User loss reason identification method and device, equipment, storage medium and program

Abstract

The invention discloses a user loss reason identification method and device, equipment, a storage medium and a program. Acquiring behavior characteristic data of the user; inputting the behavior characteristic data into a pre-trained fusion recognition model, determining a characteristic weight corresponding to each behavior characteristic based on the behavior characteristic data through the fusion recognition model, and calculating a mean value of each characteristic weight to obtain a loss coefficient of the user; when the loss coefficient is smaller than or equal to a preset loss threshold value, determining that the user is a potential loss user; inputting the behavior characteristic data of the user into a pre-trained loss reason identification model, performing clustering operation on the behavior characteristic data according to the characteristic types of the loss reasons, and outputting probability values corresponding to a plurality of loss reasons; and determining the loss reason corresponding to the maximum probability value as a target loss reason of the user. Therefore, the method can accurately identify the potential churn users, and can accurately predict the churn reason of the potential churn users.

Description

Technical Field

[0001] The present application belongs to the technical field of machine learning, and in particular, relates to a method, device, equipment, storage medium and program for identifying causes of user churn. Background Art

[0002] During the operation of telecommunications operators, users may be lost due to dissatisfaction with tariffs, dissatisfaction with bills, or poor broadband quality.

[0003] In relevant user churn cause identification technologies, whether a user is a potential churn user is often judged based on a single model. For potential churn users, when identifying the churn cause of the user, the churn cause is often identified based on the user's abnormal behavior or current problems. The user's abnormal behavior or current problems are generally occasional reasons, not the real reasons for the user's churn. Therefore, the problem of inaccurate identified reasons for user churn may arise. Summary of the invention

[0004] The embodiments of the present application provide a method, apparatus, device, storage medium and program for identifying the causes of user churn, which can identify potential churn users based on a fusion identification model and a churn cause identification model, and accurately predict the target churn causes of the user.

[0005] In a first aspect, an embodiment of the present application provides a method for identifying causes of user churn, the method comprising:

[0006] Obtain user behavior characteristic data;

[0007] The behavior feature data is input into the pre-trained fusion recognition model, and the feature weight corresponding to each behavior feature is determined by the fusion recognition model based on the behavior feature data, and the mean of each feature weight is calculated to obtain the user's churn coefficient;

[0008] When the churn coefficient is less than or equal to the preset churn threshold, the user is determined to be a potential churn user;

[0009] Input the user's behavior feature data into the pre-trained churn reason identification model, perform clustering operations on the behavior feature data according to the feature types of the churn reasons, and output the probability values ​​corresponding to multiple churn reasons;

[0010] The churn reason corresponding to the largest probability value is determined as the target churn reason of the user.

[0011] Furthermore, before inputting the behavior feature data into the pre-trained fusion recognition model, the method further includes:

[0012] Using preset first training data, respectively training a preset proximity algorithm model, a random forest algorithm model, a naive Bayes algorithm model, an adaptive boosting algorithm model, and a logistic regression algorithm model, wherein the first training data includes first behavioral feature data of each of a plurality of users and a churn label indicating whether each user has churned;

[0013] The voting algorithm is used to fuse the trained proximity algorithm model, random forest algorithm model, naive Bayes algorithm model, adaptive boosting algorithm model and logistic regression algorithm model to obtain a fusion recognition model.

[0014] Furthermore, the trained proximity algorithm model, random forest algorithm model, naive Bayes algorithm model, adaptive enhancement algorithm model and logistic regression algorithm model are fused to obtain a fusion recognition model, including:

[0015] The fused model is used as the fusion recognition model to be trained;

[0016] Inputting the first behavior feature data of each of the multiple users into the fusion recognition model to be trained, outputting the first training weight of each first behavior feature data, calculating the mean of each first training weight, and obtaining the churn coefficient training value of each user;

[0017] By comparing the churn coefficient training value of each user with the churn threshold value, a training result for each user is obtained, and the training result includes whether the corresponding user is predicted to be a churned user or a non-churned user;

[0018] By comparing the training results and the churn labels, when the training results are inconsistent with the churn labels, the parameters of the fusion recognition model to be trained are adjusted, and the first behavior feature data of multiple users are input into the fusion recognition model to be trained for training until the training results output by the fusion recognition model to be trained are consistent with the churn labels, thereby obtaining a fusion recognition model that has completed training.

[0019] Furthermore, after obtaining the trained fusion recognition model, the method further includes:

[0020] Inputting the first training data into the trained fusion recognition model to obtain the churn coefficient of each user;

[0021] Determine the number of single-coefficient users corresponding to each churn coefficient, and determine the number of single-coefficient churn users with a churn label of churned from the users corresponding to each churn coefficient;

[0022] The ratio of the number of single coefficient churn users to the number of single coefficient users is used as the churn rate of the corresponding churn coefficient;

[0023] Based on each churn coefficient and the corresponding churn rate, a target churn coefficient corresponding to a churn rate reduction when the reduction amplitude is less than a preset amplitude is determined during the churn rate reduction process, and the churn threshold is determined using the target churn coefficient.

[0024] Furthermore, before inputting the user's behavior feature data into the pre-trained churn cause identification model, the method further includes:

[0025] Acquire second training data, the second training data including second behavior feature data of each of a plurality of lost users and a reason label indicating an actual loss reason of each user;

[0026] Inputting the second behavior feature data of each of the multiple lost users into the loss cause identification model to be trained, and outputting the predicted cause of each user;

[0027] By comparing each predicted reason and the reason label, when each predicted reason is inconsistent with the reason label, the parameters of the churn cause identification model to be trained are adjusted, and the second behavior feature data of multiple churned users are input into the churn cause identification model to be trained for training until the predicted reason output by the churn cause identification model to be trained is consistent with the reason label, thereby obtaining a completely trained churn cause identification model.

[0028] Furthermore, after determining the churn reason corresponding to the maximum probability value as the target churn reason of the user, the method further includes:

[0029] Select a retention behavior corresponding to the target churn reason from a preset retention strategy library, wherein the retention strategy library includes each churn reason and the corresponding retention behavior;

[0030] Perform retention actions.

[0031] In a second aspect, an embodiment of the present application provides a device for identifying causes of user churn, the device comprising:

[0032] An acquisition module is used to acquire user behavior feature data;

[0033] A churn coefficient calculation module, used to input the behavior feature data into a pre-trained fusion recognition model, determine the feature weight corresponding to each behavior feature based on the behavior feature data through the fusion recognition model, and calculate the mean of each feature weight to obtain the churn coefficient of the user;

[0034] A user determination module, configured to determine that the user is a potential churn user when the churn coefficient is less than or equal to a preset churn threshold;

[0035] An output module, used to input the user's behavior feature data into a pre-trained churn reason identification model, perform clustering operations on the behavior feature data according to the feature types of the churn reasons, and output probability values ​​corresponding to each of the multiple churn reasons;

[0036] The reason determination module is used to determine the churn reason corresponding to the maximum probability value as the target churn reason of the user.

[0037] In a third aspect, an embodiment of the present application provides a device, the device comprising:

[0038] a processor and a memory storing computer program instructions;

[0039] When the processor executes the computer program instructions, the method for identifying the cause of user churn as described in any of the above items is implemented.

[0040] In a fourth aspect, an embodiment of the present application provides a computer storage medium, on which computer program instructions are stored. When the computer program instructions are executed by a processor, a method for identifying the cause of user churn as described in any of the above items is implemented.

[0041] In a fifth aspect, an embodiment of the present application provides a method for identifying causes of user churn as described in any of the preceding items, wherein when instructions in a program product are executed by a processor of a device, the device executes the method.

[0042] The user churn cause identification method, apparatus, device, storage medium and program of the embodiments of the present application can predict, for any user, whether the user has a churn risk, that is, whether the user is a potential churn user, based on multiple behavioral feature data obtained for the user, and when the user has a churn risk, that is, when the user is judged to be a potential churn user, the churn cause of the user can be accurately predicted.

[0043] Specifically, after the behavioral feature data is input into the fusion recognition model, since the fusion recognition model integrates multiple algorithm models, the randomness in the prediction process can be reduced, so that the fusion recognition model can accurately output the user's churn coefficient. By comparing the churn coefficient with the churn threshold, it can be accurately determined whether the user is a potential churn user.

[0044] Furthermore, when any user is a potential churn user, after the behavioral feature data is input into the churn cause identification model for clustering operation, the accurate target churn cause can be predicted from multiple churn causes, thereby avoiding the error of judging the churn cause of the user by a single accidental cause. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] In order to more clearly illustrate the technical solution of the embodiments of the present application, the following is a brief introduction to the drawings required for use in the embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0046] Figure 1 It is a flowchart of a method for identifying causes of user churn provided in an embodiment of the present application;

[0047] Figure 2 It is a schematic diagram of the process of model fusion provided in the embodiment of the present application;

[0048] Figure 3 It is a schematic diagram of the relationship between the churn rate and the churn coefficient provided in the embodiment of the present application;

[0049] Figure 4 It is a flowchart of user identification and loss reasons provided by an embodiment of the present application;

[0050] Figure 5 It is a structural schematic diagram of a device for identifying causes of user churn provided in an embodiment of the present application;

[0051] Figure 6 It is a structural schematic diagram of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0052] The features and exemplary embodiments of various aspects of the present application will be described in detail below. In order to make the purpose, technical solutions and advantages of the present application clearer, the present application will be further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are only intended to explain the present application, rather than to limit the present application. For those skilled in the art, the present application can be implemented without the need for some of these specific details. The following description of the embodiments is only to provide a better understanding of the present application by illustrating the examples of the present application.

[0053] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the statement "include..." do not exclude the presence of other identical elements in the process, method, article or device including the elements.

[0054] As described in the background technology section, the related user churn cause identification technology is still difficult to meet the needs of actual work.

[0055] In the process of operation of telecommunications operators, user loss may occur due to various reasons, such as dissatisfaction with tariffs, dissatisfaction with bills, or poor broadband quality.

[0056] In related user churn reason identification technologies, when identifying the reasons for user churn, the reasons are often only identified based on the user's abnormal behavior or current problems. For example, if the user complains about poor broadband signal, the user is identified as likely to churn due to poor signal.

[0057] However, the user's abnormal behavior or current problems are generally incidental reasons and not the real reasons for the user's churn. For example, when a user complains about poor broadband signals, he or she will not churn due to poor broadband signals, but will churn due to the need to network with other family members.

[0058] Therefore, the current method of identifying the causes of user churn may result in inaccurately identified causes of user churn.

[0059] On the other hand, in the process of implementing the present application, it was found that when identifying whether existing users have a risk of churn, the judgment on whether the user has a risk of churn is often based on a single model. However, each single model has randomness in the operation process. That is to say, when using a single model to identify users, inaccurate identification results may be caused by accidental factors.

[0060] In order to solve the problems of the prior art, the embodiments of the present application provide a method, apparatus, device, storage medium and program for identifying the causes of user churn.

[0061] The following is a detailed description of the method for identifying the cause of user churn provided in the embodiment of the present application in conjunction with the accompanying drawings.

[0062] Figure 1 A flow chart of a method for identifying causes of user churn provided by an embodiment of the present application is shown.

[0063] refer to Figure 1 A method for identifying the cause of user churn in one embodiment of the present application specifically includes the following steps S101 to S105:

[0064] S101. Obtain user behavior feature data.

[0065] The user's behavior feature data includes multiple behavior features, which may specifically include data for describing multiple behaviors of the user when using the operator's services, and data for describing behaviors with loss risks indicated by the user.

[0066] In one example, for a user to be identified, the behavior characteristic data of the user may be a plurality of pre-set data or data extracted from the recorded historical behavior of the user.

[0067] S102: input the behavior feature data into a pre-trained fusion recognition model, determine the feature weight corresponding to each behavior feature based on the behavior feature data through the fusion recognition model, and calculate the mean of each feature weight to obtain the churn coefficient of the user.

[0068] Among them, the pre-trained fusion recognition model can be a model used to identify whether a user is a potential churn user.

[0069] In this embodiment, the behavior feature data obtained above can be input into the fusion recognition model. Specifically, the data in the behavior feature data used to describe multiple behaviors of the user when using the operator's service can be input into the fusion recognition model.

[0070] Furthermore, the fusion recognition model can be used to extract features from the behavior feature data input therein to obtain a plurality of corresponding behavior features.

[0071] Furthermore, based on the obtained multiple corresponding behavior features, a fusion recognition model can be used to identify the feature weights of each behavior feature.

[0072] Furthermore, based on the identified initial feature weights of the user, the values ​​of the initial feature weights may be standardized, wherein the standardization may specifically be, for example, normalization processing, so that the feature weights after normalizing the initial feature weights may be obtained.

[0073] Furthermore, the mean of each feature weight may be calculated, and the calculated mean of each feature weight may be determined as the churn coefficient of the user.

[0074] S103: When the churn coefficient is less than or equal to a preset churn threshold, determine that the user is a potential churn user.

[0075] In this embodiment, based on the churn coefficient calculated in the aforementioned steps, the churn coefficient may be compared with a preset churn threshold.

[0076] Further, based on the above comparison, if the churn coefficient of the user to be identified is less than or equal to the churn threshold, the user may be determined as a potential churn user, that is, the user has a churn risk.

[0077] In the embodiment of the present application, users with churn risk may be regarded as potential churn users.

[0078] S104: Input the user's behavior feature data into a pre-trained churn reason identification model, perform clustering operations on the behavior feature data according to the feature types of the churn reasons, and output probability values ​​corresponding to multiple churn reasons.

[0079] Among them, the churn reason identification model is used to identify the churn reasons that cause potential churn users to have churn risks.

[0080] In this embodiment, based on the judgment of the user to be identified in the aforementioned steps, when the user is judged to be a potential churn user, the behavioral characteristic data obtained above can be input into the churn cause identification model. Specifically, the data used to describe the behavior with churn risk shown by the user can be input into the churn cause identification model.

[0081] Furthermore, the churn cause identification model can be used to extract features from the behavioral feature data input therein to obtain a plurality of corresponding behavioral features.

[0082] There are multiple churn reasons preset, and each behavioral feature can be used as a feature type of a corresponding churn reason.

[0083] Furthermore, based on the obtained multiple corresponding feature types, the churn cause identification model can be used to perform clustering operations on each feature type input therein, and output the probability value of each churn cause.

[0084] S105: Determine the churn reason corresponding to the maximum probability value as the target churn reason of the user.

[0085] In this embodiment, based on the probability values ​​of the various churn reasons determined in the aforementioned steps, a churn reason with the largest probability value can be determined as the target churn reason that causes the potential churn user to churn, that is, the reason why the potential churn user has a churn risk.

[0086] Based on this, in this embodiment, for any user, it is possible to determine whether the user is a potential churn user by utilizing a fusion recognition model based on the multiple behavioral feature data of the user obtained. By utilizing the fusion recognition model, the randomness in the judgment process can be reduced, so that the fusion recognition model can accurately output the churn coefficient of the user, thereby accurately determining whether the user is a potential churn user.

[0087] When the user is a potential churn user, the behavioral feature data is input into the churn cause identification model for clustering operation. By setting multiple churn reasons, the accurate target churn cause can be predicted from the multiple churn reasons, thereby avoiding the error of judging the user's churn cause by a single accidental reason.

[0088] In another embodiment of the present application, for the user to be identified, in order to obtain the behavior characteristic data of the user, wherein the behavior characteristic data may include multiple first behavior characteristic data and multiple second behavior characteristic data, in order to obtain the multiple first behavior characteristic data and multiple second behavior characteristic data, the multiple first behavior data and multiple second behavior data of the user can be first obtained, and the multiple first behavior data and the multiple second behavior data can be preprocessed respectively, and the multiple first behavior data after the data preprocessing are used as the first behavior characteristic data, and the second behavior data after the data preprocessing are used as the second behavior characteristic data.

[0089] In one example, multiple first behavior data may be data used to describe multiple behaviors of the user when using the operator's services, including, for example: whether international roaming service is activated, whether international long-distance service is activated, whether broadband service is activated, whether converged group service is activated, whether home network service is available, traffic usage, voice usage, number of calls during the day, number of calls at night, duration of calls during the day, duration of calls at night, daytime traffic usage, nighttime traffic usage, daytime residence, nighttime residence, number of calls on different networks and duration of calls on different networks, etc.; in addition, it may also be multiple data used to describe relevant information of the user, including: network access time, user gender and user age, etc.

[0090] In one instance, multiple second behavior data can be data used to describe the user's behavior with loss risk, including, for example: whether the traffic usage has increased, whether the voice usage has increased, whether there are fee complaints, whether there are bill complaints, whether there are broadband complaints, whether there are television complaints, whether there are complaints about network signals, the number of dissatisfactions, the number of calls on different networks, whether there is broadband, whether there is a home network, etc.; in addition, it can also be multiple data used to describe relevant information of the user, including: network access time, user gender and user age, etc.

[0091] Among them, each second behavior data can correspondingly serve as a feature type of a loss reason leading to user loss, for example, an increase in traffic usage corresponds to the reason of insufficient traffic, an increase in voice usage corresponds to the reason of insufficient voice volume, fee complaints correspond to the reason of dissatisfaction with tariffs, bill complaints correspond to the reason of dissatisfaction with bills, broadband complaints correspond to the reason of poor broadband quality, TV complaints correspond to the reason of poor TV quality, complaints about network signals correspond to the reason of poor network signals, many times of dissatisfaction correspond to the reason of poor service quality, many times of calls on different networks correspond to the reason of demand for calls on different networks, the presence of broadband corresponds to the reason of demand for broadband, and the presence of a home network corresponds to the reason of demand for home networking.

[0092] In one example, when data preprocessing is performed on the plurality of first behavior data and the plurality of second behavior data determined in the aforementioned example, abnormal data therein may be removed first, and then redundant data therein may be removed.

[0093] The abnormal data may specifically include data with missing specific data values ​​and data with abnormal data values, and the redundant data may include data irrelevant to the user and data with low relevance to the user.

[0094] In one example, a data analysis tool, such as Statistical Package for the Social Sciences (SPSS) software, can be used to remove abnormal data such as missing data values ​​and abnormal data values ​​appearing in the plurality of first behavior data and the plurality of second behavior data, respectively.

[0095] Furthermore, redundant data that is irrelevant to the user or has low relevance to the user and appears in the multiple first behavior data and the multiple second behavior data can be eliminated by respectively calculating the Pearson correlation coefficient and significance level of each data in the multiple first behavior data and the multiple second behavior data.

[0096] In one example, multiple first behavior data after abnormal data and redundant data are removed are feature extracted and used as first behavior feature data; multiple second behavior data after abnormal data and redundant data are removed are feature extracted and used as second behavior feature data.

[0097] Based on this, in this embodiment, by obtaining the user's first behavior data and multiple second behavior data, the first behavior characteristic data and the second behavior characteristic data constructed based on the first behavior data and multiple second behavior data can more comprehensively characterize the user's situation when online.

[0098] Among them, by respectively eliminating abnormal data and redundant data in the first behavior data and multiple second behavior data, the obtained first behavior characteristic data and second behavior characteristic data can be made more reasonable, and when the first behavior characteristic data and the second behavior characteristic data are input into the model for processing, interference with the model is avoided.

[0099] In another embodiment of the present application, before the behavioral feature data is input into the pre-trained fusion recognition model for recognition, the first training data can be used to train multiple preset single models to be trained, and the trained single models can be fused to obtain a fusion recognition model.

[0100] The first training data includes first behavior feature data of each of the multiple users in the historical data. In addition, the first training data also includes churn labels, which include churned labels and non-churned labels indicating whether each user has churned.

[0101] Among them, the multiple single models can specifically be multiple independent models for identifying whether a user has a risk of churn, which can be, for example: a proximity algorithm model (K-Nearest Neighbours model, KNN model), a random forest algorithm model, a naive Bayes algorithm model, an adaptive boosting algorithm model (Adaboost model) and a logistic regression algorithm model, etc.

[0102] In a specific example, Figure 2 As shown, based on the above-mentioned preset first training data, the first training data is respectively input into the KNN model, the random forest algorithm model, the naive Bayes algorithm model, the Adaboost model and the logistic regression algorithm model, so as to train each model respectively and obtain the trained KNN model, the random forest algorithm model, the naive Bayes algorithm model, the Adaboost model and the logistic regression algorithm model.

[0103] Furthermore, a voting algorithm may be used to fuse the trained KNN model, random forest algorithm model, naive Bayes algorithm model, Adaboost model and logistic regression algorithm model to obtain a fusion recognition model.

[0104] Based on this, in this embodiment, a plurality of separate single models are fused by utilizing a voting algorithm, thereby reducing the randomness when using any single model alone to identify whether a user has a risk of churn, and making the recognition result output by the fused recognition model after fusing multiple single models more accurate.

[0105] On the other hand, using a fusion recognition model instead of a single model also avoids the problem of a large deviation between the recognition results of a single model and the actual situation.

[0106] In another embodiment of the present application, in the process of fusing the proximity algorithm model, the random forest algorithm model, the naive Bayes algorithm model, the adaptive enhancement algorithm model and the logistic regression algorithm model to obtain the fused recognition model, after fusing each single model, the fused model can be used as the fused recognition model to be trained, and the fused recognition model to be trained can be trained and optimized to obtain a completed fused recognition model.

[0107] In one example, the first training data obtained in the aforementioned embodiment can be input into the fusion recognition model to be trained, and for each user in the first training data, the fusion recognition model to be trained is used to output the first training weight corresponding to each first behavior feature data of the user.

[0108] Furthermore, each first training weight may be standardized so that the value of each first training weight is converted into a range greater than 0 and less than or equal to 1.

[0109] Specifically, for each user's multiple first training weights, each first training weight can be standardized using the following formula (1):

[0110]

[0111] Among them, A n represents any initial feature weight, A min represents the initial feature weight with the smallest value among all initial feature weights, A max represents the initial feature weight with the largest value among all initial feature weights, B n Indicates A n The corresponding feature weights after normalization.

[0112] Furthermore, based on the first training weights after the standardization process, the average of the first training weights after the standardization process can be calculated according to the following formula (2), and the average is used as the churn coefficient training value of the user:

[0113]

[0114] Wherein, n represents the number of the first training weights after normalization, and X represents the loss coefficient training value.

[0115] Based on this, after comparing the churn coefficient training value of each user with the churn threshold respectively, the users corresponding to the churn coefficient training value less than or equal to the churn threshold can be predicted as churned users, and the users corresponding to the churn coefficient training value greater than the churn threshold can be predicted as non-churned users, and the predicted results are used as training results.

[0116] Further, based on the churn labels in the first training data, the training results may be compared with the churn labels.

[0117] Among them, if the prediction result of any user in the training result is inconsistent with its actual churn label, it can be considered that the training of the fusion recognition model has not been completed, and the parameters of the fusion recognition model to be trained are adjusted.

[0118] Furthermore, the first training data is input into the adjusted fusion recognition model, and prediction is performed until the predicted training result is consistent with the loss label, and it can be considered that the fusion recognition model has been completely trained.

[0119] Based on this, in this embodiment, through the first behavior feature data of each user pre-set in the first training data and the corresponding churn labels of each user, a more comprehensive first training data can be used to train the fusion recognition model to be trained, and the churn labels can be used to verify whether the trained fusion recognition model meets the requirements, and when the training results are consistent with the churn labels, a fusion recognition model that has completed training is obtained.

[0120] In another embodiment of the present application, after obtaining the trained fusion recognition model, the churn coefficient of each user can be outputted based on the first training data using the trained fusion recognition model, and the churn rate of each churn coefficient can be calculated, thereby determining the target churn coefficient when the change in the churn rate begins to stabilize based on the change relationship between each churn coefficient and the corresponding churn rate, and using the target churn coefficient to adjust the churn threshold.

[0121] The churn rate is used to indicate the proportion of users who have actually churned among all users with the same churn rate.

[0122] In one example, after the first training data is input into the trained fusion recognition model, the trained fusion recognition model outputs a corresponding churn coefficient for each user.

[0123] Each churn coefficient may correspond to one or more users in the first training data, that is, there are one or more users with the same churn coefficient.

[0124] Furthermore, for each churn coefficient, the number of users corresponding to the churn coefficient is determined and used as the number of users with a single coefficient.

[0125] Furthermore, the churn labels in the first training data can be used to determine the number of users with churn labels among the users corresponding to the churn coefficient, and used as the number of churned users with a single coefficient, wherein the users with churn labels are the actual churned users.

[0126] Furthermore, for each churn coefficient, the ratio of the number of single-coefficient churn users corresponding to the churn coefficient to the number of single-coefficient users may be taken as the churn rate corresponding to the churn coefficient.

[0127] Based on this, the churn rate corresponding to each churn coefficient can be determined.

[0128] Furthermore, based on the determined churn rates corresponding to the various churn coefficients, a variation relationship between the churn coefficients and the churn rates may be constructed.

[0129] Furthermore, based on the changing relationship between the churn coefficient and the churn rate, the churn coefficient when the churn rate decreases to a stable level during the process of decreasing the churn rate can be determined.

[0130] The situation where the churn rate decreases to a stable state can be regarded as a situation where the churn rate decreases by less than a preset amount during the process of decreasing.

[0131] Further, based on the churn rate when the reduction amplitude is less than a preset amplitude, a churn coefficient corresponding to the churn rate is used as a target churn coefficient.

[0132] Furthermore, the obtained target churn coefficient is used to replace the current churn threshold, thereby adjusting the current churn threshold.

[0133] In a specific example, the first training data includes, for example, behavioral feature data of 1,000 users and their respective churn labels.

[0134] After the first training data is input into the trained fusion recognition model, the trained fusion recognition model outputs multiple loss coefficients, wherein each loss coefficient may be, for example, 0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8 and 0.9.

[0135] Furthermore, when the fusion recognition model predicts that the churn coefficient of 100 users is 0.4, the number of single-coefficient users with a churn coefficient of 0.4 is 100, of which 20 users have churn labels, that is, the number of single-coefficient churn users is 20.

[0136] Furthermore, when calculating the churn rate of a churn coefficient of 0.4, the ratio of 20 single-coefficient churn users to 100 single-coefficient users (20%) can be used as the churn rate of a churn coefficient of 0.4, that is, the ratio of 20 to 100 (20%) can be used as the churn rate of a churn coefficient of 0.4.

[0137] Based on this, the churn rates of 0.1, 0.2, 0.3, 0.5, 0.6, 0.7, 0.8 and 0.9 can be determined respectively.

[0138] Furthermore, based on the determined churn coefficients and their corresponding churn rates, the following can be constructed: Figure 3 The changing relationship between the churn coefficient and the churn rate is shown.

[0139] Among them, Figure 3 As shown, as the churn rate gradually decreases, the churn coefficient gradually increases. That is to say, there is a negative correlation between the churn coefficient and the churn rate. The larger the churn rate, the smaller the churn coefficient, and the smaller the churn rate, the larger the churn coefficient.

[0140] It can be seen that in the process of gradually decreasing churn rate, when the churn rate drops to 10%, it begins to stabilize, and when the churn rate drops to 10%, the churn coefficient is 0.5, then 0.5 is the target churn coefficient.

[0141] Further, the determined target churn coefficient of 0.5 is used as the churn threshold, or replaces the current preset churn threshold, thereby adjusting the current churn threshold.

[0142] Based on this, in this embodiment, based on the churn coefficients output by the trained fusion recognition model, by calculating the number of single-coefficient users corresponding to each churn coefficient and the number of single-coefficient churn users therein, the churn rate corresponding to each churn coefficient can be determined by using the proportion of single-coefficient churn users in single-coefficient users, thereby establishing a changing relationship between the churn coefficient and the churn rate.

[0143] Furthermore, the target churn coefficient when the churn rate tends to be stable can be determined. After adjusting the churn threshold using the target churn coefficient, the adjusted churn threshold can accurately represent the dividing point between a sharp drop in churn rate and a stable churn rate. That is to say, when the churn coefficient exceeds the churn threshold, the churn rate changes stably and small. When the churn coefficient is less than or equal to the churn threshold, the churn rate is high and the changes are unstable.

[0144] In another embodiment of the present application, based on the churn threshold preset in the aforementioned embodiment, after the first behavioral feature data of the user to be identified is input into the trained fusion recognition model and the fusion recognition model outputs the churn coefficient of the user, the churn coefficient of the user can be compared with the churn threshold. When the churn coefficient is less than or equal to the churn threshold, the user has a churn risk, and when the churn coefficient is greater than the churn threshold, the user does not have a churn risk.

[0145] The preset churn threshold may be, for example, a churn threshold set according to demand, or may be, for example, the middle number or median of a churn coefficient, or may be, for example, an adjusted churn threshold in the above-described embodiment.

[0146] In one example, for any user to be identified, after the churn coefficient of the user is obtained using the trained fusion recognition model, the churn coefficient of the user can be compared with the churn threshold because the churn threshold can represent the dividing point between a sharp drop in churn rate and a stable churn rate.

[0147] Furthermore, by comparing the churn coefficient with the churn threshold, when the churn coefficient of the user is less than or equal to the churn threshold, it can be considered that the churn probability of the user is large, and the churn probability is not stable. The churn probability will still increase significantly due to a slight change in the churn coefficient. Based on this, the user can be determined to have a churn risk.

[0148] When the churn coefficient of the user is greater than the churn threshold, it can be considered that the churn probability of the user is small and stable, and the churn probability will not increase significantly due to a slight change in the churn coefficient. Based on this, the user can be determined to have no churn risk.

[0149] Based on this, in this embodiment, by comparing the churn coefficient and churn rate of the user to be identified, it is possible to directly determine that the user has a churn risk when the churn coefficient is less than or equal to the churn threshold.

[0150] In another embodiment of the present application, before inputting the user's behavioral characteristic data into the pre-trained churn cause identification model, the churn cause identification model to be trained can be trained using preset second training data to obtain a trained churn cause identification model.

[0151] The second training data includes second behavior feature data of multiple lost users in historical data. In addition, the second training data also includes reason labels, which include lost labels and non-lost labels indicating whether each user has lost.

[0152] In one example, the preset second training data may be input into the churn cause identification model to be trained, and for each user in the second training data, the churn cause identification model to be trained is used to predict the probability value of each churn cause.

[0153] Furthermore, for each user, the churn reason with the largest probability value among the churn reasons corresponding to the user is used as the predicted reason for the churn of the user.

[0154] Furthermore, based on the reason labels in the second training data, the predicted reasons of each user may be compared with the reason labels.

[0155] When the predicted reason of any user is inconsistent with its actual reason label, it can be considered that the training of the churn reason identification model has not been completed, and the parameters of the churn reason identification model to be trained are adjusted.

[0156] Furthermore, the second training data is input into the adjusted churn cause identification model, and prediction is performed until the predicted cause is consistent with the cause label, and it can be considered that the trained churn cause identification model is obtained.

[0157] Based on this, in this embodiment, by pre-setting the second behavioral feature data of each churned user in the second training data and the corresponding reason labels of each user, a more comprehensive second training data can be used to train the churn cause identification model to be trained, and the reason labels can be used to verify whether the trained churn cause identification model meets the requirements, and when the predicted cause is consistent with the reason label, a trained churn cause identification model is obtained.

[0158] In another embodiment of the present application, for users with churn risk, after determining the target churn reason of the user, the retention strategy library can be queried, and the corresponding retention behavior can be selected and executed from the retention strategy library, thereby retaining the identified users.

[0159] The retention strategy library includes various churn reasons and the retention behavior corresponding to each churn reason. Each retention behavior is used to retain users according to the corresponding churn reason.

[0160] In one example, for a user identified as having churn risk, after predicting a target churn reason that causes the user to have churn risk, a churn reason that is consistent with the target churn reason may be queried from a retention strategy library.

[0161] Further, after determining the churn cause that is consistent with the target churn cause, a retention behavior corresponding to the churn cause may be determined, and the retention behavior may be executed.

[0162] In a specific example, the retention strategy library may include a corresponding relationship table as shown in Table 1 below.

[0163] Table 1. Correspondence table.

[0164] Causes of loss Retention Behavior Dissatisfied with the tariff Appropriate discounts and offers on rates Insufficient voice volume Free trial voice pack Long-distance call requirements Notify the number voice traffic nationwide Not satisfied with the bill Add SMS bill push and fee deduction reminders, and make appropriate compensation for phone charges Poor broadband quality Free on-site inspection and maintenance, low-cost replacement of new equipment Poor network signal Improve network quality and add regional network equipment Poor service quality Provide SMS care, improve services in a timely manner, and give star-level service privileges Home networking requirements Push discounts on family packages

[0165] As can be seen from Table 1, the reasons for loss in the retention strategy library specifically include: dissatisfaction with tariffs, insufficient voice volume, need for long-distance calls, dissatisfaction with bills, poor broadband quality, poor network signals, poor service quality and home networking needs.

[0166] Among them, corresponding to each reason for loss, targeted retention behavior is set.

[0167] Based on this, when the predicted reason for target loss is insufficient voice volume, the corresponding retention behavior can be determined: giving away a trial voice package.

[0168] Furthermore, in order to retain the user, a retention action of giving the user a trial voice package may be performed.

[0169] Based on this, in this embodiment, based on the preset retention strategy library, it is possible to determine accurate retention behavior when predicting the target loss reason. By executing the retention behavior, it is possible to accurately retain the identified users with loss risk.

[0170] In another embodiment of the present application, Figure 4 The overall process of retaining any user is shown, which specifically includes steps S401 to S406.

[0171] Among them, Figure 4 As shown, for any user to be identified, S401 may be executed to input a fusion identification model.

[0172] In this step, a plurality of first behavior feature data of the user may be input into a fusion recognition model, so as to identify the churn coefficient of the user using the fusion recognition model.

[0173] Further, execute S402 to determine the loss coefficient.

[0174] In this step, the preset loss threshold is 0.5, and the loss threshold is used to compare with the loss coefficient obtained in S401.

[0175] Based on this, when it is determined that the churn coefficient of the user is greater than 0.5, it is determined that the user does not have a churn risk, and S403 is further executed without identifying the churn reason, thereby completing the identification of the user.

[0176] When it is determined that the churn coefficient of the user is less than or equal to 0.5, it is determined that the user has a churn risk, and S404, target churn reason identification, is further performed.

[0177] In this step, the second behavior characteristic data of the user may be input into a churn cause identification model, and the churn cause identification model may be used to predict the target churn cause of the user.

[0178] Further, after predicting the target loss reason, S405 is executed to match the retention behavior.

[0179] In this step, a retention behavior corresponding to the target loss reason is selected from a preset retention strategy library.

[0180] Further, after the retention behavior is matched, S406 is executed to execute the retention behavior.

[0181] In this step, the retention behavior determined in S405 is accurately performed on the user, thereby completing the retention of the user.

[0182] Based on this, in this embodiment, for any user, after inputting multiple first behavioral feature data of the user into the fusion recognition model, the churn coefficient of the user can be identified, so that it can be achieved to accurately judge whether the user has a churn risk based on the comparison between the churn coefficient and the churn threshold.

[0183] Furthermore, when the user is at risk of churn, by inputting the user's multiple second behavioral feature data into the churn cause identification model for prediction, the accurate target churn cause can be predicted for the user, thereby accurately locating the retention behavior used to retain the user from the retention strategy library, and accurately retaining the user by executing the retention behavior on the user.

[0184] Based on the same inventive concept, corresponding to any of the above-mentioned embodiments and methods, an embodiment of the present application further provides a device for identifying causes of user churn.

[0185] refer to Figure 5 The user loss cause identification device comprises:

[0186] Acquisition module 501, used to acquire user behavior feature data;

[0187] The churn coefficient calculation module 502 is used to input the behavior feature data into the pre-trained fusion recognition model, determine the feature weight corresponding to each behavior feature based on the behavior feature data through the fusion recognition model, and calculate the mean of each feature weight to obtain the user's churn coefficient;

[0188] A user determination module 503, configured to determine that a user is a potential churn user when the churn coefficient is less than or equal to a preset churn threshold;

[0189] Output module 504, used to input the user's behavior feature data into the pre-trained churn reason identification model, perform clustering operation on the behavior feature data according to the feature type of the churn reason, and output the probability value corresponding to each of the multiple churn reasons;

[0190] The reason determination module 505 is used to determine the churn reason corresponding to the maximum probability value as the target churn reason of the user.

[0191] In one embodiment, before inputting the behavior feature data into the pre-trained fusion recognition model, the churn coefficient calculation module 502 is further used to:

[0192] Using preset first training data, respectively training a preset proximity algorithm model, a random forest algorithm model, a naive Bayes algorithm model, an adaptive boosting algorithm model, and a logistic regression algorithm model, wherein the first training data includes first behavioral feature data of each of a plurality of users and a churn label indicating whether each user has churned;

[0193] The voting algorithm is used to fuse the trained proximity algorithm model, random forest algorithm model, naive Bayes algorithm model, adaptive boosting algorithm model and logistic regression algorithm model to obtain a fusion recognition model.

[0194] In one embodiment, the churn coefficient calculation module 502 fuses the trained proximity algorithm model, random forest algorithm model, naive Bayes algorithm model, adaptive boosting algorithm model and logistic regression algorithm model to obtain a fusion recognition model, including:

[0195] The fused model is used as the fusion recognition model to be trained;

[0196] Inputting the first behavior feature data of each of the multiple users into the fusion recognition model to be trained, outputting the first training weight of each first behavior feature data, calculating the mean of each first training weight, and obtaining the churn coefficient training value of each user;

[0197] By comparing the churn coefficient training value of each user with the churn threshold value, a training result for each user is obtained, and the training result includes whether the corresponding user is predicted to be a churned user or a non-churned user;

[0198] By comparing the training results and the churn labels, when the training results are inconsistent with the churn labels, the parameters of the fusion recognition model to be trained are adjusted, and the first behavior feature data of multiple users are input into the fusion recognition model to be trained for training until the training results output by the fusion recognition model to be trained are consistent with the churn labels, thereby obtaining a fusion recognition model that has completed training.

[0199] In one embodiment, after obtaining the trained fusion recognition model, the loss coefficient calculation module 502 is further used to:

[0200] Inputting the first training data into the trained fusion recognition model to obtain the churn coefficient of each user;

[0201] Determine the number of single-coefficient users corresponding to each churn coefficient, and determine the number of single-coefficient churn users with a churn label of churned from the users corresponding to each churn coefficient;

[0202] The ratio of the number of single coefficient churn users to the number of single coefficient users is used as the churn rate of the corresponding churn coefficient;

[0203] Based on each churn coefficient and the corresponding churn rate, a target churn coefficient corresponding to a churn rate reduction when the reduction amplitude is less than a preset amplitude is determined during the churn rate reduction process, and the churn threshold is determined using the target churn coefficient.

[0204] In one embodiment, before inputting the user's behavior feature data into the pre-trained churn cause identification model, the output module 504 is further used to:

[0205] Acquire second training data, the second training data including second behavior feature data of each of a plurality of lost users and a reason label indicating an actual loss reason of each user;

[0206] Inputting the second behavior feature data of each of the multiple lost users into the loss cause identification model to be trained, and outputting the predicted cause of each user;

[0207] By comparing each predicted reason and the reason label, when each predicted reason is inconsistent with the reason label, the parameters of the churn cause identification model to be trained are adjusted, and the second behavior feature data of multiple churned users are input into the churn cause identification model to be trained for training until the predicted reason output by the churn cause identification model to be trained is consistent with the reason label, thereby obtaining a completely trained churn cause identification model.

[0208] In one embodiment, after determining the churn cause corresponding to the maximum probability value as the target churn cause of the user, the cause determination module 505 is further configured to:

[0209] Select a retention behavior corresponding to the target churn reason from a preset retention strategy library, wherein the retention strategy library includes each churn reason and the corresponding retention behavior;

[0210] Perform retention actions.

[0211] For the convenience of description, the above devices are described in terms of functions and are divided into various modules. Of course, when implementing the embodiments of the present application, the functions of each module can be implemented in the same or multiple software and / or hardware.

[0212] The device of the above embodiment is used to implement the corresponding user churn cause identification method in any of the above embodiments, and has the beneficial effects of the corresponding method embodiment, which will not be described in detail here.

[0213] Based on the same inventive concept, corresponding to any of the above-mentioned embodiments and methods, an embodiment of the present application also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, a method for identifying the cause of user churn as in any of the above-mentioned embodiments is implemented.

[0214] Figure 6 A schematic diagram of the hardware structure of an electronic device provided in an embodiment of the present application is shown.

[0215] The electronic device may include a processor 601 and a memory 602 storing computer program instructions.

[0216] Specifically, the processor 601 may include a central processing unit (CPU), or an application specific integrated circuit (ASIC), or may be configured to implement one or more integrated circuits of the embodiments of the present application.

[0217] The memory 602 may include a large capacity memory for data or instructions. By way of example and not limitation, the memory 602 may include a hard disk drive (HDD), a floppy disk drive, a flash memory, an optical disk, a magneto-optical disk, a magnetic tape, or a universal serial bus (USB) drive or a combination of two or more of these. Where appropriate, the memory 602 may include a removable or non-removable (or fixed) medium. Where appropriate, the memory 602 may be inside or outside the electronic device. In a particular embodiment, the memory 602 is a non-volatile solid-state memory.

[0218] The memory 602 may include a read-only memory (ROM), a random access memory (RAM), a magnetic disk storage medium device, an optical storage medium device, a flash memory device, an electrical, optical or other physical / tangible memory storage device. Therefore, in general, the memory includes one or more tangible (non-transitory) computer-readable storage media (e.g., memory devices) encoded with software including computer-executable instructions, and when the software is executed (e.g., by one or more processors), it is operable to perform the operations described with reference to the method according to an aspect of the present disclosure.

[0219] The processor 601 implements any one of the methods for identifying the cause of user churn in the above embodiments by reading and executing the computer program instructions stored in the memory 602 .

[0220] In one example, the electronic device may further include a communication interface 603 and a bus 610. Figure 6 As shown, the processor 601, the memory 602, and the communication interface 603 are connected via a bus 610 and communicate with each other.

[0221] The communication interface 603 is mainly used to implement communication between various modules, devices, units and / or equipment in the embodiments of the present application.

[0222] Bus 610 includes hardware, software or both, and the parts of electronic equipment are coupled to each other. For example, but not limitation, bus may include accelerated graphics port (Accelerated Graphics Port, AGP) or other graphics bus, enhanced industry standard architecture (Extended Industry Standard Architecture, EISA) bus, front-side bus (FSB), hypertransmission (Hyper Transport, HT) interconnection, industry standard architecture (Industry Standard Architecture, ISA) bus, infinite bandwidth interconnection, low pin count (LPC) bus, memory bus, micro channel architecture (MCA) bus, peripheral component interconnection (PCI) bus, PCI-Express (PCI-X) bus, serial advanced technology attachment (SATA) bus, video electronics standard association local (VLB) bus or other suitable bus or two or more of these combinations. In appropriate cases, bus 610 may include one or more buses. Although the present application embodiment describes and shows a specific bus, the application considers any suitable bus or interconnection.

[0223] The electronic device can execute the user churn cause identification method in the embodiment of the present application based on the fusion identification model and the user churn cause identification model, thereby realizing the combination of Figure 1 and Figure 4 Describe the method for identifying the causes of user churn.

[0224] In addition, in combination with the user churn cause identification method in the above embodiment, the present application embodiment can provide a computer storage medium for implementation. The computer storage medium stores computer program instructions; when the computer program instructions are executed by a processor, any one of the user churn cause identification methods in the above embodiment is implemented.

[0225] An embodiment of the present application also provides a computer program product, including a computer program, which, when executed, implements any one of the user churn cause identification methods in the above embodiments.

[0226] It should be clear that the present application is not limited to the specific configuration and processing described above and shown in the figures. For the sake of simplicity, a detailed description of the known method is omitted here. In the above embodiments, several specific steps are described and shown as examples. However, the method process of the present application is not limited to the specific steps described and shown, and those skilled in the art can make various changes, modifications and additions, or change the order between the steps after understanding the spirit of the present application.

[0227] The functional blocks shown in the above-described block diagram can be implemented as hardware, software, firmware or a combination thereof. When implemented in hardware, it can be, for example, an electronic circuit, an application specific integrated circuit (ASIC), appropriate firmware, a plug-in, a function card, etc. When implemented in software, the elements of the present application are programs or code segments that are used to perform the required tasks. The program or code segment can be stored in a machine-readable medium, or transmitted on a transmission medium or a communication link by a data signal carried in a carrier wave. "Machine-readable medium" can include any medium capable of storing or transmitting information. Examples of machine-readable media include electronic circuits, semiconductor memory devices, ROM, flash memory, erasable ROM (EROM), floppy disks, CD-ROMs, optical disks, hard disks, optical fiber media, radio frequency (RF) links, etc. The code segment can be downloaded via a computer network such as the Internet, an intranet, etc.

[0228] It should also be noted that the exemplary embodiments mentioned in this application describe some methods or systems based on a series of steps or devices. However, this application is not limited to the order of the above steps, that is, the steps can be performed in the order mentioned in the embodiment, or in a different order from the embodiment, or several steps can be performed simultaneously.

[0229] Aspects of the present disclosure are described above with reference to the flowchart and / or block diagram of the method, device (system) and computer program product according to the embodiment of the present disclosure. It should be understood that each box in the flowchart and / or block diagram and the combination of each box in the flowchart and / or block diagram can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device to produce a machine so that these instructions executed by the processor of the computer or other programmable data processing device enable the implementation of the function / action specified in one or more boxes of the flowchart and / or block diagram. Such a processor can be, but is not limited to, a general-purpose processor, a special-purpose processor, a special application processor, or a field programmable logic circuit. It can also be understood that each box in the block diagram and / or flowchart and the combination of boxes in the block diagram and / or flowchart can also be implemented by dedicated hardware that performs a specified function or action, or can be implemented by a combination of dedicated hardware and computer instructions.

[0230] The above is only a specific implementation of the present application. Those skilled in the art can clearly understand that for the convenience and simplicity of description, the specific working processes of the systems, modules and units described above can refer to the corresponding processes in the aforementioned method embodiments, and will not be repeated here. It should be understood that the protection scope of the present application is not limited to this. Any technician familiar with the technical field can easily think of various equivalent modifications or replacements within the technical scope disclosed in this application, and these modifications or replacements should be included in the protection scope of this application.

Claims

1. A method for identifying the cause of user churn, characterized in that: include: Obtain user behavior characteristic data; The behavior feature data is input into a pre-trained fusion recognition model, and the feature weight corresponding to each behavior feature is determined by the fusion recognition model based on the behavior feature data, and the mean of each feature weight is calculated to obtain the churn coefficient of the user; When the churn coefficient is less than or equal to a preset churn threshold, determining the user as a potential churn user; Inputting the user's behavior feature data into a pre-trained churn reason identification model, performing clustering operations on the behavior feature data according to the feature types of the churn reasons, and outputting probability values ​​corresponding to each of the multiple churn reasons; The churn reason corresponding to the maximum probability value is determined as the target churn reason of the user.

2. The method for identifying the cause of user loss according to claim 1, characterized in that: Before inputting the behavior feature data into the pre-trained fusion recognition model, the method further includes: Using preset first training data, respectively training a preset proximity algorithm model, a random forest algorithm model, a naive Bayes algorithm model, an adaptive boosting algorithm model, and a logistic regression algorithm model, wherein the first training data includes first behavioral feature data of a plurality of users and a churn label indicating whether each user has churned; A voting algorithm is used to fuse the trained proximity algorithm model, random forest algorithm model, naive Bayes algorithm model, adaptive enhancement algorithm model and logistic regression algorithm model to obtain the fusion recognition model.

3. The method for identifying the cause of user loss according to claim 2, characterized in that: The training completed proximity algorithm model, random forest algorithm model, naive Bayes algorithm model, adaptive enhancement algorithm model and logistic regression algorithm model are integrated to obtain the integrated recognition model, including: The fused model is used as the fusion recognition model to be trained; Inputting the first behavior feature data of each of the multiple users into the fusion recognition model to be trained, outputting the first training weight of each first behavior feature data, calculating the mean of each first training weight, and obtaining the churn coefficient training value of each user; By comparing the churn coefficient training value of each user with the churn threshold value, a training result for each user is obtained, wherein the training result includes whether the corresponding user is predicted to be a churned user or a non-churned user; By comparing the training results and the churn labels, when the training results are inconsistent with the churn labels, the parameters of the fusion recognition model to be trained are adjusted, and the first behavior feature data of multiple users are input into the fusion recognition model to be trained for training until the training results output by the fusion recognition model to be trained are consistent with the churn labels, thereby obtaining a fusion recognition model that has completed training.

4. The method for identifying the cause of user loss according to claim 3, characterized in that: After obtaining the trained fusion recognition model, the method further includes: Inputting the first training data into the trained fusion recognition model to obtain the churn coefficient of each user; Determine the number of single-coefficient users corresponding to each churn coefficient, and determine the number of single-coefficient churn users with a churn label of churned from the users corresponding to each churn coefficient; Using the ratio of the number of single coefficient churn users to the number of single coefficient users as the churn rate of the corresponding churn coefficient; Based on each churn coefficient and the corresponding churn rate, a target churn coefficient corresponding to a churn rate reduction when the reduction amplitude is less than a preset amplitude is determined during the churn rate reduction process, and the churn threshold is determined using the target churn coefficient.

5. The method for identifying the cause of user loss according to claim 1, characterized in that: Before inputting the user's behavior feature data into the pre-trained churn cause identification model, the method further includes: Acquire second training data, wherein the second training data includes second behavior feature data of each of a plurality of lost users and a reason label indicating an actual loss reason of each user; Inputting the second behavior feature data of each of the multiple lost users into the loss cause identification model to be trained, and outputting the predicted cause of each user; By comparing each predicted reason with the reason label, when each predicted reason is inconsistent with the reason label, the parameters of the churn cause identification model to be trained are adjusted, and the second behavior feature data of multiple churned users are input into the churn cause identification model to be trained for training until the predicted reason output by the churn cause identification model to be trained is consistent with the reason label, thereby obtaining a completely trained churn cause identification model.

6. The method for identifying the cause of user loss according to claim 1, characterized in that: After determining the churn reason corresponding to the maximum probability value as the target churn reason of the user, the method further includes: Selecting a retention behavior corresponding to the target churn reason from a preset retention strategy library, wherein the retention strategy library includes each churn reason and the corresponding retention behavior; Execute the retention behavior.

7. A device for identifying the cause of user loss, characterized in that: The device comprises: An acquisition module is used to acquire user behavior feature data; A churn coefficient calculation module, used to input the behavior feature data into a pre-trained fusion recognition model, determine the feature weight corresponding to each behavior feature based on the behavior feature data through the fusion recognition model, and calculate the mean of each feature weight to obtain the churn coefficient of the user; A user determination module, configured to determine that the user is a potential churn user when the churn coefficient is less than or equal to a preset churn threshold; An output module, used to input the user's behavior feature data into a pre-trained churn reason identification model, perform clustering operations on the behavior feature data according to the feature types of the churn reasons, and output probability values ​​corresponding to each of the multiple churn reasons; The reason determination module is used to determine the churn reason corresponding to the maximum probability value as the target churn reason of the user.

8. A device, characterized in that The device comprises: a processor and a memory storing computer program instructions; When the processor executes the computer program instructions, the method for identifying the cause of user churn as described in any one of claims 1-6 is implemented.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer program instructions, and when the computer program instructions are executed by a processor, the method for identifying the cause of user churn according to any one of claims 1 to 6 is implemented.

10. A program product, characterized in that When the instructions in the program product are executed by a processor of a device, the device executes the method for identifying causes of user churn as described in any one of claims 1 to 6.

Images