CN119515399A

CN119515399A - Customer loss retrieval method and device, computer equipment and storage medium

Info

Publication number: CN119515399A
Application number: CN202411295857.0A
Authority: CN
Inventors: 李雪丽; 朱艳乔; 沈嘉良
Original assignee: Ping An Technology Shenzhen Co Ltd
Current assignee: Ping An Technology Shenzhen Co Ltd
Priority date: 2024-09-14
Filing date: 2024-09-14
Publication date: 2025-02-25

Abstract

The application belongs to the field of artificial intelligence and financial science and technology, and relates to a customer loss retrieval method, which comprises the steps of predicting according to customer characteristic data of a target customer through a trained prediction model, wherein the prediction model comprises a first prediction sub-model and a second prediction sub-model, predicting the customer characteristic data through the first prediction sub-model and the second prediction sub-model respectively to obtain a first prediction result and a second prediction result, obtaining a customer retrieval success rate through weighted summation, and then making a decision according to customer state information and the customer retrieval success rate through a trained Markov decision model to obtain an optimal customer retrieval strategy. The application also provides a device for recovering the customer loss, computer equipment and a storage medium. In addition, the present application relates to blockchain technology, in which customer characteristic data may be stored. The application can improve the prediction accuracy of the retrieval success rate of the client and provide an accurate retrieval strategy for the client.

Description

Customer loss retrieval method and device, computer equipment and storage medium

Technical Field

The present application relates to the field of artificial intelligence and financial technology, and in particular, to a customer loss retrieval method, apparatus, computer device, and storage medium.

Background

In recent years, with the rapid popularization of the mobile internet, the custom of the customer behavior accelerates the online migration, and the rapid development of financial science and technology, artificial intelligence and big data application technology provides possibility for personal customer intelligent service. For banks, customer resources are one of the most important resources for commercial banks, and preventing customer churn has become an important measure for commercial banks in face of vigorous competition. Along with the population structure change and the rapid development of Internet quasi credit card products such as Ant, beijing Dong Baibar and the like, the traditional credit card market is changed from an increment market to an stock market, and higher requirements are put on the fine operation of commercial banks, wherein the effective recovery of losing customers is particularly important.

The traditional customer retrieval mainly adopts modes of telephone, short message, letter, customer manager visit and the like, the strategy mainly depends on manual experience, the degree of automation and intellectualization is low, and the problems of indiscriminate retrieval, incapability of controlling marketing cost, excessive disturbance to users and the like exist, so that the retrieval customer is positioned inaccurately, and an accurate retrieval strategy cannot be provided.

Disclosure of Invention

The embodiment of the application aims to provide a method, a device, computer equipment and a storage medium for recovering customer loss, which are used for solving the technical problems that in the prior art, the recovered customer is positioned inaccurately in the process of recovering the customer, and an accurate recovery strategy cannot be provided.

In order to solve the above technical problems, the embodiment of the present application provides a method for recovering customer loss, which adopts the following technical scheme:

Acquiring client characteristic data of a target client, and constructing a client characteristic vector according to the client characteristic data;

inputting the customer feature vector into a trained prediction model, wherein the prediction model comprises a first prediction sub-model and a second prediction sub-model;

Processing the client characteristic data through the first predictor model to obtain client state information and a first prediction result, and inputting the client state information into the second predictor model;

Processing the client feature vector and the client state information through the second predictor model to obtain a second prediction result;

carrying out weighted summation on the first prediction result and the second prediction result to obtain a customer retrieval success rate;

and inputting the client state information and the client retrieval success rate into a trained Markov decision model to make a decision, so as to obtain an optimal retrieval strategy of the target client.

In order to solve the technical problems, the embodiment of the application also provides a device for recovering the customer loss, which adopts the following technical scheme:

The acquisition module is used for acquiring the client characteristic data of the target client and constructing a client characteristic vector according to the client characteristic data;

the input module is used for inputting the client feature vector into a trained prediction model, and the prediction model comprises a first prediction sub-model and a second prediction sub-model;

The first prediction module is used for processing the client characteristic data through the first prediction sub-model to obtain client state information and a first prediction result, and inputting the client state information into the second prediction sub-model;

the second prediction module is used for processing the client feature vector and the client state information through the second prediction sub-model to obtain a second prediction result;

The weighting module is used for carrying out weighted summation on the first prediction result and the second prediction result to obtain a customer retrieval success rate;

And the decision module is used for inputting the client state information and the client retrieval success rate into a trained Markov decision model to make decisions so as to obtain an optimal retrieval strategy of the target client.

In order to solve the above technical problems, the embodiment of the present application further provides a computer device, which adopts the following technical schemes:

the computer device includes a memory having stored therein computer readable instructions which when executed by the processor implement the steps of the customer churn retrieval method as described above.

In order to solve the above technical problems, an embodiment of the present application further provides a computer readable storage medium, which adopts the following technical schemes:

the computer readable storage medium has stored thereon computer readable instructions which when executed by a processor implement the steps of the customer churn retrieval method as described above.

Compared with the prior art, the embodiment of the application has the following main beneficial effects:

the application provides a customer loss retrieval method, which predicts according to customer characteristic data of a target customer through a trained prediction model, wherein the prediction model comprises a first prediction sub-model and a second prediction sub-model, the customer characteristic data is predicted through the first prediction sub-model and the second prediction sub-model respectively to obtain a first prediction result and a second prediction result, a customer retrieval success rate is obtained through weighted summation, then a decision is made according to customer state information and the customer retrieval success rate through a trained Markov decision model to obtain an optimal customer retrieval strategy, the prediction accuracy of the customer retrieval success rate can be effectively improved, the retrieved customer can be more accurately positioned by combining the customer state information and the customer retrieval success rate obtained through the prediction model, and an accurate retrieval strategy is provided for the customer, so that the customer retrieval is more efficient, intelligent and controllable.

Drawings

In order to more clearly illustrate the solution of the present application, a brief description will be given below of the drawings required for the description of the embodiments of the present application, it being apparent that the drawings in the following description are some embodiments of the present application, and that other drawings may be obtained from these drawings without the exercise of inventive effort for a person of ordinary skill in the art.

FIG. 1 is an exemplary system architecture diagram in which the present application may be applied;

FIG. 2 is a flow chart of one embodiment of a method of customer churn retrieval in accordance with the present application;

FIG. 3 is a schematic diagram of one embodiment of a customer churn retrieval apparatus in accordance with the present application;

FIG. 4 is a schematic structural diagram of one embodiment of a computer device in accordance with the present application.

Detailed Description

Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs, the terms used in the description herein are used for the purpose of describing particular embodiments only and are not intended to limit the application, and the terms "comprising" and "having" and any variations thereof in the description of the application and the claims and the above description of the drawings are intended to cover non-exclusive inclusions. The terms first, second and the like in the description and in the claims or in the above-described figures, are used for distinguishing between different objects and not necessarily for describing a sequential or chronological order.

Reference herein to "an embodiment" means that a particular feature, structure, or characteristic described in connection with the embodiment may be included in at least one embodiment of the application. The appearances of such phrases in various places in the specification are not necessarily all referring to the same embodiment, nor are separate or alternative embodiments mutually exclusive of other embodiments. Those of skill in the art will explicitly and implicitly appreciate that the embodiments described herein may be combined with other embodiments.

In order to make the person skilled in the art better understand the solution of the present application, the technical solution of the embodiment of the present application will be clearly and completely described below with reference to the accompanying drawings.

The embodiment of the application can acquire and process the related data based on the artificial intelligence technology. Wherein artificial intelligence (ARTIFICIAL INTELLIGENCE, AI) is the theory, method, technique, and application system that uses a digital computer or a digital computer-controlled machine to simulate, extend, and expand human intelligence, sense the environment, acquire knowledge, and use knowledge to obtain optimal results.

Artificial intelligence infrastructure technologies generally include technologies such as sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, big data processing technologies, operation/interaction systems, mechatronics, and the like. The artificial intelligence software technology mainly comprises a computer vision technology, a robot technology, a biological recognition technology, a voice processing technology, a natural language processing technology, machine learning/deep learning and other directions.

The present application provides a method for customer churn retrieval, which can be applied to a system architecture 100 as shown in fig. 1, where the system architecture 100 may include terminal devices 101, 102, 103, a network 104 and a server 105. The network 104 is used as a medium to provide communication links between the terminal devices 101, 102, 103 and the server 105. The network 104 may include various connection types, such as wired, wireless communication links, or fiber optic cables, among others.

The user may interact with the server 105 via the network 104 using the terminal devices 101, 102, 103 to receive or send messages or the like. Various communication client applications, such as a web browser application, a shopping class application, a search class application, an instant messaging tool, a mailbox client, social platform software, etc., may be installed on the terminal devices 101, 102, 103.

The terminal devices 101, 102, 103 may be various electronic devices having a display screen and supporting web browsing, including but not limited to smartphones, tablet computers, electronic book readers, MP3 players (Moving Picture Experts Group Audio Layer III, dynamic video expert compression standard audio plane 3), MP4 (Moving Picture Experts Group Audio Layer IV, dynamic video expert compression standard audio plane 4) players, laptop and desktop computers, and the like.

The server 105 may be a server providing various services, such as a background server providing support for pages displayed on the terminal devices 101, 102, 103.

It should be noted that, the method for recovering the customer loss provided in the embodiment of the present application is generally executed by a server/terminal device, and accordingly, the device for recovering the customer loss is generally disposed in the server/terminal device.

It should be understood that the number of terminal devices, networks and servers in fig. 1 is merely illustrative. There may be any number of terminal devices, networks, and servers, as desired for implementation.

With continued reference to FIG. 2, a flow chart of one embodiment of a method of customer churn retrieval according to the present application is shown, comprising the steps of:

step S201, obtaining the customer characteristic data of the target customer, and constructing a customer characteristic vector according to the customer characteristic data.

The target client is a client needing to be recovered, and the client characteristic data comprises, but is not limited to, client information, account information, consumption records, repayment records, credit investigation conditions, historical contact persons, other time sequence data and the like, wherein the client information comprises age, gender, education, occupation, group value grade and the like; the account information comprises card opening time, card grade, amount, account opening mode and the like, the consumption records comprise consumption time, consumption times, consumption amount, amount utilization rate and the like, the credit investigation conditions comprise credit scores, effective credit card numbers, effective amount, check times and the like, the history touch information comprises touch date, touch time, touch mode, whether the touch is connected, whether the touch is self, whether the touch is authorized to participate, whether the task is taken, the final task completion degree and the like, and the other time sequence data comprise longitude and latitude information, app login, wifi information and the like.

In this embodiment, the electronic device (e.g., the server/terminal device shown in fig. 1) on which the client churn retrieval method operates may acquire the client characteristic data of the target client through a wired connection manner or a wireless connection manner. It should be noted that the wireless connection may include, but is not limited to, 3G/4G connection, wiFi connection, bluetooth connection, wiMAX connection, zigbee connection, UWB (ultra wideband) connection, and other now known or later developed wireless connection.

In this embodiment, after the client feature data of the target client is obtained, the client feature data is subjected to data cleaning and preprocessing, so as to ensure the integrity and consistency of the data. And (5) obtaining standardized customer characteristic data through data cleaning and preprocessing. And extracting features related to the customer churn retrieval problem according to the standardized customer feature data to construct a customer feature vector suitable for the input model.

For example, missing values are detected and duplicated data is deleted by using Pandas library in Python, and then data normalization processing, such as Z-score normalization of numerical characteristics of age, consumption amount, etc., is performed to eliminate the influence of dimension.

It is emphasized that to further ensure the privacy and security of the customer characteristic data, the customer characteristic data may also be stored in a node of a blockchain.

The blockchain is a novel application mode of computer technologies such as distributed data storage, point-to-point transmission, consensus mechanism, encryption algorithm and the like. The blockchain (Blockchain), essentially a de-centralized database, is a string of data blocks that are generated in association using cryptographic methods, each of which contains information from a batch of network transactions for verifying the validity (anti-counterfeit) of its information and generating the next block. The blockchain may include a blockchain underlying platform, a platform product services layer, an application services layer, and the like.

Step S202, inputting the customer feature vector into a trained prediction model, wherein the prediction model comprises a first prediction sub-model and a second prediction sub-model.

The prediction model is a trained model and is used for predicting the retrieval success rate of the target user. The first predictor model can adopt an LSTM model, LSTM (long short term memory network) is a time recurrent neural network, solves the problems of gradient elimination and explosion of RNN, can effectively process long-term time sequence data, and the LSTM can record the last client characteristic and influence the subsequent model prediction. The second predictor model may employ XGBoost models, XGBoost is an optimization of Boosting algorithm that integrates weak classifiers into a strong classifier. The XGBoost algorithm generates a residual error of a new tree fitting the previous tree through continuous iteration, the precision is continuously improved along with the increase of iteration times, and finally an optimal weak classifier, namely a XGBoost model, is determined by a linear search method.

In step S203, the client feature vector is processed through the first predictor model to obtain client state information and a first prediction result, and the client state information is input into the second predictor model.

The first predictor model comprises at least two LSTM layers, a full-connection layer and an output layer, the client feature vector is input into the first predictor model, the client feature vector is processed through the at least two LSTM layers, the full-connection layer and the output layer which are connected, the time sequence feature and the change trend of a target client are captured, the hidden state representation of the client feature and a first predictor result are output, the hidden state representation of the client feature is client state information, and the first predictor result is the client retrieval success rate of the first predictor model for predicting according to the client feature vector.

In some optional implementations of this embodiment, the step of processing, by the first predictor model, the client feature vector to obtain client state information and a first prediction result includes:

Feature extraction is sequentially carried out on the customer feature vectors through each LSTM layer, so that feature hiding states with different scales are obtained;

Inputting the characteristic hiding states of different scales into the full-connection layer for splicing to obtain client state information;

and predicting the client state information input and output layer to obtain a first prediction result.

In this embodiment, each LSTM layer processes information of different time scales to capture features of different time scales, and the input of the next LSTM layer is the output and input customer feature vector of the previous LSTM layer. For example, the first predictor model may include two LSTM layers, three LSTM layers, or four LSTM layers, which may be specifically set according to actual needs.

The client feature vector is processed through the first predictor model, so that long-term dependency relationship between input data can be captured, the efficiency and accuracy of time sequence feature extraction are improved, and the prediction efficiency and accuracy are improved.

In some alternative implementations, the residual connection between different LSTM layers may help the gradient flow more effectively in the deep network, preventing the gradient vanishing problem.

In some optional implementations of this embodiment, each LSTM layer includes a forgetting gate, an input gate, and an output gate, where the step of sequentially extracting features of the client feature vector through each LSTM layer to obtain feature hiding states of different scales includes:

calculating forgetting gate output characteristics through a forgetting gate based on the characteristic hiding state of the last moment and the client characteristic vector of the current moment;

Calculating the output characteristics of the forgetting gate, the characteristic hiding state at the last moment and the client characteristic vector at the current moment through the input gate to obtain an input gating signal and a candidate memory cell state;

updating the memory cell state according to the forgetting gate output characteristic, the input gating signal and the candidate memory cell state;

Calculating an output gating signal through an output gate according to the characteristic hiding state of the last moment and the client characteristic vector of the current moment, and outputting the characteristic hiding state of the current moment based on the output gating signal and the memory cell state;

And combining the characteristic hiding states of the current moment output by each LSTM layer into characteristic hiding states of different scales.

The calculation formula of the forgetting gate is as follows:

f_t＝σ(W_xfx_t+W_hfh_t-1+b_f);

Wherein f _t is the forgetting gate output characteristic, sigma is the sigmoid function, W _xf、W_hf and b _f are the weight and bias of the forgetting gate respectively, x _t is the client characteristic vector at the current moment, and h _t-1 is the characteristic hiding state at the last moment.

The input gate is calculated as follows:

i_t＝σ(W_xix_t+W_hih_t-1+b_i);

Wherein i _t is an input gate control signal, W _xi、W_hi and b _i are the weight and bias of the input gate respectively, x _t is a client feature vector at the current moment, and h _t-1 is a feature hiding state at the last moment.

The calculation formula of the candidate memory cell state is as follows:

Wherein, tan is hyperbolic tangent function, tan range is [ -1,1], W _xc、W_hc and b _c are respectively the weight and bias of the state of the candidate memory cell, x _t is the client feature vector at the current moment, and h _t-1 is the feature hiding state at the last moment.

The updated memory cell state calculation formula is as follows:

the characteristic hiding state formula at the current moment is as follows:

u_t＝σ(W₀[h_t-1,x_t]+b₀);

Wherein, Hiding the state for the candidate feature at the current time.

And splicing the characteristic hiding states of each LSTM layer at the current moment to obtain characteristic hiding states of different scales.

In the embodiment, the LSTM layer overcomes the problems of gradient elimination and gradient explosion existing in model prediction by using a mechanism constructed by an input gate, a forgetting gate and an output gate, and can capture long-term dependency in a sequence, realize long-term memory and improve the accuracy of prediction.

And S204, processing the client feature vector and the client state information through a second predictor model to obtain a second prediction result.

In the embodiment, a XGBoost model is adopted as a second predictor model, feature extraction is carried out on a customer feature vector and customer state information through the XGBoost model to obtain M basic features, M is a positive integer, the M basic features are added into a XGBoost model according to a stepwise regression algorithm to carry out iterative calculation to obtain an iterative calculation result, the iterative calculation result comprises contribution scores corresponding to the basic features, the contribution scores are arranged in the order from large to small from the iterative calculation result, N contribution scores before ranking are selected, the basic features corresponding to the N contribution scores are taken as N target features, N < M, N is a positive integer, and the second prediction result is output based on the target features and the corresponding contribution scores.

And the second prediction result is a customer retrieval success rate obtained by predicting the XGBoost model according to the customer characteristic vector and the customer state information.

The second predictor model is used for predicting the combination of the client state information and the client feature vector output by the first predictor model, so that the operation speed of the predictor model can be effectively improved, and the prediction precision and stability are improved.

And step S205, carrying out weighted summation on the first prediction result and the second prediction result to obtain the recovery success rate of the client.

In this embodiment, the first prediction result and the second prediction result are combined with different weights by the inverse error method to construct a prediction model, so that the advantages of each model can be integrated, fluctuation of the sample caused by the accident on a single model is reduced, the error of the whole combined model tends to be reduced, and the overall prediction precision is improved.

Specifically, the calculation formula of the customer retrieval success rate Q _t is as follows:

Q_t＝W₁f_1t+W₂f_2t;

Wherein t represents time, t=1, 2,3. F _1t denotes a first prediction result, f _2t denotes a second prediction result, W ₁ and W ₂ denote weight coefficients of the first and second predictor models, respectively, and ω ₁ and ω ₂ denote errors of the first and second predictor models, respectively.

And S206, inputting the client state information and the client retrieval success rate into a trained Markov decision model to make a decision, and obtaining an optimal retrieval strategy of the target client.

Wherein the Markov decision model is derived based on defining a Markov decision process, which may be described by tuples (S, A, P, Q, r). Wherein S is a finite state set, A is a finite policy set, P is a state transition probability, Q is a corresponding value return (rewarding function), and r is a discount factor between 0 and 1, so as to ensure that the discount is larger when the rewarding is further from the current moment.

In some optional implementations of this embodiment, before the step of inputting the client state information and the client retrieval success rate into the trained markov decision model for decision making, the method includes:

Customer retrieval is regarded as a multi-stage decision problem, a Markov decision process is defined, and the Markov decision process comprises a finite state set, a finite strategy set, state transition probabilities and rewarding functions;

determining a state space of a multi-stage decision problem by defining the state space to obtain a finite state set of a client in different stages;

Defining an action space according to the state characteristics of the client, and determining a retrieval strategy selectable in each state to obtain a limited strategy set retrieved by the client;

Defining state transition probability by analyzing the state characteristics of the client and the retrieval strategy to obtain the state transition probability in the current state;

defining a reward function, and acquiring reward values of different retrieval strategies under different states;

And solving the established Markov decision process according to the finite state set, the finite strategy set, the state transition probability and the rewarding function by setting a discount factor and utilizing a strategy iteration algorithm to obtain an optimal customer retrieval strategy so as to obtain a trained Markov decision model.

The state space is a set of all possible states, the state space is obtained according to the client state information output by the first predictor model, and each state s corresponds to an action space a, namely a retrieval strategy.

The state transition probability is the probability of transition from state s to s' after performing action a for each state s and action a, and the calculation formula is as follows:

Wherein, The probability of a state migrating from S to S' given the current state S _t =s, the current retrieval policy a _t =a, is represented.

The goal of reinforcement learning is to find a retrieval strategy that maximizes the cumulative expected value return (prize value) given a markov process, expressed as:

Wherein k represents the iteration number, Q ^* (s, a) represents the expected maximum sum of the reward values of k subsequent iterations by adopting a retrieval strategy a on the basis of the state s, pi represents the client retrieval strategy, and the optimal client retrieval strategy is obtained when the expected maximum sum is the expected maximum sum.

In this embodiment, the reward function retrieves the success rate for the customer output by the predictive model. For example, the customer retrieval success rate output by the prediction model is introduced as a reward function, the adjustment of the customer retrieval policy on the t+1st day is tried on the t th day, and under the finite policy set a (for example, the time of the touch and the touch mode), the customer retrieval success rate Q (a) corresponding to the customer retrieval policy a can be predicted, and the customer retrieval policy corresponding to the maximum expected retrieval success rate is the optimal customer retrieval policy recommended on the t+1st day.

Based on a trained prediction model, a corresponding customer retrieval success rate can be predicted according to a customer feature vector, meanwhile, a Markov decision process is defined by applying reinforcement learning theory, the prediction model is introduced, the customer retrieval success rate corresponding to the maximum value return (rewarding value) is found, in the training process, the optimal decision model can be continuously trained, the theoretical optimal strategy is continuously approached, the known optimal customer retrieval operation is always executed, the customer retrieval effect and reliability are greatly improved, and especially when the model training is very reliable through time accumulation, the execution strategy is optimal, and the automatic intelligent retrieval of customers is completely realized.

In this embodiment, after obtaining a trained markov decision model, the client state information and the client retrieval success rate are input into the trained markov decision model, and then an optimal retrieval strategy corresponding to the target client can be obtained.

According to the application, the prediction accuracy of the customer retrieval success rate can be effectively improved by predicting the customer feature vector through the prediction model, and the decision is made according to the customer feature vector and the customer retrieval success rate by combining with the Markov decision model, so that the optimal customer retrieval strategy is obtained, the customer needing retrieval can be more accurately positioned, and the accurate retrieval strategy is provided for the customer, so that the customer retrieval is more efficient, intelligent and controllable, and meanwhile, the retrieval of the target customer by service personnel can be promoted, and the working efficiency of the service personnel is improved.

In some alternative implementations, before the step of inputting the customer feature vector into the trained predictive model, further comprises:

acquiring customer characteristic factors under all service scenes, and carrying out characteristic selection on the customer characteristic factors to obtain key customer characteristics;

carrying out feature engineering on the key customer features to obtain normalized customer feature data, and randomly dividing the normalized customer feature data into a training set and a testing set;

Inputting the training set into a pre-constructed neural network model, wherein the neural network model comprises an LSTM model and a XGBoost model;

Training the LSTM model through a training set to obtain a predicted time sequence feature and a first predicted probability;

Inputting the predicted time sequence characteristics and the training set into XGBoost models to obtain second predicted probabilities;

Carrying out weighted summation on the first prediction probability and the second prediction probability to obtain a prediction retrieval probability;

Calculating a loss value between a predicted retrieval probability and an actual retrieval probability based on a preset loss function;

adjusting model parameters of the neural network model according to the loss value, and continuing iterative training until convergence to obtain a model to be verified;

Inputting the test set into the model to be verified to obtain a verification result, and determining the model to be verified as a final prediction model when the verification result meets preset conditions.

In this embodiment, the methods of comprehensive saturation, correlation, PSI, etc. perform feature selection on the client feature factors used for modeling. Wherein saturation (Variance Inflation Factor, VIF) is a measure of the multiple collinearity between features. Features are considered saturated if there is a high degree of correlation between them, i.e. they provide information that has a large overlap. The formula for the calculation of VIF is as follows:

Wherein VIF _j represents the saturation of the jth customer feature factor; And when the jth customer characteristic factor is taken as a response variable, determining coefficients (R-squared) of the multiple linear regression model taking other characteristics as independent variables.

The relevance is to calculate the relevance coefficient between the client feature factors, such as pearson relevance coefficient, and the like, reserve the features with higher relevance, and remove the features with low relevance. PSI (Population Stability Index) is used to evaluate the stability of the client feature factor in different sample subsets, the lower the PSI value, the more stable. The data set is divided into a plurality of sample subsets, the importance score of each customer feature factor in the different subsets is calculated, and then the PSI value is calculated, wherein the calculation formula is as follows:

Where Importance _i is the feature Importance score in the i-th subset, μ averages the Importance scores, and n is the number of subsets.

And carrying out feature engineering on the key customer features obtained by feature selection, wherein the feature engineering comprises over normalization, binning, one-hot/target coding conversion and the like. Where normalization is the scaling of data to a specific range, typically 0 to 1, or to a distribution with a specific mean and standard deviation, common normalization methods include min-max normalization, Z-score normalization. The binning is to divide continuous numerical data into a plurality of sections (bins), and represent the values in each section by the average value or median value in the bin, so that the influence of abnormal values can be reduced, and the generalization capability of the model can be possibly improved.

And randomly dividing normalized customer characteristic data obtained by the characteristic engineering into a training set and a testing set according to a preset proportion, wherein the training set is the testing set=8:2.

In this embodiment, the preset loss function uses an average absolute percentage error (Mean Absolute Percentage Error, MAPE) and a root mean square error (Root Mean Square Error, RMSE), where MAPE is an average of the percentages of the difference between the predicted value (predicted retrieval probability) and the actual value (actual retrieval probability), which is generally used to measure the accuracy of the prediction, and the calculation formula is as follows:

Wherein, And y _t is the predicted retrieval probability and the actual retrieval probability at the time t respectively, and N is the predicted total number of times.

RMSE is the square root of the average of the squares of the observed value (predicted retrieval probability) and the true value (actual retrieval probability) deviation, is a standard measure of the prediction error, and is calculated by the following formula:

And calculating a loss value based on a preset loss function, adjusting model parameters according to the loss value, and continuing to perform iterative training until the model converges, wherein the condition that the convergence condition is met can be that the loss value does not change significantly or the iteration times reach the preset times.

Inputting the test set into the model to be verified to obtain verification prediction probability, calculating the prediction precision of the model according to the verification prediction probability and the actual retrieval probability, and taking the prediction precision as a verification result. And if the prediction precision is smaller than the preset threshold value, the prediction precision of the model is not high, and the number of samples is required to be increased, or model parameters are required to be modified for retraining so as to improve the prediction precision.

According to the application, the acquired customer characteristic factors are subjected to characteristic selection and characteristic engineering, and the model is trained, so that the quality of data preparation can be effectively improved, the robustness of model learning and prediction of the recovery success rate of customers can be increased, and the anti-interference capability and generalization capability of the model are enhanced, thereby improving the prediction accuracy.

In some optional implementations of this embodiment, the step of inputting the predicted timing feature and the training set into the XGBoost model to obtain the second predicted probability includes:

Initializing a model, and setting XGBoost basic parameters of the model, wherein the basic parameters at least comprise a learning rate, a maximum depth of a tree and an objective function;

Constructing a first tree by adopting a greedy algorithm, splitting each node of the tree according to the input predicted time sequence characteristics and the importance of normalized customer characteristic data in a training set so as to maximize an objective function of information gain, and outputting a first predicted value;

Constructing a second tree based on the predicted time sequence characteristics, the training set and a new target variable, and outputting a second predicted value, wherein the new target variable is the residual error of the first tree, namely the difference value between the true value and the first predicted value;

Iteratively constructing a residual tree in the XGBoost model, constructing a next tree by using the residual error of the previous tree as a new target variable in each iteration until the last tree, namely a kth tree, to obtain a kth predicted value;

and carrying out weighted summation on the first predicted value to the kth predicted value to obtain a second predicted probability of the XGBoost model.

The expression of the objective function of XGBoost model is as follows:

Wherein, Representing a loss function; And y _t is the predicted retrieval probability and the actual retrieval probability at time T respectively, Ω (f _k) represents a regularization term, f _k represents the kth tree, T represents the number of leaves in the tree, λ represents a regularization parameter, ω is a leaf weight.

During training of the XGBoost model, a new tree is built at each iteration and added to the XGBoost model, enabling the built tree to minimize the objective function of the XGBoost model. In the ith iteration, let f _i(x_t) be the tree generated by the t sample in the ith iteration, then the objective function is:

in the formula, The probability is retrieved for the prediction of the target timing (including the predicted timing characteristics and training set) at time t for the ith iteration.

By training the XGBoost model, the characteristics of the characteristic factors of the clients are fully utilized, key information related to retrieval of the clients is effectively extracted, the retrieved characteristics of the clients are more comprehensively captured, and the accuracy and reliability of prediction are improved.

The application is operational with numerous general purpose or special purpose computer system environments or configurations. Such as a personal computer, a server computer, a hand-held or portable device, a tablet device, a multiprocessor system, a microprocessor-based system, a set top box, a programmable consumer electronics, a network PC, a minicomputer, a mainframe computer, a distributed computing environment that includes any of the above systems or devices, and the like. The application may be described in the general context of computer-executable instructions, such as program modules, being executed by a computer. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform particular tasks or implement particular abstract data types. The application may also be practiced in distributed computing environments where tasks are performed by remote processing devices that are linked through a communications network. In a distributed computing environment, program modules may be located in both local and remote computer storage media including memory storage devices.

Those skilled in the art will appreciate that implementing all or part of the above described methods may be accomplished by computer readable instructions stored in a computer readable storage medium that, when executed, may comprise the steps of the embodiments of the methods described above. The storage medium may be a nonvolatile storage medium such as a magnetic disk, an optical disk, a Read-Only Memory (ROM), or a random access Memory (Random Access Memory, RAM).

It should be understood that, although the steps in the flowcharts of the figures are shown in order as indicated by the arrows, these steps are not necessarily performed in order as indicated by the arrows. The steps are not strictly limited in order and may be performed in other orders, unless explicitly stated herein. Moreover, at least some of the steps in the flowcharts of the figures may include a plurality of sub-steps or stages that are not necessarily performed at the same time, but may be performed at different times, the order of their execution not necessarily being sequential, but may be performed in turn or alternately with other steps or at least a portion of the other steps or stages.

With further reference to fig. 3, as an implementation of the method shown in fig. 2, the present application provides an embodiment of a customer churn retrieval apparatus, which corresponds to the method embodiment shown in fig. 2, and which is particularly applicable to various electronic devices.

As shown in fig. 3, the apparatus 300 for customer churn retrieval according to this embodiment includes an acquisition module 301, an input module 302, a first prediction module 303, a second prediction module 304, a weighting module 305, and a decision module 306. Wherein:

the obtaining module 301 is configured to obtain client feature data of a target client, and construct a client feature vector according to the client feature data;

The input module 302 is configured to input the client feature vector into a trained prediction model, where the prediction model includes a first predictor model and a second predictor model;

The first prediction module 303 is configured to process the client feature data through the first prediction sub-model to obtain client state information and a first prediction result, and input the client state information into the second prediction sub-model;

The second prediction module 304 is configured to process the client feature vector and the client state information through the second predictor model to obtain a second prediction result;

The weighting module 305 is configured to perform weighted summation on the first prediction result and the second prediction result, so as to obtain a customer retrieval success rate;

the decision module 306 is configured to input the client state information and the client retrieval success rate into a trained markov decision model to make a decision, so as to obtain an optimal retrieval policy of the target client.

Based on the device 300 for recovering the customer loss, the prediction accuracy of the recovery success rate of the customer can be effectively improved by predicting the customer according to the customer feature vector through the prediction model, and the decision is made according to the customer feature vector and the recovery success rate of the customer by combining with the Markov decision model so as to obtain the optimal customer recovery strategy, so that the customer needing to be recovered can be more accurately positioned, and the accurate recovery strategy is provided for the customer, so that the customer recovery is more efficient, intelligent and controllable, meanwhile, the recovery of the target customer by service personnel can be promoted, and the working efficiency of the service personnel is improved.

In some alternative implementations, the first predictor model includes at least two LSTM layers, a full link layer, and an output layer, and the first predictor module 303 includes a feature extraction sub-module, a stitching sub-module, and a predictor sub-module, wherein:

the feature extraction submodule is used for sequentially carrying out feature extraction on the client feature vectors through each LSTM layer to obtain feature hiding states of different scales;

The splicing sub-module is used for inputting the characteristic hiding states with different scales into the full-connection layer for splicing to obtain client state information;

and the prediction sub-module is used for inputting the client state information into the output layer for prediction to obtain a first prediction result.

In some optional implementations of this embodiment, each of the LSTM layers includes a forget gate, an input gate, and an output gate, and the feature extraction submodule is further configured to:

Calculating forgetting gate output characteristics through the forgetting gate based on the characteristic hiding state of the last moment and the client characteristic vector of the current moment;

Calculating the forgetting gate output characteristics, the characteristic hiding state at the last moment and the client characteristic vector at the current moment through the input gate to obtain an input gating signal and a candidate memory cell state;

Calculating an output gating signal through the output gate according to the characteristic hiding state of the last moment and the client characteristic vector of the current moment, and outputting the characteristic hiding state of the current moment based on the output gating signal and the memory cell state;

And splicing the characteristic hiding states of the current moment output by each LSTM layer to obtain characteristic hiding states of different scales.

In some alternative implementations, the second predictor model employs a XGBoost model, and the second prediction module 304 is further configured to:

Extracting features of the client feature vector and the client state information through the XGBoost model to obtain M basic features, wherein M is a positive integer;

according to a stepwise regression algorithm, M basic features are added into the XGBoost models for iterative computation, so that an iterative computation result is obtained, wherein the iterative computation result contains contribution scores corresponding to the basic features;

from the iterative calculation result, arranging the contribution scores in order from large to small, selecting N contribution scores before ranking, and taking basic features corresponding to the N contribution scores as N target features, wherein N < M, N is a positive integer;

and outputting a second prediction result based on the target feature and the corresponding contribution score.

In some alternative implementations, the customer churn retrieval apparatus 300 further includes a definition decision module for:

regarding customer retrieval as a multi-stage decision problem, defining a Markov decision process, wherein the Markov decision process comprises a finite set of states, a finite set of strategies, a state transition probability and a reward function;

Determining a state space of the multi-stage decision problem by defining the state space to obtain a finite state set of the client in different stages;

and solving the established Markov decision process according to the finite state set, the finite strategy set, the state transition probability and the reward function by setting a discount factor and utilizing a strategy iteration algorithm to obtain an optimal customer retrieval strategy so as to obtain a trained Markov decision model.

By defining a Markov decision process, decision is made according to the client feature vector and the client retrieval success rate so as to obtain an optimal client retrieval strategy, so that the client needing retrieval can be positioned more accurately, and an accurate retrieval strategy is provided for the client, so that the client retrieval is more efficient, intelligent and controllable.

In some alternative implementations, the customer churn retrieval apparatus 300 further includes a training module comprising:

The feature selection sub-module is used for acquiring the client feature factors under all service scenes and carrying out feature selection on the client feature factors to obtain key client features;

the feature engineering sub-module is used for carrying out feature engineering on the key customer features to obtain normalized customer feature data, and randomly dividing the normalized customer feature data into a training set and a testing set;

an input sub-module for inputting the training set into a pre-constructed neural network model, wherein the neural network model comprises an LSTM model and a XGBoost model;

The first training sub-module is used for training the LSTM model through the training set to obtain a predicted time sequence characteristic and a first predicted probability;

The second training sub-module is used for inputting the predicted time sequence characteristics and the training set into the XGBoost model to obtain a second predicted probability;

the weighted summation sub-module is used for carrying out weighted summation on the first prediction probability and the second prediction probability to obtain a prediction retrieval probability;

The calculation sub-module is used for calculating a loss value between the prediction retrieval probability and the actual retrieval probability based on a preset loss function;

the iteration sub-module is used for adjusting model parameters of the neural network model according to the loss value, and continuing to iterate training until convergence to obtain a model to be verified;

And the verification sub-module is used for inputting the test set into the model to be verified to obtain a verification result, and determining the model to be verified as a final prediction model when the verification result accords with a preset condition.

By carrying out feature selection and feature engineering on the obtained customer feature factors and training the model, the quality of data preparation can be effectively improved, the robustness of model learning and prediction customer retrieval success rate can be increased, and the anti-interference capability and generalization capability of the model are enhanced, so that the prediction accuracy is improved.

In this embodiment, the second training sub-module is further configured to:

constructing a first tree by adopting a greedy algorithm, splitting each node of the tree according to the input predicted time sequence characteristics and the importance of normalized customer characteristic data in the training set so as to maximize an objective function of information gain, and outputting a first predicted value;

constructing a second tree based on the predicted time sequence characteristics, the training set and a new target variable, and outputting a second predicted value, wherein the new target variable is a residual error of the first tree, namely a difference value between a true value and a first predicted value;

Iteratively constructing the rest tree in the XGBoost model, and constructing a next tree by using the residual error of the previous tree as a new target variable in each iteration until the last tree, namely a kth tree, to obtain a kth predicted value;

In order to solve the technical problems, the embodiment of the application also provides computer equipment. Referring specifically to fig. 4, fig. 4 is a basic structural block diagram of a computer device according to the present embodiment.

The computer device 4 comprises a memory 41, a processor 42, a network interface 43 communicatively connected to each other via a system bus. It should be noted that only computer device 4 having components 41-43 is shown in the figures, but it should be understood that not all of the illustrated components are required to be implemented and that more or fewer components may be implemented instead. It will be appreciated by those skilled in the art that the computer device herein is a device capable of automatically performing numerical calculation and/or information processing according to a preset or stored instruction, and its hardware includes, but is not limited to, a microprocessor, an Application SPECIFIC INTEGRATED Circuit (ASIC), a Programmable gate array (Field-Programmable GATE ARRAY, FPGA), a digital Processor (DIGITAL SIGNAL Processor, DSP), an embedded device, and the like.

The computer equipment can be a desktop computer, a notebook computer, a palm computer, a cloud server and other computing equipment. The computer equipment can perform man-machine interaction with a user through a keyboard, a mouse, a remote controller, a touch pad or voice control equipment and the like.

The memory 41 includes at least one type of readable storage medium including flash memory, hard disk, multimedia card, card memory (e.g., SD or DX memory, etc.), random Access Memory (RAM), static Random Access Memory (SRAM), read Only Memory (ROM), electrically Erasable Programmable Read Only Memory (EEPROM), programmable Read Only Memory (PROM), magnetic memory, magnetic disk, optical disk, etc. In some embodiments, the storage 41 may be an internal storage unit of the computer device 4, such as a hard disk or a memory of the computer device 4. In other embodiments, the memory 41 may also be an external storage device of the computer device 4, such as a plug-in hard disk, a smart memory card (SMART MEDIA CARD, SMC), a Secure Digital (SD) card, a flash memory card (FLASH CARD) or the like, which are provided on the computer device 4. Of course, the memory 41 may also comprise both an internal memory unit of the computer device 4 and an external memory device. In this embodiment, the memory 41 is typically used to store an operating system and various application software installed on the computer device 4, such as computer readable instructions of a customer churn-up method, etc. Further, the memory 41 may be used to temporarily store various types of data that have been output or are to be output.

The processor 42 may be a central processing unit (Central Processing Unit, CPU), controller, microcontroller, microprocessor, or other data processing chip in some embodiments. The processor 42 is typically used to control the overall operation of the computer device 4. In this embodiment, the processor 42 is configured to execute computer readable instructions stored in the memory 41 or process data, such as computer readable instructions for executing the customer churn retrieval method.

The network interface 43 may comprise a wireless network interface or a wired network interface, which network interface 43 is typically used for establishing a communication connection between the computer device 4 and other electronic devices.

According to the method, the steps of the method for recovering the customer loss in the embodiment are realized when the processor executes the computer readable instructions stored in the memory, the prediction accuracy of the customer recovery success rate can be effectively improved by predicting the customer characteristic vector through the prediction model, and the decision is made according to the customer characteristic vector and the customer recovery success rate by combining with the Markov decision model so as to obtain the optimal customer recovery strategy, so that the customer needing to be recovered can be more accurately positioned, and the accurate recovery strategy is provided for the customer, so that the customer recovery is more efficient, intelligent and controllable, meanwhile, the recovery of the target customer by service personnel can be promoted, and the working efficiency of the service personnel is improved.

The application also provides another embodiment, namely a computer readable storage medium, wherein the computer readable storage medium stores computer readable instructions, the computer readable instructions can be executed by at least one processor, so that the at least one processor executes the steps of the method for recovering the customer loss, the prediction accuracy of the customer recovery success rate can be effectively improved through predicting according to the customer feature vector by a prediction model, and a Markov decision model is combined, and decisions are made according to the customer feature vector and the customer recovery success rate to obtain an optimal customer recovery strategy, so that the customer needing to be recovered can be more accurately positioned, and an accurate recovery strategy is provided for the customer recovery, so that the customer recovery is more efficient, intelligent and controllable, meanwhile, the recovery of the target customer by a business person can be promoted, and the work efficiency of the business person is improved.

From the above description of the embodiments, it will be clear to those skilled in the art that the above-described embodiment method may be implemented by means of software plus a necessary general hardware platform, but of course may also be implemented by means of hardware, but in many cases the former is a preferred embodiment. Based on such understanding, the technical solution of the present application may be embodied essentially or in a part contributing to the prior art in the form of a software product stored in a storage medium (e.g. ROM/RAM, magnetic disk, optical disk) comprising instructions for causing a terminal device (which may be a mobile phone, a computer, a server, an air conditioner, or a network device, etc.) to perform the method according to the embodiments of the present application.

It is apparent that the above-described embodiments are only some embodiments of the present application, but not all embodiments, and the preferred embodiments of the present application are shown in the drawings, which do not limit the scope of the patent claims. This application may be embodied in many different forms, but rather, embodiments are provided in order to provide a thorough and complete understanding of the present disclosure. Although the application has been described in detail with reference to the foregoing embodiments, it will be apparent to those skilled in the art that modifications may be made to the embodiments described in the foregoing description, or equivalents may be substituted for elements thereof. All equivalent structures made by the content of the specification and the drawings of the application are directly or indirectly applied to other related technical fields, and are also within the scope of the application.

Claims

1. A method for customer churn retrieval, comprising the steps of:

2. The method for customer churn retrieval according to claim 1, wherein said first predictor model comprises at least two LSTM layers, a full link layer and an output layer, and said step of processing said customer feature vector by said first predictor model to obtain customer state information and a first prediction result comprises:

Extracting features of the client feature vectors through each LSTM layer in sequence to obtain feature hiding states with different scales;

inputting the characteristic hiding states with different scales into the full-connection layer for splicing to obtain client state information;

And inputting the client state information into the output layer for prediction to obtain a first prediction result.

3. The method for customer churn retrieval according to claim 2, wherein each LSTM layer includes a forget gate, an input gate and an output gate, and the step of extracting features of the customer feature vector sequentially by each LSTM layer to obtain feature hiding states of different scales includes:

4. The method of claim 1, wherein the second predictor model is a XGBoost model, and wherein the step of processing the client feature vector and the client state information by the second predictor model to obtain a second prediction result comprises:

5. The method of customer churn retrieval according to claim 1, wherein prior to said step of inputting said customer state information and said customer retrieval success rate into a trained markov decision model for decision making, comprising:

6. The method of customer churn retrieval of claim 1, further comprising, prior to said step of inputting said customer feature vector into a trained predictive model:

carrying out feature engineering on the key client features to obtain normalized client feature data, and randomly dividing the normalized client feature data into a training set and a testing set;

training the LSTM model through the training set to obtain a predicted time sequence characteristic and a first predicted probability;

Inputting the predicted time sequence characteristics and the training set into the XGBoost model to obtain a second predicted probability;

calculating a loss value between the predicted retrieval probability and the actual retrieval probability based on a preset loss function;

Inputting the test set into the model to be verified to obtain a verification result, and determining the model to be verified as a final prediction model when the verification result meets a preset condition.

7. The method of customer churn retrieval of claim 6 wherein said step of inputting said predicted timing characteristics and said training set into said XGBoost model to obtain a second predicted probability comprises:

8. An apparatus for customer churn retrieval, comprising:

9. A computer device comprising a memory having stored therein computer readable instructions which when executed by a processor implement the steps of the customer churn retrieval method of any one of claims 1 to 7.

10. A computer readable storage medium having stored thereon computer readable instructions which when executed by a processor implement the steps of the customer churn retrieval method of any one of claims 1 to 7.