Intelligent dialing system for post-loan management
Through the RNN recurrent neural network model prediction and the gray wolf optimization algorithm to optimize the dialing time period, the problems of low efficiency of collection information transmission and concentrated connection rate in the post-loan intelligent dialing system are solved, and the quality of post-loan management and collection efficiency are improved.
Patent Information
- Application Number
- CN202411210680.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-30
- Publication Date
- 2024-12-20
AI Technical Summary
The traditional post-loan intelligent dialing system has the risk of collecting information being flooded by information, and it is unable to receive customers' repayment intentions in time. The telephone connection rate is concentrated during peak periods, resulting in low dialing configuration efficiency and affecting the quality of post-loan management.
The RNN recurrent neural network model is used to predict the connection rate of different time periods, and the customer's dialing time period is optimized through the gray wolf optimization algorithm to improve the connection rate of intelligent dialing and form the optimal dialing strategy.
The connection rate and post-loan management quality of intelligent dialing are improved, the dialing is ensured during the high connection rate period, the resource allocation is optimized, the waste of dialing time is reduced, and the collection efficiency is enhanced.
Smart Images
Figure CN119168765A_ABST
Abstract
Description
Technical Field
[0001] The invention relates to the field of intelligent dialing, and in particular to an intelligent dialing system for post-loan management. Background Art
[0002] Post-loan management is the last link for banks to reduce overdue non-performing loans and improve asset quality. Post-loan smart dialing is to continue to track and monitor the customer's capital operation after the loan is issued and collect debts through telephone calls, so as to timely reduce or resolve risks and ensure the safety of credit funds.
[0003] For a long time, post-loan smart dialing has been a weak link in bank credit management and has many problems. The traditional manual dialing method is inefficient and there are risks of violent debt collection and personal information leakage, which affects the collection rate and credit fund security. The existing post-loan smart dialing management can monitor customers' overdue account information and contact customers through robots. It can also use artificial intelligence and machine learning technology to predict user behavior and interests, and use the most appropriate method of telephone, SMS, WeChat or Alipay to perform smart debt collection through appropriate pressure based on the prediction results.
[0004] The above methods have improved the efficiency of post-loan management to a certain extent. However, the methods of sending information such as SMS, WeChat or Alipay cannot ensure that customers receive it in time, and there is a risk that the collection information will be overwhelmed by information, and it is impossible to perceive the customer's willingness to repay. Therefore, telephone collection has always been the most direct and most popular method. However, since the call connection rate has a time regularity, and the number of telephone seats is limited, the peak period of call connection is concentrated. Therefore, how to configure telephone dialing to improve the call connection rate is the key to improving the quality of post-loan management. Summary of the invention
[0005] The present invention aims to provide an intelligent dialing system for post-loan management, which predicts the connection rate of intelligent dialing in different time periods through the RNN recurrent neural network model. Based on the prediction results, the customer's dialing time period is used as the optimization target through the Grey Wolf optimization algorithm to obtain the optimal intelligent dialing time period configuration, so as to improve the intelligent dialing connection rate and thus improve the quality of post-loan management.
[0006] In order to achieve the above object, the present invention adopts the following technical scheme:
[0007] An intelligent dialing system for post-loan management, comprising:
[0008] The connection rate prediction model creation module is used to take user portrait data as input, take the connection rate in different time periods as output, and train the RNN recurrent neural network model to obtain the connection rate prediction model;
[0009] The connection rate prediction module obtains the user portrait data of the user to be intelligently dialed, pre-processes the user portrait data and inputs it into the connection rate prediction model to obtain the connection rate prediction results for different time periods;
[0010] The dialing strategy optimization module is used to predict the connection rate results in different time periods. The customer's dialing time period is used as the optimization target through the gray wolf optimization algorithm, and a dialing strategy optimization model is created. The wolf pack position is iteratively updated, and the final gray wolf α is used as the optimal dialing strategy.
[0011] The principle and advantages of this solution are: in actual applications, the user's mobile phone connection rate is affected by many factors, such as age, occupation, living habits, etc. The user portrait information takes into account various factors to improve the prediction accuracy. The RNN recurrent neural network model has outstanding performance in time series prediction problems. The RNN recurrent neural network model is convenient for predicting the connection rate in different time periods, so that intelligent dialing can be performed in time periods with high connection rates to improve dialing efficiency. However, due to the number of agent phones and resource limitations, the peak time period of the call connection rate is concentrated. How to configure the dialing time directly affects the connection rate of intelligent dialing. Therefore, the customer's dialing time period is optimized through the Gray Wolf Optimization Algorithm model to further improve the overall intelligent dialing connection rate and promote the improvement of post-loan management quality.
[0012] Preferably, as an improvement, the pretreatment includes:
[0013] Data collection submodule, used to collect user portrait data;
[0014] Data cleaning submodule, which is used to remove missing values and outliers and fill in missing values;
[0015] The feature encoding submodule uses One-hot encoding to convert user portrait data into binary numerical vectors.
[0016] Technical effect: The large amount of label data in the user portrait is feature encoded to facilitate conversion into a data format that can be processed by the RNN recurrent neural network model, and to facilitate the fusion prediction of multi-label information.
[0017] Preferably, as an improvement, the step of creating a dialing strategy optimization model includes:
[0018] The initialization submodule is used to initialize the gray wolf population, and initialize the convergence factor, distance coefficient vector and position coefficient vector;
[0019] A fitness function determination submodule is used to determine the fitness function according to the dial-up connection rate and the payment collection rate;
[0020] The constraint condition determination submodule is used to determine the constraint condition based on the call duration and the connection rate.
[0021] Technical effect: It is easy to integrate the calling strategy into the gray wolf optimization algorithm, simulate the gray wolf leadership level and hunting mechanism in nature, and accurately obtain the optimization results.
[0022] Preferably, as an improvement, the initialization of the gray wolf population includes: obtaining a target optimization time period, predicting the connection rate of each user to be intelligently dialed in the target optimization time period, and taking the combination of users performing intelligent dialing in the target optimization time period as a solution, taking each solution as a gray wolf, and obtaining an initialized gray wolf population Xi (i=1, 2..., n).
[0023] Technical effect: It is convenient to optimize the configuration of dial-up users when the dial-up time is determined.
[0024] Preferably, as an improvement, the constraint conditions include:
[0025] ∑max(t i )≤t m
[0026] J i ≥J j
[0027] Among them, t i is the call duration of customer i, t m The duration of the target optimization period, J i is the predicted connection rate of customer i, J j The predicted call completion rate threshold within the target optimization time period.
[0028] Technical effect: It is easy to make optimal decisions under limited seat resources, can reduce the number of interfering gray wolves in the initialized gray wolf population, and improve the quality of the dialing strategy optimization model.
[0029] Preferably, as an improvement, the fitness function is:
[0030]
[0031] Among them, μ j is the constant coefficient for the connection rate adjustment, is the average connection rate, φ h is the constant coefficient for adjusting the repayment rate, H h is the expected amount of repayment, H′ h is the amount to be recovered;
[0032]
[0033] Among them, h is the frequency of payment after connection, h0 It is the payment collection frequency threshold after connection.
[0034] Technical effect: Through the fitness function, the gray wolf in the model can move according to the connection rate and collection rate, so that the connection rate and collection rate of the final result are optimal.
[0035] Preferably, as an improvement, the iterative update of the wolf pack position is:
[0036] The distance calculation formula between the wolf pack and the three best gray wolves a, β and δ is:
[0037] D α =|C 1 *X α -X|;
[0038] D β =|C 2 *X β -X|;
[0039] D δ =|C 3 *X δ -X|;
[0040] D α , D β , D δ represents the distance between other ω wolves and the best three gray wolves α, β and δ; X α , X β and X δ represent the positions of α, β and δ wolves respectively, X represents the current position of the gray wolf individual; C 1 , C 2 and C 3 is the position coefficient vector, C 1 , C 2 and C 3 All are random numbers between 0 and 2;
[0041] The gray wolf's position update formula is:
[0042] X 1＝ X α -A 1 D α ;
[0043] X 2＝ X β -A 2 D β ;
[0044] X 3＝ X δ -A 3 D δ ;
[0045]
[0046] Among them, A 1 , A 2 and A 3 is the distance coefficient vector; ε is a constant, R 1 , R 2 , R 3 Dynamically modify the weight for the gray wolf position, specifically:
[0047]
[0048] Technical effect: Can better balance the algorithm's global search and local development processes.
[0049] Preferably, as an improvement, after each round of iterative updating of the wolf pack's position, the positions of the best three gray wolves α, β and δ are also corrected through a competitive strategy, and the competitive strategy is: grid division is performed based on the current position of the gray wolf, a grid is randomly selected, the best gray wolf in the grid competes with the α, β and δ wolves, and re-orders them to obtain the new best three gray wolves α, β and δ.
[0050] Technical effect: During the iteration process, the movement of the remaining gray wolves is affected by α, β, and δ, and the algorithm cannot avoid local optimal situations. The global performance of the algorithm can be improved through competitive strategies.
[0051] Preferably, as an improvement, the dialing strategy optimization module also obtains customers who have not performed smart dialing during the target optimization time period based on the optimization results, obtains the time period with the highest predicted connection rate for customers who have not performed smart dialing based on the connection rate prediction model, and uses this time period as a new target optimization time period for customers who have not performed smart dialing.
[0052] Technical effect: It can fully cover the users waiting for smart dialing and avoid missed calls. BRIEF DESCRIPTION OF THE DRAWINGS
[0053] Figure 1 It is a structural diagram of an intelligent dialing system for post-loan management;
[0054] Figure 2 Schematic diagram of the execution flow of the dialing strategy optimization model. DETAILED DESCRIPTION
[0055] The following is further described in detail through specific implementation methods:
[0056] Embodiment 1
[0057] The embodiment is basically as shown in the attached Figure 1 As shown, an intelligent dialing system for post-loan management includes:
[0058] The connection rate prediction model creation module is used to take the user portrait data as input, the connection rate in different time periods as output, and train the RNN recurrent neural network model to obtain the connection rate prediction model. The user portrait data includes but is not limited to occupation, age, address, gender, income, and consumption and shopping. The user portrait data is composed of several data with the best correlation and within the threshold range obtained through cluster analysis. When training the RNN recurrent neural network model, the historical user portraits and the connection rate data in different time periods are used as training data, and the user portrait data is regarded as a series of features, which will have different effects on the call connection rate in different time periods. The user portrait data is received at the input layer of the RNN recurrent neural network model; LSTM or GRU units are used to process sequence data in the RNN layer; the extracted features are processed in the fully connected layer, and the predicted value of the call connection rate is output; the model is trained using a labeled data set, and after the training is completed, new user portrait data is input, and the model will output the predicted value of the call connection rate in different time periods.
[0059] The connection rate prediction module obtains the user portrait data of the user to be intelligently dialed, pre-processes the user portrait data and inputs it into the connection rate prediction model to obtain the connection rate prediction results for different time periods. The user portrait data is converted into a data form that can be processed by the RNN recurrent neural network model through pre-processing. The pre-processing includes: a data collection sub-module for collecting user portrait data; a data cleaning sub-module for removing missing values and outliers and filling missing values; a feature encoding sub-module, which uses One-hot encoding to convert user portrait data into a binary numerical vector.
[0060] The dialing strategy optimization module is used to predict the connection rate results in different time periods. The customer's dialing time period is used as the optimization target through the gray wolf optimization algorithm, and a dialing strategy optimization model is created. The wolf pack position is iteratively updated, and the final gray wolf α is used as the optimal dialing strategy.
[0061] The creation of the dialing strategy optimization model includes:
[0062] Initialize the submodule, obtain the target optimization time period, predict the connection rate of each user to be intelligently dialed in the target optimization time period, and take the combination of users who perform intelligent dialing in the target optimization time period as a solution. Take each solution as a gray wolf, initialize the gray wolf population Xi (i=1, 2..., n), and initialize the convergence factor, distance coefficient vector and position coefficient vector.
[0063] The fitness function determination submodule is used to determine the fitness function according to the dial-up connection rate and the payment collection rate; the fitness function is:
[0064]
[0065] Among them, μ j is the constant coefficient for the connection rate adjustment, is the average connection rate, φ h is the constant coefficient for adjusting the repayment rate, H h is the expected amount of repayment, H′ h is the amount to be recovered;
[0066]
[0067] Among them, h is the frequency of payment after connection, h 0 is the frequency threshold of payment collection after connection. Through the fitness function, the gray wolf in the model can move according to the connection rate and payment collection rate, so that the connection rate and payment collection rate of the final result are optimal.
[0068] The constraint condition determination submodule is used to determine the constraint conditions based on the call duration and the connection rate. The constraint conditions include:
[0069] ∑max(t i )≤t m
[0070] J i ≥J j
[0071] Among them, t i is the call duration of customer i, t m The duration of the target optimization period, J i is the predicted connection rate of customer i, J j Predict the connection rate threshold for the target optimization time period. Through constraints, ensure that dialing can still be completed within the target optimization time period under the influence of call time, and at the same time, the connection rate of customers who dial within the target time period reaches the threshold level.
[0072] like Figure 2 As shown in FIG. 1 , after the dialing strategy optimization model is created, the model is iteratively updated to obtain the final grey wolf α as the optimal dialing strategy.
[0073] Iterative update of the wolf pack position includes: the distance calculation formula between the wolf pack and the best three gray wolves a, β and δ is:
[0074] D α =|C 1 *X α -X|;
[0075] D β =|C 2 *X β -X|;
[0076] Dδ =|C 3 *X δ -X|;
[0077] D α , D β , D δ represents the distance between other ω wolves and the best three gray wolves α, β and δ; X α , X β and X δ represent the positions of α, β and δ wolves respectively, X represents the current position of the gray wolf individual; C 1 , C 2 and C 3 is the position coefficient vector, C 1 , C 2 and C 3 All are random numbers between 0 and 2;
[0078] The gray wolf's position update formula is:
[0079] X 1 =X α -A 1 D α ;
[0080] X 2 =X β -A 2 D β ;
[0081] X 3 =X δ -A 3 D δ ;
[0082]
[0083] Among them, A 1 , A 2 and A 3 is the distance coefficient vector; ε is a constant, R 1 , R 2 , R 3 Dynamically modify the weight for the gray wolf position, specifically:
[0084]
[0085] Update the convergence factor, distance coefficient vector and position coefficient vector. The calculation formula of the convergence factor is:
[0086]
[0087] Among them, t is the current iteration number, t max is the maximum number of iterations;
[0088] The distance coefficient vector is calculated as:
[0089] A=2ar 1 -a;
[0090] The calculation formula of the position coefficient vector is:
[0091] C = 2ar 2 ;
[0092] r 1 、r 2 is a random vector in [0, 1].
[0093] After each round of iterative updating of the wolf pack's position, the positions of the three best gray wolves α, β and δ are also corrected through a competitive strategy. The competitive strategy is: grid division based on the current position of the gray wolf, randomly selecting a grid, and the best gray wolf in the grid competes with α, β and δ wolves, and re-ordering to obtain the new best three gray wolves α, β and δ. Since the movement of the remaining gray wolves is affected by α, β and δ during the iteration process, it is inevitable that the algorithm will appear in the local optimal situation. The competitive strategy can improve the globality of the algorithm.
[0094] Based on the optimization results, customers who have not performed smart dialing in the target optimization time period are obtained, and based on the connection rate prediction model, the time period with the highest predicted connection rate for customers who have not performed smart dialing is obtained, and this time period is used as the new target optimization time period for customers who have not performed smart dialing, and the dialing optimization configuration of this time period is continued in the new target optimization time period.
[0095] Embodiment 2
[0096] The difference from the first embodiment is that in the first embodiment, the target optimization time period is determined first, and then the customers who dial in within the target optimization time period are optimized, which is suitable for short-term optimization. In this embodiment, it is considered that post-loan management is a continuous process, and the availability of resources such as seats is different in different time periods, so waste of resources needs to be avoided.
[0097] The initialization of the gray wolf population includes: obtaining all current customers waiting for smart dialing, predicting the time period with the highest connection rate of all current customers waiting for smart dialing based on the connection rate prediction model, obtaining the congestion situation of each time period, and the congestion situation is determined according to the call duration and the number of connected customers; judging whether the congestion situation exceeds the threshold, obtaining all time periods in which the congestion situation exceeds the threshold and the time period in which the congestion situation does not exceed the threshold, optimizing the customer configuration in the time period in which the congestion situation exceeds the threshold, obtaining customers who have not performed smart dialing, and allocating customers who have not performed smart dialing to the time period in which the congestion situation does not exceed the threshold. The allocation plan is regarded as a solution, and each solution is regarded as a gray wolf, to obtain the initialization of the gray wolf population Xi (i=1, 2..., n).
[0098] The above is only an embodiment of the present invention, and the common knowledge such as the known specific technical solutions and / or characteristics in the solution is not described in detail here. It should be pointed out that for those skilled in the art, without departing from the technical solution of the present invention, several modifications and improvements can be made, which should also be regarded as the protection scope of the present invention, and these will not affect the effect of the implementation of the present invention and the practicality of the patent. The scope of protection required by this application shall be based on the content of its claims, and the specific implementation methods and other records in the specification can be used to interpret the content of the claims.
Claims
1. An intelligent dialing system for post-loan management, characterized in that: include: The connection rate prediction model creation module is used to take user portrait data as input, take the connection rate in different time periods as output, and train the RNN recurrent neural network model to obtain the connection rate prediction model; The connection rate prediction module obtains the user portrait data of the user to be intelligently dialed, pre-processes the user portrait data and inputs it into the connection rate prediction model to obtain the connection rate prediction results for different time periods; The dialing strategy optimization module is used to predict the connection rate results in different time periods. The customer's dialing time period is used as the optimization target through the gray wolf optimization algorithm, and a dialing strategy optimization model is created. The wolf pack position is iteratively updated, and the final gray wolf α is used as the optimal dialing strategy.
2. The intelligent dialing system for post-loan management according to claim 1, characterized in that: The pre-processing comprises: Data collection submodule, used to collect user portrait data; Data cleaning submodule, which is used to remove missing values and outliers and fill missing values; The feature encoding submodule uses One-hot encoding to convert user portrait data into binary numerical vectors.
3. The intelligent dialing system for post-loan management according to claim 1, characterized in that: The creation of the dialing strategy optimization model includes: The initialization submodule is used to initialize the gray wolf population, and initialize the convergence factor, distance coefficient vector and position coefficient vector; A fitness function determination submodule is used to determine the fitness function according to the dial-up connection rate and the payment collection rate; The constraint condition determination submodule is used to determine the constraint condition based on the call duration and the connection rate.
4. The intelligent dialing system for post-loan management according to claim 3, characterized in that: The initialization of the gray wolf population includes: obtaining a target optimization time period, predicting the connection rate of each user to be intelligently dialed in the target optimization time period, and taking the combination of users who perform intelligent dialing in the target optimization time period as a solution, taking each solution as a gray wolf, and obtaining an initialized gray wolf population Xi (i=1, 2..., n).
5. The intelligent dialing system for post-loan management according to claim 3, characterized in that: The constraints include: Σmax(t i )≤t m J i ≥J j Among them, t i is the call duration of customer i, t m The duration of the target optimization period, J i is the predicted connection rate of customer i, J j The predicted call completion rate threshold within the target optimization time period.
6. The intelligent dialing system for post-loan management according to claim 3, characterized in that: The fitness function is: Among them, μ j is the constant coefficient for the connection rate adjustment, is the average connection rate, φ h is the constant coefficient for the collection rate adjustment, H h is the expected amount of repayment, H′ h is the amount to be recovered; Among them, h is the frequency of collection after connection, and h0 is the threshold of collection frequency after connection.
7. The intelligent dialing system for post-loan management according to claim 1, characterized in that: The iterative update of the wolf pack position is: The distance calculation formula between the wolf pack and the three best gray wolves α, β and δ is: D α =|C1*X α -X|; D β =|C2*X β -X|; D δ =|C3*X δ -X|; D α , D β , D δ represents the distance between other ω wolves and the best three gray wolves α, β and δ; X α , X β and X δ Represent the positions of α, β and δ wolves respectively, X represents the position of the current gray wolf individual; C1, C2 and C3 are position coefficient vectors, and C1, C2 and C3 are all random numbers between 0 and 2; The gray wolf's position update formula is: X1=X α -A1D α ; X2=X β -A2D β ; <h2 style=";text-align:left;direction:ltr">X3=X<h2 style=";text-align:left;direction:ltr"> δ <h2 style=";text-align:left;direction:ltr"> -A3D<h2 style=";text-align:left;direction:ltr"> δ <h2 style=";text-align:left;direction:ltr"> ; Among them, A1, A2 and A3 are distance coefficient vectors; ε is a constant, and R1, R2, R3 are the dynamic modification weights of the gray wolf position. Specifically:
8. The intelligent dialing system for post-loan management according to claim 7, characterized in that: After each round of iterative updating of the wolf pack's position, the positions of the three best gray wolves α, β and δ are also corrected through a competitive strategy. The competitive strategy is: grid division is performed based on the current position of the gray wolf, a grid is randomly selected, the best gray wolf in the grid competes with the α, β and δ wolves, and the order is re-ordered to obtain the new best three gray wolves α, β and δ.
9. The intelligent dialing system for post-loan management according to claim 8, characterized in that: The dialing strategy optimization module also obtains customers who have not performed smart dialing in the target optimization time period based on the optimization results, obtains the time period with the highest predicted connection rate for customers who have not performed smart dialing based on the connection rate prediction model, and uses the time period as a new target optimization time period for customers who have not performed smart dialing.