Marketing sensitive personnel analysis method and device, electronic equipment and storage medium
By constructing a marketing gain model based on A/B testing and multiple learning algorithms, we can identify marketing-sensitive groups, solve the problem of resource waste in traditional marketing methods, and achieve precise marketing targeting.
Patent Information
- Application Number
- CN202511050078.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-07-29
- Publication Date
- 2025-11-14
AI Technical Summary
Traditional marketing methods cannot achieve precise marketing targeting, resulting in a significant waste of marketing resources on ineffective audiences.
By integrating various historical customer characteristics to construct a training sample set, a marketing gain model is trained based on A/B experiments and multiple learning algorithms and meta-learners. The model with the highest evaluation index is selected for marketing-sensitive audience analysis.
Effectively identify marketing-sensitive groups, avoid wasting marketing resources, and improve the efficiency of marketing campaigns.
Smart Images
Figure CN120952852A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of artificial intelligence technology, and more specifically, to a method, apparatus, electronic device, and storage medium for analyzing marketing-sensitive groups. Background Technology
[0002] With the development of mobile internet and artificial intelligence technologies, the marketing market is becoming increasingly complex and volatile, and consumers' consumption concepts and behaviors have also undergone tremendous changes. Faced with these diverse and complex consumption patterns and behaviors, marketers find that the traditional "broad-based" approach, targeting all users without differentiation, fails to achieve precise marketing. This results in a significant waste of marketing resources being spent on groups that are already insensitive to the campaign. Summary of the Invention
[0003] In view of this, this application provides a marketing-sensitive audience analysis method, apparatus, electronic device, and storage medium for identifying marketing-sensitive audiences to prevent marketers from wasting a large amount of marketing resources on ineffective audiences.
[0004] To achieve the above objectives, the following solution is proposed:
[0005] A marketing-sensitive audience analysis method, applied to electronic devices, is used to analyze marketing-sensitive audiences based on a marketing gain prediction model constructed using A / B experiments and a marketing gain model. The marketing-sensitive audience analysis method includes the following steps:
[0006] A training sample set is constructed by integrating various historical feature information of customers;
[0007] Based on A / B experiments, multiple marketing gain models were obtained by training the models using various learning algorithms and meta-learners.
[0008] The prediction model with the highest evaluation index among the multiple prediction models is selected as the optimal model;
[0009] Based on the optimal model, predictive processing is performed on the characteristic information of multiple marketing targets of the customer to determine whether the customer is a sensitive person who can influence the customer.
[0010] Optionally, the various historical characteristics include customer identity information, recent login preferences, device model of the login terminal, duration of financial instrument application, and age.
[0011] Optionally, the various learning algorithms include the LightGBM algorithm, the RandomForest algorithm, and the CatBoost algorithm.
[0012] Optionally, the multi-meta-learners include S-Learner, T-Learner, and X-Learner.
[0013] Optionally, selecting the prediction model with the highest evaluation index among the multiple marketing gain models as the optimal model includes the following steps:
[0014] Calculate the evaluation metrics for each of the aforementioned marketing gain models;
[0015] The marketing gain model with the highest evaluation index is determined as the so-called optimal model.
[0016] Optionally, the evaluation index is the Qini coefficient.
[0017] A marketing-sensitive audience analysis device, applied to an electronic device, is used to analyze marketing-sensitive audiences based on a marketing gain prediction model constructed using A / B experiments and a marketing gain model. The marketing-sensitive audience analysis device includes:
[0018] The sample construction module is configured to build a training sample set by integrating various historical feature information of the customer.
[0019] The model training module is configured to train models based on A / B experiments, using multiple learning algorithms and meta-learners to obtain multiple marketing gain models.
[0020] The model selection module is configured to select the prediction model with the highest evaluation index among the multiple marketing gain models as the optimal model.
[0021] The analysis and execution module is configured to perform predictive processing based on the optimal marketing target feature information of the customer to determine whether the customer is a sensitive person.
[0022] Optionally, the model selection module is configured as follows:
[0023] Calculate the evaluation metrics for each of the aforementioned marketing gain models;
[0024] The marketing gain model with the highest evaluation index is determined as the so-called optimal model.
[0025] An electronic device includes at least one processor and a memory connected to the processor, wherein:
[0026] The storage device is used to store computer programs or instructions;
[0027] The processor is used to execute the computer program or instructions to enable the electronic device to implement the marketing-sensitive person analysis method as described above.
[0028] A computer-readable storage medium is applied to an electronic device, the storage medium carrying one or more computer programs that can be executed by the electronic device, thereby enabling the electronic device to implement the marketing-sensitive person analysis method as described above.
[0029] As can be seen from the above technical solution, this application discloses a method, apparatus, electronic device, and storage medium for analyzing marketing-sensitive individuals. This method and apparatus are applied to an electronic device for analyzing marketing-sensitive individuals based on a marketing gain prediction model constructed using A / B experiments and a marketing gain model. Specifically, it involves constructing a training sample set by integrating various historical characteristic information of customers; training the model based on A / B experiments using multiple learning algorithms and meta-learners to obtain multiple marketing gain models; selecting the prediction model with the highest evaluation index among the multiple prediction models as the optimal model; and performing predictive processing on multiple marketing target characteristic information of the customer based on the optimal model to determine whether the customer is a marketing-sensitive individual. By identifying marketing-sensitive individuals, marketers can avoid wasting marketing resources by applying a large amount of marketing resources to ineffective individuals, thus preventing the waste of marketing resources. Attached Figure Description
[0030] To more clearly illustrate the technical solutions in the embodiments of this application 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 only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0031] Figure 1 This is a flowchart illustrating a marketing-sensitive personnel analysis method according to an embodiment of this application;
[0032] Figure 2 This is a block diagram of a marketing-sensitive personnel analysis device according to an embodiment of this application;
[0033] Figure 3 This is a block diagram of an electronic device according to an embodiment of this application. Detailed Implementation
[0034] 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 are only some embodiments of this application, and not all embodiments. Based on the embodiments of 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.
[0035] Figure 1 This is a flowchart illustrating a marketing-sensitive personnel analysis method according to an embodiment of this application.
[0036] like Figure 1 As shown, the marketing-sensitive personnel analysis method provided in this embodiment is applied to an electronic device. It is used to analyze marketing-sensitive individuals based on a marketing gain prediction model constructed using A / B experiments and a marketing gain model, thereby identifying individuals sensitive to marketing strategies. This electronic device can be understood as a computer, server, or cloud platform with data computing and information processing capabilities. The marketing-sensitive personnel analysis method includes the following steps:
[0037] S1. Construct a training sample set by integrating various historical feature information of customers.
[0038] This training sample set is used to train subsequent prediction models. The various historical features include, but are not limited to, customer identity information, recent login preferences, device model of the login terminal, duration of financial instrument use, and age. Customer identity information refers to an individual's name, age, occupation, address, and contact information. Login preferences refer to the characteristics of their login behavior on selected marketing websites, such as login time and frequency.
[0039] S2. Train multiple marketing gain models based on various learning algorithms and meta-learners.
[0040] Specifically, based on A / B testing, multiple marketing gain models were trained using various learning algorithms and meta-learners. These learning algorithms included LightGBM, RandomForest, and CatBoost, while the meta-learners included S-Learner, T-Learner, and X-Learner.
[0041] A / B testing, also known as split testing or controlled experiments, is a method for evaluating the impact of a variable on test results by randomly assigning test subjects to two or more different experimental groups. It is commonly used in fields such as user experience, marketing, and product feature testing.
[0042] A / B testing is a common experimental method used to compare the effectiveness of different strategies or versions. In marketing, A / B testing typically divides users into two groups: an experimental group and a control group, each receiving different marketing strategies or product versions. The effectiveness of the strategy is then evaluated by comparing metrics such as the achievement rates of the two groups. For example, in one round of A / B testing, the marketing experimental group achieved a 0.8775% achievement rate in browsing the "Positioning Returns" page of the mobile banking app, while the control experimental group achieved only 0.458%, a significant difference. By combining multiple rounds of A / B testing with the Uplift marketing gain model, marketing strategies can be further optimized to improve effectiveness.
[0043] The uplift model (marketing gain model), also known as the reverse causal model or individual effect model, is a statistical model used to predict the effects of interventions on individual decision-making. Based on A / B experiments, it measures the impact of marketing strategies, behavioral interventions, or any other actions on individual behavior, predicting the causal effect of an intervention on an individual's state or behavior. The uplift model predicts the incremental value, i.e., the lift portion, calculated using the following formula:
[0044] Lift=P(buy / treatment)-P(buy / no treatment)
[0045] Traditional models typically predict the target directly:
[0046] Outcome = P(buy / treatment)
[0047] Mathematically, this can be expressed as the difference between two conditional probabilities, as shown below:
[0048] P(Yi / Xi,Ti=1)-P(Yi / Xi,Ti=0)
[0049] Where Y represents the outcome (such as user browsing, clicks, conversions, etc.), X represents user-dimensional features, and T represents marketing variables (1 represents intervention, 0 represents no intervention). Therefore, the calculated probability difference represents the changes in users under conditions of intervention and no intervention.
[0050] The core idea of the Uplift model is to measure the impact of an intervention on individual behavior by comparing two groups of individuals (e.g., a marketing group and a control group, where marketing is used for the intervention and the control group does not). Unlike traditional predictive models, which focus on predicting the behavior of a single individual, the Uplift model focuses on predicting the behavioral changes resulting from the intervention. Uplift models are typically built based on experimental or observational data. In experimental settings, individuals are randomly assigned to marketing and control groups, and their behavioral differences are observed. In observational data, individuals are identified by whether they engaged in a particular marketing activity to determine which individuals received the marketing.
[0051] The Uplift model development process includes feature selection, causal inference, modeling, and evaluation. In the feature selection phase, features relevant to marketing effectiveness need to be chosen; these features may include basic individual attributes, consumer behavior, and intervention-related factors. In the causal inference phase, statistical methods are used to evaluate causal effects. In the modeling phase, commonly used modeling methods include logistic regression, random forests, and gradient boosting trees. Finally, in the evaluation phase, evaluation metrics are used to assess the model's accuracy and predictive ability.
[0052] Uplift models are typically used to estimate CATE / ITE, which assess the effect of a treatment or intervention. CATE refers to the average treatment effect of a treatment or intervention given the observed variables. Its core concept is conditional averaging, i.e., the average treatment effect given specific conditions. ITE is more individualized, referring to the effect of a treatment or intervention on a single individual. In some causal inference tasks, we are more concerned with the potential difference in outcomes for each individual given or without the treatment. Based on these models, we can better understand the impact of a specific treatment or intervention on different individuals or different situations. Basic Uplift modeling is implemented through the following three meta-learners:
[0053] (1) Two-Model (T-LEARNER).
[0054] For a single treatment, the dual-model approach involves building separate models for the experimental and control groups, and then subtracting the predictions from the two models to obtain the lift. For multiple treatments, separate models are built for each experimental and control group, and then the predictions for each experimental group are subtracted from those for the control group to obtain the lift for each experimental group.
[0055] Advantages: The advantage of dual models is that they are simple and intuitive, and can reuse common machine learning models (LR, Tree Model, NN).
[0056] Disadvantages: Insufficient data utilization, poor fit to the differences between the two groups (i.e., lift signal), and the model error is easily amplified.
[0057] (2)One-Model(S-LEARNER).
[0058] The single-model approach involves adding the intervention attribute W to the sample features of both the experimental and control groups, merging the experimental and control groups, and training them using the same model.
[0059] Compared to dual-model approaches, single-model approaches offer several advantages: Data utilization is more efficient during model training. Modeling is simpler, requiring only a simple logistic regression or tree model (random forest, XGboost, LightGBM). It allows for the imposition of strict monotonic constraints on the processing variable or other variables, something dual-model approaches cannot do.
[0060] (3) X-Learner model.
[0061] X-Learner is a fusion of T-Learner and S-Learner, and its steps are as follows:
[0062] First, X-Learner will perform predictive modeling on the experimental group and the control group respectively, resulting in two pre-trained models;
[0063] Next, X-Learner uses the prediction results from the first step to predict the data for the control group using the experimental group's model, and simultaneously uses the control group's model to predict the data for the experimental group. Then, it calculates the difference between these predictions and the actual results to obtain the estimated effect.
[0064] In the third step, X-Learner uses preprocessed variables to model the predicted effects of the forecasts calculated in the second step.
[0065] Finally, X-Learner uses the model learned in step three to predict the predicted individual effects, obtaining an estimate of the predicted effect for each individual.
[0066] The main advantage of X-Learner is that it addresses the imbalance between the control and experimental groups. However, a major drawback of X-Learner is its use of multiple models, which introduces cumulative error. Furthermore, X-Learner results also depend on the accuracy of the predictive models; poor performance of these models can lead to inaccurate estimates of the predicted effects.
[0067] The machine learning process is implemented based on the following three learning algorithms.
[0068] (1) LightGBM algorithm.
[0069] LightGBM, short for Light Grandient Boosting Machine, is a Gradient Boosting Decision Tree (GBDT) algorithm framework that supports highly efficient parallel training and features faster training speed, lower memory consumption, better accuracy, and distributed processing capabilities for handling massive amounts of data quickly.
[0070] (2) Random Forest Algorithm.
[0071] Random Forest is a parallel ensemble learning method, an extension of decision trees. It is based on decision trees, and multiple trees form a forest. The randomness lies in the randomness of the selection of the splitting attribute. When training machine learning, Random Forest also uses sampling with replacement to add sample perturbation. At the same time, it introduces an attribute perturbation. When splitting attributes during the training of decision trees, Random Forest first randomly selects a subset containing K attributes from the candidate attribute set, and then selects the optimal splitting attribute from this subset.
[0072] (3) Catoost algorithm.
[0073] CatBoost is a machine learning library open-sourced by Russian giant Yandex in 2017. It is a GBDT framework based on decision trees (oblivious trees), with fewer parameters, support for categorical variables, and high accuracy. Its main pain point is the efficient and reasonable processing of categorical features, as its name suggests. CatBoost consists of categorical features and gradient boosting. In addition, CatBoost also solves the problems of gradient bias and prediction shift, thereby reducing overfitting and improving the accuracy and generalization ability of the algorithm.
[0074] S3. The marketing gain model with the highest evaluation index is selected as the optimal model.
[0075] This involves determining the optimal model by evaluating multiple marketing gain models. Specifically, the model with the highest evaluation index among the multiple prediction models is selected as the optimal model. The evaluation index selected in this application is the Qini coefficient. That is, the Qini coefficient is determined by evaluating the Qini curve of each marketing gain model, and then the optimal model is selected based on this coefficient.
[0076] The Qini curve is a tool used to evaluate the performance of uplift models. It is based on the assumption that customer responses are random in the absence of any intervention. Therefore, if we rank customers according to the degree to which they are likely to be affected by the intervention, and start applying the intervention to the most affected customers, we expect that initially, the intervention will significantly outperform random intervention. Then, as the intervention is applied to more and more customers, this outperformance gradually decreases, and eventually, when all customers are affected, the intervention's effect is the same as that of random intervention.
[0077] The Qini curve reflects this relationship by showing the proportion of intervention (from 0 to 1) on the horizontal axis and the effect of intervention (gain relative to random intervention) on the vertical axis. An ideal Qini curve starts at the origin, rises rapidly, then flattens out, and ends in the upper right corner. The closer a model's Qini curve is to the ideal curve, the better the model's performance.
[0078] A key characteristic of the Qini curve is its ability to visually represent model performance, rather than simply providing a single number. However, this also means that the Qini curve can be more difficult to interpret and understand than some other evaluation metrics, such as precision, recall, or the area under the ROC curve. Below is the formula for calculating the Qini curve:
[0079] Qini=nt,1(φ) / Nt-nc,1(φ) / Nc
[0080] Where, n t,1 (φ) and n c,1 (φ) represents the number of people with an outcome of 1 in the control group and the control group, respectively. The fraction φ represents the ratio of the observed population to the target population. Nt and Nc represent the total number of people in the experimental group and the control group (independent of φ).
[0081] S4. Analyze customers based on this optimal model.
[0082] This involves using the optimal model to predict and process multiple marketing target characteristics of a customer to determine whether the customer is a sensitive individual to influence. These multiple marketing target characteristics include, but are not limited to, the customer's identity information, recent login preferences, device model of the login terminal, duration of use of financial instruments, and age.
[0083] As can be seen from the above technical solution, this embodiment provides a method for analyzing marketing-sensitive individuals. This method is applied to electronic devices and is used to analyze marketing-sensitive individuals based on a marketing gain prediction model constructed based on A / B experiments and a marketing gain model. Specifically, it involves constructing a training sample set by integrating various historical characteristic information of customers; training the model based on A / B experiments using multiple learning algorithms and meta-learners to obtain multiple marketing gain models; selecting the prediction model with the highest evaluation index among the multiple prediction models as the optimal model; and performing prediction processing on multiple marketing target characteristic information of the customer based on the optimal model to determine whether the customer is a marketing-sensitive individual. By identifying marketing-sensitive individuals, marketers can avoid applying a large amount of marketing resources to ineffective individuals, thereby avoiding the waste of marketing resources.
[0084] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.
[0085] Although the operations are described in a specific order, this should not be construed as requiring these operations to be performed in the specific order shown or in a sequential order. In certain environments, multitasking and parallel processing may be advantageous.
[0086] It should be understood that the steps described in the method embodiments of this disclosure may be performed in different orders and / or in parallel. Furthermore, the method embodiments may include additional steps and / or omit the steps shown. The scope of this disclosure is not limited in this respect.
[0087] Computer program code for performing the operations of this disclosure can be written in one or more programming languages or a combination thereof, including but not limited to object-oriented programming languages such as Java, Smalltalk, and C++, as well as conventional procedural programming languages such as C or similar languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer.
[0088] Figure 2 This is a block diagram of a marketing-sensitive personnel analysis device according to an embodiment of this application.
[0089] like Figure 2As shown, the marketing-sensitive personnel analysis device provided in this embodiment is applied to an electronic device for analyzing marketing-sensitive groups based on a marketing gain prediction model constructed using A / B experiments and a marketing gain model, thereby identifying individuals sensitive to marketing strategies. This electronic device can be understood as a computer, server, or cloud platform with data computing and information processing capabilities. The marketing-sensitive personnel analysis device includes a sample construction module 10, a model training module 20, a model selection module 30, and an analysis execution module 40.
[0090] The sample construction module is used to build a training sample set by integrating various historical feature information of customers.
[0091] This training sample set is used to train subsequent prediction models. The various historical features include, but are not limited to, customer identity information, recent login preferences, device model of the login terminal, duration of financial instrument use, and age. Customer identity information refers to an individual's name, age, occupation, address, and contact information. Login preferences refer to the characteristics of their login behavior on selected marketing websites, such as login time and frequency.
[0092] The model training module trains multiple marketing gain models each month using various learning algorithms and meta-learners.
[0093] Specifically, based on A / B testing, multiple marketing gain models were trained using various learning algorithms and meta-learners. These learning algorithms included LightGBM, RandomForest, and CatBoost, while the meta-learners included S-Learner, T-Learner, and X-Learner.
[0094] A / B testing, also known as split testing or controlled experiments, is a method for evaluating the impact of a variable on test results by randomly assigning test subjects to two or more different experimental groups. It is commonly used in fields such as user experience, marketing, and product feature testing.
[0095] A / B testing is a common experimental method used to compare the effectiveness of different strategies or versions. In marketing, A / B testing typically divides users into two groups: an experimental group and a control group, each receiving different marketing strategies or product versions. The effectiveness of the strategy is then evaluated by comparing metrics such as the achievement rates of the two groups. For example, in one round of A / B testing, the marketing experimental group achieved a 0.8775% achievement rate in browsing the "Positioning Returns" page of the mobile banking app, while the control experimental group achieved only 0.458%, a significant difference. By combining multiple rounds of A / B testing with the Uplift marketing gain model, marketing strategies can be further optimized to improve effectiveness.
[0096] The uplift model (marketing gain model), also known as the reverse causal model or individual effect model, is a statistical model used to predict the effects of interventions on individual decision-making. Based on A / B experiments, it measures the impact of marketing strategies, behavioral interventions, or any other actions on individual behavior, predicting the causal effect of an intervention on an individual's state or behavior. The uplift model predicts the incremental value, i.e., the lift portion, calculated using the following formula:
[0097] Lift=P(buy / treatment)-P(buy / no treatment)
[0098] Traditional models typically predict the target directly:
[0099] Outcome = P(buy / treatment)
[0100] Mathematically, this can be expressed as the difference between two conditional probabilities, as shown below:
[0101] P(Yi / Xi,Ti=1)-P(Yi / Xi,Ti=0)
[0102] Where Y represents the outcome (such as user browsing, clicks, conversions, etc.), X represents user-dimensional features, and T represents marketing variables (1 represents intervention, 0 represents no intervention). Therefore, the calculated probability difference represents the changes in users under conditions of intervention and no intervention.
[0103] The core idea of the Uplift model is to measure the impact of an intervention on individual behavior by comparing two groups of individuals (e.g., a marketing group and a control group, where marketing is used for the intervention and the control group does not). Unlike traditional predictive models, which focus on predicting the behavior of a single individual, the Uplift model focuses on predicting the behavioral changes resulting from the intervention. Uplift models are typically built based on experimental or observational data. In experimental settings, individuals are randomly assigned to marketing and control groups, and their behavioral differences are observed. In observational data, individuals are identified by whether they engaged in a particular marketing activity to determine which individuals received the marketing.
[0104] The Uplift model development process includes feature selection, causal inference, modeling, and evaluation. In the feature selection phase, features relevant to marketing effectiveness need to be chosen; these features may include basic individual attributes, consumer behavior, and intervention-related factors. In the causal inference phase, statistical methods are used to evaluate causal effects. In the modeling phase, commonly used modeling methods include logistic regression, random forests, and gradient boosting trees. Finally, in the evaluation phase, evaluation metrics are used to assess the model's accuracy and predictive ability.
[0105] Uplift models are typically used to estimate CATE / ITE, which assess the effect of a treatment or intervention. CATE refers to the average treatment effect of a treatment or intervention given the observed variables. Its core concept is conditional averaging, i.e., the average treatment effect given specific conditions. ITE is more individualized, referring to the effect of a treatment or intervention on a single individual. In some causal inference tasks, we are more concerned with the potential difference in outcomes for each individual given or without the treatment. Based on these models, we can better understand the impact of a specific treatment or intervention on different individuals or different situations. Basic Uplift modeling is implemented through the following three meta-learners:
[0106] (1) Two-Model (T-LEARNER).
[0107] For a single treatment, the dual-model approach involves building separate models for the experimental and control groups, and then subtracting the predictions from the two models to obtain the lift. For multiple treatments, separate models are built for each experimental and control group, and then the predictions for each experimental group are subtracted from those for the control group to obtain the lift for each experimental group.
[0108] Advantages: The advantage of dual models is that they are simple and intuitive, and can reuse common machine learning models (LR, Tree Model, NN).
[0109] Disadvantages: Insufficient data utilization, poor fit to the differences between the two groups (i.e., lift signal), and the model error is easily amplified.
[0110] (2)One-Model(S-LEARNER).
[0111] The single-model approach involves adding the intervention attribute W to the sample features of both the experimental and control groups, merging the experimental and control groups, and training them using the same model.
[0112] Compared to dual-model approaches, single-model approaches offer several advantages: Data utilization is more efficient during model training. Modeling is simpler, requiring only a simple logistic regression or tree model (random forest, XGboost, LightGBM). It allows for the imposition of strict monotonic constraints on the processing variable or other variables, something dual-model approaches cannot do.
[0113] (3) X-Learner model.
[0114] X-Learner is a fusion of T-Learner and S-Learner, and its steps are as follows:
[0115] First, X-Learner will perform predictive modeling on the experimental group and the control group respectively, resulting in two pre-trained models;
[0116] Next, X-Learner uses the prediction results from the first step to predict the data for the control group using the experimental group's model, and simultaneously uses the control group's model to predict the data for the experimental group. Then, it calculates the difference between these predictions and the actual results to obtain the estimated effect.
[0117] In the third step, X-Learner uses preprocessed variables to model the predicted effects of the forecasts calculated in the second step.
[0118] Finally, X-Learner uses the model learned in step three to predict the predicted individual effects, thus obtaining an estimate of the predicted effect for each individual.
[0119] The main advantage of X-Learner is that it addresses the imbalance between the control and experimental groups. However, a major drawback of X-Learner is its use of multiple models, which introduces cumulative error. Furthermore, X-Learner results also depend on the accuracy of the predictive models; poor performance of these models can lead to inaccurate estimates of the predicted effects.
[0120] The machine learning process is implemented based on the following three learning algorithms.
[0121] (1) LightGBM algorithm.
[0122] LightGBM, short for Light Grandient Boosting Machine, is a Gradient Boosting Decision Tree (GBDT) algorithm framework that supports highly efficient parallel training and features faster training speed, lower memory consumption, better accuracy, and distributed processing capabilities for handling massive amounts of data quickly.
[0123] (2) Random Forest Algorithm.
[0124] Random Forest is a parallel ensemble learning method, an extension of decision trees. It is based on decision trees, and multiple trees form a forest. The randomness lies in the randomness of the selection of the splitting attribute. When training machine learning, Random Forest also uses sampling with replacement to add sample perturbation. At the same time, it introduces an attribute perturbation. When splitting attributes during the training of decision trees, Random Forest first randomly selects a subset containing K attributes from the candidate attribute set, and then selects the optimal splitting attribute from this subset.
[0125] (3) Catoost algorithm.
[0126] CatBoost is a machine learning library open-sourced by Russian giant Yandex in 2017. It is a GBDT framework based on decision trees (oblivious trees), with fewer parameters, support for categorical variables, and high accuracy. Its main pain point is the efficient and reasonable processing of categorical features, as its name suggests. CatBoost consists of categorical features and gradient boosting. In addition, CatBoost also solves the problems of gradient bias and prediction shift, thereby reducing overfitting and improving the accuracy and generalization ability of the algorithm.
[0127] The model selection module is used to select the marketing gain model with the highest evaluation index as the optimal model.
[0128] This involves determining the optimal model by evaluating multiple marketing gain models. Specifically, the model with the highest evaluation index among the multiple prediction models is selected as the optimal model. In this application, the evaluation index chosen is the Qini coefficient. That is, the Qini coefficient is determined by evaluating the Qini curve of each marketing gain model, and then the optimal model is selected based on this coefficient.
[0129] The Qini curve is a tool used to evaluate the performance of uplift models. It is based on the assumption that customer responses are random in the absence of any intervention. Therefore, if we rank customers according to the degree to which they are likely to be affected by the intervention, and start applying the intervention to the most affected customers, we expect that initially, the intervention will significantly outperform random intervention. Then, as the intervention is applied to more and more customers, this outperformance gradually decreases, and eventually, when all customers are affected, the intervention's effect is the same as that of random intervention.
[0130] The Qini curve reflects this relationship by showing the proportion of intervention (from 0 to 1) on the horizontal axis and the effect of intervention (gain relative to random intervention) on the vertical axis. An ideal Qini curve starts at the origin, rises rapidly, then flattens out, and ends in the upper right corner. The closer a model's Qini curve is to the ideal curve, the better the model's performance.
[0131] A key characteristic of the Qini curve is its ability to visually represent model performance, rather than simply providing a single number. However, this also means that the Qini curve can be more difficult to interpret and understand than some other evaluation metrics, such as precision, recall, or the area under the ROC curve. Below is the formula for calculating the Qini curve:
[0132] Qini=nt,1(φ) / Nt-nc,1(φ) / Nc
[0133] Where, n t,1 (φ) and n c,1 (φ) represents the number of people with an outcome of 1 in the control group and the control group, respectively. The fraction φ represents the ratio of the observed population to the target population. Nt and Nc represent the total number of people in the experimental group and the control group (independent of φ).
[0134] The analysis execution module is used to analyze customers based on this best model.
[0135] This involves using the optimal model to predict and process multiple marketing target characteristics of a customer to determine whether the customer is a sensitive individual to influence. These multiple marketing target characteristics include, but are not limited to, the customer's identity information, recent login preferences, device model of the login terminal, duration of use of financial instruments, and age.
[0136] As can be seen from the above technical solution, this embodiment provides a marketing-sensitive personnel analysis device. This device is applied to electronic devices and is used to analyze marketing-sensitive personnel based on a marketing gain prediction model constructed based on A / B experiments and a marketing gain model. Specifically, it constructs a training sample set by integrating various historical characteristic information of customers; trains the model based on A / B experiments using multiple learning algorithms and multiple meta-learners to obtain multiple marketing gain models; selects the prediction model with the highest evaluation index among the multiple prediction models as the optimal model; and performs prediction processing on multiple marketing target characteristic information of customers based on the optimal model to determine whether the customer is a marketing-sensitive person. By identifying marketing-sensitive personnel, marketers can avoid applying a large amount of marketing resources to ineffective groups, thereby avoiding the waste of marketing resources.
[0137] The units described in the embodiments of this disclosure can be implemented in software or in hardware. The name of a unit does not necessarily limit the unit itself; for example, the first acquisition unit can also be described as "a unit that acquires at least two Internet Protocol addresses".
[0138] The functions described above in this document can be performed, at least in part, by one or more hardware logic components. For example, exemplary types of hardware logic components that can be used, without limitation, include: Field Programmable Gate Arrays (FPGAs), Application-Specific Integrated Circuits (ASICs), Application Standard Products (ASSPs), System-on-Chip (SoCs), Complex Programmable Logic Devices (CPLDs), and so on.
[0139] Figure 3 This is a block diagram of an electronic device according to an embodiment of this application.
[0140] The following is for reference. Figure 3 This document illustrates a structural diagram suitable for implementing the electronic device in the embodiments of this disclosure. The terminal device in the embodiments of this disclosure may include, but is not limited to, mobile terminals such as mobile phones, laptops, digital broadcast receivers, PDAs (personal digital assistants), PADs (tablet computers), PMPs (portable multimedia players), in-vehicle terminals (e.g., in-vehicle navigation terminals), and fixed terminals such as digital TVs and desktop computers. This electronic device is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of this disclosure.
[0141] The electronic device may include a processing unit (e.g., a central processing unit, a graphics processing unit, etc.) 301, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 302 or a program loaded from an input device 306 into a random access memory (RAM) 303. The RAM also stores various programs and data required for the operation of the electronic device. The processing unit, ROM, and RAM are interconnected via a bus 304. An input / output (I / O) interface 305 is also connected to the bus 304.
[0142] Typically, the following devices can be connected to the I / O interface: input devices including, for example, touchscreens, touchpads, keyboards, mice, cameras, microphones, accelerometers, gyroscopes, etc.; output devices 307 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; storage devices 308 including, for example, magnetic tapes, hard disks, etc.; and communication devices 309. Communication device 309 allows the electronic device to communicate wirelessly or wiredly with other devices to exchange data. Although electronic devices with various devices are shown in the figures, it should be understood that it is not required to implement or possess all of the devices shown. More or fewer devices may be implemented or possessed alternatively.
[0143] This application also provides an embodiment of a computer-readable storage medium.
[0144] The aforementioned computer-readable storage medium is applied to an electronic device and carries one or more computer programs. When these programs are executed by the electronic device, the device performs marketing-sensitive audience analysis based on a marketing gain prediction model constructed using A / B testing and a marketing gain model. Specifically, this involves constructing a training sample set by integrating various historical customer characteristics; training the model using multiple learning algorithms and meta-learners based on A / B testing to obtain multiple marketing gain models; selecting the model with the highest evaluation index among the multiple prediction models as the optimal model; and using the optimal model to predict multiple marketing target characteristic information of the customer to determine whether the customer is a sensitive target audience. By identifying marketing-sensitive audiences, marketers can avoid wasting marketing resources by applying them to ineffective groups.
[0145] It should be noted that the computer-readable medium described above in this disclosure can be a computer-readable signal medium or a computer-readable storage medium, or any combination of the two. A computer-readable storage medium can be, for example,—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof.
[0146] In this disclosure, a computer-readable storage medium can be any tangible medium that contains or stores a program that can be used by or in connection with an instruction execution system, apparatus, or device. In this disclosure, a computer-readable signal medium can include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium can also be any computer-readable medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to: wires, optical fibers, RF (radio frequency), etc., or any suitable combination thereof.
[0147] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other.
[0148] Although preferred embodiments of the present invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of the embodiments of the present invention.
[0149] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or terminal device that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or terminal device. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or terminal device that includes said element.
[0150] The technical solution provided by the present invention has been described in detail above. Specific examples have been used to illustrate the principle and implementation of the present invention. The description of the above embodiments is only for the purpose of helping to understand the method and core idea of the present invention. At the same time, for those skilled in the art, there will be changes in the specific implementation and application scope based on the idea of the present invention. Therefore, the content of this specification should not be construed as a limitation of the present invention.
Claims
1. A method for analyzing marketing-sensitive individuals, applied to electronic devices, for analyzing marketing-sensitive groups based on a marketing gain prediction model constructed using A / B experiments and a marketing gain model, characterized in that... The method for analyzing marketing-sensitive demographics includes the following steps: A training sample set is constructed by integrating various historical feature information of customers; Based on A / B experiments, multiple marketing gain models were obtained by training the models using various learning algorithms and meta-learners. The prediction model with the highest evaluation index among the multiple prediction models is selected as the optimal model; Based on the optimal model, predictive processing is performed on the characteristic information of multiple marketing targets of the customer to determine whether the customer is a sensitive person who can influence the customer.
2. The marketing-sensitive personnel analysis method as described in claim 1, characterized in that, The various historical characteristics include customer identity information, recent login preferences, device model of the login terminal, duration of use of financial instruments, and age.
3. The marketing-sensitive personnel analysis method as described in claim 1, characterized in that, The various learning algorithms include the LightGBM algorithm, the RandomForest algorithm, and the CatBoost algorithm.
4. The marketing-sensitive personnel analysis method as described in claim 1, characterized in that, The multi-meta-learners include S-Learner, T-Learner, and X-Learner.
5. The marketing-sensitive audience analysis method as described in claim 1, characterized in that, The step of selecting the prediction model with the highest evaluation index among the multiple marketing gain models as the optimal model includes the following steps: Calculate the evaluation metrics for each of the aforementioned marketing gain models; The marketing gain model with the highest evaluation index is determined as the so-called optimal model.
6. The marketing-sensitive audience analysis method as described in claim 5, characterized in that, The evaluation index is the Qini coefficient.
7. A marketing-sensitive personnel analysis device, applied to electronic devices, for analyzing marketing-sensitive populations based on a marketing gain prediction model constructed using A / B experiments and a marketing gain model, characterized in that, The marketing-sensitive audience analysis device includes: The sample construction module is configured to build a training sample set by integrating various historical feature information of the customer. The model training module is configured to train models based on A / B experiments, using multiple learning algorithms and meta-learners to obtain multiple marketing gain models. The model selection module is configured to select the prediction model with the highest evaluation index among the multiple marketing gain models as the optimal model. The analysis and execution module is configured to perform predictive processing based on the optimal marketing target feature information of the customer to determine whether the customer is a sensitive person.
8. The marketing-sensitive audience analysis device as described in claim 7, characterized in that, The model selection module is configured as follows: Calculate the evaluation metrics for each of the aforementioned marketing gain models; The marketing gain model with the highest evaluation index is determined as the so-called optimal model.
9. An electronic device, characterized in that, The electronic device includes at least one processor and a memory connected to the processor, wherein: The storage device is used to store computer programs or instructions; The processor is used to execute the computer program or instructions to enable the electronic device to implement the marketing-sensitive personnel analysis method as described in any one of claims 1 to 6.
10. A computer-readable storage medium for use in electronic devices, characterized in that, The storage medium carries one or more computer programs that can be executed by the electronic device, thereby enabling the electronic device to implement the marketing-sensitive personnel analysis method as described in any one of claims 1 to 6.