Customer loss early warning method and device, storage medium, program product and computer equipment

Abstract

The application discloses a customer loss early warning method and device, a storage medium, a program product and computer equipment. The method comprises the following steps: obtaining behavior data of a target customer, and determining key behavior characteristics corresponding to the behavior data; determining behavior mode change information of the target customer in a target period based on the key behavior characteristics; in the case that the behavior mode change information meets a preset mode deviation condition, determining risk index information of the target customer based on the behavior mode change information, the key behavior characteristics, reference characteristic information associated with the key behavior characteristics; obtaining an early warning strategy based on the key behavior characteristics, the risk index information and a first risk trend direction corresponding to the risk index information, and historical loss-related data of a customer group to which the target customer belongs; and pushing a retention information message corresponding to the early warning strategy to the target customer. Thus, the loss risk early warning accuracy can be improved, and the pushing accuracy of the retention information message pushed to the customer with a high loss risk can be improved.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and in particular to a customer churn early warning method, device, storage medium, program product, and computer equipment. Background Technology

[0002] In related technologies, in order to provide early warnings for customer churn, a threshold corresponding to a certain dimension of customer data can be configured. This allows for the issuance of an alert when the real-time data for that dimension reaches the corresponding threshold, thus facilitating customer retention efforts.

[0003] However, the accuracy of churn risk warnings for this type of technology is low, making it difficult to accurately identify customers at high risk of churn, and it is even prone to misidentification. It may also make it difficult to push accurate and effective retention information messages to customers at high risk of churn. Summary of the Invention

[0004] To address the aforementioned technical issues, this application proposes a customer churn early warning method, apparatus, storage medium, program product, and computer equipment, which can improve the accuracy of churn risk early warning and enhance the accuracy of push notifications of retention information messages to customers at high churn risk.

[0005] In a first aspect, embodiments of this application provide a customer churn early warning method, including: Acquire behavioral data of target customers and determine the key behavioral characteristics corresponding to the behavioral data; Based on the key behavioral characteristics, determine the behavioral pattern changes of the target customer within the target time period; If the behavioral pattern change information meets the preset pattern deviation conditions, the risk index information of the target customer is determined based on the behavioral pattern change information, the key behavioral characteristics, and the benchmark characteristic information associated with the key behavioral characteristics. Based at least on the key behavioral characteristics, the risk index information and its corresponding first risk trend direction, and the historical churn data of the target customer's customer group, an early warning strategy is obtained. Push retention information messages corresponding to the early warning strategy to the target customers.

[0006] Optionally, determining the risk index information of the target customer based on the behavioral pattern change information, the key behavioral characteristics, and the benchmark characteristic information associated with the key behavioral characteristics includes: Based on the key behavioral characteristics, determine the behavioral characteristic parameters of the target customer during the target time period; Based on the behavioral characteristic parameters, at least one trend period is determined from the target time period, wherein the at least one trend period includes a period in which the parameter changes of the behavioral characteristic parameters continuously decay and / or a period in which the parameter changes in a stepwise manner decay. Based on the rate of change of the behavioral feature parameters in each trend period, preprocessed feature data is determined, wherein the preprocessed feature data includes at least a portion of the rate of change extreme value information of the parameter change rate; Based on the key behavioral characteristics, the baseline characteristic information, the behavioral pattern change information, and the preprocessed characteristic data, the risk index information of the target customer is determined.

[0007] Optionally, determining the risk index information of the target customer based on the key behavioral features, the baseline feature information, the behavioral pattern change information, and the preprocessed feature data includes: Based on the key behavioral features and the baseline feature information, feature offset information matching the preprocessed feature data is determined; Based on the feature offset information, the behavioral pattern change information, and the preprocessed feature data, the first churn risk information of the target customer is predicted. Based on the feature weights associated with the key behavioral characteristics, the first churn risk information is weighted to obtain the second churn risk information. Based on the second loss risk information and the preset characteristics of different risk levels, the risk index information is determined.

[0008] Optionally, the method of obtaining an early warning strategy based at least on the key behavioral characteristics, the risk index information and its corresponding first risk trend direction, and the historical churn data of the target customer's customer group includes: Determine the degree of variation of the key behavioral features; If the degree of variation of the key behavioral features is greater than a preset degree of variation threshold, the target customer is marked as having a tendency to churn. Furthermore, based on the cumulative number of times the target customer is marked as having a tendency to churn and the type of churn tendency behavior corresponding to the cumulative number of times, the marking distribution features are determined. When the marked distribution characteristics meet the risk distribution conditions, the tendency influence factor is determined based on the cumulative number of times and the cumulative number of times threshold, as well as the churn tendency behavior type and the preset churn tendency behavior type set; Based on the aforementioned tendency influencing factors and the historical churn-related data, a machine learning model is invoked to generate current churn risk threshold information. Determine the second risk trend direction corresponding to the current critical churn risk standard information, and determine the directional matching relationship between the first risk trend direction and the second risk trend direction; The early warning strategy is generated based at least on the directional matching relationship and the risk index information, combined with the historical churn-related data and / or the current churn risk threshold information.

[0009] Optionally, the historical data on data loss includes information on historical risk thresholds for data loss. The early warning strategy is generated based at least on the directional matching relationship and the risk index information, combined with the historical churn-related data and / or the current churn risk threshold information, including: When the directional matching relationship indicates that the first risk trend direction and the second risk trend direction are in opposite directions, if the risk index information matches the historical loss risk critical standard information, a general early warning strategy is obtained, wherein the early warning strategy includes the general early warning strategy; And / or, When the directional matching relationship indicates that the first risk trend direction and the second risk trend direction are in a directional intersection relationship, a reference risk index information matching the risk index information is determined based on the historical churn risk critical standard information and the current churn risk critical standard information. If the risk index information matches the reference risk index information, a fusion early warning strategy is generated. The early warning strategy includes the fusion early warning strategy, which is formed by fusing multiple general early warning strategies. And / or, When the directional matching relationship indicates that the first risk trend direction and the second risk trend direction are in the same direction, if the risk index information matches the current churn risk threshold information, then at least based on the basic information and behavioral characteristics of the target customer, a target early warning strategy for the target customer is generated through a large model, wherein the early warning strategy includes the target early warning strategy.

[0010] Optionally, the method further includes: Construct a causal relationship graph that matches the behavioral data; Based on the causal relationship diagram, the causal effect value corresponding to the key behavioral feature is determined, wherein the causal effect value characterizes the correlation between the key behavioral feature and customer churn; The step of generating a target early warning strategy for the target customer based at least on the target customer's basic information and behavioral characteristics through a large model includes: Based on the causal effect value, the basic information and behavioral characteristics of the target customer, the target early warning strategy is generated through a large model.

[0011] Secondly, embodiments of this application provide a customer churn early warning device, comprising: The data processing module is used to acquire the behavioral data of the target customer and determine the key behavioral features corresponding to the behavioral data. The behavior pattern recognition module is used to determine the behavior pattern change information of the target customer within a target time period based on the key behavioral features. The risk index determination module is used to determine the risk index information of the target customer based on the behavioral pattern change information, the key behavioral characteristics, and the benchmark characteristic information associated with the key behavioral characteristics, when the behavioral pattern change information meets the preset pattern deviation conditions. The early warning strategy acquisition module is used to acquire an early warning strategy based at least on the key behavioral characteristics, the risk index information and its corresponding first risk trend direction, and the historical churn data of the target customer's customer group. The message push module is used to push retention information messages corresponding to the early warning strategy to the target customers.

[0012] Thirdly, embodiments of this application provide a non-transitory computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the method described in any of the above-mentioned embodiments.

[0013] Fourthly, embodiments of this application provide a computer program product, including computer instructions that, when executed by a processor, implement the steps of the method described in any of the above-described embodiments.

[0014] Fifthly, embodiments of this application provide a computer device including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor executes the computer program to implement the steps of the method described in any of the preceding claims.

[0015] In summary, the embodiments of this application have at least the following beneficial effects: By employing the embodiments of this application, behavioral data of target customers is acquired, and key behavioral characteristics corresponding to the behavioral data are determined. Based on the key behavioral characteristics, behavioral pattern change information of the target customer within a target time period is determined. When the behavioral pattern change information meets preset pattern deviation conditions, risk index information of the target customer is determined based on the behavioral pattern change information, the key behavioral characteristics, and benchmark feature information associated with the key behavioral characteristics. At least based on the key behavioral characteristics, the risk index information and its corresponding first risk trend direction, and historical churn-related data of the target customer's customer group, an early warning strategy is obtained. Retention information messages corresponding to the early warning strategy are pushed to the target customer. In this way, key behavioral characteristics can be identified based on customer behavioral data to improve the accuracy of behavioral pattern change identification. When behavioral model deviation is identified, the risk index of the target customer is accurately analyzed by comprehensively considering behavioral pattern change information, key behavioral characteristics, and benchmark feature information. Furthermore, based on the key behavioral characteristics, the risk index information and its risk trend direction, as well as historical churn-related data of the target customer's customer group, are further considered to accurately obtain early warning strategies, thereby improving the accuracy of churn risk early warning and the accuracy of pushing retention information messages to customers with high churn risk. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating the customer churn early warning method provided in the embodiments of this application; Figure 2 This is a schematic diagram of the customer churn early warning device provided in the embodiments of this application; Figure 3 This is a schematic diagram of the computer device provided in the embodiments of this application. Detailed Implementation

[0017] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments / examples are only a part of the embodiments / examples of this application, and not all of the embodiments / examples. Based on the embodiments / examples in this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0018] In the description of this application, the terms "first," "second," "third," etc., are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Therefore, a feature defined with "first," "second," "third," etc., may explicitly or implicitly include one or more of that feature. In the description of this application, unless otherwise stated, "multiple" means two or more. In the description of this application, the term "comprising" and its variations are open-ended, meaning "including but not limited to." The term "based on" means "at least partially based on." The term "according to" means "at least partially according to." The term "one embodiment / example" means "at least one embodiment / example"; the term "another embodiment / example" means "at least one additional embodiment / example"; the term "some embodiments / examples" means "at least some embodiments / examples."

[0019] In the description of this application, it should be noted that, unless otherwise expressly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection between two components. Those skilled in the art can understand the specific meaning of the above terms in this application based on the specific circumstances.

[0020] In the description of this application, it should be noted that, unless otherwise defined, all technical and scientific terms used in this application have the same meaning as commonly understood by one of ordinary skill in the art. The terminology used in this application is for the purpose of describing specific embodiments only and is not intended to limit the application. Those skilled in the art can understand the specific meaning of the above terms in this application according to the specific circumstances.

[0021] Firstly, see [the following] Figure 1 The diagram shows a flowchart of a customer churn warning method provided in an embodiment of this application. The customer churn warning method can be applied to a computer device with data processing capabilities. The method includes steps S101-S105, as detailed below.

[0022] S101, acquire the target customer's behavioral data and determine the key behavioral characteristics corresponding to the behavioral data.

[0023] In some examples, behavioral data can be multi-dimensional dynamic behavioral data of the target customer, such as at least one of the following: web browsing path data, user call data with customer service (e.g., emotional information contained in the call voice), text ticket data submitted by the user (e.g., complaint topic information contained in the text ticket), social network relationship data with other customers, etc.

[0024] In some examples, key behavioral features can be obtained by feature extraction from behavioral data, for example, by extracting behavioral attributes from behavioral data that are strongly correlated with customer churn events.

[0025] S102, Based on the key behavioral characteristics, determine the behavioral pattern change information of the target customer within the target time period.

[0026] In some examples, the target time period can be a preset behavior analysis period. Key behavioral characteristics within this period are detected to obtain behavioral characteristic parameters and behavioral pattern change trends. Based on these trends, the behavior pattern type is output. The behavioral characteristic parameters can reflect the specific values ​​and / or statistics of key behavioral characteristics of the target customer within the behavior analysis period. The behavioral pattern change trend can describe the directional changes of key behavioral characteristics over time (e.g., increase, decrease, increased volatility, etc.). It is easy to understand that this behavioral pattern change trend can be obtained through statistical analysis of the behavioral characteristic parameters or the key behavioral characteristic.

[0027] In some examples, behavioral pattern change information can be used to describe the dynamic evolution of key behavioral characteristics of a target customer within a target time period. This dynamic evolution may include behavioral intensity, frequency, persistence, and / or direction of change.

[0028] S103, if the behavior pattern change information meets the preset pattern deviation conditions, the risk index information of the target customer is determined based on the behavior pattern change information, the key behavior characteristics, and the benchmark characteristic information associated with the key behavior characteristics.

[0029] In some examples, key behavioral features may contain multiple features. In this case, the acquisition of benchmark feature information may include: acquiring normal behavioral data of the target customer and / or its customer group during the normal interaction cycle, detecting and calculating the mean data of the features corresponding to each feature in the key behavioral features in the normal behavioral data, thereby obtaining the benchmark feature parameters associated with each feature in the key behavioral features.

[0030] In some examples, preset mode deviation conditions can be used to determine whether the current behavior significantly deviates from the normal interaction mode. The preset mode deviation conditions used as the basis for judgment may include at least one of the following: whether the degree of deviation between the real-time parameters of the behavior feature and the baseline feature parameters exceeds the deviation degree threshold, whether the trend of behavior pattern change has an abnormal direction, and whether the rate of change of the trend of behavior pattern change exceeds the rate of change threshold.

[0031] In some examples, the behavioral pattern change information may include the direction of the behavioral pattern change trend. This direction can indicate the current abnormal behavioral characteristics present in the key behavioral features. Then, a difference analysis can be performed on the current abnormal behavioral characteristics and the benchmark features corresponding to the current abnormal behavioral characteristics in the benchmark feature information. Based on the analyzed differences, the risk index information of the target customer can be determined. Generally speaking, the larger the analyzed differences, the higher the risk corresponding to the risk index information.

[0032] S104. At least based on the key behavioral characteristics, the risk index information and its corresponding first risk trend direction, and the historical churn data of the target customer's customer group, an early warning strategy is obtained.

[0033] In some examples, a pre-trained early warning model can be used to determine an early warning strategy based on key behavioral features, risk index information, the first risk trend direction, and historical churn data. This early warning model can be a trained model capable of predicting using key behavioral features, risk index information, the first risk trend direction, and historical churn data as input and an early warning strategy as output. During training, sample key behavioral features, sample risk index information, sample risk trend direction, and historical churn data can be used as sample data (this sample data also carries a corresponding expected strategy label, which represents the expected early warning strategy). The model's predicted early warning strategy generated based on this sample data is obtained. Based on the difference between this predicted early warning strategy and the expected early warning strategy represented by the label, a general loss function is used to calculate the loss value. A general training algorithm (e.g., gradient descent) is then used to train the model based on this loss value, so that the trained model possesses the aforementioned capabilities. For example, the model may include an input representation layer, a feature fusion and representation layer, and a prediction output layer. The input representation layer can receive data from the input model and convert the data into the desired feature vector form. For example, the input representation layer can use word embeddings or pre-trained language models (such as bidirectional language representation models based on the Transformer architecture) to generate semantic vectors, and / or use embedding layers to generate dense vectors. The feature fusion and representation layer can be used to fuse the feature vectors converted by the input representation layer. For example, the feature fusion and representation layer can implement the fusion through fully connected layers or attention mechanism layers. The prediction output layer can be used to generate prediction results based on the fused features. For example, the prediction output layer can use a softmax layer to output the probability distribution for different categories, and then output the prediction result based on the probability (for example, it can output the top one or more classification results with the highest probability as the prediction result).

[0034] Understandably, key behavioral characteristics (CQCs) are behavioral attributes extracted from dynamic behavioral data across all dimensions and strongly correlated with customer churn events (such as a sudden drop in login frequency, a surge in customer service complaints, and interruption of core function usage). These characteristics directly reflect the specific manifestations of abnormal customer behavior and serve as the basis for developing intervention measures. For example, if CQCs indicate that a customer has not used core services recently, the retention strategy should focus on reactivating the use of that service.

[0035] Understandably, the risk index is a numerical risk score calculated by combining key behavioral characteristics and baseline characteristics (such as the average behavior of customers within a normal interaction cycle) after behavioral pattern changes meet preset deviation conditions. This index can reflect the severity of current customer churn risk and provide a basis for the response strength of early warning strategies. For example, a high-risk index may trigger high-priority intervention (such as dedicated customer service intervention), while a low-risk index may only trigger lightweight outreach (such as coupon push).

[0036] Understandably, the first risk trend direction is derived from the trend of behavioral pattern changes and is used to identify whether the risk is rising, stable, or mitigating. This directional information can be used to determine the urgency of intervention and the direction of dynamic strategy adjustments. For example, if the risk trend is rising rapidly, strong intervention can be implemented immediately; if the trend has slowed down, an observation and gentle guidance strategy can be adopted to avoid excessive interference.

[0037] Understandably, historical churn data for the target customer group (such as groups using the same product package, groups located in the same region, and / or at the same lifecycle stage) can include the actual churn rate of this group under similar behavioral patterns, typical churn time windows, and the success rate of effective retention methods. Historical churn data can provide an empirical basis for strategy selection, ensuring that early warning strategies are not only based on the current state of individuals but also conform to the behavioral patterns of the group, thereby improving the universality, effectiveness, and success rate of the strategies.

[0038] Thus, in this embodiment, key behavioral characteristics can be used to determine the content to be intervened (content targeting), risk index information can be used to determine the intensity of the intervention (resource investment level), the first risk trend direction can be used to determine when to intervene and / or how to adjust (time series and dynamism), and customer group historical churn related data can be used to determine which intervention is more likely to succeed (strategy effectiveness assurance), thereby providing a multi-dimensional and accurate basis for obtaining early warning strategies and effectively improving the accuracy of early warning strategy acquisition.

[0039] S105, push retention information messages corresponding to the early warning strategy to the target customer.

[0040] In some examples, the alert strategy may include the customer type to which the target customer belongs, and the retention information message may include at least one of the following: personalized offer messages (such as exclusive discounts, coupons for spending a certain amount, double points, extension of free trials, etc., applicable to customer types that are sensitive to price or cost), and service improvement or upgrade information messages (such as introductions to newly launched features, announcements of service process optimization, notifications of the opening of exclusive service channels, etc., applicable to customer types that are at increased risk of churn due to poor experience).

[0041] In one optional implementation, determining the risk index information of the target customer based on the behavioral pattern change information, the key behavioral characteristics, and the benchmark characteristic information associated with the key behavioral characteristics includes: Based on the key behavioral characteristics, the behavioral characteristic parameters of the target customer during the target time period are determined; wherein, the behavioral characteristic parameters can be used to reflect the specific values ​​and / or statistics of the key behavioral characteristics of the target customer during the behavioral analysis period. Based on the behavioral characteristic parameters, at least one trend period is determined from the target time period, wherein the at least one trend period includes a period in which the parameter changes of the behavioral characteristic parameters continuously decay and / or a period in which the parameter changes in a stepwise manner decay. Based on the rate of change of the behavioral feature parameters in each trend period, preprocessed feature data is determined, wherein the preprocessed feature data includes at least a portion of the rate of change extreme value information of the parameter change rate; Based on the key behavioral characteristics, the baseline characteristic information, the behavioral pattern change information, and the preprocessed characteristic data, the risk index information of the target customer is determined.

[0042] It should be noted that parameter change decay refers to the trend of the value of the corresponding behavioral characteristic parameter decreasing gradually over a continuous period of time or time window. This decreasing process generally has a consistent direction (i.e., downward overall) and continuity (not accidental fluctuations).

[0043] In some examples, at least one trend period may include a period of sustained decay of parameter change in a behavioral characteristic parameter. This sustained decay period can refer to any one or more behavioral characteristic parameters corresponding to the trend period exhibiting a sustained decay trend during that trend period. Thus, the at least partial rate extremum information may include the maximum value of the parameter decay rate during the sustained decay period, where the parameter decay rate indicates the rate of decrease / reduction of the value of the corresponding behavioral characteristic parameter.

[0044] In some examples, at least one trend period may include a step-decline period for the parameter changes of a behavioral characteristic parameter. This step-decline period can refer to any one or more behavioral characteristic parameters corresponding to the trend period exhibiting significant, discrete numerical drops at multiple time points, with each drop followed by relative stability at a new lower level, forming a step-like or stair-like descent pattern. Thus, the at least partial rate extremum information may include the maximum value of the parameter decay rate during the step-decline period, where the parameter decay rate can be determined by the ratio between the value of each drop and the duration of that drop (i.e., the time elapsed from the first drop to the next).

[0045] In some examples, at least one trend period includes a period of continuous decay of parameter change and a period of stepped decay of parameter change, such that the at least partial rate maximum / minimum information may include at least one of the following: the maximum value of parameter decay rate during the period of continuous decay of parameter change, and the maximum value of parameter decay rate during the period of stepped decay of parameter change.

[0046] In this embodiment, at least one trend period can be identified based on behavioral characteristic parameters, and rate extremum information is introduced to help determine the urgency of intervention (customers with high decay rates should be prioritized for intervention, while customers with low decay rates can be addressed with automated light-touch interventions), thereby achieving precise resource allocation and maximizing operational efficiency. Furthermore, by setting identification conditions for trend periods (such as minimum duration period and decay amplitude threshold), occasional behavioral fluctuations caused by holidays, temporary events, etc., can be effectively filtered out, reducing false alarms. Sensitive capture of step-like abrupt changes can also avoid underreporting of high-risk customers that are ignored by general models due to non-continuous behavioral decline. Furthermore, combining continuous decay periods of parameter changes with step-like decay periods of parameter changes can significantly improve the signal-to-noise ratio and recall capability of the early warning system.

[0047] In one optional implementation, determining the risk index information of the target customer based on the key behavioral characteristics, the baseline characteristic information, the behavioral pattern change information, and the preprocessed characteristic data includes: Based on the key behavioral features and the baseline feature information, feature offset information matching the preprocessed feature data is determined; Based on the feature offset information, the behavioral pattern change information, and the preprocessed feature data, the first churn risk information of the target customer is predicted. Based on the feature weights associated with the key behavioral characteristics, the first churn risk information is weighted to obtain the second churn risk information. Based on the second loss risk information and the preset characteristics of different risk levels, the risk index information is determined.

[0048] In some examples, feature weights can be determined by detecting the impact weight of each key behavioral feature on customer churn outcomes, thereby obtaining feature weight values. Specifically, the GBDT (Gradient Boosting Decision Tree) algorithm can be used to quantify the contribution of each key behavioral feature in customer churn prediction, thus clarifying the weight value corresponding to each feature and forming feature weights associated with the key behavioral features. For example, in predicting customer churn in the telecommunications industry, the GBDT algorithm can learn from a large amount of historical customer data and discover that the feature of package tariff complaints frequently appears in the node splits of multiple rounds of the decision tree and has high discriminative power. Therefore, a high weight value can be assigned to this feature to accurately identify its high impact on customer churn.

[0049] In some examples, feature offset information matching the preprocessed feature data can be predicted based on the key behavioral features and the baseline feature information using particle swarm optimization and / or the maximum entropy principle.

[0050] Specifically, the parameters of the feature shift prediction model can be optimized by combining it with PSO (Particle Swarm Optimization). PSO simulates the particle search process in the solution space, using the prediction accuracy of the feature shift prediction model as the fitness function. By iteratively updating the position and velocity of the particles, the model's ability to capture the amount of feature shift information can be maximized, thereby improving the accuracy of the feature shift information. For example, when predicting mobile data traffic shift parameters, PSO can adjust key parameters in the model such as the weight of the data traffic decay rate and / or the behavioral pattern deviation coefficient, enabling the model to more accurately fit the change pattern from the baseline feature parameters (e.g., normal 30GB monthly data traffic) to the shifted parameters (e.g., subsequent 20GB monthly data traffic), thus reducing uncertainty in the prediction.

[0051] In addition, the maximum entropy principle can be introduced to further optimize the churn prediction process. By maximizing the information entropy during the prediction of feature offset parameters, the comprehensiveness of the probability distribution information retained by the feature offset prediction model when dealing with deviations from unknown behavioral patterns can be improved, thereby enhancing its adaptability to various potential offset scenarios.

[0052] In some examples, key behavioral features may include multiple features. The first churn risk information may include individual churn risk coefficients corresponding to each feature in the key behavioral features. That is, each individual churn risk coefficient can be used to indicate the churn risk corresponding to a feature in the key behavioral features. Correspondingly, feature offset information may contain feature offsets corresponding to each feature in the key behavioral features, behavioral pattern change information may contain change information corresponding to each feature in the key behavioral features, and preprocessed feature data may contain the maximum and minimum values ​​of parameter change rates corresponding to each feature in the key behavioral features. Thus, each feature in the key behavioral features can be called a type of key behavioral feature. In this way, risk mapping rules for each type of key behavioral feature can be predefined to map the feature offsets, change information, and maximum and minimum values ​​of parameter change rates to risk coefficients according to the corresponding risk mapping rules. For example, if the feature offset of monthly active days is lower than 40% of the baseline, the corresponding individual churn risk coefficient can be assigned a value of 0.85. It is easy to understand that various risk mapping rules can be derived based on historical churn sample statistics, and this application embodiment does not specifically limit this.

[0053] In some examples, different risk level characteristics can include the numerical ranges of different risk levels. Thus, based on the second loss risk information and the preset numerical ranges of different risk levels, the second loss risk information can be converted to the corresponding risk level to obtain the risk index to be judged, i.e., the final risk index information. Here, this risk index information can be mapped to preset levels such as low risk, medium risk, and high risk, thereby providing a quantitative basis for the generation of subsequent early warning strategies.

[0054] In one optional implementation, the step of obtaining an early warning strategy based at least on the key behavioral characteristics, the risk index information and its corresponding first risk trend direction, and the historical churn data of the target customer's customer group includes: Determine the degree of variation of the key behavioral features; for example, the degree of variation can be determined by the parameter change rate and / or feature offset information, such as the parameter change rate and / or feature offset information. If the degree of variation of the key behavioral features is greater than a preset degree of variation threshold, the target customer is marked as having a tendency to churn. Furthermore, based on the cumulative number of times the target customer is marked as having a tendency to churn and the type of churn tendency behavior corresponding to the cumulative number of times, the marking distribution features are determined. When the marked distribution characteristics meet the risk distribution conditions, the tendency influence factor is determined based on the cumulative number of times and the cumulative number of times threshold, as well as the churn tendency behavior type and the preset churn tendency behavior type set; Based on the aforementioned tendency influencing factors and the historical churn-related data, a machine learning model is invoked to generate current churn risk threshold information. Determine the second risk trend direction corresponding to the current critical churn risk standard information, and determine the directional matching relationship between the first risk trend direction and the second risk trend direction; The early warning strategy is generated based at least on the directional matching relationship and the risk index information, combined with the historical churn-related data and / or the current churn risk threshold information.

[0055] In some examples, when the degree of variation of the key behavioral feature is less than or equal to a preset degree of variation threshold, a general early warning strategy can be obtained, wherein the early warning strategy includes the general early warning strategy.

[0056] In some examples, historical churn data may include statistics such as the actual churn rate of similar customers within the target customer's customer base under similar behavioral patterns, the distribution of typical churn times, and / or the success rate of effective interventions.

[0057] In some examples, the preset variability threshold can be associated with the target customer's customer group. For example, it can be the industry average variability of the target customer's industry. The average variability of each key behavioral characteristic of different customers in the industry can be collected and the mean can be calculated to obtain the industry average variability of that key behavioral characteristic, which can be used as the corresponding preset variability threshold.

[0058] In some examples, the cumulative number of times a target customer is marked as having churn-prone behavior refers to the cumulative number of times a target customer has been marked as having churn-prone behavior throughout history. In response to each time an alert policy is generated for that target customer, the cumulative number of times for that target customer can be reset to zero, or the cumulative number of times can be kept from being reset to zero.

[0059] In some examples, the cumulative number of times a target customer is marked as having churn-prone behavior (i.e., the number of times a churn-prone behavior mark is triggered) and the type of churn-prone behavior corresponding to each mark (i.e. the covered behavioral dimensions, such as price-sensitive type, service interruption type, and / or customer service complaint surge type) can be continuously recorded. In this way, the mark distribution characteristics can be obtained by statistically analyzing the cumulative number and the corresponding churn-prone behavior types, so that the mark distribution characteristics can be used to indicate the statistical distribution parameters of the target customer being marked as having churn-prone behavior.

[0060] In some examples, risk distribution conditions may include at least one of the following: the labeling distribution characteristics are characterized by high-frequency triggering characteristics (such as being labeled more than 5 times in the past 30 days), or the labeling distribution characteristics are characterized by churn tendency behavior types covering multiple behavioral type dimensions (such as simultaneously involving two or more dimensions such as traffic usage, bill payment, and customer service interaction). In such cases, the target customer can be classified as a high churn tendency group, and the tendency influencing factors of the target customer can be further determined.

[0061] In some examples, to more accurately characterize the urgency of churn for target customers who meet the risk distribution conditions, a propensity influence factor corresponding to that target customer can be further calculated. This propensity influence factor may include two influencing factors: Impact Factor 1: Determined by the ratio between the cumulative number of occurrences and the cumulative number of occurrences threshold. For example, if the cumulative number of occurrences is 6 and the cumulative number of occurrences threshold is 4, then Impact Factor 1 = 6 / 4 = 1.5.

[0062] Impact Factor 2: Determined by the ratio between the scope of influence of churn-prone behavior types (i.e., the number of churn-prone behavior types that have been triggered) and the total number of all behavioral dimensions in the preset churn-prone behavior type set. For example, if churn-prone behavior types cover a total of 4 behavioral dimensions and the total number of behavioral dimensions is 5, then Impact Factor 2 = 4 / 5 = 0.8.

[0063] In some examples, the propensity factor can be used to indicate the urgency of churn for target customers who meet the risk distribution conditions. Historical churn-related data can include historical churn risk threshold information. Thus, the propensity factor and historical churn-related data can be input into a machine learning model to obtain the current churn risk threshold information output by the machine learning model. The current churn risk threshold information can be generated by the machine learning model automatically adjusting the historical churn risk threshold information according to the churn urgency indicated by the propensity factor.

[0064] In some examples, techniques such as SVR (Support Vector Regression) can be used to build / train machine learning models. Mathematically, SVR achieves effective fitting of nonlinear relationships by finding an optimal hyperplane in a high-dimensional feature space that minimizes the error between the predicted and true values ​​within a tolerable range. Specific implementations may include: 1. Feature standardization: Before model training, standardize the information of the tendency influencing factors and the critical standard of historical loss risk (such as Z-score or Min-Max normalization) to eliminate the difference in the scale between different features and improve the convergence stability of the model. 2. Hyperparameter Optimization: Key parameters of the SVR, including the kernel function and penalty coefficient C, are tuned using methods such as grid search; among them, For the kernel function, RBF (Radial Basis Function) can be selected to enhance the model's ability to express complex nonlinear modes; The penalty coefficient C can be used to control the balance between the model's tolerance for error and its generalization ability; 3. Model Training: The SVR model can be trained using historical customer samples (including sample tendency influencing factors, historical churn risk threshold information and corresponding expected churn risk threshold information) to establish a mapping relationship from input features to output target (i.e., current churn risk threshold information).

[0065] In some examples, the churn risk threshold information in the embodiments of this application may include at least one of the following: numerical risk threshold (which can be used to determine whether the target customer has reached the risk threshold that requires triggering early warning intervention), segmented or graded threshold (which can be represented as multiple threshold intervals, corresponding to different risk levels), and combination of conditional rules (i.e. rule-based threshold; in some business scenarios, the churn risk threshold information can be represented as logical rules, for example: if the number of monthly active days is <3 and the number of customer service complaints is ≥2 and the package usage rate is <40%, then it is determined to be high risk).

[0066] In one optional implementation, the historical churn-related data includes historical churn risk threshold information; The early warning strategy is generated based at least on the directional matching relationship and the risk index information, combined with the historical churn-related data and / or the current churn risk threshold information, including: When the directional matching relationship indicates that the first risk trend direction and the second risk trend direction are in opposite directions, if the risk index information matches the historical loss risk critical standard information, a general early warning strategy is obtained, wherein the early warning strategy includes the general early warning strategy.

[0067] In some examples, when all the values ​​in the risk index information are equal to all the values ​​in the historical churn risk threshold information, it can be determined that the risk index information matches the historical churn risk threshold information.

[0068] And / or, When the directional matching relationship indicates that the first risk trend direction and the second risk trend direction are in a directional intersection relationship, a reference risk index information matching the risk index information is determined based on the historical churn risk critical standard information and the current churn risk critical standard information. If the risk index information matches the reference risk index information, a fusion early warning strategy is generated. The early warning strategy includes the fusion early warning strategy, which is formed by fusing multiple general early warning strategies.

[0069] In some examples, the historical churn risk threshold information and the current churn risk threshold information can be numerically decomposed in parallel with the risk dimension to which the risk index information belongs, thereby obtaining a reference risk index. When all the values ​​in the risk index information are equal to all the values ​​in the reference risk index information, it can be determined that the risk index information matches the reference risk index information.

[0070] And / or, When the directional matching relationship indicates that the first risk trend direction and the second risk trend direction are in the same direction, if the risk index information matches the current churn risk threshold information, then at least based on the basic information and behavioral characteristics of the target customer, a target early warning strategy for the target customer is generated through a large model, wherein the early warning strategy includes the target early warning strategy.

[0071] In some examples, when all the values ​​in the risk index information are equal to all the values ​​in the current churn risk threshold information, it can be determined that the risk index information matches the current churn risk threshold information.

[0072] In some examples, the target warning strategy can be obtained by inputting warning prompts, basic information and behavioral features into a large model. Here, warning prompts can be used to prompt the large model to analyze the basic information and behavioral features in order to generate the target warning strategy.

[0073] In some examples, the first risk trend direction can refer to the trend direction reflecting the evolution of the target customer churn risk, derived from information on changes in the target customer's behavior patterns during the target period. The second risk trend direction can refer to the direction of threshold change reflected by the current churn risk threshold information relative to the historical churn risk threshold information.

[0074] By comparing these two directions, three typical scenarios can be identified: Same direction relationship (e.g., both are rising): The two trends are completely consistent (e.g., both are rising or falling); Inverse relationship: The two trends are completely opposite (e.g., the rising part in one shows a falling part in the other). Directional intersection relationship: The trends of the two are partially consistent.

[0075] In some examples, the target early warning strategy can be a customized early warning strategy. To generate a customized early warning strategy, you can first build an internal database of successful customer retention cases (which may include retention measures and actual effect data for different customer types), and organize high-quality customer service dialogue records to extract effective communication scripts and service action materials to form rich training data.

[0076] Then, a personalized retention strategy content generation engine is built based on the large model. During the fine-tuning of the large model, training data such as customer characteristics, retention measures, effect feedback, and high-quality customer service dialogue records from the case library are input, allowing the large model to learn the mapping relationship between different customer characteristics and the optimal retention strategy. When the conditions for generating a customized strategy are triggered, basic customer information (such as consumption level, business preferences, and historical complaint records) and behavioral characteristics (such as high-frequency tariff complaints and a sudden drop in service usage frequency) can be input into the engine. This allows the large model to quickly generate a personalized customer retention plan by analyzing the retention paths of similar customers in the case library and combining it with the communication strategies learned after fine-tuning.

[0077] For example, for high-spending customers who frequently complain about network quality, the engine can generate customized strategies that include complimentary premium network acceleration services and dedicated account managers for network optimization follow-ups. This improves the relevance of the strategy to the customer's pain points and increases the success rate of customer retention.

[0078] In an optional implementation, the method further includes: Construct a causal relationship graph that matches the behavioral data; Based on the causal relationship diagram, the causal effect value corresponding to the key behavioral feature is determined, wherein the causal effect value characterizes the correlation between the key behavioral feature and customer churn; The step of generating a target early warning strategy for the target customer based at least on the target customer's basic information and behavioral characteristics through a large model includes: Based on the causal effect value, the basic information and behavioral characteristics of the target customer, the target early warning strategy is generated through a large model.

[0079] In some examples, using traffic usage data from behavioral data as an example, a causal relationship graph can be constructed to characterize the potential causal path between declining traffic usage and customer churn. For example, the following two causal relationship graphs can be constructed: (1) poor network quality → declining traffic usage → customer churn, (2) competitor's traffic packages are more affordable → declining traffic usage → customer churn.

[0080] In some examples, a causal relationship diagram can be used to determine the causal effect value for each key behavioral feature. This causal effect value can be used to characterize the intrinsic causal relationship between the key behavioral feature and customer churn. For example: For customers whose traffic usage decreases due to poor network quality, the causal effect value is high, indicating that network quality problems are the main cause. For customers whose data usage decreased due to the attraction of competing packages, the causal effect value is also high, but the underlying causes are different.

[0081] In this way, we can not only clarify the correlation between features and churn results, but also uncover the underlying causal mechanism of features causing churn, providing a more accurate basis for the formulation of subsequent tiered early warning strategies.

[0082] In some examples, based on the causal effect value, the target customer's basic information, and behavioral characteristics, a large model can be used to generate targeted early warning strategies for the target customer. For example, for customers whose data usage has decreased due to poor network quality, network service optimization can be prioritized to reduce the risk of churn; for customers whose data usage has decreased due to the attractiveness of competing packages, a more competitive package strategy can be launched.

[0083] In some examples, to further improve the effectiveness and long-term returns of the early warning strategy, this embodiment may also introduce a reinforcement learning framework, as detailed below.

[0084] 1. Define actions and rewards Treat routine customer maintenance measures (such as reminders of package offers and general customer service follow-ups) as actions, and reward customers' subsequent feedback (such as retention rates after accepting offers and satisfaction with follow-ups). For example: Actions: Send promotional reminders and arrange for customer service to follow up; Rewards: Whether customers accept the offer and continue to stay, and their satisfaction rating during follow-up visits.

[0085] 2. Update the policy network Reinforcement learning frameworks can continuously learn the long-term rewards of different actions for specific types of customers. For example: For ordinary risk customers who are highly sensitive to pricing, strategy networks can gradually explore measures such as personalized discount reminders combined with light pricing explanations and more personalized intelligent customer service follow-ups, maximizing the long-term returns of general maintenance measures while avoiding causing customer resentment.

[0086] Thus, this embodiment can autonomously discover which retention strategy yields the highest long-term return for a specific type of ordinary risk customer, thereby pushing the optimal general customer maintenance measures in a targeted manner, which can improve customer retention rate while avoiding resource waste and customer resentment.

[0087] Secondly, correspondingly, this application embodiment also provides a customer churn early warning device, which can implement all the processes of the customer churn early warning method provided in the above embodiments.

[0088] See Figure 2 The diagram shows a schematic representation of a customer churn warning device 200 provided in an embodiment of this application. The customer churn warning device 200 includes: The data processing module 201 is used to acquire the behavioral data of the target customer and determine the key behavioral features corresponding to the behavioral data. The behavior pattern recognition module 202 is used to determine the behavior pattern change information of the target customer within a target time period based on the key behavioral features; The risk index determination module 203 is used to determine the risk index information of the target customer based on the behavior pattern change information, the key behavior characteristics, and the benchmark characteristic information associated with the key behavior characteristics, when the behavior pattern change information meets the preset pattern deviation conditions. The early warning strategy acquisition module 204 is used to acquire an early warning strategy based at least on the key behavioral characteristics, the risk index information and its corresponding first risk trend direction, and the historical churn-related data of the target customer's customer group. The message push module 205 is used to push retention information messages corresponding to the early warning strategy to the target customer.

[0089] In one optional implementation, determining the risk index information of the target customer based on the behavioral pattern change information, the key behavioral characteristics, and the benchmark characteristic information associated with the key behavioral characteristics includes: Based on the key behavioral characteristics, determine the behavioral characteristic parameters of the target customer during the target time period; Based on the behavioral characteristic parameters, at least one trend period is determined from the target time period, wherein the at least one trend period includes a period in which the parameter changes of the behavioral characteristic parameters continuously decay and / or a period in which the parameter changes in a stepwise manner decay. Based on the rate of change of the behavioral feature parameters in each trend period, preprocessed feature data is determined, wherein the preprocessed feature data includes at least a portion of the rate of change extreme value information of the parameter change rate; Based on the key behavioral characteristics, the baseline characteristic information, the behavioral pattern change information, and the preprocessed characteristic data, the risk index information of the target customer is determined.

[0090] In one optional implementation, determining the risk index information of the target customer based on the key behavioral characteristics, the baseline characteristic information, the behavioral pattern change information, and the preprocessed characteristic data includes: Based on the key behavioral features and the baseline feature information, feature offset information matching the preprocessed feature data is determined; Based on the feature offset information, the behavioral pattern change information, and the preprocessed feature data, the first churn risk information of the target customer is predicted. Based on the feature weights associated with the key behavioral characteristics, the first churn risk information is weighted to obtain the second churn risk information. Based on the second loss risk information and the preset characteristics of different risk levels, the risk index information is determined.

[0091] In one optional implementation, the step of obtaining an early warning strategy based at least on the key behavioral characteristics, the risk index information and its corresponding first risk trend direction, and the historical churn data of the target customer's customer group includes: Determine the degree of variation of the key behavioral features; If the degree of variation of the key behavioral features is greater than a preset degree of variation threshold, the target customer is marked as having a tendency to churn. Furthermore, based on the cumulative number of times the target customer is marked as having a tendency to churn and the type of churn tendency behavior corresponding to the cumulative number of times, the marking distribution features are determined. When the marked distribution characteristics meet the risk distribution conditions, the tendency influence factor is determined based on the cumulative number of times and the cumulative number of times threshold, as well as the churn tendency behavior type and the preset churn tendency behavior type set; Based on the aforementioned tendency influencing factors and the historical churn-related data, a machine learning model is invoked to generate current churn risk threshold information. Determine the second risk trend direction corresponding to the current critical churn risk standard information, and determine the directional matching relationship between the first risk trend direction and the second risk trend direction; The early warning strategy is generated based at least on the directional matching relationship and the risk index information, combined with the historical churn-related data and / or the current churn risk threshold information.

[0092] In one optional implementation, the historical churn-related data includes historical churn risk threshold information; The early warning strategy is generated based at least on the directional matching relationship and the risk index information, combined with the historical churn-related data and / or the current churn risk threshold information, including: When the directional matching relationship indicates that the first risk trend direction and the second risk trend direction are in opposite directions, if the risk index information matches the historical loss risk critical standard information, a general early warning strategy is obtained, wherein the early warning strategy includes the general early warning strategy; And / or, When the directional matching relationship indicates that the first risk trend direction and the second risk trend direction are in a directional intersection relationship, a reference risk index information matching the risk index information is determined based on the historical churn risk critical standard information and the current churn risk critical standard information. If the risk index information matches the reference risk index information, a fusion early warning strategy is generated. The early warning strategy includes the fusion early warning strategy, which is formed by fusing multiple general early warning strategies. And / or, When the directional matching relationship indicates that the first risk trend direction and the second risk trend direction are in the same direction, if the risk index information matches the current churn risk threshold information, then at least based on the basic information and behavioral characteristics of the target customer, a target early warning strategy for the target customer is generated through a large model, wherein the early warning strategy includes the target early warning strategy.

[0093] In an optional embodiment, the apparatus further includes a causal processing module, the causal processing module being used to: Construct a causal relationship graph that matches the behavioral data; Based on the causal relationship diagram, the causal effect value corresponding to the key behavioral feature is determined, wherein the causal effect value characterizes the correlation between the key behavioral feature and customer churn; The step of generating a target early warning strategy for the target customer based at least on the target customer's basic information and behavioral characteristics through a large model includes: Based on the causal effect value, the basic information and behavioral characteristics of the target customer, the target early warning strategy is generated through a large model.

[0094] Thirdly, embodiments of this application provide a non-transitory computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the method described in any of the above-mentioned embodiments.

[0095] Fourthly, embodiments of this application provide a computer program product, including computer instructions that, when executed by a processor, implement the steps of the method described in any of the above-described embodiments.

[0096] Fifthly, embodiments of this application provide a computer device including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor executes the computer program to implement the steps of the method described in any of the preceding claims.

[0097] See Figure 3 The computer device in this embodiment includes a processor 301, a memory 302, and a computer program stored in the memory 302 and executable on the processor 301, such as a customer churn warning program. When the processor 301 executes the computer program, it implements the steps in the various customer churn warning method embodiments described above, for example... Figure 1 The steps S101-S105 are shown.

[0098] For example, the computer program may be divided into one or more modules / units, which are stored in the memory 302 and executed by the processor 301 to complete this application. The one or more modules / units may be a series of computer program instruction segments capable of performing a specific function, which describe the execution process of the computer program in the computer device.

[0099] The computer device may be a desktop computer, laptop, handheld computer, or cloud server, etc. The computer device may include, but is not limited to, a processor 301 and a memory 302. Those skilled in the art will understand that the schematic diagram is merely an example of a computer device and does not constitute a limitation on the computer device. It may include more or fewer components than shown, or combine certain components, or different components. For example, the computer device may also include input / output devices, network access devices, buses, etc.

[0100] The processor 301 can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor, or the processor 301 can be any conventional processor. The processor 301 is the control center of the computer device, connecting various parts of the entire computer device through various interfaces and lines.

[0101] The memory 302 can be used to store the computer programs and / or modules. The processor 301 implements various functions of the computer device by running or executing the computer programs and / or modules stored in the memory 302 and calling the data stored in the memory 302. The memory 302 may mainly include a program storage area and a data storage area. The program storage area may store the operating system, at least one application program required for a function (such as sound playback function, image playback function, etc.), etc.; the data storage area may store data created according to the use of the mobile phone (such as audio data, phonebook, etc.). In addition, the memory 302 may include high-speed random access memory, and may also include non-volatile memory, such as hard disk, memory, plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0102] Wherein, if the modules / units integrated into the computer device are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments can also be implemented by a computer program instructing related hardware. The computer program can be stored in a non-transitory computer-readable storage medium. When the computer program is executed by the processor 301, it can implement the steps of the various method embodiments described above. Wherein, the computer program includes computer program code, which can be in the form of source code, object code, executable file, or some intermediate form, etc. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording medium, USB flash drive, portable hard drive, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signal, telecommunication signal, and software distribution medium, etc.

[0103] In summary, the embodiments of this application have at least the following beneficial effects: By employing the embodiments of this application, behavioral data of target customers is acquired, and key behavioral characteristics corresponding to the behavioral data are determined. Based on the key behavioral characteristics, behavioral pattern change information of the target customer within a target time period is determined. When the behavioral pattern change information meets preset pattern deviation conditions, risk index information of the target customer is determined based on the behavioral pattern change information, the key behavioral characteristics, and benchmark feature information associated with the key behavioral characteristics. At least based on the key behavioral characteristics, the risk index information and its corresponding first risk trend direction, and historical churn-related data of the target customer's customer group, an early warning strategy is obtained. Retention information messages corresponding to the early warning strategy are pushed to the target customer. In this way, key behavioral characteristics can be identified based on customer behavioral data to improve the accuracy of behavioral pattern change identification. When behavioral model deviation is identified, the risk index of the target customer is accurately analyzed by comprehensively considering behavioral pattern change information, key behavioral characteristics, and benchmark feature information. Furthermore, based on the key behavioral characteristics, the risk index information and its risk trend direction, as well as historical churn-related data of the target customer's customer group, are further considered to accurately obtain early warning strategies, thereby improving the accuracy of churn risk early warning and the accuracy of pushing retention information messages to customers with high churn risk.

[0104] Through the above description of the embodiments, those skilled in the art can clearly understand that this application can be implemented by means of software plus necessary hardware platforms, or it can be implemented entirely by hardware. Based on this understanding, all or part of the technical solutions of this application that contribute to the background technology can be embodied in the form of a software product. This computer software product can be stored in a storage medium, such as ROM (Read-Only Memory) / RAM (Random Access Memory), magnetic disk, optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in various embodiments or some parts of the embodiments of this application.

[0105] The above description is the preferred embodiment of this application. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of this application, and these improvements and modifications are also considered to be within the scope of protection of this application.

Claims

1. A customer churn early warning method, characterized in that, include: Acquire behavioral data of target customers and determine the key behavioral characteristics corresponding to the behavioral data; Based on the key behavioral characteristics, determine the behavioral pattern changes of the target customer within the target time period; If the behavioral pattern change information meets the preset pattern deviation conditions, the risk index information of the target customer is determined based on the behavioral pattern change information, the key behavioral characteristics, and the benchmark characteristic information associated with the key behavioral characteristics. Based at least on the key behavioral characteristics, the risk index information and its corresponding first risk trend direction, and the historical churn data of the target customer's customer group, an early warning strategy is obtained. Push retention information messages corresponding to the early warning strategy to the target customers.

2. The method according to claim 1, characterized in that, The determination of the risk index information of the target customer based on the behavioral pattern change information, the key behavioral characteristics, and the benchmark characteristic information associated with the key behavioral characteristics includes: Based on the key behavioral characteristics, determine the behavioral characteristic parameters of the target customer during the target time period; Based on the behavioral characteristic parameters, at least one trend period is determined from the target time period, wherein the at least one trend period includes a period in which the parameter changes of the behavioral characteristic parameters continuously decay and / or a period in which the parameter changes in a stepwise manner decay. Based on the rate of change of the behavioral feature parameters in each trend period, preprocessed feature data is determined, wherein the preprocessed feature data includes at least a portion of the rate of change extreme value information of the parameter change rate; Based on the key behavioral characteristics, the baseline characteristic information, the behavioral pattern change information, and the preprocessed characteristic data, the risk index information of the target customer is determined.

3. The method according to claim 2, characterized in that, The process of determining the risk index information of the target customer based on the key behavioral characteristics, the baseline characteristic information, the behavioral pattern change information, and the preprocessed characteristic data includes: Based on the key behavioral features and the baseline feature information, feature offset information matching the preprocessed feature data is determined; Based on the feature offset information, the behavioral pattern change information, and the preprocessed feature data, the first churn risk information of the target customer is predicted. Based on the feature weights associated with the key behavioral characteristics, the first churn risk information is weighted to obtain the second churn risk information. Based on the second loss risk information and the preset characteristics of different risk levels, the risk index information is determined.

4. The method according to claim 1, characterized in that, The method for obtaining an early warning strategy, based at least on the key behavioral characteristics, the risk index information and its corresponding first risk trend direction, and the historical churn data of the target customer's customer group, includes: Determine the degree of variation of the key behavioral features; If the degree of variation of the key behavioral features is greater than a preset degree of variation threshold, the target customer is marked as having a tendency to churn. Furthermore, based on the cumulative number of times the target customer is marked as having a tendency to churn and the type of churn tendency behavior corresponding to the cumulative number of times, the marking distribution features are determined. When the marked distribution characteristics meet the risk distribution conditions, the tendency influence factor is determined based on the cumulative number of times and the cumulative number of times threshold, as well as the churn tendency behavior type and the preset churn tendency behavior type set; Based on the aforementioned tendency influencing factors and the historical churn-related data, a machine learning model is invoked to generate current churn risk threshold information. Determine the second risk trend direction corresponding to the current critical churn risk standard information, and determine the directional matching relationship between the first risk trend direction and the second risk trend direction; The early warning strategy is generated based at least on the directional matching relationship and the risk index information, combined with the historical churn-related data and / or the current churn risk threshold information.

5. The method according to claim 4, characterized in that, The historical data related to data loss includes information on historical risk threshold standards for data loss. The early warning strategy is generated based at least on the directional matching relationship and the risk index information, combined with the historical churn-related data and / or the current churn risk threshold information, including: When the directional matching relationship indicates that the first risk trend direction and the second risk trend direction are in opposite directions, if the risk index information matches the historical loss risk critical standard information, a general early warning strategy is obtained, wherein the early warning strategy includes the general early warning strategy; And / or, When the directional matching relationship indicates that the first risk trend direction and the second risk trend direction are in a directional intersection relationship, a reference risk index information matching the risk index information is determined based on the historical churn risk critical standard information and the current churn risk critical standard information. If the risk index information matches the reference risk index information, a fusion early warning strategy is generated. The early warning strategy includes the fusion early warning strategy, which is formed by fusing multiple general early warning strategies. And / or, When the directional matching relationship indicates that the first risk trend direction and the second risk trend direction are in the same direction, if the risk index information matches the current churn risk threshold information, then at least based on the basic information and behavioral characteristics of the target customer, a target early warning strategy for the target customer is generated through a large model, wherein the early warning strategy includes the target early warning strategy.

6. The method according to claim 5, characterized in that, The method further includes: Construct a causal relationship graph that matches the behavioral data; Based on the causal relationship diagram, the causal effect value corresponding to the key behavioral feature is determined, wherein the causal effect value characterizes the correlation between the key behavioral feature and customer churn; The step of generating a target early warning strategy for the target customer based at least on the target customer's basic information and behavioral characteristics through a large model includes: Based on the causal effect value, the basic information and behavioral characteristics of the target customer, the target early warning strategy is generated through a large model.

7. A customer churn early warning device, characterized in that, include: The data processing module is used to acquire the behavioral data of the target customer and determine the key behavioral features corresponding to the behavioral data. The behavior pattern recognition module is used to determine the behavior pattern change information of the target customer within a target time period based on the key behavioral features. The risk index determination module is used to determine the risk index information of the target customer based on the behavioral pattern change information, the key behavioral characteristics, and the benchmark characteristic information associated with the key behavioral characteristics, when the behavioral pattern change information meets the preset pattern deviation conditions. The early warning strategy acquisition module is used to acquire an early warning strategy based at least on the key behavioral characteristics, the risk index information and its corresponding first risk trend direction, and the historical churn data of the target customer's customer group. The message push module is used to push retention information messages corresponding to the early warning strategy to the target customers.

8. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method described in any one of claims 1-6.

9. A computer program product comprising computer instructions, characterized in that, When the computer instructions are executed by the processor, they implement the method described in any one of claims 1-6.

10. A computer device, characterized in that, The method includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor, when executing the computer program, implements the method of any one of claims 1-6.

Images