Off-grid prediction method and device based on user clustering, equipment and storage medium
By segmenting users and using machine learning algorithms to build a churn prediction model, the problem of operators being unable to accurately predict user churn has been solved, improving the efficiency and accuracy of customer retention.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- CHINA UNITED NETWORK COMM GRP CO LTD
- Filing Date
- 2021-12-15
- Publication Date
- 2026-06-05
AI Technical Summary
Operators lack understanding of the current lifecycle of their overall customer base, leading to customer churn. Existing technologies lack efficient solutions for predicting and analyzing customer churn.
By creating an churn prediction model, users are grouped, and feature selection and clustering are performed using random forest, Adaboost, and XGBoost algorithms to build an churn prediction model that predicts the churn trend of users in each group.
It enables efficient and accurate analysis of user churn trends, providing operators with sufficient time to conduct customer retention activities and reduce customer loss.
Smart Images
Figure CN116266219B_ABST
Abstract
Description
Technical Field
[0001] This disclosure relates to the field of mobile communication technology, and in particular to a user-segmentation-based churn prediction method, a user-segmentation-based churn prediction device, a terminal device, and a computer-readable storage medium. Background Technology
[0002] With the rapid development of mobile communication services, major operators face a fiercely competitive environment under the national policy of speeding up network speeds and reducing fees. Users' demands on operators are also increasing. When users are dissatisfied with their current tariffs and services, they will choose to switch operators, resulting in churn. For telecom operators, user churn causes significant losses to their business. However, most operators lack understanding of the overall customer lifecycle, and their customer service teams are slow to recognize customers with poor product experiences, missing the ideal window for recovery and ultimately leading to customer loss. Therefore, effectively predicting user churn tendencies has become a key factor for operators in maintaining their customer base. Summary of the Invention
[0003] This disclosure provides a method, apparatus, terminal device, and computer-readable storage medium for predicting churn based on user segmentation, in order to at least address the problem of current operators not understanding the overall customer lifecycle status, leading to customer churn.
[0004] To address the aforementioned objectives, this disclosure provides a user segmentation-based churn prediction method, comprising:
[0005] Create an off-grid prediction model;
[0006] Users are segmented to obtain user segmentation results; and,
[0007] Based on the churn prediction model, the churn trend of users in each segment of the user segmentation results is predicted.
[0008] In one implementation, creating an off-grid prediction model includes:
[0009] Collect sample data;
[0010] The sample data is divided into training set data and validation set data;
[0011] A grid search combination of several parameters of the model is generated based on the training set data;
[0012] Based on the validation set data, cross-validation is performed on the grid search combination of several parameters to obtain the validation results;
[0013] Based on the verification results, the optimal parameter grid search combination is selected; and an off-grid prediction model is established based on the optimal parameter grid search combination.
[0014] In one implementation, user segmentation includes:
[0015] Divide user data into data for each user across several dimensions;
[0016] Feature filtering is performed on the data of each user across several dimensions to obtain the feature data of each user across several dimensions; and,
[0017] Users are grouped based on their characteristic data across several dimensions.
[0018] In one implementation, before segmenting user data into data for each user across several dimensions, the method further includes:
[0019] Obtain raw user data; and preprocess the raw user data to obtain user data.
[0020] In one implementation, feature filtering is performed on the data of each user across several dimensions, including:
[0021] Standardize the data for each user across several dimensions; and,
[0022] Feature filtering was performed on the standardized data of each user across several dimensions.
[0023] In one implementation, the step of performing feature filtering on the standardized data of each user across several dimensions includes:
[0024] The random forest algorithm, Adaboost algorithm, and XGBoost algorithm, or any combination thereof, are used to perform feature filtering on the standardized data of each user across several dimensions.
[0025] In one implementation, users are grouped based on feature data of each user across several dimensions, including:
[0026] Divide users into several categories; and,
[0027] Users are grouped according to the categories based on their feature data across several dimensions.
[0028] In one implementation, several categories are defined regarding user groups, including:
[0029] Perform KMeans clustering on the data of each user across several dimensions to obtain the clustering results;
[0030] The silhouette coefficient method is used to determine the number of categories in each of the several dimensions based on the clustering results; and,
[0031] By randomly combining the number of categories in each dimension, several categories related to user segmentation are obtained.
[0032] To address the aforementioned issues, this disclosure also provides an churn prediction device based on user segmentation, comprising:
[0033] The model building module is set up to create an off-grid prediction model.
[0034] The user segmentation module is configured to segment users into groups and obtain the user segmentation results; and...
[0035] The prediction module is configured to predict the churn trend of users in each segment of the user segmentation results based on the churn prediction model.
[0036] To address the aforementioned issues, this disclosure also provides a terminal device, including a memory and a processor. The memory stores a computer program, and when the processor runs the computer program stored in the memory, the processor executes the aforementioned user-segmentation-based churn prediction method.
[0037] To address the aforementioned problems, this disclosure also provides a computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the processor performs the aforementioned user-segmentation-based churn prediction method.
[0038] According to the user segmentation-based churn prediction method, apparatus, terminal equipment, and computer-readable storage medium provided in this disclosure, a churn prediction model is created, users are segmented to obtain user segmentation results, and finally, the churn trend of users in each segment is predicted based on the churn prediction model. This disclosure, by segmenting users and then predicting churn for each user segment based on the created churn prediction model, can efficiently and accurately analyze the churn trend of operator users, allowing sufficient time for user retention activities, helping marketers develop feasible customer retention plans, effectively avoiding customer churn, and has broad industry application prospects.
[0039] Other features and advantages of this disclosure will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the disclosure. The objects and other advantages of this disclosure may be realized and obtained by means of the structures particularly pointed out in the description, claims and drawings. Attached Figure Description
[0040] The accompanying drawings are provided to further understand the technical solutions of this disclosure and constitute a part of the specification. They are used together with the embodiments of this disclosure to explain the technical solutions of this disclosure and do not constitute a limitation on the technical solutions of this disclosure.
[0041] Figure 1 A flowchart illustrating an churn prediction method based on user segmentation provided in this embodiment of the disclosure;
[0042] Figure 2 for Figure 1 A flowchart illustrating step S101;
[0043] Figure 3 A flowchart illustrating another user-segmentation-based churn prediction method provided in this disclosure embodiment;
[0044] Figure 4 This is an example diagram illustrating the classification of user groups in this embodiment of the disclosure;
[0045] Figure 5 A schematic diagram of the structure of an off-grid prediction device based on user segmentation provided in an embodiment of this disclosure;
[0046] Figure 6 This is a schematic diagram of the structure of a terminal device provided in an embodiment of this disclosure. Detailed Implementation
[0047] To make the objectives, technical solutions, and advantages of the embodiments of this disclosure clearer, the specific implementation methods of this disclosure will be described in detail below with reference to the accompanying drawings. It should be understood that the specific implementation methods described herein are for illustration and explanation only and are not intended to limit this disclosure.
[0048] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this disclosure are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence; furthermore, in the absence of conflict, the embodiments and features in the embodiments of this disclosure can be arbitrarily combined with each other.
[0049] In the following description, the use of suffixes such as “module,” “part,” or “unit” to denote elements is solely for the purpose of illustrative purposes and has no specific meaning in itself. Therefore, “module,” “part,” or “unit” may be used interchangeably.
[0050] To address the aforementioned issues, this disclosure utilizes user lifecycle data to construct a customer segmentation and grading system tailored to specific business needs. It predicts churn based on user segments, allowing sufficient time for customer retention activities and providing a wealth of valuable information to help marketers develop feasible customer retention strategies. This approach has broad industry application prospects.
[0051] Please refer to Figure 1 , Figure 1 This is a flowchart illustrating an off-network prediction method based on user segmentation provided in an embodiment of the present disclosure, including steps S101-S103.
[0052] In step S101, an off-grid prediction model is created.
[0053] In one implementation, an churn prediction model that integrates multiple prediction algorithms, such as random forest, Adaboost, and XGBoost, can be established by deploying multiple prediction algorithms in the model to predict whether a user has a churn trend in a classification manner.
[0054] Furthermore, this embodiment establishes an off-grid prediction model by training optimal parameters, making the prediction results of the off-grid prediction model more accurate, such as... Figure 2 As shown, step S101 creates the off-grid prediction model, including steps S101a-S101f:
[0055] In S101a, sample data is collected.
[0056] Specifically, massive amounts of operator data can be collected as sample data, including operator data with diverse characteristics such as different user genders, age groups, years of use, and consumption levels.
[0057] In S101b, the sample data is divided into training set data and validation set data.
[0058] The training set data is used to train the model parameters, and the validation set data is used to verify the model parameters trained on the training set, in order to improve the accuracy of the model and establish the optimal model. The model parameters are the parameter grid search combination.
[0059] In some embodiments, after feature filtering of the sample data as described in steps S102a-S102c of the following embodiments, the training set data and the validation set data are then divided to improve the accuracy of the data.
[0060] In S101c, a grid search combination of several parameters of the model is generated based on the training set data.
[0061] Specifically, a grid search is used to iterate through the features and prediction results of each user in the training set data, where the prediction results include both offline and online outcomes. In essence, the grid search runs through all the parameters used to obtain all grid search combinations of parameters, where the parameters are the user features in the sample data. Furthermore, random forest, Adaboost, and XGBoost algorithms can be used to obtain the parameters that need to be adjusted.
[0062] In S101d, cross-validation is performed on the several parameter grid search combinations based on the validation set data to obtain the validation results.
[0063] Specifically, cross-validation calculates the prediction errors of the validation set data samples, records their squared sums, and continues this process until all samples have been predicted once. The squared prediction errors of each sample are then summed to obtain the validation results.
[0064] In S101e, the optimal parameter grid search combination is selected based on the verification results; and in S101f, an off-grid prediction model is established based on the optimal parameter grid search combination.
[0065] It is understandable that the parameter grid search combination that minimizes the forecast error is the optimal parameter grid search combination.
[0066] In step S102, users are grouped to obtain user grouping results.
[0067] To facilitate the prediction of churn among a large number of operator customers, this embodiment segments users and then predicts churn for each user segment. Specifically, user segmentation can be achieved based on the full lifecycle data of each user, where the full lifecycle data refers to all user data generated by the operator. The following embodiments of this disclosure provide specific examples of user segmentation, which will not be repeated here.
[0068] In step S103, the churn trend of users in each group in the user grouping results is predicted based on the churn prediction model.
[0069] Compared to related technologies, which lack solutions for predicting and analyzing user churn, making it impossible to grasp user churn trends, or which directly analyze churn data from a massive number of users, resulting in a complex and inefficient churn prediction process, this embodiment achieves hierarchical churn prediction by grouping users, making the churn prediction process simpler and more efficient.
[0070] Please refer to Figure 3 , Figure 3 This embodiment of the present disclosure provides a user segmentation-based churn prediction method. Based on the previous embodiment, this embodiment illustrates a specific user segmentation method. Step S102 specifically includes the following steps S102a-S102c.
[0071] In step S103a, the user data is divided into data for each user in several dimensions.
[0072] In this embodiment, user data is divided into three dimensions: basic information, usage, and cost, as shown in Table 1 below:
[0073]
[0074]
[0075] Table 1
[0076] It should be noted that the above-mentioned dimension field data are only examples, and this disclosure is not limited to the above data.
[0077] In one implementation, to improve the integrity and accuracy of user data, the following steps are included before step S103a:
[0078] Obtain raw user data; and preprocess the raw user data to obtain user data.
[0079] The original user data refers to the raw data collected from the operator's backend without any processing. Preprocessing can include data cleaning, filtering, and imputation operations. For example, firstly, all field data is filtered out from the original user data. All field data is traversed, and fields with more than 50% missing values are deleted. The remaining field data is imputed with mean, median, or mode. Character variables can be converted into numerical variables using one-hot encoding. Variables with a correlation coefficient exceeding 0.8 are filtered and deleted to make the final user data more complete and accurate.
[0080] In step S103b, feature filtering is performed on the data of each user in several dimensions to obtain the feature data of each user in several dimensions.
[0081] In this embodiment, feature filtering is performed based on the importance of data in each dimension. The feature variables for each dimension after feature filtering are shown below, with variable importance from left to right, as shown in Table 2 below:
[0082]
[0083] Table 2
[0084] Furthermore, this embodiment first standardizes the data and then performs feature filtering to eliminate the influence of dimensions and improve the accuracy of feature filtering. Feature filtering is performed on the data of each user in several dimensions, including the following steps:
[0085] The data of each user in several dimensions is standardized; and the standardized data of each user in several dimensions is then used for feature filtering.
[0086] Specifically, Z-score standardization is performed on the data for each dimension to eliminate the influence of units, and then feature selection is performed. Furthermore, any one or any combination of the following algorithms can be used to perform feature selection on the standardized data of each user in several dimensions: Random Forest; Adaboost algorithm; XGBoost algorithm.
[0087] This embodiment takes the use of random forest, Adaboost, and XGBoost algorithms on three standardized dimensions as an example. The feature selection for each dimension is performed by calculating the mean. The average value of the coefficients of several important factors of the algorithm is taken as the feature importance. Features with a weight greater than 80% can be taken as user grouping feature data.
[0088] In step S103c, users are grouped based on their feature data in several dimensions.
[0089] In one implementation, users are grouped based on feature data of each user across several dimensions, including the following steps:
[0090] Divide users into several categories; and,
[0091] Users are grouped according to the categories based on their feature data across several dimensions.
[0092] For example, users can be categorized into eight groups: low spending, low frequency of voice usage, high activity; low spending, high frequency of voice usage, low activity; low spending, high frequency of voice usage, high activity; high spending, low frequency of voice usage, low activity; high spending, low frequency of voice usage, high activity; high spending, high frequency of voice usage, low activity; and high spending, high frequency of voice usage, high activity. After categorizing, each user can be further subdivided into these eight groups based on their characteristic data across various dimensions.
[0093] In one implementation, classifying users into several categories includes the following steps:
[0094] Perform KMeans clustering on the data of each user across several dimensions to obtain the clustering results;
[0095] The silhouette coefficient method is used to determine the number of categories in each of the several dimensions based on the clustering results; and,
[0096] By randomly combining the number of categories in each dimension, several categories related to user segmentation are obtained.
[0097] Specifically, in this embodiment, KMeans clustering is performed on the feature data (factors) of the three dimensions respectively. The silhouette coefficient method is used to find the optimal number of user groups that can be divided into in each dimension, thereby predicting the number of categories for each user in the three dimensions. Finally, the user groups in each dimension are cross-permuted and combined to find the total category to which each user belongs, thus completing the user grouping.
[0098] In this embodiment, KMeans clustering is performed on the cost dimension, usage dimension, and basic information dimension respectively. The number of categories with the largest silhouette coefficient among 2 to 10 categories is selected as the optimal number of categories for each dimension. It is understood that the silhouette coefficient is a way to evaluate the quality of clustering. This embodiment calculates the silhouette coefficient of each feature vector in the clustering results to determine the number of categories for each dimension.
[0099] For example, calculations show that the optimal number of categories for the cost dimension, usage dimension, and basic information dimension is 2: the cost dimension is divided into categories 0 and 1, with category 0 representing low consumption and category 1 representing high consumption. To more intuitively illustrate the differences between users across the three dimensions, this embodiment uses PCA (Principal Component Analysis) to reduce the feature factors of each dimension to two dimensions, depicting the differences between each category of users, thus more clearly showing the differences between user groups.
[0100] Furthermore, based on the classification results of customer behavior across the three dimensions, a random permutation and combination method is used to randomly combine the categories under the three dimensions, resulting in 2*2*2, a total of 8 different user categories. The specific combination process is combined with... Figure 4 As shown in Table 3 below, this general category can be used to distinguish all users. Users within each general category have different characteristics in terms of cost, usage, and basic information, but users within the same general category share the same characteristics. This lays the foundation for user marketing and churn prediction after segmentation.
[0101]
[0102]
[0103] Table 3
[0104] In practical applications, combined with soft voting, it can be concluded that the method of using the churn prediction model in this embodiment to predict the churn trend of users in user groups is more conducive to churn early warning analysis for each type of user group. Referring to Table 4 below, the model performance before and after user grouping is compared, and the F1 score of the user churn model before and after grouping is calculated.
[0105]
[0106] Table 4
[0107] The F1 score for predicted churn tendency of categorized users was 74.9%, an improvement of 2.3% compared to the F1 score of uncategorized users (72.6%). This indicates that the categorized user churn prediction is effective and can improve the accuracy of prediction. In particular, for users in categories 1 and 7, the prediction after categorization is more conducive to predicting user churn trends, indicating that the behavioral characteristics of these user categories have a more significant impact on user churn.
[0108] Based on the same technical concept, this disclosure also provides an churn prediction device based on user segmentation, such as... Figure 5 As shown, the device includes a model building module 51, a clustering module 52, and a prediction module 53, wherein,
[0109] Model building module 51 is configured to create an off-grid prediction model;
[0110] The user segmentation module 52 is configured to segment users into groups and obtain user segmentation results; and...
[0111] The prediction module 53 is configured to predict the churn trend of users in each segment of the user segmentation results based on the churn prediction model.
[0112] In one implementation, the model building module 51 includes:
[0113] The data acquisition unit is configured to collect sample data.
[0114] The first partitioning unit is configured to divide the sample data into training set data and validation set data;
[0115] The generation unit is configured to generate a grid search combination of several parameters of the model based on the training set data;
[0116] A verification unit is configured to perform cross-validation on the grid search combination of several parameters based on the verification set data to obtain a verification result.
[0117] Select a modeling unit, which is configured to select the optimal parameter grid search combination based on the verification results; and establish an off-grid prediction model based on the optimal parameter grid search combination.
[0118] In one embodiment, the grouping module 52 includes:
[0119] The second partitioning unit is configured to divide user data into data for each user across several dimensions.
[0120] A filtering unit is configured to perform feature filtering on data for each user across several dimensions, thereby obtaining feature data for each user across several dimensions; and,
[0121] The segmentation unit is configured to segment users based on their feature data across several dimensions.
[0122] In one embodiment, the clustering module 52 further includes:
[0123] An acquisition unit is configured to acquire raw user data; and a preprocessing unit is configured to preprocess the raw user data to obtain user data.
[0124] In one implementation, the filtering unit is specifically configured to: standardize the data of each user in several dimensions; and perform feature filtering on the standardized data of each user in several dimensions.
[0125] In one implementation, the filtering unit uses any one or any combination of the following algorithms to perform feature filtering on the standardized data of each user across several dimensions: Random Forest; Adaboost algorithm; XGBoost algorithm.
[0126] In one implementation, the grouping unit is specifically configured to divide users into several categories; and to group users according to the categories based on the feature data of each user in several dimensions.
[0127] In one embodiment, the second partitioning unit includes:
[0128] The clustering subunit is configured to perform KMeans clustering on data for each user across several dimensions to obtain the clustering results.
[0129] The sub-units are divided by using the silhouette coefficient method to divide the number of categories in each of the several dimensions based on the clustering results; and,
[0130] The combination subunit is configured to randomly combine the number of categories in each dimension to obtain several categories related to user groups.
[0131] Based on the same technical concept, this disclosure also provides a terminal device, such as... Figure 6 As shown, the terminal device includes a memory 61 and a processor 62. The memory 61 stores a computer program. When the processor 62 runs the computer program stored in the memory 61, the processor 62 executes the user-segmentation-based churn prediction method.
[0132] Based on the same technical concept, embodiments of this disclosure also provide a computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the processor executes the aforementioned user-segmentation-based churn prediction method.
[0133] It will be understood by those skilled in the art that all or some of the steps, systems, or apparatuses disclosed above, and their functional modules / units, can be implemented as software, firmware, hardware, or suitable combinations thereof. In hardware implementations, the division between functional modules / units mentioned in the above description does not necessarily correspond to the division of physical components; for example, a physical component may have multiple functions, or a function or step may be performed collaboratively by several physical components. Some or all physical components may be implemented as software executed by a processor, such as a central processing unit, digital signal processor, or microprocessor, or as hardware, or as an integrated circuit, such as an application-specific integrated circuit (ASIC). Such software may be distributed on a computer-readable medium, which may include computer storage media (or non-transitory media) and communication media (or transient media). As is known to those skilled in the art, the term computer storage media includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information (such as computer-readable instructions, data structures, program modules, or other data). Computer storage media include, but are not limited to, RAM, ROM, EEPROM, flash memory or other memory technologies, CD-ROM, digital versatile disc (DVD) or other optical disc storage, magnetic cartridges, magnetic tape, disk storage or other magnetic storage devices, or any other medium that can be used to store desired information and can be accessed by a computer. Furthermore, it is well known to those skilled in the art that communication media typically contain computer-readable instructions, data structures, program modules, or other data in modulated data signals such as carrier waves or other transmission mechanisms, and may include any information delivery medium.
[0134] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this disclosure, and are not intended to limit them. Although this disclosure has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this disclosure.
Claims
1. A method for predicting churn based on user segmentation, characterized in that, include: Create an off-grid prediction model; Users are segmented to obtain user segmentation results; and, Based on the aforementioned churn prediction model, the churn trend of users in each of the user segmentation results is predicted respectively. The process of segmenting users to obtain user segmentation results includes: User data is divided into data for each user across several dimensions, including basic information, usage, and cost. Feature filtering is performed on the data of each user in several dimensions to obtain the feature data of each user in several dimensions. Perform KMeans clustering on the data of each user across several dimensions to obtain the clustering results; The silhouette coefficient method is used to classify the number of categories in each of the several dimensions based on the clustering results; The number of categories in each dimension is randomly combined to obtain several categories of user groups. These categories include low-spending, low-frequency voice usage, high-activity users; low-spending, high-frequency voice usage, low-activity users; low-spending, high-frequency voice usage, high-activity users; high-spending, low-frequency voice usage, low-activity users; high-spending, low-frequency voice usage, high-activity users; and high-spending, high-frequency voice usage, high-activity users. Based on the characteristic data of each user in several dimensions, users are grouped according to the categories to obtain user grouping results.
2. The method according to claim 1, characterized in that, Creating an off-grid prediction model includes: Collect sample data; The sample data is divided into training set data and validation set data; A grid search combination of several parameters of the model is generated based on the training set data; Based on the validation set data, cross-validation is performed on the grid search combination of several parameters to obtain the validation results; Based on the verification results, the optimal parameter grid search combination is selected; and an off-grid prediction model is established based on the optimal parameter grid search combination.
3. The method according to claim 1, characterized in that, Before dividing user data into data for each user across several dimensions, the following is also included: Obtain raw user data; and preprocess the raw user data to obtain user data.
4. The method according to claim 1, characterized in that, For each user, feature filtering is performed on data across several dimensions, including: Standardize the data for each user across several dimensions; and, Feature filtering was performed on the standardized data of each user across several dimensions.
5. The method according to claim 4, characterized in that, The step of performing feature filtering on each user's standardized data across several dimensions includes: The random forest algorithm, Adaboost algorithm, and XGBoost algorithm, or any combination thereof, are used to perform feature filtering on the standardized data of each user across several dimensions.
6. An churn prediction device based on user segmentation, characterized in that, include: The model building module is set up to create an off-grid prediction model. The user segmentation module is configured to segment users into groups and obtain the user segmentation results; and... The prediction module is configured to predict the churn trend of users in each segment of the user segmentation results based on the churn prediction model. The clustering module is specifically configured as follows: The process of segmenting users to obtain user segmentation results includes: User data is divided into data for each user across several dimensions, including basic information, usage, and cost. Feature filtering is performed on the data of each user in several dimensions to obtain the feature data of each user in several dimensions. Perform KMeans clustering on the data of each user across several dimensions to obtain the clustering results; The silhouette coefficient method is used to classify the number of categories in each of the several dimensions based on the clustering results; The number of categories in each dimension is randomly combined to obtain several categories of user groups. These categories include low-spending, low-frequency voice usage, high-activity users; low-spending, high-frequency voice usage, low-activity users; low-spending, high-frequency voice usage, high-activity users; high-spending, low-frequency voice usage, low-activity users; high-spending, low-frequency voice usage, high-activity users; and high-spending, high-frequency voice usage, high-activity users. Based on the characteristic data of each user in several dimensions, users are grouped according to the categories to obtain user grouping results.
7. A terminal device, characterized in that, It includes a memory and a processor, wherein the memory stores a computer program, and when the processor runs the computer program stored in the memory, the processor executes the user-segmentation-based churn prediction method according to any one of claims 1 to 5.
8. A computer-readable storage medium, characterized in that, It stores a computer program, which, when executed by a processor, performs an off-network prediction method based on user grouping according to any one of claims 1 to 5.