User stability analysis method and system
By using the Self-Organizing Map (SOM) algorithm to cluster and analyze the differences in features of user data, the problem of inaccurate user stability analysis caused by individual feature differences is solved, and a more accurate user stability assessment is achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- CHINA MOBILE FINANCIAL TECHNOLOGY CO LTD
- Filing Date
- 2021-05-06
- Publication Date
- 2026-06-05
Smart Images
Figure CN115392328B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of data mining technology, and in particular to a user stability analysis method and system. Background Technology
[0002] User stability is a key indicator across many industries. For example, in the financial industry, higher user stability generally indicates a stronger repayment ability.
[0003] Existing user stability analysis methods mainly include:
[0004] Cluster-based user identification and marketing methods: These methods aim to classify customer groups using unsupervised machine learning algorithms such as K-Means, analyze user behavior characteristics based on the clustering results, and customize user operation and marketing strategies.
[0005] Choking probability analysis based on machine learning models: Choking probability analysis based on machine learning models aims to predict the likelihood of users churning based on user behavior by using machine learning models and collected user on-network and churn data.
[0006] K-Means is an unsupervised clustering algorithm. Given a sample set, it divides the set into K clusters based on the distance between samples, aiming to make points within a cluster as close together as possible while maximizing the distance between clusters. Its main drawback is that, for the obtained cluster centers, it merges the closest clusters based on the distances between the k clusters, thus reducing the number of cluster centers and consequently the number of clusters. Furthermore, if the data is imbalanced, the clustering effect will be poor.
[0007] Churn probability analysis based on machine learning models uses data mining algorithms to predict user churn, and then implements customer retention strategies and care programs for users with churn tendencies. Based on similar individual characteristics and the reasoning ability of machine learning models, it determines the user's key characteristics. Its main drawback is that it is susceptible to problems with unusual datasets or unstable individual characteristics, potentially leading to significant biases in the clustering results.
[0008] Given the inherent instability and seasonality of user communication data, relying solely on individual characteristics to assess stability (including churn probability) can lead to significant biases. Therefore, addressing the issues of poor clustering results caused by imbalanced individual data or unstable individual characteristics, and the resulting biases in user analysis due to poor clustering, are urgent problems that need to be solved. Summary of the Invention
[0009] The user stability analysis method and system provided by this invention are used to solve the above-mentioned problems existing in the prior art, and can solve the problems of poor clustering effect caused by individual characteristic differences and inaccurate user stability analysis caused by poor clustering effect.
[0010] This invention provides a user stability analysis method, comprising:
[0011] Cluster the full set of users at different time intervals after the update to obtain the optimal pair with the highest similarity between cluster categories of adjacent different time intervals;
[0012] Based on the optimal pairing with the highest similarity, a user group that meets the preset conditions is determined, and the user group is added to the target user group to update the target user group;
[0013] Users who meet the preset conditions are removed to update all users in different time series until the number of all users in different time series after the update is less than a preset threshold, at which point the update stops.
[0014] The updated target user group is divided into an observation group and an experimental group, and a feature difference analysis is performed on the observation group and the experimental group to analyze the stability of the target user group.
[0015] The preset conditions include users who appear in all time series of the preset observation period in the optimal pairing.
[0016] According to a user stability analysis method provided by the present invention, the step of clustering all users at different updated time series to obtain the optimal pairing with the highest similarity between cluster categories of adjacent different time series includes:
[0017] The self-organizing map (SOM) clustering algorithm is used to cluster all users at different time series after the update, and the cluster categories of all users at different time series after the update are obtained.
[0018] Based on the preset category vector generation rules and the clustering categories, obtain the category vectors of clustering categories at different time sequences;
[0019] Based on the category vectors of the clustering categories at different time sequences, the similarity of the clustering categories at different time sequences is obtained, and based on the similarity of the clustering categories at different time sequences, the optimal pair with the highest similarity between adjacent clustering categories at different time sequences is determined.
[0020] According to a user stability analysis method provided by the present invention, the step of obtaining category vectors of cluster categories at different time intervals based on preset category vector generation rules and the cluster categories includes:
[0021] The first user belonging to the first category in the cluster is labeled with a first preset number;
[0022] Other users in the cluster that do not belong to the first category are labeled with a second preset number;
[0023] Based on the annotation results, obtain the category vectors of the clustering categories at different time sequences.
[0024] According to a user stability analysis method provided by the present invention, obtaining the similarity of cluster categories in different time periods based on the category vectors of the cluster categories in different time periods includes:
[0025] Based on the category vectors of the cluster categories at different time intervals and the total number of users, the similarity of the cluster categories at different time intervals is obtained; or
[0026] The similarity of the clustering categories in different time periods is obtained based on the category vectors of the clustering categories in different time periods, the total number of users, and the preset weights.
[0027] According to a user stability analysis method provided by the present invention, the step of dividing the updated target user group into an observation group and an experimental group includes:
[0028] Based on the target stability analysis requirements, each user in the target user group is classified and labeled.
[0029] Users belonging to the first preset category are defined as the experimental group.
[0030] Users belonging to the second preset category are defined as the observation group.
[0031] According to a user stability analysis method provided by the present invention, the step of performing characteristic difference analysis on the observation group and the experimental group to analyze the stability of the target user group includes:
[0032] The training set samples and training labels are input into a preset binary classification model for training to obtain a target prediction model;
[0033] The test set samples are input into the target prediction model to predict the stability of the target user group;
[0034] The training set samples include the observation group and the experimental group, which are in a first preset proportion.
[0035] The test set sample includes the observation group and the experimental group in a second preset ratio;
[0036] The training labels are determined by classifying and labeling each user in the training set samples according to the target stability analysis requirements.
[0037] According to a user stability analysis method provided by the present invention, the step of performing characteristic difference analysis on the observation group and the experimental group to analyze the stability of the target user group further includes:
[0038] The differential integrated moving average autoregressive (ARIMA) algorithm is used to analyze the characteristic differences between the observation group and the experimental group to analyze the stability of the target user group.
[0039] The present invention also provides a user stability analysis system, comprising: a data acquisition module, a first update module, a second update module, and a user analysis module;
[0040] The data acquisition module is used to cluster all users in different time series after the update, so as to obtain the optimal pair with the highest similarity between cluster categories of adjacent different time series.
[0041] The first update module is used to determine a user group that meets preset conditions based on the optimal pairing with the highest similarity, and add the user group to the target user group to update the target user group;
[0042] The second update module is used to remove user groups that meet the preset conditions to update all users in different time series until the number of all users in different time series after the update is less than a preset threshold, and then stop updating.
[0043] The user analysis module is used to divide the updated target user group into an observation group and an experimental group, and to perform feature difference analysis on the observation group and the experimental group to analyze the stability of the target user group.
[0044] The preset conditions include users who appear in all time series of the preset observation period in the optimal pairing.
[0045] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of any of the user stability analysis methods described above.
[0046] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of any of the user stability analysis methods described above.
[0047] The user stability analysis method and system provided by this invention clusters users in different time periods and identifies the customer group based on the data with the highest similarity. This addresses the problems of poor clustering results due to individual characteristic differences and inaccurate user stability analysis caused by poor clustering results. Attached Figure Description
[0048] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.
[0049] Figure 1 This is a flowchart illustrating the user stability analysis method provided by the present invention;
[0050] Figure 2 This is a schematic diagram of clustering and grouping provided by the present invention;
[0051] Figure 3 This is a schematic diagram of the user stability analysis system provided by the present invention;
[0052] Figure 4 This is a schematic diagram of the physical structure of the electronic device provided by the present invention. Detailed Implementation
[0053] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.
[0054] Figure 1 This is a flowchart illustrating the user stability analysis method provided by the present invention, as shown below. Figure 1 As shown, the method includes:
[0055] S1. Cluster all users in different time series after the update to obtain the optimal pair with the highest similarity between cluster categories of adjacent different time series.
[0056] S2. Based on the best pairing with the highest similarity, determine the user group that meets the preset conditions, and add the user group to the target user group to update the target user group;
[0057] S3. Remove users who meet the preset conditions to update all users in different time periods until the number of all users in different time periods after the update is less than the preset threshold, then stop updating.
[0058] S4. Divide the updated target user group into an observation group and an experimental group, and conduct a feature difference analysis between the observation group and the experimental group to analyze the stability of the target user group.
[0059] The preset conditions include users who appear in all time series of the preset observation period in the optimal pairing.
[0060] It should be noted that the above method can be implemented by computer equipment.
[0061] Optionally, the user stability analysis method provided by the present invention includes clustering users within different billing periods, selecting the categories with the highest similarity for secondary delineation, labeling users in all secondary delineated users as observation group and experimental group, and performing inter-group characteristic difference analysis to achieve stability analysis of the target user group. The specific implementation is as follows:
[0062] Optionally, the user stability analysis method provided by this invention can analyze stability indicators under different business scenarios, such as user stability analysis under scenarios like consumption downgrade, churn, consumption upgrade, and new service promotion. Specifically, it can analyze all users at different time periods, such as all mobile users with and without consumption downgrades, all users churned and active, all users with and without consumption upgrades, and all users who have subscribed to new services and those who haven't, to achieve user stability analysis under scenarios like consumption downgrade, churn, consumption upgrade, and new service promotion.
[0063] Clustering calculations are performed on user data with different billing periods. Based on specific needs, clusters belonging to specific user groups are selected. Then, within the clusters of different billing periods, the category containing the specific user group with the highest cluster similarity is found. Specifically, for example... Figure 2 As shown:
[0064] Performance Period: Cluster analysis is performed on all users across different time periods (billing period 1 to billing period n) after the update. Based on the similarity of cluster categories across different time periods, the customer groups with the closest similarity (the optimal pairing with the highest similarity) are identified. The user data for different time periods is obtained by dividing user data for different billing periods into time-series segments, such as by "month" or "quarter".
[0065] Observation period: Users who appear in all time series corresponding to the optimal pair with the highest similarity among cluster categories in adjacent time series within the preset observation period are formed into a user group, and this user group is added to the target user group to update the target user group. Then, this user group is removed from the full user list in different time series to update the full user list in different time series.
[0066] Repeat the above process to update the target user group and the total number of users at different time intervals until the number of total users at different time intervals after the update is less than the preset threshold. Stop updating the total number of users at different time intervals and obtain the updated target user group based on the last update result.
[0067] The updated target user group is divided into an experimental group and an observation group. Behavioral characteristics of the two groups are analyzed separately. For example, the analysis results can be comprehensively evaluated and operational observation indicators can be set to achieve stability analysis of the target user group.
[0068] The user stability analysis method provided by this invention clusters users in different time periods and identifies the customer group based on the data with the highest similarity. This addresses the problems of poor clustering results due to individual characteristic differences and inaccurate user stability analysis caused by poor clustering results.
[0069] Furthermore, in one embodiment, step S1 may specifically include:
[0070] S11. Based on the self-organizing map (SOM) clustering algorithm, cluster the full users in different time series after the update to obtain the clustering categories of the full users in different time series after the update.
[0071] S12. Based on the preset category vector generation rules and clustering categories, obtain the category vectors of clustering categories at different time series;
[0072] S13. Based on the category vectors of cluster categories in different time series, obtain the similarity of cluster categories in different time series, and determine the optimal pair with the highest similarity between adjacent cluster categories in different time series based on the similarity of cluster categories in different time series.
[0073] Optionally, the Self-Organizing Maps (SOM) algorithm is used to cluster the full users of different time series after the update, so as to obtain the cluster categories of the full users of different time series after the update. The SOM algorithm has low requirements for the selection of the initial centroid and does not need to consider the number of clusters of the full users of each time series in advance.
[0074] After clustering all users at different time series, a category vector is generated based on the user groups within the cluster category of all users at each time series and the preset category vector generation rules.
[0075] Based on the category vectors of the cluster categories obtained at different time intervals, the similarity of the cluster categories at different time intervals is obtained, and based on the similarity of the cluster categories at different time intervals, the optimal pair with the highest similarity between adjacent cluster categories at different time intervals is determined.
[0076] For example, the clustering category vector corresponding to the clustering categories of all users from time series 1 to time series n is x. 1i ~x ni , where x 1i x represents the category vector corresponding to time series 1. ni Let represent the category vector corresponding to time series n, and let represent the i-th user in each time series. Then, the similarity of cluster categories in different time series is calculated, and the optimal pair of users with the highest similarity among cluster categories of adjacent different time series is obtained.
[0077] The user stability analysis method provided by this invention uses clustering algorithms such as SOM to perform clustering calculations on data from different time series. Based on specific needs, it selects the clusters in which specific user groups belong and finds the category containing the specific user group with the highest cluster similarity among the clusters from different time series. This solves the problems of poor clustering effect and large subsequent bias caused by individual data differences and imbalance.
[0078] Furthermore, in one embodiment, step S12 may specifically include:
[0079] S121. Label the first user belonging to the first category in the cluster with a first preset number;
[0080] S122. Label other users in the cluster category who do not belong to the first category with the second preset number;
[0081] S123. Based on the annotation results, obtain the category vectors of clustering categories at different time series.
[0082] Optionally, after clustering is completed, a category vector is generated based on the user groups within the cluster category. The first user in the cluster category belonging to the first category is labeled with a first preset number, such as the number "1", and other users in the cluster category not belonging to the first category are labeled with a second preset number, such as the number "0".
[0083] For example, when labeling users in the cluster categories of time series 1, if the i-th user in time series 1 belongs to the first category, the label value corresponding to the i-th user is "1"; otherwise, if the i-th user in time series 1 does not belong to the first category, the label value corresponding to the i-th user is "0". Based on this, the labeling of each user in time series 1 is completed.
[0084] For example, if we arbitrarily select 10 users as the analysis subjects, and assume that only the 1st, 2nd, and 7th users fall into the first category, then the category vector for that category would be:
[0085] The user stability analysis method provided by this invention lays the foundation for obtaining the similarity of clustering categories at different time series by acquiring the category vectors of users at different time series.
[0086] Furthermore, in one embodiment, obtaining the similarity of clustering categories at different time sequences based on the category vectors of clustering categories at different time sequences in step S13 may specifically include:
[0087] S131. Based on the category vectors of cluster categories and the total number of users at different time intervals, obtain the similarity of cluster categories at different time intervals; or
[0088] S132. Based on the category vectors, total number of users, and preset weights of the cluster categories in different time series, obtain the similarity of the cluster categories in different time series.
[0089] Optionally, based on the results of clustering the full user base at different time series, the similarity S of the cluster categories at different time series can be cross-analyzed:
[0090]
[0091] Where n is the total number of selected users, These represent the category vectors of all users in different time series. When two category vectors are exactly the same, the similarity approaches infinity. When two categories are completely different (without any shared users), the similarity is 0.
[0092] To address different target stability analysis needs, the similarity WS of cluster categories at different time periods can be obtained based on the following formula, using preset weights, category vectors of cluster categories at different time periods, and the selected total number of users:
[0093]
[0094] Among them, w k Represents n i Weight allocation for samples belonging to category k.
[0095] The user stability analysis method provided by this invention calculates the similarity of all users at different time series after clustering, and uses an algorithm to delineate user data according to different characteristics, accurately aggregating data with high similarity and defining user groups with more commonalities.
[0096] Furthermore, in one embodiment, step S4, which divides the updated target user group into an observation group and an experimental group, may specifically include:
[0097] S41. Based on the target stability analysis requirements, classify and label each user in the target user group.
[0098] S42. Users belonging to the first preset category are the experimental group customers;
[0099] S43. Users belonging to the second preset category are designated as the observation group.
[0100] Optionally, based on the target stability analysis requirements, each user in the target user group is classified and labeled. For example, to analyze the stability of users leaving the network, each user in the obtained target user group is classified and labeled. Specifically, users belonging to the first preset category (e.g., leaving the network) are labeled as "1" and used as the experimental group, while users belonging to the second preset category (e.g., online) are labeled as "0" and used as the observation group.
[0101] It should be noted that the first and second preset categories can be freely set according to different stability analysis needs. For example, when performing stability analysis on user consumption downgrade, the first preset category is set to consumption downgrade, and the second preset category is set to consumption not downgraded; as another example, when performing stability analysis on user consumption upgrade, the first preset category is set to consumption upgrade, and the second preset category is set to consumption not upgraded; and as yet another example, when performing stability analysis on user new business promotion, the first preset category is set to new business application, and the second preset category is set to no new business application.
[0102] The user stability analysis method provided by this invention classifies the customer groups into experimental and observation groups based on similarity, analyzes the differences in behavioral characteristics between the two groups, makes a comprehensive evaluation, and sets operational observation indicators, thereby reducing the bias in user stability analysis.
[0103] Furthermore, in one embodiment, step S4, which involves performing a characteristic difference analysis on the observation group and the experimental group to analyze the stability of the target user group, may include:
[0104] S44. Input the training set samples and training labels into the preset binary classification model for training to obtain the target prediction model;
[0105] S45. Input the test set samples into the target prediction model to predict the stability of the target user group;
[0106] The training set samples include a first preset proportion of the observation group and the experimental group.
[0107] The test set sample includes a second preset proportion of the observation group and the experimental group;
[0108] The training labels were determined by classifying and labeling each user in the training set samples according to the requirements of target stability analysis.
[0109] Optionally, the experimental group and observation group obtained above are divided according to a preset ratio to obtain an experimental group and observation group with a first preset ratio and a second preset ratio. The experimental group and observation group with the first preset ratio are used as training set samples, and the experimental group and observation group with the second preset ratio are used as test set samples. The first preset ratio and the second preset ratio can be set to 0.8 and 0.2 respectively, or can be set according to actual needs; this invention does not impose specific limitations on this.
[0110] Based on the requirements of target stability analysis, each user in the training set sample is classified and labeled to obtain training labels. For example, when analyzing the stability of user churn, users who are churned are labeled with the number "1" and users who are still online are labeled with the number "0" to obtain training labels.
[0111] The training set samples and training labels are input into a preset binary classification model (e.g., a support vector machine (SVM) model) for training. Training is stopped when the loss value of the SVM model is less than a preset value, and the trained SVM model is obtained and used as the target prediction model.
[0112] The test set samples are fed into the target prediction model to predict the stability of the target user group (e.g., churn stability).
[0113] The user stability analysis method provided by this invention can perform user stability analysis for different business scenarios (such as consumption downgrade, churn, consumption upgrade, and new business promotion) based on the obtained target prediction model.
[0114] Furthermore, in one embodiment, step S4, which involves performing a characteristic difference analysis on the observation group and the experimental group to analyze the stability of the target user group, may specifically include:
[0115] S46. Based on the Autoregressive Integrated Moving Average (ARIMA) algorithm, the characteristic differences between the observation group and the experimental group are analyzed to analyze the stability of the target user group.
[0116] Optionally, the ARIMA algorithm can be used to analyze the differences in the temporal characteristics of the observation group and the experimental group of users in order to analyze the stability of the target user group.
[0117] The user stability analysis method provided by this invention analyzes user stability based on the ARIMA algorithm, which can reduce bias and improve the accuracy and reliability of user stability analysis.
[0118] The user stability analysis system provided by the present invention is described below. The user stability analysis system described below can be referred to in correspondence with the user stability analysis method described above.
[0119] Figure 3 This is a schematic diagram of the user stability analysis system provided by the present invention, as shown below. Figure 3 As shown, it includes: a data acquisition module 310, a first update module 311, a second update module 312, and a user analysis module 313;
[0120] The data acquisition module 310 is used to cluster all users in different time series after the update to obtain the optimal pair with the highest similarity between cluster categories of adjacent different time series.
[0121] The first update module 311 is used to determine the user group that meets the preset conditions based on the best pair with the highest similarity, and add the user group to the target user group to update the target user group;
[0122] The second update module 312 is used to remove user groups that meet preset conditions to update all users at different time sequences until the number of all users at different time sequences after the update is less than a preset threshold, and then stop updating.
[0123] User analysis module 313 is used to divide the updated target user group into observation group and experimental group, and to perform characteristic difference analysis on the observation group and experimental group to analyze the stability of the target user group.
[0124] The preset conditions include users who appear in all time series of the preset observation period in the optimal pairing.
[0125] The user stability analysis system provided by this invention clusters users within different time periods and identifies the customer group based on the data with the highest similarity. This addresses the problems of poor clustering results due to individual characteristic differences and inaccurate user stability analysis caused by poor clustering results.
[0126] Furthermore, in one embodiment, the data acquisition module 310 may further include: a first data acquisition submodule, a second data acquisition submodule, and a third data acquisition submodule.
[0127] The first data acquisition submodule is used to cluster the full users at different time series after the update based on the self-organizing map (SOM) clustering algorithm, and obtain the clustering categories of the full users at different time series after the update.
[0128] The second data acquisition submodule is used to obtain the category vectors of clustering categories at different time series based on preset category vector generation rules and clustering categories;
[0129] The third data acquisition submodule is used to obtain the similarity of cluster categories in different time series based on the category vectors of cluster categories in different time series, and to determine the optimal pair with the highest similarity between adjacent cluster categories in different time series based on the similarity of cluster categories in different time series.
[0130] The user stability analysis system provided by this invention uses clustering algorithms such as SOM to perform clustering calculations on data from different time series. Based on specific needs, it selects the clusters in which specific user groups belong and finds the category containing the specific user group with the highest cluster similarity among the clusters from different time series. This solves the problems of poor clustering effect and large subsequent bias caused by individual data differences and imbalance.
[0131] Furthermore, in one embodiment, the second data acquisition submodule can also be used for:
[0132] The first user belonging to the first category in the cluster is labeled with a first preset number;
[0133] Users who do not belong to the first cluster category are labeled with a second preset number; and
[0134] Based on the annotation results, obtain the category vectors of clustering categories at different time series.
[0135] The user stability analysis system provided by this invention lays the foundation for obtaining the similarity of clustering categories at different time series by acquiring the category vectors of users at different time series.
[0136] Furthermore, in one embodiment, the third data acquisition submodule can also be used to obtain the similarity of cluster categories in different time series based on the category vectors of cluster categories in different time series and the total number of users; or
[0137] The similarity of the clustering categories at different time intervals is obtained based on the category vectors of the clustering categories at different time intervals, the total number of users, and the preset weights.
[0138] The user stability analysis system provided by this invention calculates the similarity of all users at different time series after clustering, and uses an algorithm to delineate user data based on different characteristics, accurately aggregating data with high similarity and defining user groups with more commonalities.
[0139] Furthermore, in one embodiment, the user analysis module 313 may specifically include: a classification submodule, an experimental group customer acquisition submodule, and an observation group customer acquisition submodule;
[0140] The classification submodule is used to classify and label users in the target user group according to the target stability analysis requirements;
[0141] The experimental group customer acquisition submodule is used to identify users belonging to the first preset category as the experimental group customer.
[0142] The observation group customer acquisition submodule is used to identify users belonging to the second preset category as the observation group customer.
[0143] The user stability analysis system provided by this invention classifies the customer groups identified by similarity into an experimental group and an observation group, analyzes the differences in behavioral characteristics between the two groups, makes a comprehensive evaluation, sets operational observation indicators, and reduces the bias in user stability analysis.
[0144] Furthermore, in one embodiment, the user analysis module 313 can also be used to input training set samples and training labels into a preset binary classification model for training, in order to obtain a target prediction model; and
[0145] The test set samples are input into the target prediction model to predict the stability of the target user group;
[0146] The training set samples include a first preset proportion of the observation group and the experimental group.
[0147] The test set sample includes a second preset proportion of the observation group and the experimental group;
[0148] The training labels were determined by classifying and labeling each user in the training set samples according to the requirements of target stability analysis.
[0149] The user stability analysis system provided by this invention can perform user stability analysis for different business scenarios (such as consumption downgrade, churn, consumption upgrade, and new business promotion) based on the obtained target prediction model.
[0150] Furthermore, in one embodiment, the user analysis module 313 can also be used to perform characteristic difference analysis on the observation group and the experimental group based on the differential integrated moving average autoregressive (ARIMA) algorithm, so as to analyze the stability of the target user group. The user stability analysis system provided by the present invention analyzes user stability based on the ARIMA algorithm, which can reduce bias and improve the accuracy and reliability of user stability analysis.
[0151] Figure 4 This is a schematic diagram of the physical structure of an electronic device provided by the present invention, such as... Figure 4 As shown, the electronic device may include a processor 410, a communication interface 411, a memory 412, and a bus 413. The processor 410, communication interface 411, and memory 412 communicate with each other via the bus 413. The processor 410 can call logical instructions from the memory 412 to execute the following methods:
[0152] Cluster the full set of users at different time intervals after the update to obtain the optimal pair with the highest similarity between cluster categories of adjacent different time intervals;
[0153] Based on the best pairing with the highest similarity, user groups that meet the preset conditions are identified and added to the target user group to update the target user group;
[0154] Users who meet the preset conditions will be removed to update all users at different time periods until the number of all users at different time periods after the update is less than the preset threshold, at which point the update will stop.
[0155] The updated target user group is divided into an observation group and an experimental group, and a feature difference analysis is performed on the observation group and the experimental group to analyze the stability of the target user group.
[0156] The preset conditions include users who appear in all time series of the preset observation period in the optimal pairing.
[0157] Furthermore, the logical instructions in the aforementioned memory can be implemented as software functional units and sold or used as independent products, and can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer power supply (which may be a personal computer, server, or network power supply, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.
[0158] Furthermore, this invention discloses a computer program product, which includes a computer program stored on a non-transitory computer-readable storage medium. The computer program includes program instructions, and when these instructions are executed by a computer, the computer can execute the user stability analysis methods provided in the above-described method embodiments, such as including:
[0159] Cluster the full set of users at different time intervals after the update to obtain the optimal pair with the highest similarity between cluster categories of adjacent different time intervals;
[0160] Based on the best pairing with the highest similarity, user groups that meet the preset conditions are identified and added to the target user group to update the target user group;
[0161] Users who meet the preset conditions will be removed to update all users at different time periods until the number of all users at different time periods after the update is less than the preset threshold, at which point the update will stop.
[0162] The updated target user group is divided into an observation group and an experimental group, and a feature difference analysis is performed on the observation group and the experimental group to analyze the stability of the target user group.
[0163] The preset conditions include users who appear in all time series of the preset observation period in the optimal pairing.
[0164] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to perform the user stability analysis methods provided in the above embodiments, including, for example:
[0165] Cluster the full set of users at different time intervals after the update to obtain the optimal pair with the highest similarity between cluster categories of adjacent different time intervals;
[0166] Based on the best pairing with the highest similarity, user groups that meet the preset conditions are identified and added to the target user group to update the target user group;
[0167] Users who meet the preset conditions will be removed to update all users at different time periods until the number of all users at different time periods after the update is less than the preset threshold, at which point the update will stop.
[0168] The updated target user group is divided into an observation group and an experimental group, and a feature difference analysis is performed on the observation group and the experimental group to analyze the stability of the target user group.
[0169] The preset conditions include users who appear in all time series of the preset observation period in the optimal pairing.
[0170] The system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.
[0171] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., including several instructions to cause a computer power supply (which may be a personal computer, server, or network power supply, etc.) to execute the methods described in various embodiments or some parts of the embodiments.
[0172] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention 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 of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.
Claims
1. A user stability analysis method, characterized in that, include: Clustering is performed on all users in different time series after the update to obtain the optimal pair with the highest similarity between cluster categories of adjacent different time series; wherein, the full user data in different time series is obtained by dividing the user data of different billing periods into time series. Based on the optimal pairing with the highest similarity, a user group that meets the preset conditions is determined, and the user group is added to the target user group to update the target user group; Users who meet the preset conditions are removed to update all users in different time series until the number of all users in different time series after the update is less than a preset threshold, at which point the update stops. The updated target user group is divided into an observation group and an experimental group, and a feature difference analysis is performed on the observation group and the experimental group to analyze the stability of the target user group. The preset conditions include users who appear in all time series of the preset observation period in the optimal pairing; The step of clustering all users across different time periods after the update to obtain the optimal pairing with the highest similarity between cluster categories of adjacent different time periods includes: The self-organizing map (SOM) clustering algorithm is used to cluster all users at different time series after the update, and the cluster categories of all users at different time series after the update are obtained. Based on the preset category vector generation rules and the clustering categories, obtain the category vectors of clustering categories at different time sequences; Based on the category vectors of the clustering categories at different time sequences, the similarity of the clustering categories at different time sequences is obtained, and based on the similarity of the clustering categories at different time sequences, the optimal pair with the highest similarity between adjacent clustering categories at different time sequences is determined. The user stability analysis method is used to analyze the stability of mobile users' consumption downgrades, churns, consumption upgrades, and new business promotions.
2. The user stability analysis method according to claim 1, characterized in that, The step of obtaining the category vectors of clustering categories at different time intervals according to the preset category vector generation rules and the clustering categories includes: The first user belonging to the first category in the cluster is labeled with a first preset number; Other users in the cluster that do not belong to the first category are labeled with a second preset number; Based on the annotation results, obtain the category vectors of the clustering categories at different time sequences.
3. The user stability analysis method according to claim 1, characterized in that, The step of obtaining the similarity of cluster categories at different time intervals based on the category vectors of the cluster categories at different time intervals includes: Based on the category vectors of the cluster categories at different time intervals and the total number of users, the similarity of the cluster categories at different time intervals is obtained; or The similarity of the clustering categories in different time periods is obtained based on the category vectors of the clustering categories in different time periods, the total number of users, and the preset weights.
4. The user stability analysis method according to claim 1, characterized in that, The process of dividing the updated target user group into an observation group and an experimental group includes: Based on the target stability analysis requirements, each user in the target user group is classified and labeled. Users belonging to the first preset category are defined as the experimental group. Users belonging to the second preset category are defined as the observation group.
5. The user stability analysis method according to claim 4, characterized in that, The characteristic difference analysis of the observation group and the experimental group to analyze the stability of the target user group includes: The training set samples and training labels are input into a preset binary classification model for training to obtain a target prediction model; The test set samples are input into the target prediction model to predict the stability of the target user group; The training set samples include the observation group and the experimental group, which are in a first preset proportion. The test set sample includes the observation group and the experimental group in a second preset ratio; The training labels are determined by classifying and labeling each user in the training set samples according to the target stability analysis requirements.
6. The user stability analysis method according to any one of claims 1-5, characterized in that, The method of performing characteristic difference analysis on the observation group and the experimental group to analyze the stability of the target user group also includes: The differential integrated moving average autoregressive (ARIMA) algorithm is used to analyze the characteristic differences between the observation group and the experimental group to analyze the stability of the target user group.
7. A user stability analysis system, characterized in that, include: The module consists of a data acquisition module, a first update module, a second update module, and a user analysis module. The data acquisition module is used to cluster the updated full user data for different time periods to obtain the optimal pair with the highest similarity among cluster categories of adjacent different time periods; wherein, the full user data for different time periods is obtained by dividing the user data for different billing periods into time periods; The first update module is used to determine a user group that meets preset conditions based on the optimal pairing with the highest similarity, and add the user group to the target user group to update the target user group; The second update module is used to remove user groups that meet the preset conditions to update all users in different time series until the number of all users in different time series after the update is less than a preset threshold, and then stop updating. The user analysis module is used to divide the updated target user group into an observation group and an experimental group, and to perform feature difference analysis on the observation group and the experimental group to analyze the stability of the target user group. The preset conditions include users who appear in all time series of the preset observation period in the optimal pairing; The data acquisition module is specifically used for: The self-organizing map (SOM) clustering algorithm is used to cluster all users at different time series after the update, and the cluster categories of all users at different time series after the update are obtained. Based on the preset category vector generation rules and the clustering categories, obtain the category vectors of clustering categories at different time sequences; Based on the category vectors of the clustering categories at different time sequences, the similarity of the clustering categories at different time sequences is obtained, and based on the similarity of the clustering categories at different time sequences, the optimal pair with the highest similarity between adjacent clustering categories at different time sequences is determined. The user stability analysis system is used to perform stability analysis on mobile users' consumption downgrades, churns, consumption upgrades, and new business promotions.
8. An electronic device comprising a processor and a memory storing a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the user stability analysis method according to any one of claims 1 to 6.
9. A processor-readable storage medium, characterized in that, The processor-readable storage medium stores a computer program that causes the processor to perform the steps of the user stability analysis method according to any one of claims 1 to 6.