according to historical overdue customer information; specifically including: the initial unit sends the operators, features, and feature values of the overdue customer information into the rule seed pool unit, and the rule seed pool unit randomly selects a set of operators, features, and feature values to generate rule seeds and repeats this operation to generate a large number of rule seeds, forming a rule seed pool; Generate initial risk rules according to the rule seeds and calculate the fitness corresponding to the initial risk rules; specifically including: randomly select a random number of rule seeds from the rule seed pool unit to form a rule seed sequence, and randomly add rule seed operators and the calculation priorities of the rule seed operators in the rule seed sequence to generate initial risk rules; Iterate the initial risk rules as the initial population of the genetic programming algorithm to obtain risk rules and the fitness corresponding to each risk rule; The calculation formula for the fitness is: s(i,t) = ω 1 c' i + ω 2 g' i + ω 3 z' i Among them, s(i, t) is the fitness of the i-th risk rule in the t-th generation population, c' i is the normalized risk rule complexity term, g' i is the normalized utility term, z' i is the normalized integrity term, ω 1 、ω 2 、ω 3 are weight factors, and ω 1 +ω 2 +ω 3 = 1; The expression for the customer risk level is: Among them, P i is the risk level of the i-th customer, w j is the weight of the j-th risk rule, δ ij indicates whether the i-th customer hits the j-th risk rule, and n is the number of risk rules; Recommend collection actions according to the customer risk level to obtain collection actions corresponding to different customer risk levels.

2. The method for recommending collection actions according to claim 1, characterized in that, Recommend collection actions according to the customer risk level to obtain collection actions corresponding to different customer risk levels, specifically including: Sort and classify the customer risk levels to obtain multiple customer risk grades; Determine the collection actions corresponding to different customer risk levels according to the multiple customer risk grades; the collection actions include legal litigation, telephone collection, intelligent outbound call collection, SMS collection, and no collection for the time being.

3. A collection action recommendation system, characterized in that, it includes: A data acquisition module for obtaining overdue customer information; A data cleaning module for performing data cleaning on the overdue customer information; the overdue customer information includes basic information, loan information, behavior information, overdue information, and previous collection information; the data cleaning includes outlier deletion and missing value filling; Algorithm processing module, which is used to generate risk rules and their weights through a genetic programming algorithm. The specific process is as follows: Generate rule sub - expressions based on historical overdue customer information; specifically including: The initial unit sends the operators, features, and feature values of the overdue customer information to the rule sub - expression pool unit. The rule sub - expression pool unit randomly selects any group of operators, features, and feature values to generate rule sub - expressions, and repeats this operation to generate a large number of rule sub - expressions, forming a rule sub - expression pool; Generate initial risk rules based on the rule sub - expressions and calculate the fitness corresponding to the initial risk rules; specifically including: Randomly select a random number of rule sub - expressions from the rule sub - expression pool unit to form a rule sub - expression sequence, and randomly add rule sub - expression operators and the calculation priorities of the operators in the rule sub - expression sequence to generate the initial risk rules; Use the initial risk rules as the initial population of the genetic programming algorithm for iteration to obtain risk rules and the fitness corresponding to each risk rule; The formula for calculating the fitness is: s(i,t) = ω 1 c' i +ω 2 g' i +ω 3 z' i ; where s(i,t) is the fitness of the i - th risk rule in the t - th generation population, c' i is the normalized risk rule complexity term, g' i is the normalized utility term, z' i is the normalized integrity term, ω 1 、ω 2 、ω 3 are weights factor, and ω 1 + ω 2 + ω 3 = 1; The expression of the customer risk level is: where P i is the risk level of the i-th customer, w j is the weight of the j-th risk rule, δ ij indicates whether the i-th customer hits the j-th risk rule, and n is the number of risk rules; A collection action recommendation module for calculating the customer risk level according to the overdue customer information using risk rules; the risk rules are determined using a genetic programming algorithm based on historical overdue customer information; recommend collection actions according to the customer risk level to obtain collection actions corresponding to different customer risk levels.

4. An electronic device, characterized in that, it includes: One or more processors; A storage device on which one or more programs are stored; When the one or more programs are executed by the one or more processors, the one or more processors implement the method according to any one of claims 1 to 2.

5. A computer storage medium, It is characterized in that a computer program is stored thereon, wherein when the computer program is executed by a processor, the method described in any one of claims 1 to 2 is implemented.

Images