A method, device, equipment and storage medium for distributing clue data

Abstract

This paper provides a method, device, equipment and storage medium for distributing clue data, wherein the method includes: determining the clue vector of the clue data to be distributed according to the input vector format of the business retention probability model, the clue data to be distributed includes: business withdrawal information and business-related customer behavior information; inputting the clue vector to be distributed into the business retention probability model, calculating the retention success probability of the clue data to be distributed, the business retention probability model is trained by historical clue samples and their retention success probabilities, and the historical clue samples include clue data related to preset businesses that have been retained; according to the retention success probability of the clue data to be distributed, the level of the clue data to be distributed is divided according to the probability interval of each level; according to the priority order of the level of the clue data to be distributed, the clue data to be distributed is distributed to the corresponding processing personnel. This paper can reduce the labor cost of withdrawal and retention, and improve the retention efficiency and retention rate.

Description

Technical Field

[0001] The present invention relates to the field of business distribution technology, and in particular to a clue data distribution method, device, equipment and storage medium. Background Art

[0002] In the prior art, in the case of a customer cancelling a service (such as cancelling an insurance policy), in order to reduce the number of cancellations, it is often necessary to perform a business retention operation on the customer who initiated the cancellation request.

[0003] In the current customer cancellation business retention process, the lead data is directly and randomly distributed to the processing personnel for processing. The random distribution method is only applicable when there are sufficient processing personnel. Because the lead data is not differentiated, there are problems with low retention efficiency and success rate.

[0004] In addition, when there is a shortage of processing personnel, valuable lead data cannot be processed in a timely manner, causing a large amount of valuable lead data to accumulate, thereby reducing retention processing efficiency, leading to low retention rates and a large number of customer business losses.

[0005] How to distribute valuable lead data efficiently and timely to improve the efficiency of lead data distribution and retention rate is an urgent problem to be solved. Summary of the invention

[0006] This article aims to solve the problem in the existing technology that in the process of retaining customers' business cancellation, the lead data is not distributed in a timely manner, which reduces the efficiency of retention processing, leads to low retention rate and a large number of customer losses.

[0007] In order to solve the above technical problems, a first aspect of this document provides a clue data distribution method, including:

[0008] Determine the to-be-allocated clue vector of the to-be-allocated clue data according to the input vector format of the business retention probability model, wherein the to-be-allocated clue data includes: business withdrawal information and business-related customer behavior information;

[0009] Input the lead vector to be assigned into the business retention probability model to calculate the retention success probability of the lead data to be assigned, wherein the business retention probability model is trained by historical lead samples and their retention success probabilities, and the historical lead samples include lead data related to the preset business and on which retention operations have been performed;

[0010] According to the probability of success in retaining the clue data to be allocated, the levels of the clue data to be allocated are divided according to the probability intervals of each level;

[0011] Distribute the lead data to be assigned to the corresponding processing personnel according to the priority order of the lead data to be assigned.

[0012] As a further embodiment of this invention, the business retention probability model training process includes:

[0013] Aggregate and digitize the features in the historical clue samples to obtain the historical clue sample vector;

[0014] Digitally process the retention results corresponding to the historical clue sample vectors to obtain the retention success probability of the historical clue sample vectors;

[0015] The business retention probability model is trained by using historical clue sample vectors and their retention success probability, using k-fold cross validation and grid search algorithms.

[0016] As a further embodiment of this invention, before aggregating and digitizing the features in the historical clue samples, the following steps are further included:

[0017] Counting the missing rate of features in historical clue samples, and deleting historical clue samples whose missing rate of features is greater than a predetermined value;

[0018] Determine whether each feature in the historical clue sample exceeds the value range. If so, limit the feature to the value range.

[0019] As a further embodiment of this invention, the historical clue sample includes continuous features and categorical features, each categorical feature includes multiple categories, wherein the features in the historical clue sample are aggregated and digitized to obtain a historical clue sample vector, including:

[0020] The following processing is performed on the historical clue samples: determine the data volume distribution of each category in each categorical feature, and set the category with data volume less than a predetermined threshold as a new category; perform interval division processing on each continuous feature;

[0021] Digitally process the processed historical clue samples to obtain characteristic values ​​of the historical clue samples;

[0022] Aggregating the feature values ​​of the historical clue samples to obtain aggregated feature values;

[0023] According to the eigenvalues ​​of the historical clue samples and the aggregated eigenvalues, a clue sample vector is obtained.

[0024] As a further embodiment of this invention, the business retention probability model is trained by using the historical clue sample vector and its retention success probability, using the k-fold cross validation and grid search algorithm, including:

[0025] a. Divide the clue sample vectors into k groups of clue sample vectors, and perform a classifier training process for each group of clue sample vectors, wherein each classifier training process includes: taking one group of clue sample vectors as a validation set and the remaining groups of clue sample vectors as a training set; training the classifier using the training set and the retention success probability of each clue sample vector in the training set; inputting each clue sample vector in the validation set into the trained classifier respectively, and calculating the retention success probability of each clue sample vector in the validation set; and calculating the performance evaluation index value of the classifier for the validation set according to the retention success probability of each clue sample vector in the validation set;

[0026] b. averaging the performance evaluation index values ​​obtained by training k groups of clue sample vectors k times in step a to obtain an average performance evaluation index value;

[0027] c. Determine whether the parameter values ​​of the classifier have been adjusted according to the parameter adjustment strategy. If not, execute step d; if yes, execute step e;

[0028] d. Adjust the parameters of the classifier according to the preset parameter adjustment strategy, and then return to continue executing steps a to c;

[0029] e. Filter out the parameter value of the classifier corresponding to the highest average performance evaluation index value;

[0030] f. Determine the business retention probability model using the selected classifier parameter values;

[0031] g. Use the clue sample vector and its retention success probability to train the business retention probability model.

[0032] As a further embodiment of this invention, the clue data distribution method further includes:

[0033] According to the data of the leads to be assigned and the portraits of the processing personnel, establish the corresponding relationship between the data of the leads to be assigned and the processing personnel;

[0034] Distributing the clue data to be allocated to corresponding processing personnel in the order of priority levels of the clue data to be allocated is further as follows:

[0035] In the order of the priority levels of the clue data to be assigned, and according to the correspondence between the clue data to be assigned and the processing personnel, the clue data to be assigned are distributed to the corresponding processing personnel.

[0036] As a further embodiment of this invention, the clue data distribution method further includes:

[0037] Count the amount of lead data that has not been processed by the processing staff;

[0038] Calculate the remaining processing capacity of each processing personnel based on the processing personnel's upper limit processing capacity and the amount of unprocessed clue data;

[0039] If the remaining processing capacity of each processing personnel is less than the preset value;

[0040] Then readjust the probability intervals of each level.

[0041] A second aspect of the present invention provides a clue data distribution device, comprising:

[0042] A conversion module, for determining a to-be-allocated lead vector of to-be-allocated lead data according to a vector format inputted into a business retention probability model, wherein the to-be-allocated lead data includes: business cancellation information and business-related customer behavior information;

[0043] A calculation module, used for inputting the lead vector to be assigned into a business retention probability model, and calculating the retention success probability of the lead data to be assigned, wherein the business retention probability model is trained by historical lead samples and their retention success probabilities, and the historical lead samples include lead data related to the preset business and on which retention operations have been performed;

[0044] A level classification module, used to classify the level of the lead data to be assigned according to the probability intervals of each level according to the success probability of retaining the lead data to be assigned;

[0045] The distribution module is used to distribute the clue data to be distributed to corresponding processing personnel according to the priority order of the levels of the clue data to be distributed.

[0046] The third aspect of this document provides a computer device, including a memory, a processor, and a computer program stored in the memory, and when the computer program is executed by the processor, the instructions of the clue data distribution method described in any of the aforementioned embodiments are executed.

[0047] A fourth aspect of the present invention provides a computer storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor of a computer device, the computer program executes instructions of the clue data distribution method described in any of the aforementioned embodiments.

[0048] The clue data distribution method and device provided in this article establish a business cancellation retention probability model by learning historical clue samples and their retention success probabilities; predict the retention success probability of the clue data to be assigned by the business cancellation retention probability model; divide the level of the clue data to be assigned according to the probability intervals of each level according to the retention success probability of the clue data to be assigned; distribute the clue data to be assigned to the corresponding processing personnel according to the priority order of the level of the clue data to be assigned, which can replace the manual random allocation method, save the required manpower cost, and can realize the effective allocation of the clue data to be assigned under the condition of limited manpower, and retain customers with a high probability of business cancellation retention in time, thereby improving the retention efficiency and retention rate.

[0049] In order to make the above and other purposes, features and advantages of this article more obvious and easy to understand, the following specifically cites preferred embodiments and describes them in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] In order to more clearly illustrate the technical solutions in the embodiments of this article or the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of this article. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0051] Figure 1 A first flow chart of the clue data distribution method according to the embodiment of this invention is shown;

[0052] Figure 2 A flowchart showing the business retention probability model training process of the embodiment of this article is shown;

[0053] Figure 3 A flowchart of the historical clue sample feature aggregation and digitization processing process in the embodiment of this article is shown;

[0054] Figure 4 A flowchart showing a process of training a business retention probability model using a k-fold cross validation and a grid search algorithm in an embodiment of this invention;

[0055] Figure 5 A second flow chart of the clue data distribution method according to the embodiment of this invention is shown;

[0056] Figure 6 A third flow chart of the clue data distribution method of the embodiment of this article is shown;

[0057] Figure 7 A fourth flow chart of the clue data distribution method of the embodiment of this document is shown;

[0058] Figure 8A structural diagram of a clue data distribution device according to an embodiment of this invention is shown;

[0059] Fig. 9 A flowchart showing the training process of the policy cancellation and retention probability model in the embodiment of this article is shown;

[0060] Fig.10 A flowchart of the method for distributing policy cancellation clue data according to an embodiment of this invention is shown;

[0061] Fig.11 A structural diagram of a computer device according to an embodiment of this invention is shown.

[0062] Description of the accompanying symbols:

[0063] 810, conversion module;

[0064] 820, calculation module;

[0065] 830, level classification module;

[0066] 840, distribution module;

[0067] 1102. Computer equipment;

[0068] 1104. Processor;

[0069] 1106. Memory;

[0070] 1108, driving mechanism;

[0071] 1110, input / output module;

[0072] 1112. Input device;

[0073] 1114. Output device;

[0074] 1116. Presentation equipment;

[0075] 1118. Graphical user interface;

[0076] 1120, network interface;

[0077] 1122. Communication link;

[0078] 1124. Communication bus. DETAILED DESCRIPTION

[0079] The following will be combined with the drawings in the embodiments of this article to clearly and completely describe the technical solutions in the embodiments of this article. Obviously, the described embodiments are only part of the embodiments of this article, not all of the embodiments. Based on the embodiments of this article, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this article.

[0080] It should be noted that the terms "including" and "having" and any variations thereof in the specification, claims and drawings of this document are intended to cover non-exclusive inclusions. For example, a process, method, apparatus, product or equipment that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units that are not explicitly listed or are inherent to these processes, methods, products or equipment.

[0081] This specification provides method operation steps as described in the embodiments or flow charts, but more or fewer operation steps may be included based on routine or non-creative work. The order of steps listed in the embodiments is only one way of executing the order of many steps and does not represent the only execution order. When the system or device product is executed in practice, it can be executed in the order of the method shown in the embodiments or the drawings or in parallel.

[0082] It should be noted that the clue data distribution method and device in this article can be used in the insurance field. The clue data is the clue data for initiating the cancellation business, and the business is, for example, insurance business. It can also be used in any field other than the insurance field. This article does not limit the application field of the clue data distribution method and device. The customer cancellation business information and business-related customer behavior information involved in this article are all data authorized by the customer or fully authorized by all parties.

[0083] In one embodiment of the present invention, a method for distributing lead data is provided to solve the problem that in the process of retaining customers who have withdrawn from the service in the prior art, lead data distribution is not timely, thereby reducing the efficiency of retention processing, resulting in low retention rate and a large number of customer losses. Specifically, Figure 1 As shown, the clue data distribution method includes:

[0084] Step 110, determining a to-be-allocated lead vector of the to-be-allocated lead data according to the input vector format of the business retention probability model, wherein the to-be-allocated lead data includes the business cancellation information and the business-related customer behavior information;

[0085] Step 120, inputting the lead vector to be assigned into the business retention probability model, and calculating the retention success probability of the lead data to be assigned, wherein the business retention probability model is trained by historical lead samples and their retention success probabilities, and the historical lead samples include lead data related to the preset business and on which retention operations have been performed;

[0086] Step 130, according to the retention success probability of the lead data to be allocated, the level of the lead data to be allocated is divided according to the probability interval of each level;

[0087] Step 140: distribute the clue data to be assigned to corresponding processing personnel according to the priority order of the levels of the clue data to be assigned.

[0088] In detail, the data of leads to be allocated comes from customers who have initiated a request for a withdrawal service but have not yet carried out a retention operation. The withdrawal service information includes the time of initiation of the withdrawal service, the initiation channel, basic business information, and basic customer information. Taking the insurance business as an example, the initiation channels include but are not limited to Weinsurance, 360, Baidu, etc. The basic business information includes but is not limited to premium information, insurance time, withdrawal time, insurance start time, insured city, claim information, etc. The basic customer information includes age, gender, whether there is a social security identity, etc. Business-related customer behavior information includes whether the customer has initiated a withdrawal service operation before initiating the withdrawal service request, whether the retention is successful, etc. Taking the insurance business as an example, business-related customer behavior information includes whether the withdrawal operation is carried out, whether the withdrawal retention is successful, whether the policy has been claimed, etc. This paper can improve the accuracy of the prediction of the probability of successful retention by incorporating business-related customer behavior information into the input vector.

[0089] The lead data to be assigned contains many features (variables). The input vector of the business retention probability model specifies the features and specific positions input to the business retention probability model. According to the input vector format of the business retention probability model, the input vector that meets the business retention probability model can be determined.

[0090] In order to ensure the training speed of the business retention probability model, the business retention probability model described in this article can use the LightGBM model. LightGBM is a gradient boosting framework that uses a decision tree-based learning algorithm. LightGBM implements the GBDT (Gradient Boosting Decision Tree) algorithm to efficiently train multiple decision trees in parallel, and has the advantages of fast training speed, low memory consumption, strong prediction ability, support for distribution, and the ability to quickly process massive data. It is widely used in industrial practice. The specific training process of the business retention probability model refers to the subsequent embodiments and will not be described in detail here.

[0091] The preset business for training the business retention probability model is specified according to the demand, and this document does not limit this. Any business that needs to be retained can be used as the preset business described in this document.

[0092] The variables included in the clue data that has been retained are the same as those in the clue data to be allocated. The retention success probability of the historical clue sample is determined according to the retention result. Specifically, for example, if the retention result of a historical clue sample is a retention success, then its corresponding retention success probability is 1. If the retention result of a historical clue sample is a retention failure, then its corresponding retention success probability is 0. For the definition of the retention result, please refer to the subsequent embodiments and will not be described in detail here.

[0093] The probability intervals of each level can be set according to needs, and this article does not limit its specific values. The higher the probability of retention, the higher the priority corresponding to the level. In some implementations, only two level probability intervals may be included, a level with a high probability of successful retention and a level with a low probability of successful retention. The two levels are distributed to different processing personnel for different retention operations. In specific implementation, the retention operation can be achieved by, for example, making a phone call, visiting the door, etc., which is not limited in this article. The processing personnel can be the business responsible personnel, or they can be personnel who specialize in handling retention operations. Through the implementation of steps 130 and 140, this article can improve the allocation efficiency of the clue data to be allocated.

[0094] This embodiment calculates the success probability of retaining the lead data to be assigned by using the business retention probability model trained based on historical lead samples and their success probability of retaining, divides the level of the lead data to be assigned according to the success probability of retaining the lead data to be assigned, and distributes the lead data to be assigned to corresponding processing personnel in the order of the level priority of the lead data to be assigned. It can replace the manual random allocation method, save the required manpower cost, and can achieve effective allocation of the lead data to be assigned under the condition of limited manpower, and retain customers with a high probability of business withdrawal and retention in time, thereby improving the retention efficiency and retention rate.

[0095] In one embodiment of this invention, Figure 2 As shown, the business retention probability model training process includes:

[0096] Step 201, aggregate and digitize the features in the historical clue samples to obtain a historical clue sample vector;

[0097] Step 202, digitally process the retention result corresponding to the historical clue sample vector to obtain the retention success probability of the historical clue sample vector;

[0098] Step 203 , using the historical clue sample vectors and their retention success probabilities, a business retention probability model is trained using a k-fold cross validation and grid search algorithm.

[0099] Step 201 can prevent dimensionality disasters, reduce model training time, enhance model generalization capabilities, reduce overfitting, and create more meaningful and usable features. Historical clue samples can extract clue data from different channels from the business database according to preset requirements. Specifically, before extracting data from the business database, the data related to modeling can be manually screened out based on the meaning of all features in the relevant tables understood by the processing personnel, and then the features with unbalanced data distribution can be deleted. For example, the policyholder customer type field has two category values ​​1 and 2, and the number ratio of these two categories is 1000000:1, then the field can be deleted.

[0100] The historical clue samples described in this article include: categorical features and continuous features. Taking insurance business as an example, categorical features include gender, insurance channel (Weibo, 360, Baidu), insurance sales plan (Weibo million medical insurance, 360 universal medical insurance, etc.), etc., and continuous features include age, etc.

[0101] Specifically, Figure 3 As shown, the implementation process of step 201 includes:

[0102] The following steps 301 and 302 are performed on the historical clue samples:

[0103] Step 301, determining the data volume distribution of each category in each category-type feature, and setting the category with a data volume less than a predetermined threshold as a new category. The name of the new category is, for example, other categories, and the name of the new category can be set according to demand;

[0104] Step 302, performing interval division processing on each continuous feature;

[0105] Step 303, digitally processing the categorical features processed in step 301 and the continuous features processed in step 302 to obtain feature values ​​of the historical clue samples;

[0106] Step 304, performing aggregation processing on the feature values ​​of the historical clue samples to obtain an aggregated feature value;

[0107] Step 305: Obtain a clue sample vector according to each feature value and aggregated feature value of the historical clue samples.

[0108] The predetermined threshold in step 301 can be set according to actual conditions, for example, to 100. Taking the insurance business as an example, there are 119 insurance sales plans in the clue sample, of which the number of insurance policies corresponding to 100 sales plan categories is less than 100. In this case, these 100 sales plan categories can be represented as "other insurance types".

[0109] In step 302, the interval division method can be set according to actual needs, and this document does not limit this. Taking age as an example, for example, every 10 years is divided into an interval, such as [0-10], [10-20], [20-30], [30-40], [40-50]...etc.

[0110] When step 303 is implemented, the data obtained in step 301 and step 302 can be digitized using processing methods such as LabelEncoder, get_dummies or OneHotEncoder. Among them, LabelEncoder converts text-type category data into numerical values. For example, there are three channels in all the data: Weibo, 360, and Baidu. After being processed by LabelEncoder, these three category features become 0, 1, and 2 respectively. For example, the age range of 0-10 years old is 1, 10-20 years old is 2, etc. get_dummies is a way to perform one-hot encoding. OneHotEncoder converts digital or text-type category data into one or more columns of data with only 0 and 1. For example, channel features can be converted to Weibo: [1,0,0], 360: [0,1,0], Baidu: [0,0,1].

[0111] In step 304, the feature value aggregation processing of the historical clue samples can be implemented according to the aggregation features set by the business. When implemented, the aggregation features can be calculated using the groupby operation of pandas. Taking the insurance business as an example, the aggregation features include but are not limited to: the number of insurance purchased by the insured, the average age, maximum age, and minimum age of the insured in the insurance policy purchased by the insured, and the maximum premium, minimum premium, and average premium in the insurance policy purchased by the insured.

[0112] In step 305, the characteristic values ​​and aggregated characteristic values ​​of the historical clue samples are collected together to form a clue sample vector, and the clue sample vector constitutes the input vector format of the business retention probability model.

[0113] In step 202, the probability of successful retention includes successful retention and failed retention. When the probability of successful retention is digitized, the probability of successful retention can be marked as 1 if the retention is successful, and as 0 if the retention fails. Whether the retention is successful or not can be specified according to the business situation. Taking the insurance business as an example, when the customer applies for a refund, the text customer service will perform a refund retention operation on the customer in the form of CSS, and the telephone outbound customer service will call the customer to perform a refund retention operation. The definition of successful retention is: the customer has answered the customer service call or responded to the customer service consultation, and the difference between the refund time and the consultation time (the time of the customer service outbound call or the time of consulting the customer through CSS) is greater than 35 days (can be set according to demand), and the single-period retention premium is greater than 0, indicating that the refund retention is successful, otherwise the refund retention fails. Among them, the text customer service and the telephone outbound customer service are both processing personnel.

[0114] The above step 203 can improve the generalization ability of the model. Specifically, Figure 4 As shown, step 203 uses the historical clue sample vector and its retention success probability to train the business retention probability model using k-fold cross validation and grid search algorithm, including:

[0115] Step a. Divide the clue sample vectors into k groups of clue sample vectors, and perform a classifier training process for each group of clue sample vectors, wherein each classifier training process includes: taking one group of clue sample vectors as a verification set, and the remaining groups of clue sample vectors as training sets; training the classifier using the training set and the retention success probability of each clue sample vector in the training set; inputting each clue sample vector in the verification set into the trained classifier respectively, and calculating the retention success probability of each clue sample vector in the verification set; and calculating the performance evaluation index value of the classifier for the verification set according to the retention success probability of each clue sample vector in the verification set;

[0116] Step b. averaging the performance evaluation index values ​​obtained by training k groups of clue sample vectors k times in step a to obtain an average performance evaluation index value;

[0117] Step c, determining whether the parameter values ​​of the classifier have been adjusted according to the parameter adjustment strategy, if not, executing step d, if yes, executing step e;

[0118] Step d, adjusting the parameters of the classifier according to the preset parameter adjustment strategy, and then returning to continue to execute steps a to c;

[0119] Step e, screening out the parameter value of the classifier corresponding to the highest average performance evaluation index value;

[0120] Step f, using the parameter values ​​of the screened classifiers to determine a business retention probability model;

[0121] Step g: using the clue sample vector and its retention success probability to train the business retention probability model.

[0122] In the specific implementation, step a can use recall rate, precision rate and Fl-score as performance evaluation indicators. In order to avoid overfitting of the model, AUC (Area under the Curve of ROC) can also be used as a performance evaluation indicator. The ROC curve defines the false positive rate (FPR) as the X-axis and the true positive rate (TPR) as the Y-axis. FPR and TPR are defined as:

[0123] TPR: The ratio of samples that are correctly judged as positive among all samples that are actually positive.

[0124] FPR: The ratio of samples that are incorrectly judged as positive among all samples that are actually negative.

[0125]

[0126]

[0127] Among them, TP is the sample size with a positive actual result and a positive predicted result; FP is the sample size with a negative actual result and a positive predicted result; FN is the sample size with a positive actual result and a negative predicted result; TN is the sample size with a negative actual result and a negative predicted result.

[0128] After training the business retention probability model, the performance of the trained policy cancellation retention prediction model can be tested using a test set that did not participate in the model training. The trained policy cancellation retention prediction model can be used to obtain the probability value corresponding to each sample point in the test set. The prediction values ​​of all samples are sorted from high to low, and then each probability value is taken as the threshold in turn. All samples greater than or equal to the threshold are positive, and samples less than the threshold are negative. The FPR and TPR values ​​are calculated according to the above formula, which corresponds to a point on the ROC curve, with FPR as the horizontal coordinate and TPR as the vertical coordinate. All points are drawn out in turn according to this method, and then the points are connected in turn to obtain the ROC curve. The area under the ROC curve is the AUC value.

[0129] The k value can be selected according to the actual situation. Generally, the k value is 5. The classifier can be selected as LightGBM. The parameters of the classifier include num_iterations (number of iterations), max_depth (depth of the tree), num_leaves (number of leaf nodes), learning_rate (learning rate), etc. A step size can be set for each parameter of the classifier. The step size of each parameter can be the same or different. The specific value of the step size can be determined according to the actual situation, and this article does not limit it.

[0130] In step d, the parameter adjustment strategy specifies the adjustment order and supplement of parameters. In specific implementation, only one parameter may be adjusted at a time, or multiple parameters may be adjusted at a time. The adjustment step sizes of the parameters may be the same or different. This article does not specifically limit the parameter adjustment strategy.

[0131] In step e, when AUC is used as the performance evaluation indicator, an average performance evaluation index value will be calculated for each parameter value combination, and the larger the average performance evaluation index value, the more accurate the classifier is. The parameter value corresponding to the maximum value of the average performance evaluation index value is used as the parameter value of the allocator.

[0132] In step f, the parameter values ​​of the screened classifiers are substituted into the classifiers with unknown parameters to obtain a business retention probability model.

[0133] In one embodiment of this invention, in order to improve the accuracy of the business retention probability model, Figure 5 As shown, before the above step 201 is implemented, it also includes:

[0134] Step 2001, counting the missing rate of features in historical clue samples;

[0135] Step 2002, deleting historical clue samples whose feature missing rate is greater than a predetermined value;

[0136] Step 2003, determine whether each feature in the historical clue sample exceeds the value range, and if so, limit the feature to the value range.

[0137] The missing rate of features in step 2001 refers to the proportion of features without values ​​to the total features. The predetermined value in step 2002 can be set according to the actual situation, and this document does not limit this. When step 2003 is implemented, the feature can be set to an upper limit or a lower limit. Specifically, if the feature value exceeds the upper limit, it is set as the upper limit, and if it exceeds the lower limit, it is set as the lower limit. Through step 2003, the sample data can be normalized to prevent data entry errors from affecting the accuracy of model training.

[0138] In one embodiment of the present invention, in order to further improve the retention efficiency and success rate, the lead data distribution method includes the above steps 110 to 130, as follows: Figure 6 As shown, it also includes:

[0139] Step 1401, establishing a correspondence between the lead data to be assigned and the processing personnel according to the lead data and the processing personnel portrait;

[0140] Step 1402 , distribute the clue data to corresponding processing personnel in the order of priority levels of the clue data and according to the correspondence between the clue data to be distributed and the processing personnel.

[0141] Specifically, in step 1401, the processing personnel portrait includes: the processing personnel retention success rate, the processing personnel's professional service business type, and the processing personnel's basic information (age, gender, region, etc.). The corresponding relationship between the lead data to be assigned and the processing personnel can be established based on the professional service business type in the processing personnel portrait and the business type in the lead data.

[0142] When step 1402 is implemented, the clue data can be obtained in the order of priority levels of the clue data, and the processing personnel corresponding to the obtained clue data can be determined according to the correspondence between the clue data to be assigned and the processing personnel, and the clue data can be sent to the processing personnel.

[0143] In one embodiment of the present invention, in order to avoid the situation where the best retention period is delayed due to the heavy tasks at hand of the processing personnel, Figure 7 As shown, the clue data distribution method is Figure 1 and Figure 6 In addition to the content shown above, it also includes:

[0144] Step 150, counting the amount of clue data not processed by the processing personnel;

[0145] Step 160, calculating the remaining processing capacity of each processing personnel according to the upper limit processing capacity of the processing personnel and the amount of unprocessed clue data;

[0146] Step 170, determining whether the remaining processing capacity of each processing personnel is less than a preset value. If the remaining processing capacity of each processing personnel is less than the preset value, executing step 180. If the remaining processing capacity of each processing personnel is greater than or equal to the preset value, no processing is required.

[0147] Step 180, readjusting the probability intervals of each level to redetermine the level of the clue data to be allocated.

[0148] This embodiment adjusts the probability intervals of each level according to the remaining processing capacity of the processing personnel, so as to ensure that the clue data with a high retention probability is distributed and processed in a timely manner, thereby improving the retention efficiency and retention rate.

[0149] Based on the same inventive concept, this article also provides a clue data distribution device, as described in the following embodiments. Since the principle of solving the problem by the clue data distribution device is similar to that of the clue data distribution method, the implementation of the clue data distribution device can refer to the clue data distribution method, and the repeated parts will not be repeated.

[0150] Specifically, Figure 8 As shown, the clue data distribution device includes:

[0151] The conversion module 810 is used to determine the to-be-allocated clue vector of the to-be-allocated clue data according to the business retention probability model input vector format, wherein the to-be-allocated clue data includes: business cancellation information and business-related customer behavior information;

[0152] A calculation module 820 is used to input the lead vector to be assigned into a business retention probability model to calculate the retention success probability of the lead data to be assigned, wherein the business retention probability model is trained by historical lead samples and their retention success probabilities, and the historical lead samples include lead data related to the preset business and on which retention operations have been performed;

[0153] A level classification module 830 is used to classify the level of the clue data to be assigned according to the probability of success of retaining the clue data to be assigned and the probability interval of each level;

[0154] The distribution module 840 is used to distribute the clue data to be distributed to corresponding processing personnel according to the priority order of the levels of the clue data to be distributed.

[0155] This embodiment can replace the manual random allocation method, saving the required labor costs. It can also effectively allocate the lead data to be allocated under the condition of limited manpower, and timely retain customers with a high probability of business withdrawal, thereby improving the retention efficiency and retention rate.

[0156] In order to more clearly illustrate the technical solution of this article, the training and use process of the policy cancellation retention probability model is explained below using the policy cancellation retention as an example. It can be divided into two stages: preparation stage and application stage.

[0157] 1. Preparation

[0158] The purpose of this article is to predict the probability of successful cancellation and retention of each clue data, and to determine whether to perform cancellation and retention operations on this clue data by adjusting the threshold. The specific process is as follows Fig. 9 shown.

[0159] Step 901, according to business needs, select a preset insurance type, extract historical clue data of all customers from different channels from the business database, and the historical clue data of each customer is a historical clue sample.

[0160] Since there are many fields in the historical clue data that are irrelevant or redundant to modeling, feature data for data analysis and modeling are processed from the business database, mainly including business information and business-related customer behavior information. Among them, the cancellation business information includes policy information, time information (insurance time, cancellation time, insurance start time), insurance city, basic information of the insured and the insured (age, gender, whether there is social security status). Business-related customer behavior information includes claim information, whether the cancellation operation is carried out, whether the cancellation retention is successful, whether the policy has been claimed, etc.

[0161] Step 902: filter out the lead data of the customer who has performed the policy cancellation and retention operation from the historical lead data of the customer, and perform missing value processing and outlier processing on the filtered lead data. Specifically:

[0162] The missing values ​​were processed as follows: the missing rate of each feature was calculated, and features with a missing rate greater than 70% were deleted.

[0163] The outliers are processed as follows: samples whose feature values ​​are too large, too small, or do not conform to the data distribution of the business scenario can be directly deleted or replaced with the average value of the feature or with the upper and lower limits.

[0164] In step 903, the clue data obtained in step 902 are divided into two categories, successful policy cancellation and retention and failed policy cancellation and retention, according to the definition of whether the policy cancellation and retention are successful, and are marked as 1 and 0 respectively.

[0165] In step 904, feature engineering is performed on the clue data in step 902.

[0166] Specifically, feature engineering processing includes aggregating and digitizing clue data. Discretize continuous feature data, for example, divide age features into multiple age intervals, 0-10 years old is 1, 10-20 years old is 2, etc. Analyze and process the data of categorical features first, that is, first determine the data volume distribution of each category in the categorical feature, and for each categorical feature, filter out categories with less data volume than the predetermined threshold, set the filtered categories to the same category, and then use LabelEncoder, OneHotEncoder or get_dummies and other methods for digitization. For example, the city where the insured is located, "Wuhan City", can be digitized to 1. Construct some aggregate features based on the purchase information of the policyholder and the insured, such as the average age and maximum age of the policyholder.

[0167] In step 905, the feature data subjected to feature engineering in step 904 is randomly divided into two parts according to a certain ratio, one of which is used as a test set for testing the performance of the model, and the other is used as a training set for training the model. It should be noted that the amount of data in the test set and the training set can be the same or different, and this is not limited in this article.

[0168] In step 906, LightGBM is used to train the classification model, and the parameters of LightGBM are adjusted using the five-fold cross validation and grid search methods, and the AUC value on the test set is used as the performance evaluation indicator of this classification model. The specific implementation steps include:

[0169] First, for each parameter to be adjusted in LightGBM, determine the parameter range and adjust the parameters in sequence according to the step size. The core parameters to be adjusted include num_iterations, max_depth, num_leaves, learning_rate, etc.

[0170] Then, the training set used for training the model in the above step 905 is equally divided into 5 groups of clue sample vectors, each group of clue sample vectors is used as a validation set, and the clue sample vectors of the remaining groups are used as training sets to train the LightGBM model, and the parameter with the highest average AUC value is selected from all the parameters to be selected as the optimal parameter;

[0171] Finally, all parameters to be adjusted are adjusted according to the above steps, and N (N is the number of parameter adjustments) optimal parameters will be obtained. After fixing the optimal parameters of the LightGBM model, the model is trained with the training set in step 905, and the AUC value of the test set on the model is used as the indicator for the final evaluation model. The trained model is the policy cancellation and retention probability model.

[0172] (II) Application stage

[0173] like Fig.10 As shown, the clue data distribution process includes:

[0174] Step 1001, for a batch of clue data that have applied for cancellation of insurance, the cancellation and retention success probability of each clue data can be obtained through the cancellation and retention probability model trained in the preparation stage. The higher the cancellation and retention success probability value, the more likely the clue data is to be successfully retained.

[0175] In step 1002, the business can flexibly set the probability interval, and the clue data is divided into multiple levels, for example: greater than 0.9 is a Class A clue, [0.9, 0.8] is a Class B clue, [0.8, 0.7] is a Class C clue, [0.7, 0.6] is a Class D clue, [0.6, 0.5] is a Class E clue, and [0.5, 0] is a Class F clue.

[0176] In step 1003, the business can distribute the leads according to the lead level in combination with the existing manpower, for example, performing cancellation and retention operations for Class A leads first.

[0177] This embodiment uses the LightGMB method to build a model, uses the five-fold cross-validation and grid search methods to adjust parameters, and uses the AUC value as a performance evaluation indicator, which can avoid overfitting of the model and improve the generalization ability of the model.

[0178] This embodiment uses artificial intelligence methods to establish a policy cancellation retention probability model, classifies the clue data for policy cancellation retention, and intelligently distributes the clue data. The method is intelligently applied to the policy cancellation retention system of the insurance industry, replacing the manual random distribution of clues, greatly reducing the manpower cost of policy cancellation retention, and improving the retention rate, solving the defects of the current random distribution of clue data that consumes a lot of manpower and has an uncertain retention rate.

[0179] In one embodiment of the present invention, a computer device is also provided for implementing the lead data distribution method and the process of establishing a business retention probability model. Fig.11 As shown, the computer device 1102 may include one or more processors 1104, such as one or more central processing units (CPUs), each of which may implement one or more hardware threads. The computer device 1102 may also include any memory 1106, which is used to store any kind of information such as code, settings, data, etc. Non-limitingly, for example, the memory 1106 may include any one or more combinations of the following: any type of RAM, any type of ROM, flash memory device, hard disk, optical disk, etc. More generally, any memory may use any technology to store information. Further, any memory may provide volatile or non-volatile retention of information. Further, any memory may represent a fixed or removable component of the computer device 1102. In one case, when the processor 1104 executes an associated instruction stored in any memory or a combination of memories, the computer device 1102 may perform any operation of the associated instruction. The computer device 1102 also includes one or more drive mechanisms 1108 for interacting with any memory, such as a hard disk drive mechanism, an optical disk drive mechanism, etc.

[0180] The computer device 1102 may also include an input / output module 1110 (I / O) for receiving various inputs (via input device 1112) and for providing various outputs (via output device 1114). A specific output mechanism may include a presentation device 1116 and an associated graphical user interface 1118 (GUI). In other embodiments, the input / output module 1110 (I / O), input device 1112, and output device 1114 may not be included, and the computer device 1102 may be used as a computer device in a network. The computer device 1102 may also include one or more network interfaces 1120 for exchanging data with other devices via one or more communication links 1122. One or more communication buses 1124 couple the components described above together.

[0181] The communication link 1122 may be implemented in any manner, for example, through a local area network, a wide area network (e.g., the Internet), a point-to-point connection, etc., or any combination thereof. The communication link 1122 may include any combination of hardwired links, wireless links, routers, gateway functions, name servers, etc. governed by any protocol or combination of protocols.

[0182] Corresponds to Figure 1-Figure 7 The method in this article also provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the steps of the above method are executed.

[0183] The embodiment of the present invention also provides a computer readable instruction, wherein when the processor executes the instruction, the program therein causes the processor to execute the following Figures 1 to 7 The method shown.

[0184] It should be understood that in the various embodiments of this document, the size of the serial numbers of the above-mentioned processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this document.

[0185] It should also be understood that in the embodiments of this article, the term "and / or" is only a description of the association relationship of the associated objects, indicating that three relationships may exist. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. In addition, the character " / " in this article generally indicates that the associated objects before and after are in an "or" relationship.

[0186] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in terms of function in the above description. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this article.

[0187] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0188] In the several embodiments provided herein, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic, for example, the division of the units is only a logical function division, and there may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, devices or units, or can be electrical, mechanical or other forms of connection.

[0189] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the embodiments of this article.

[0190] In addition, each functional unit in each embodiment of this invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above integrated unit may be implemented in the form of hardware or in the form of software functional unit.

[0191] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this article is essentially or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of this article. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program codes.

[0192] Specific embodiments are used in this article to illustrate the principles and implementation methods of this article. The description of the above embodiments is only used to help understand the methods and core ideas of this article. At the same time, for general technicians in this field, according to the ideas of this article, there will be changes in the specific implementation methods and application scopes. In summary, the content of this specification should not be understood as a limitation on this article.

Claims

1. A clue data distribution method, characterized in that: include: Determine the to-be-allocated clue vector of the to-be-allocated clue data according to the input vector format of the business retention probability model, wherein the to-be-allocated clue data includes: business cancellation information and business-related customer behavior information; Input the to-be-assigned clue vector into the business retention probability model, and calculate the retention success probability of the to-be-assigned clue data, wherein the business retention probability model is trained by historical clue samples and their retention success probabilities, and the historical clue samples include clue data related to the preset business and on which retention operations have been performed; According to the probability of success of retaining the clue data to be allocated, the level of the clue data to be allocated is divided according to the probability intervals of each level; Distribute the clue data to be assigned to corresponding processing personnel according to the priority order of the levels of the clue data to be assigned; The business retention probability model training process includes: Aggregating and digitally processing the features in the historical clue samples to obtain a historical clue sample vector; Digitally processing the retention result corresponding to the historical clue sample vector to obtain the retention success probability of the historical clue sample vector; Using the historical clue sample vector and its retention success probability, the business retention probability model is trained using k-fold cross validation and grid search algorithms; Using the historical clue sample vector and its retention success probability, the business retention probability model is trained using k-fold cross validation and grid search algorithms, including: a. Divide the clue sample vector into k groups of clue sample vectors, and perform a classifier training process for each group of clue sample vectors, wherein each classifier training process includes: taking one group of clue sample vectors as a validation set and the remaining groups of clue sample vectors as a training set; using the training set and the retention success probability of each clue sample vector in the training set to train the classifier; inputting each clue sample vector in the validation set into the trained classifier respectively, and calculating the retention success probability of each clue sample vector in the validation set; and calculating the performance evaluation index value of the classifier for the validation set according to the retention success probability of each clue sample vector in the validation set; b. averaging the performance evaluation index values ​​obtained by training k groups of clue sample vectors k times in step a to obtain an average performance evaluation index value; c. Determine whether the parameter values ​​of the classifier have been adjusted according to the parameter adjustment strategy. If not, execute step d; if yes, execute step e; d. Adjust the parameters of the classifier according to the preset parameter adjustment strategy, and then return to continue executing steps a to c; e. Filter out the parameter value of the classifier corresponding to the highest average performance evaluation index value; f. Determine the business retention probability model using the selected classifier parameter values; g. Use the clue sample vector and its retention success probability to train the business retention probability model.

2. The clue data distribution method according to claim 1, characterized in that: Before aggregating and digitizing the features in the historical clue samples, the following steps are also included: Counting the missing rate of features in the historical clue samples, and deleting the historical clue samples whose feature missing rate is greater than a predetermined value; It is determined whether each feature in the historical clue sample exceeds a value range. If so, the feature is limited to within the value range.

3. The clue data distribution method according to claim 1, characterized in that: The historical clue samples include continuous features and categorical features, each categorical feature includes multiple categories, wherein the features in the historical clue samples are aggregated and digitized to obtain a historical clue sample vector, including: The following processing is performed on the historical clue samples: determine the data volume distribution of each category in each categorical feature, and set the category with data volume less than a predetermined threshold as a new category; perform interval division processing on each continuous feature; Digitally process the processed historical clue samples to obtain characteristic values ​​of the historical clue samples; Aggregating the feature values ​​of the historical clue samples to obtain aggregated feature values; A clue sample vector is obtained according to the feature values ​​and aggregated feature values ​​of the historical clue samples.

4. The clue data distribution method according to claim 1, characterized in that: Also includes: According to the data of the leads to be assigned and the portraits of the processing personnel, the corresponding relationship between the data of the leads to be assigned and the processing personnel is established; Distributing the clue data to be allocated to corresponding processing personnel in the order of priority levels of the clue data to be allocated is further as follows: The clue data to be assigned are distributed to corresponding processing personnel in the order of the priority levels of the clue data to be assigned and according to the corresponding relationship between the clue data to be assigned and the processing personnel.

5. The clue data distribution method according to claim 1, characterized in that: Also includes: Count the amount of lead data that has not been processed by the processing staff; Calculate the remaining processing capacity of each processing personnel based on the processing personnel's upper limit processing capacity and the amount of unprocessed clue data; If the remaining processing capacity of each processing personnel is less than the preset value; Then readjust the probability intervals of each level.

6. A clue data distribution device, characterized in that: include: A conversion module, for determining a to-be-allocated lead vector of to-be-allocated lead data according to a vector format inputted into a business retention probability model, wherein the to-be-allocated lead data includes: business cancellation information and business-related customer behavior information; A calculation module, used for inputting the to-be-assigned clue vector into the business retention probability model, and calculating the retention success probability of the to-be-assigned clue data, wherein the business retention probability model is trained by historical clue samples and their retention success probabilities, and the historical clue samples include clue data related to the preset business and for which retention operations have been performed; A level classification module, used to classify the level of the clue data to be assigned according to the probability of success of retaining the clue data to be assigned and according to each level probability interval; A distribution module, used for distributing the clue data to be distributed to corresponding processing personnel according to the priority order of the levels of the clue data to be distributed; The business retention probability model training process includes: Aggregating and digitally processing the features in the historical clue samples to obtain a historical clue sample vector; Digitally processing the retention result corresponding to the historical clue sample vector to obtain the retention success probability of the historical clue sample vector; Using the historical clue sample vector and its retention success probability, the business retention probability model is trained using k-fold cross validation and grid search algorithms; Using the historical clue sample vector and its retention success probability, the business retention probability model is trained using k-fold cross validation and grid search algorithms, including: a. Divide the clue sample vector into k groups of clue sample vectors, and perform a classifier training process for each group of clue sample vectors, wherein each classifier training process includes: taking one group of clue sample vectors as a validation set and the remaining groups of clue sample vectors as a training set; using the training set and the retention success probability of each clue sample vector in the training set to train the classifier; inputting each clue sample vector in the validation set into the trained classifier respectively, and calculating the retention success probability of each clue sample vector in the validation set; and calculating the performance evaluation index value of the classifier for the validation set according to the retention success probability of each clue sample vector in the validation set; b. averaging the performance evaluation index values ​​obtained by training k groups of clue sample vectors k times in step a to obtain an average performance evaluation index value; c. Determine whether the parameter values ​​of the classifier have been adjusted according to the parameter adjustment strategy. If not, execute step d; if yes, execute step e; d. Adjust the parameters of the classifier according to the preset parameter adjustment strategy, and then return to continue executing steps a to c; e. Filter out the parameter value of the classifier corresponding to the highest average performance evaluation index value; f. Determine the business retention probability model using the selected classifier parameter values; g. Use the clue sample vector and its retention success probability to train the business retention probability model.

7. A computer device comprising a memory, a processor, and a computer program stored in the memory, characterized in that: When the computer program is executed by the processor, the computer program executes the instructions of the method according to any one of claims 1 to 5.

8. A computer storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor of a computer device, the computer program executes the instructions of the method according to any one of claims 1 to 5.

Images