Target service processing method and device, electronic equipment and computer program product

Abstract

The application provides a target service processing method and device, electronic equipment and computer program product, and belongs to the field of data processing. The method comprises the following steps: obtaining historical data related to a target service, and preprocessing the historical data; extracting features related to a set index of the target service from the preprocessed data, and determining key features based on a feature selection algorithm; generating a machine learning model for predicting the set index according to the key features; performing parameter balancing training using multiple sub-models, performing weighted integrated processing according to the training results of the multiple sub-models to obtain an integrated prediction model for predicting the set index; inputting the features of a target service to be processed into the integrated prediction model to obtain the predicted value of the set index, and performing priority sorting on the target service to be processed according to the predicted value.

Description

Technical Field

[0001] This application relates to the field of data processing, and more particularly to a method, apparatus, electronic device, and computer program product for processing a target business. Background Technology

[0002] Currently, in the fields of marketing automation and data analytics, the management of marketing campaign priorities mainly relies on strategies based on fixed rules or human experience. While these methods are easy to implement, they struggle to dynamically adapt to market changes, fluctuations in consumer behavior, and the complex relationships between different marketing campaigns. This leads to irrational allocation of marketing resources, limited campaign success rates, and an overall failure to optimize marketing efficiency and effectiveness. Summary of the Invention

[0003] This application provides a method, apparatus, electronic device, and computer program product for processing target business, which can solve the problems of unreasonable allocation of marketing resources and limited success rate of activities caused by the difficulty of existing methods to dynamically adapt to market changes and fluctuations in consumer behavior when used for marketing activity management.

[0004] In a first aspect, embodiments of this application provide a method for processing a target service, the method comprising the following steps: Acquire historical data related to the target business and preprocess the historical data; Features related to the set indicators of the target business are extracted from the preprocessed data, and key features are determined based on the feature selection algorithm; Based on the aforementioned key features, determine the model structure and parameter settings of the sub-model; Multiple sub-models are trained with parameter balancing, and the training results of the multiple sub-models are weighted and integrated to obtain an integrated prediction model for predicting the set index. The characteristics of the target business to be processed are input into the integrated prediction model to obtain the predicted values ​​of its set indicators, and the target business to be processed is prioritized according to the predicted values.

[0005] Secondly, embodiments of this application provide a processing apparatus for a target service, the apparatus comprising the following: The data preprocessing module is used to acquire historical data related to the target business and preprocess the historical data. The feature extraction module is used to extract features related to the set indicators of the target business from the preprocessed data, and to determine key features based on the feature selection algorithm. The model generation module is used to determine the model structure and parameter settings of the sub-model based on the key features. The training ensemble module is used to perform parameter balancing training on multiple sub-models, and to perform weighted ensemble processing based on the training results of the multiple sub-models to obtain an ensemble prediction model for predicting the set index; and, The execution module is used to input the features of the target business to be processed into the integrated prediction model, obtain the predicted values ​​of its set indicators, and prioritize the target business to be processed according to the predicted values.

[0006] Thirdly, embodiments of this application provide an electronic device including a processor, a memory, and a program or instructions stored in the memory and executable on the processor, wherein the program or instructions, when executed by the processor, implement the steps of the processing method for the target service as described in the first aspect.

[0007] Fourthly, embodiments of this application provide a computer program product, the computer program product including a computer program stored on a non-transitory computer-readable storage medium, the computer program including program instructions, the program instructions being executed by a computer to implement the steps of the processing method of the target service as described in the first aspect.

[0008] This application's embodiments automatically analyze historical data, extract key features, and generate predictive models using machine learning models. This avoids reliance on fixed rules or human experience, enabling intelligent and automated prioritization of target businesses, thus improving decision-making efficiency and objectivity. Through parameter balancing training and weighted ensemble processing of sub-models, the resulting ensemble predictive model can more accurately predict the set indicators for target businesses. Dynamic priority adjustment based on the prediction results ensures that resources are prioritized for high-potential businesses, thereby optimizing resource allocation and improving overall success rate and return on investment. Attached Figure Description

[0009] 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 recorded in this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0010] Figure 1 This is a flowchart illustrating a method for processing a target service provided in an embodiment of this application; Figure 2 This is a flowchart illustrating another method for processing a target service provided in an embodiment of this application; Figure 3 This is a flowchart illustrating a marketing activity processing method provided in an embodiment of this application; Figure 4 This is a schematic diagram of the structure of a target service processing device provided in an embodiment of this application; Figure 5 This is a flowchart illustrating another marketing activity processing method provided in an embodiment of this application; Figure 6 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0011] To enable those skilled in the art to better understand the technical solutions in this application, the technical solutions in 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 skilled in the art without creative effort should fall within the scope of protection of this application.

[0012] The terms "first," "second," etc., used in the specification and claims of this application are used to distinguish similar objects and not to describe a specific order or sequence. It should be understood that such use of data can be interchanged where appropriate so that embodiments of this application can be implemented in orders other than those illustrated or described herein. Furthermore, in the specification and claims, "and / or" indicates at least one of the connected objects, and the character " / " generally indicates that the preceding and following objects are in an "or" relationship.

[0013] With the rapid development of big data and artificial intelligence technologies, the management and decision-making of marketing campaigns are gradually shifting from traditional experience-driven to data-driven approaches. However, most existing methods for prioritizing marketing campaigns are based on fixed rules or human experience, which cannot accurately reflect market changes, consumer behavior, and the complex relationships between marketing campaigns. This leads to unreasonable allocation of marketing resources and limited marketing effectiveness.

[0014] Therefore, there is an urgent need for a method for handling target business, especially for marketing campaign management, that can achieve priority adjustment of marketing campaigns throughout their entire lifecycle based on machine learning algorithms.

[0015] The target business processing method and apparatus proposed in this application are based on a marketing activity lifecycle priority adjustment mechanism using machine learning algorithms. This mechanism collects and analyzes marketing activity data, uses machine learning algorithm models to perform deep learning and analysis on the marketing activity data, predicts the success probability of marketing activities, and dynamically adjusts the priority of activities based on the prediction results, thereby optimizing the allocation of marketing resources and maximizing marketing effectiveness.

[0016] The following is in conjunction with the appendix Figures 1 to 6The present application provides a detailed description of a target service processing method, apparatus, electronic device, and computer program product through specific embodiments and application scenarios.

[0017] Figure 1 This application illustrates a method for processing a target service according to an embodiment of the present application. This method can be executed by an electronic device, which may include a server and / or terminal devices, etc. In other words, the method can be executed by software or hardware installed on the server and / or terminal devices, and the method includes the following steps: Step 110: Obtain historical data related to the target business and preprocess the historical data.

[0018] The target business can include commercial or service activities that require prioritization or resource optimization and are targeted at specific groups of people. For example, telecom operators' user-facing data traffic marketing activities or package promotion activities, financial services' credit card promotion activities or wealth management product recommendation activities, e-commerce product promotion activities or new product launch promotion activities, etc.

[0019] The historical data can come from multiple data sources such as business databases, user behavior logs, transaction records, and channel contact information. It typically includes multi-dimensional information such as business type, time period, target object attributes, execution channels, resource investment, and historical results.

[0020] After acquiring the data, it is preprocessed, such as data cleaning, deduplication, missing value handling, outlier detection and correction, and data normalization, to improve data quality, eliminate noise, and provide a high-quality dataset with complete structure and uniform format for subsequent feature extraction and model training.

[0021] Step 120: Extract features related to the set indicators of the target business from the preprocessed data, and determine key features based on the feature selection algorithm.

[0022] The set metrics can include success rate, conversion rate, etc. For example, for telecom operators' user-facing data marketing campaigns or package promotions, the set metrics can include marketing success rate, package subscription rate, customer retention rate, etc. Features related to the set metrics of the target business are extracted from the preprocessed data. Feature extraction can cover multi-dimensional information, such as the target audience's attribute characteristics, behavioral characteristics, consumption characteristics, and time-series fluctuation characteristics, to comprehensively characterize the key factors affecting business performance.

[0023] Feature selection algorithms are used to automatically filter out the most influential subset of key features from a large number of extracted features, which are crucial to a given objective. Common types include filtering, wrapping, and embedding. By employing feature selection algorithms to evaluate and rank the importance of extracted features, and selecting the subset of key features that contribute most to the prediction model, the aim is to reduce data dimensionality, decrease model complexity, prevent overfitting, and improve the training efficiency and prediction accuracy of subsequent machine learning models.

[0024] Step 130: Determine the model structure and parameter settings of the sub-model based on the key features.

[0025] The model structure and parameter settings can be adaptively selected from a set of candidate algorithms (e.g., random forest, gradient boosting decision tree, support vector machine, neural network, etc.) based on the distribution characteristics of the data samples and the prediction target. By fine-tuning the parameters of the candidate algorithms and using methods such as cross-validation to conduct multiple rounds of training and performance evaluation on multiple candidate models, the optimal model structure and parameter settings are finally determined based on the comprehensive evaluation results.

[0026] Step 140: Perform parameter balancing training on multiple sub-models, and perform weighted ensemble processing based on the training results of the multiple sub-models to obtain an ensemble prediction model for predicting the set index.

[0027] The sub-models can be multiple basic machine learning models with specific functional focuses that are trained and used in parallel during the construction of the final ensemble prediction model. Different sub-models can be designed to focus on different aspects of the learning data, different subsets of features, or different business scenarios.

[0028] By constructing multiple sub-models and performing parameter balancing training, a high-performance prediction model is finally obtained through integration. Specifically, multiple sub-models are first constructed, and each sub-model is trained by imposing consistent parameter constraint rules to ensure that they have the conditions for parallel computation and comparability. During the training process, the actual success rate of the target business in historical data can be used as the positive sample ratio of each sub-model, while the negative sample ratio is controlled to be on the same order of magnitude to achieve sample balancing among the sub-models.

[0029] After training, weight coefficients are determined based on the preset value of the target business associated with each sub-model, and the output results of each sub-model are weighted and fused to form the final integrated prediction model. This integrated prediction model can combine the prediction capabilities of multiple sub-models and incorporate business value orientation to improve the accuracy, stability and business adaptability of the model prediction.

[0030] Step 150: Input the features of the target business to be processed into the integrated prediction model to obtain the predicted values ​​of its set indicators, and prioritize the target business to be processed according to the predicted values.

[0031] The process involves inputting relevant feature data of the target business to be processed into a pre-trained ensemble prediction model. This model, based on the weighted calculations of its multiple sub-models, outputs a predicted value for a specific metric (e.g., success rate) for that target business. Subsequently, based on this prediction, multiple target businesses are automatically prioritized. Typically, businesses with higher predicted values ​​(indicating a greater probability of success or higher business value) are assigned higher execution priority. The prioritization results can be directly used to guide the intelligent allocation of resources (e.g., budget, manpower, channels), ensuring that resources are prioritized for businesses with the best expected results, thereby maximizing overall business objectives and optimizing resource allocation.

[0032] In this embodiment, a machine learning model automatically analyzes historical data, extracts key features, and generates a predictive model, avoiding reliance on fixed rules or human experience. This enables intelligent and automated prioritization of target businesses, improving decision-making efficiency and objectivity. Through parameter balancing training and weighted ensemble processing of sub-models, the resulting ensemble predictive model can more accurately predict the set indicators for target businesses. Dynamic priority adjustment based on the prediction results ensures that resources are allocated to high-potential businesses, thereby optimizing resource allocation and improving overall success rate and return on investment.

[0033] In yet another exemplary embodiment, based on step 110 of the above embodiment, historical data related to the target business is obtained, and the historical data is preprocessed. The method of this embodiment may further include the following specific steps: The historical data is cleaned using an iterative loop cleaning method based on sample training residuals; the cleaned data is then normalized using a multilinear normalization method that combines binning ratio and dynamic jump.

[0034] The historical data acquired that is related to the target business may include at least one of the following: business type data, time period data, target object attribute data, execution channel data, resource input data, and historical performance data.

[0035] The data cleaning process employs an iterative loop cleaning method based on training residuals. In each round of cleaning, a predetermined proportion of training samples are selected and removed based on the model's training residual rate, and this process is repeated multiple times. After data cleaning, the cleaned data undergoes normalization. This normalization uses a binning ratio normalization method based on the overall data distribution, and also incorporates a dynamic skip strategy for multilinear normalization.

[0036] This embodiment uses a marketing campaign targeting specific business activities as an example, such as a mobile operator's promotional campaign targeting a group of customers. Depending on the marketing objectives and scenarios, preparation data is collected for different time periods. Specifically, data on key elements of the marketing campaign over the past year is collected, including campaign type, time, marketing timing, contact channels, target customer group, cost budget, and historical sales data. The collected data undergoes preprocessing operations such as cleaning, deduplication, and normalization. Iterative cleaning is performed using a sample training residual rate approach, gradually refining the sample size by 10% to improve accuracy. Normalization is then performed according to the overall binning ratio, and multilinear normalization is performed dynamically in conjunction with basic processing, supporting multiple training sample groups. After the above cleaning processes, noise and outliers are eliminated from the data, improving data quality.

[0037] The data cleaning method based on iterative residual rate analysis differs from traditional rule-based cleaning. It dynamically combines data cleaning with model training. First, a basic prediction model is trained using an initial sample set, and the prediction residual (the deviation between the predicted and actual values) is calculated for each sample. Based on the residual size, samples that are difficult for the current model to fit effectively are filtered and removed according to a preset ratio (e.g., removing the 10% of samples with the largest residuals in each round). These samples typically contain noise, anomalies, or data that significantly deviates from the main pattern. Subsequently, the model is retrained using the cleaned sample set, and the "training-residual calculation-sample removal" steps are repeated for multiple iterations. This cyclical filtering improves data quality, creating a training set suitable for subsequent machine learning models.

[0038] The cleaned data undergoes standardization and binning based on its overall distribution characteristics, mapping continuous values ​​to discrete intervals defined by the data distribution ratio, thus achieving scale regularization based on the data's inherent structure. Furthermore, a dynamic jump mechanism can be introduced to adaptively adjust the data processing interval and mapping function during normalization, avoiding boundary sensitivity issues caused by fixed segmentation. Through the comprehensive application of multilinear piecewise functions, flexible and robust numerical normalization is achieved for different data intervals, different sample batches, and dynamic changes during multiple training iterations. This approach, while maintaining the data distribution pattern, effectively enhances the adaptability of the normalization process to different data scenarios and iterative training needs, providing high-quality, highly consistent numerical input for subsequent feature extraction and model training.

[0039] Figure 2This illustration shows a flowchart of another method for processing a target service provided by an embodiment of this application. This method can be executed by an electronic device, which may include a server and / or terminal devices, etc. In other words, the method can be executed by software or hardware installed on the server and / or terminal devices, and includes the following steps: Step 210: Obtain historical data related to the target business and preprocess the historical data.

[0040] Step 210 can be found above. Figure 1 The specific description of step 110 in the illustrated embodiment can achieve the same technical effect, and will not be repeated here to avoid repetition.

[0041] Step 220: Based on step 120 of the above embodiment, extract features related to the set indicators of the target business from the preprocessed data, and determine key features based on a feature selection algorithm. The method of this embodiment may further include the following specific steps: From the preprocessed data, behavioral features, consumption features, and temporal fluctuation features of the target object are extracted. The temporal fluctuation features are obtained based on variance derivation and trend derivation processing of behavioral data and / or consumption data in the time dimension. Based on a feature selection algorithm with preset evaluation criteria, all extracted features are sorted, and the top-ranked features are selected as the key features according to a preset number.

[0042] The feature selection algorithm with preset evaluation criteria can be based on information theory, using indicators such as information gain, information gain ratio, or Gini coefficient to quantify the degree to which each feature reduces the uncertainty of the prediction target, and then ranking and filtering features accordingly. In the calculation of information gain ratio, weight coefficients based on business rules can be introduced to adjust the importance of features in specific categories.

[0043] The feature selection algorithm with a pre-defined evaluation criterion can also be based on the evaluation criterion of model performance. It adopts a wrapper feature selection method, which constructs multiple candidate models containing different feature subsets and uses cross-validation to evaluate the prediction performance of each candidate model (e.g., accuracy, AUC value, etc.). Finally, it selects the feature subset that makes the model perform optimal as the key features.

[0044] This embodiment uses a marketing campaign as an example to illustrate the process. Features that influence the success probability of the marketing campaign are extracted from the preprocessed data, such as user behavior characteristics, market trend characteristics, competitor characteristics, recently targeted marketing products, and recently prioritized distribution channels. Feature selection algorithms (e.g., based on information gain, based on model performance) are used to filter out features that contribute significantly to the prediction results, reducing model complexity and the risk of overfitting. Specifically, user features are extracted from basic user attributes, departure behavior, in-store behavior, service degradation behavior, consumption behavior, and dynamic temporal fluctuations. Variance-level feature derivation and trend-level feature derivation are processed. A stepwise method is used to determine feature extraction, and features are ranked according to the adjusted information gain ratio before top-level filtering. For example, user behavior before ordering and the price and value of marketing products are selected. These two types of highly correlated features that contribute significantly to the prediction results are used as inputs for subsequent machine learning model calculations.

[0045] Step 230: Determine the model structure and parameter settings of the sub-model based on the key features.

[0046] Step 240: Perform parameter balancing training on multiple sub-models, and perform weighted ensemble processing based on the training results of the multiple sub-models to obtain an ensemble prediction model for predicting the set index.

[0047] Step 250: Input the features of the target service to be processed into the integrated prediction model to obtain the predicted values ​​of its set indicators, and prioritize the target services to be processed according to the predicted values. Steps 230-250 can be found above. Figure 1 The specific descriptions of steps 130-150 in the illustrated embodiment are provided, and the same technical effects can be achieved. To avoid repetition, they will not be repeated here.

[0048] In this embodiment, by extracting features from multiple dimensions (behavior, consumption, and temporal fluctuations), especially by introducing temporal fluctuation features derived from variance and trend, the static attributes of the target object can be captured, and the dynamic changes and stability of its behavioral patterns over time can be explored. This helps to enhance the expressive power of the feature set on factors affecting business success or failure, providing a good data foundation for high-precision prediction. Employing a feature selection algorithm based on preset evaluation criteria, the algorithm can intelligently and objectively quantify and select the most predictive key features according to different data characteristics and business objectives. Through sorting and top filtering, dimensionality reduction is automatically performed from the high-dimensional feature space, accurately retaining the feature subset with the largest information content and strongest discriminative power. This reduces the complexity of subsequent machine learning models, reduces the risk of overfitting, improves the model's generalization ability and robustness, and also reduces the computational resources and time required for model training and inference, improving overall processing efficiency and making it more suitable for large-scale, real-time business scenarios.

[0049] In yet another exemplary embodiment, based on step 130 of the above embodiment, determining the model structure and parameter settings of the sub-model according to the key features, the method of this embodiment may further include the following specific steps: Based on the data sample distribution corresponding to the key features, at least one machine learning algorithm is selected from the candidate algorithm set as the base model; the parameters of the base model are tuned, and cross-validation is used to conduct multiple rounds of training and performance evaluation on multiple base models after parameter tuning; based on the results of multiple rounds of training and performance evaluation, the model structure and parameter settings of the sub-model are determined.

[0050] The sub-models can be designed to focus on learning patterns in a specific aspect of the data. For example, one sub-model might analyze user consumption behavior, while another analyzes the temporal characteristics of service interactions. Multiple sub-models make predictions from different perspectives, and their conclusions complement and balance each other. Even if a sub-model makes an error due to data noise or specific biases, the correct judgments of other sub-models can correct it during the integration process, thereby reducing the sensitivity of the overall prediction results to local data disturbances and making the final model more robust.

[0051] The candidate algorithm set can include random forests, gradient boosting decision trees, support vector machines, and neural networks. Based on the extracted features and the prediction target (e.g., the probability of activity success), a suitable machine learning algorithm is selected to construct the prediction model. This embodiment can employ various algorithm models, including but not limited to random forests, gradient boosting decision trees (GBDT), support vector machines (SVM), and deep learning models (e.g., neural networks). Random forests offer advantages such as high accuracy, the ability to handle multiple data types, ease of implementation, and high computational efficiency. Disadvantages include potentially slow performance on large datasets and limited ability to handle interactive features and nonlinear relationships. Neural networks offer advantages such as high classification accuracy, strong nonlinear modeling capabilities, parallel processing capabilities, and good robustness and generalization ability. Disadvantages include the need for large amounts of training data, difficulty in adjusting model parameters, and poor interpretability. Decision trees are easy to understand and implement, and can handle non-numerical data. Disadvantages include a tendency to overfit, poor performance on continuous variables, and potential neglect of correlations in the data. The final algorithm is determined primarily by selecting algorithms based on sample distribution and comprehensively evaluating the effects of multiple training rounds. The selected algorithm is optimized by tuning its parameters and evaluating its performance through methods such as cross-validation. The optimal model structure and parameter settings are then selected.

[0052] In this embodiment, by intelligently selecting a base model from a candidate algorithm set based on the data sample distribution, and systematically optimizing and cross-validating the selected model's parameters, it is ensured that the structure and parameters of each sub-model are highly matched with the current data characteristics and prediction targets. This overcomes the subjectivity and limitations of manual experience-based selection, enabling each sub-model to achieve optimal performance within its preset professional domain (e.g., consumer analysis, time series prediction), providing high-quality basic building blocks for subsequent integration. By having multiple sub-models focus on learning different aspects of patterns in the data, local errors or data noise in a single sub-model can be corrected by the correct judgments of other sub-models. This significantly improves the overall robustness and fault tolerance of the integrated prediction model, reduces prediction fluctuations caused by data disturbances or model biases, and ensures stability in complex and ever-changing business environments. The candidate algorithm set covers everything from high-efficiency tree models (random forests, GBDT) to highly expressive complex models (neural networks, SVM). Through multiple rounds of training and comprehensive evaluation, the most suitable algorithm can be selected for sub-tasks with different focuses, thereby optimizing overall computational efficiency and feasibility while ensuring overall prediction accuracy.

[0053] In yet another exemplary embodiment, based on step 140 of the above embodiment, parameter balancing training is performed on multiple sub-models, and weighted ensemble processing is performed based on the training results of the multiple sub-models to obtain an ensemble prediction model for predicting the set index. The method of this embodiment may further include the following specific steps: For each of the multiple sub-models, a consistent parameter constraint rule is used for training. The parameter constraint rule is used to uniformly adjust at least one key parameter of each sub-model so that each sub-model meets the parallel computing conditions.

[0054] Key parameters may include tree depth, number of leaf nodes, and learning rate.

[0055] In this embodiment, by employing consistent parameter constraint rules to uniformly adjust the key parameters of each sub-model, sub-models with different functional focuses are placed on the same training benchmark and complexity level. This eliminates performance evaluation biases caused by differences in model structure or hyperparameters, ensuring that the outputs of all sub-models are in a mathematical space that allows for fair comparison and collaborative computation, providing a foundation for subsequent integration. Parameter balancing training does not turn sub-models into homogenized models, but rather applies consistent constraints to key structural parameters while preserving their ability to learn freely and interact with features within their respective focused data dimensions or business perspectives, thus balancing uniformity and diversity. Balanced training ensures that the output deviations of each sub-model mainly stem from differences in their learned patterns, rather than differences in model capacity or training stability. This allows subsequent weighted integration processing to allocate weights more purely based on the importance (value coefficient) of the business dimension represented by each sub-model, avoiding weight distortion caused by uneven model performance.

[0056] In yet another exemplary embodiment, based on step 140 of the above embodiment, parameter balancing training is performed on multiple sub-models, and weighted ensemble processing is performed based on the training results of the multiple sub-models to obtain an ensemble prediction model for predicting the set index. The method of this embodiment may further include the following specific steps: For each sub-model, the proportion of its corresponding target business in historical data is used as the proportion of positive samples during the training of that sub-model; the proportion of negative samples used by the multiple sub-models during training is controlled to be at the same preset level.

[0057] This involves setting differentiated sample control strategies for each sub-model. For each sub-model's specific target business type, the actual success rate of that business in historical data is used as the positive sample ratio during training, ensuring that the data distribution learned by the sub-model aligns with real-world business conditions. Simultaneously, the training process for all sub-models is uniformly controlled, constraining the proportion of negative samples used by each model to the same preset level. This preserves the inherent sample distribution characteristics between different businesses while eliminating model bias caused by differences in negative sample size, providing a balanced data foundation for fair comparison and effective integration of subsequent sub-models.

[0058] In this embodiment, by using the actual success rate of each target business in historical data as the positive sample ratio of the corresponding sub-model, it can be ensured that the distribution of positive and negative samples encountered by each sub-model during the learning process is strictly consistent with its corresponding actual business scenario. This overcomes the distribution distortion problem caused by traditional fixed-ratio sampling (e.g., 1:5 or 1:10), enabling the sub-model to learn the true boundaries and patterns of business success and failure more accurately, thereby improving its discrimination ability and prediction accuracy in practical applications. While retaining the true proportion of positive samples to reflect business differences, by forcibly constraining the negative sample ratio of all sub-models to the same preset order of magnitude, key standardization control can be achieved at the input end of sub-model training. This effectively eliminates the model bias problem that may be caused by different negative sample bases for different businesses, ensuring the fairness and comparability of each sub-model in the subsequent integration process. The difference in their output scores mainly reflects the quality of pattern recognition ability, rather than the interference of the training data scale.

[0059] In yet another exemplary embodiment, based on step 140 of the above embodiment, parameter balancing training is performed on multiple sub-models, and weighted ensemble processing is performed based on the training results of the multiple sub-models to obtain an ensemble prediction model for predicting the set index. The method of this embodiment may further include the following specific steps: Determine the preset value coefficient of the target business associated with each sub-model; use the preset value coefficient as the weight of the corresponding sub-model, and perform weighted fusion calculation on the output results of the multiple sub-models to obtain the integrated prediction model.

[0060] The weighted integration process assigns different weights to the prediction results of multiple sub-models according to their importance or reliability, and then combines them in a weighted manner to form a final, better single prediction result. Each sub-model is assigned a weight that matches the importance of its associated target business. Specifically, a value coefficient is pre-set for each type of target business, which reflects the priority, expected revenue, or importance of the business. During integration, the pre-set value coefficient of the corresponding business of the sub-model is used as its weight to weight and fuse the output results of each sub-model. This ensures that the final prediction result is not only based on the statistical accuracy of the model, but also incorporates business value orientation, enabling resources to be intelligently allocated to high-value businesses.

[0061] In this embodiment, by using the preset value coefficient of the target business as the integration weight of the sub-models, the core business logic (e.g., priority, expected revenue, etc.) can be directly encoded into the algorithm decision-making process. This ensures that the final prediction result is not only an estimation of the set indicators but also a value-weighted optimized score, thereby optimizing resource allocation. After the output of each sub-model is amplified or reduced by its value coefficient, the final integration score can comprehensively consider the business value and the probability of success.

[0062] See Figure 3 This embodiment uses a marketing campaign as an example to illustrate the specifics.

[0063] Among them, the GBDT algorithm is widely used and highly effective in the field of data mining. This embodiment applies the GBDT algorithm to the construction of sub-models. GBDT itself is an ensemble algorithm, and its application to sub-models is equivalent to another ensemble. The first ensemble is the serial error calculation, and the second ensemble is the model balancing score output. To ensure that the scoring is more scientific and reasonable, GBDT needs to be trained one by one with balancing parameters. The balancing parameters mainly refer to the consistent adjustment of key parameters such as decision tree construction depth, number of leaf nodes, and learning ratio. While taking into account the model's recognition effect, it ensures that the sub-models have the conditions for parallel computation. By using GBDT and constructing the final prediction model through secondary ensemble, the sub-models can capture complex nonlinear patterns and residuals through serial training of GBDT, and the sub-models can achieve the integration of multi-view decision-making through balancing training and weighted ensemble, which helps to amplify the advantages of ensemble learning and make the final model have more powerful expressive power and higher prediction accuracy.

[0064] The ordinary data mining model uses a fixed ratio of 1:5 or 1:10 for sampling training to ensure sufficient training. However, the stability model requires that the strength of each sub-model be consistent. Therefore, in this embodiment, each sub-model uses the true ratio of each marketing behavior as the positive sample ratio, while the negative sample ratio of each sub-model is controlled at the same order of magnitude.

[0065] The sub-model comprehensive calculation uses value loss as the weighting method. The marketing sub-model is weighted according to the product value, which is called the value coefficient. This coefficient is used as the basis for the sub-model comprehensive calculation.

[0066] The trained model is validated using a validation set to evaluate its prediction accuracy and generalization ability. Based on the validation results, the model is further optimized.

[0067] The following examples illustrate the input and output of the ensemble prediction model.

[0068] The input data is shown in the table below:

[0069] The output data is shown in the table below:

[0070] The method of balancing parameters through sub-models has the following advantages: similar sub-models can be selected for similarity calculations during marketing inference, resulting in more accurate results; after nearly three months of marketing trials, the marketing success rate has increased by an average of 5.3 percentage points (5.3pp).

[0071] In yet another exemplary embodiment, based on step 150 of the above embodiment, the features of the target service to be processed are input into the integrated prediction model to obtain the predicted value of its set indicators, and the target service to be processed is prioritized according to the predicted value. The method of this embodiment may further include the following specific steps: This embodiment uses a marketing campaign as an example for specific explanation. For a new marketing campaign, its features are input into a trained ensemble prediction model to obtain the predicted success probability of the campaign. Marketing campaigns are prioritized based on the predicted success probability, with campaigns having a high predicted success probability and high marketing value being executed first, while campaigns with a low predicted success probability are given lower priority. Thresholds and strategies for priority adjustment are set according to actual conditions to ensure the rationality and effectiveness of priority adjustments.

[0072] In yet another exemplary embodiment, the method of this embodiment may further include the following specific steps: After prioritizing and executing the target services to be processed according to the predicted values, the actual execution result data of the target services is collected, and the actual execution result data is compared and analyzed with the corresponding predicted values. Based on the results of the comparison and analysis, the integrated prediction model is iteratively updated, and the iterative update includes at least one of the following operations: adjusting model parameters, updating training samples, and reselecting key features.

[0073] After the marketing campaign is executed, campaign performance data is collected and compared with the predicted results to evaluate the model's predictive accuracy. Based on the feedback, the machine learning model is iteratively optimized, updating its parameters and structure to improve its predictive accuracy and stability.

[0074] Corresponding to the target service processing method provided in the above embodiments, based on the same technical concept, this application also provides a target service processing apparatus. See [link to relevant documentation]. Figure 4 The device 400 includes a data preprocessing module 410, a feature extraction module 420, a model generation module 430, a training and integration module 440, and an execution module 450.

[0075] The data preprocessing module 410 is used to acquire historical data related to the target business and preprocess the historical data; the feature extraction module 420 is used to extract features related to the set indicators of the target business from the preprocessed data and determine key features based on a feature selection algorithm; the model generation module 430 is used to determine the model structure and parameter settings of the sub-models according to the key features; the training ensemble module 440 is used to perform parameter balancing training on multiple sub-models, perform weighted ensemble processing based on the training results of the multiple sub-models, and obtain an ensemble prediction model for predicting the set indicators; and the execution module 450 is used to input the features of the target business to be processed into the ensemble prediction model, obtain the predicted value of its set indicators, and prioritize the target business to be processed according to the predicted value.

[0076] It should be noted that the target service processing apparatus and the target service processing method provided in this application embodiment are based on the same application concept. Therefore, the specific implementation of this embodiment can refer to the implementation of the aforementioned target service processing method, and the repeated parts will not be described again.

[0077] This application proposes a marketing campaign lifecycle priority adjustment mechanism based on machine learning algorithms. It utilizes machine learning algorithms to perform deep learning and analysis on marketing campaign data, predicting the success probability of campaigns and dynamically adjusting campaign priorities based on the prediction results. This optimizes the allocation of marketing resources and maximizes marketing effectiveness. The advantage of this application lies in its ability to use machine learning algorithms for in-depth mining and analysis of marketing campaign data, automatically and accurately predicting the success probability of marketing campaigns and dynamically adjusting campaign priorities based on the prediction results. Compared to traditional priority adjustment methods based on experience or fixed rules, the method provided in this application is more scientific and objective, better adapting to market changes and user needs, improving the utilization efficiency of marketing resources, and enhancing the overall effectiveness of marketing campaigns.

[0078] The target business processing method and apparatus provided in this application, applied to marketing campaign operations, can realize a self-learning, adaptive, and automated intelligent priority adjustment system. The marketing campaign full lifecycle priority adjustment mechanism based on machine learning algorithms in this application includes data collection and preprocessing, feature extraction and selection, machine learning model construction, model training and validation, priority prediction and adjustment, and feedback and optimization steps. The machine learning model construction step includes selecting algorithms such as random forest, gradient boosting decision tree, or deep learning for modeling. The priority prediction and adjustment step prioritizes marketing campaigns based on the prediction results of the machine learning model. See also... Figure 5 The data sources include data from the entire marketing campaign lifecycle, as well as user behavior characteristics before ordering and marketing product price and value characteristics. The marketing campaign lifecycle data serves as the system's input, encompassing historical and real-time data from each stage of campaign planning, execution, and evaluation. User behavior characteristics before ordering and marketing product price and value characteristics are core feature examples extracted from the full dataset and directly input into the feature engineering stage. Core processing includes data collection and preprocessing, feature extraction and selection, and machine learning model building. Data collection and preprocessing involves acquiring, cleaning, and transforming raw data to prepare for analysis. Feature extraction and selection involves constructing and filtering features from the preprocessed data that have the greatest impact on the prediction objective (e.g., campaign success rate). Machine learning model building designs and builds a predictive model based on the selected features. Model training and validation include training the model using historical data and evaluating its performance using a validation set for optimization. The trained model is integrated or deployed into the actual marketing business system, ready for use. Priority prediction and adjustment is the system's core output function. For new marketing campaigns or target customers, the model predicts their success probability, and the system automatically makes priority ranking and resource allocation decisions accordingly. After the marketing campaign is executed, actual performance data is collected and compared with the predicted results. This feedback data is used to retrain and optimize the model, enabling it to adapt to changes and achieve continuous learning and improvement.

[0079] Corresponding to the target service processing method provided in the above embodiments, based on the same technical concept, this application also provides an electronic device for executing the above method. Figure 6 To illustrate the structure of an electronic device according to various embodiments of this application, as shown in the following diagrams... Figure 6 As shown. Electronic device 500 can vary considerably due to differences in configuration or performance, and may include one or more processors 510 and memory 520. Memory 520 may store one or more application programs or data. Memory 520 may be temporary or persistent storage. The application programs stored in memory 520 may include one or more modules (not shown), each module may include a series of computer-executable instructions for the electronic device. Furthermore, processor 510 may be configured to communicate with memory 520 and execute the series of computer-executable instructions stored in memory 520 on the electronic device.

[0080] This application also provides a computer program product, which includes a computer program stored on a non-transitory computer-readable storage medium. The computer program includes program instructions, which, when executed by a computer, implement the steps of the processing method for the target business as described above.

[0081] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, apparatus, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0082] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems, devices), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0083] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0084] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0085] In a typical configuration, an electronic device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0086] Memory may include non-persistent storage in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0087] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0088] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus 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 apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0089] Those skilled in the art will understand that embodiments of this application can be provided as methods, apparatus, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0090] The above description is merely an embodiment of this application and is not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.

[0091] It should be understood that the training and prediction processes of the artificial intelligence (AI) models involved in the various embodiments of this specification all adhere to multiple legal and compliant principles, including legal data sources, compliant data content, compliant data governance, compliant training objectives and schemes, compliant training processes, compliant training environments and tools, and compliant ethical verification of training results, and comply with the requirements of Article 5 of the Patent Law. Among them: Data source legitimacy: All datasets used for AI model training were obtained through legal means, covering three categories: publicly authorized data, data authorized by partners, and self-collected compliant data. Publicly authorized data comes from compliant data sources following open-source licenses such as Apache 2.0, with complete copyright attribution and authorization scope clearly marked, and no unauthorized open-source code or data reuse. Data authorized by partners has been subject to formal data usage agreements, clearly defining the scope, duration, and confidentiality obligations, and possessing a complete authorization chain. For self-collected data involving personal information, strict informed consent procedures have been followed, and anonymization processes (including but not limited to field masking, feature anonymization, and differential privacy technology applications) have been implemented to remove personally identifiable information, fully complying with the requirements of relevant laws and regulations such as the "Interim Measures for the Administration of Generative Artificial Intelligence Services" and the "Personal Information Protection Law."

[0092] Data content compliance: The AI ​​model's dataset undergoes multiple screenings and cleaning processes to remove all content that may violate social morality or harm public interests. It contains no obscene, pornographic, violent, discriminatory, or information that endangers national or public safety, nor does it involve the illegal acquisition or use of genetic resources. For data in sensitive fields (such as healthcare and finance), an additional privacy-preserving computation module (including federated learning and secure multi-party computation technologies) ensures that the data is "usable but not visible," avoiding compliance risks during the original data transmission process and ensuring that the data application scenarios and uses comply with public order and good morals and industry regulatory requirements.

[0093] Data governance norms: A complete data traceability system is established during the AI ​​model training process to automatically record the source, collection time, annotation process, cleaning rules, and permission allocation of training data, generating traceable compliance reports to ensure that the data is verifiable throughout its entire lifecycle. The dataset annotation process for AI models is completed by a professional human R&D team, clearly defining the proportion of human creative contributions and avoiding reliance on AI-generated data that has not undergone substantial human modification, thus meeting the examination requirements for "human main contributions" in AI patent applications.

[0094] Training objectives and plans are compliant: The training objective of the AI ​​model focuses on [specific technical scenarios that can be supplemented, such as intelligent driving decision optimization, multimodal information interaction, etc., and replaced based on specific content]. The training scheme and the final output results do not violate any mandatory provisions of laws and administrative regulations, do not harm the public interest or the legitimate rights and interests of others, and do not pose any potential risks of being used for illegal activities, infringing on privacy, or disrupting public safety. The model strictly adheres to the ethical principle of "intelligent for good".

[0095] Training process compliance: A closed-loop training framework is adopted to ensure compliance and controllability of the training process. The specific process is as follows: First, training samples are obtained through compliant data sources. After the aforementioned data cleaning and desensitization, they are input into the neural network model to generate preliminary training results. Second, an expert system is introduced to verify the preliminary results. Based on preset rules and human expert experience, the feasibility of the results is evaluated, and outputs that may pose ethical risks or compliance hazards are corrected (such as removing decision-making logic that violates public order and good morals, and adjusting model parameters that do not comply with safety regulations). Finally, the loss function weights are dynamically optimized based on expert system feedback to strengthen the model's learning of compliant results, avoid overfitting errors or non-compliant labels, and form a closed-loop control of "data input - model training - expert verification - parameter optimization - result feedback" to ensure that the entire training process complies with A5 ethical review requirements.

[0096] Training environment and tool compliance: AI model training is implemented using nationally licensed chips and compliant training platforms. All open-source frameworks and components used in the training process have obtained their corresponding licenses, and copyright statements and patent citation information are fully retained, with no instances of infringement or reuse. The training environment is built using virtual devices (containers / virtual machines) with fixed random seeds and initial parameter configurations to ensure the reproducibility of the training process. Furthermore, through access control and operation log recording, risks such as data leakage and parameter tampering during training are prevented, ensuring the security and compliance of the training process.

[0097] Training results ethical verification compliance: After the model is trained, it undergoes additional third-party ethical compliance assessment and algorithm filing review to verify that the model output does not violate social morality or harm public interests. For potentially sensitive scenarios (such as public services and intelligent decision-making), a special result verification mechanism is established to ensure that the model always complies with Article 5 of the Patent Law and relevant laws and regulations in practical applications.

[0098] In summary, the data and training process used in the AI ​​model of this specification strictly comply with the relevant provisions of Article 5 of the Patent Law and the Patent Examination Guidelines (2023 Edition), and there are no violations of laws, social ethics, public interests, or illegal use of genetic resources. It fully meets the compliance requirements for patent authorization.

Claims

1. A method for processing a target business, characterized in that, The method includes the following steps: Acquire historical data related to the target business and preprocess the historical data; Features related to the set indicators of the target business are extracted from the preprocessed data, and key features are determined based on the feature selection algorithm; Based on the aforementioned key features, determine the model structure and parameter settings of the sub-model; Multiple sub-models are trained with parameter balancing, and the training results of the multiple sub-models are weighted and integrated to obtain an integrated prediction model for predicting the set index. The characteristics of the target business to be processed are input into the integrated prediction model to obtain the predicted values ​​of its set indicators, and the target business to be processed is prioritized according to the predicted values.

2. The method according to claim 1, characterized in that, The process of acquiring historical data related to the target business and preprocessing the historical data includes the following steps: The historical data is cleaned using an iterative loop cleaning method based on sample training residuals; The cleaned data was normalized using a multilinear normalization method that combines binning ratio and dynamic jump.

3. The method according to claim 1, characterized in that, The step of extracting features related to the set indicators of the target business from the preprocessed data and determining key features based on a feature selection algorithm includes the following steps: From the preprocessed data, behavioral characteristics, consumption characteristics, and temporal fluctuation characteristics of the target object are extracted, wherein the temporal fluctuation characteristics are obtained based on variance derivation and trend derivation processing of behavioral data and / or consumption data in the time dimension; The feature selection algorithm based on preset evaluation criteria sorts all extracted features and selects the top-ranked features as the key features according to a preset number.

4. The method according to claim 1, characterized in that, The process of determining the model structure and parameter settings of the sub-model based on the key features includes the following steps: Based on the data sample distribution corresponding to the key features, at least one machine learning algorithm is selected from the candidate algorithm set as the base model; The parameters of the base model are tuned, and cross-validation is used to train and evaluate the performance of multiple base models after parameter tuning in multiple rounds. Based on the results of multiple rounds of training and performance evaluation, the model structure and parameter settings of the sub-model are determined.

5. The method according to claim 1, characterized in that, The step of performing parameter balancing training on multiple sub-models and then performing weighted ensemble processing based on the training results of the multiple sub-models to obtain an ensemble prediction model for predicting the set indicator includes the following steps: For each of the multiple sub-models, a consistent parameter constraint rule is used for training. The parameter constraint rule is used to uniformly adjust at least one key parameter of each sub-model so that each sub-model meets the parallel computing conditions.

6. The method according to claim 5, characterized in that, The step of performing parameter balancing training on multiple sub-models and weighted ensemble processing based on the training results of the multiple sub-models to obtain an ensemble prediction model for predicting the set indicator further includes the following steps: For each sub-model, the proportion of its corresponding target business in historical data is used as the proportion of positive samples during the training of that sub-model. The proportion of negative samples used by the multiple sub-models during training is controlled to be at the same preset level.

7. The method according to claim 1, characterized in that, The step of performing parameter balancing training on multiple sub-models and then performing weighted ensemble processing based on the training results of the multiple sub-models to obtain an ensemble prediction model for predicting the set indicator includes the following steps: Determine the preset value coefficient of the target business associated with each sub-model; Using the preset value coefficient as the weight of the corresponding sub-model, the output results of the multiple sub-models are weighted and fused to obtain the integrated prediction model.

8. The method according to claim 1, characterized in that, It also includes the following steps: After prioritizing and executing the target services to be processed according to the predicted values, the actual execution result data of the target services is collected, and the actual execution result data is compared and analyzed with the corresponding predicted values. Based on the results of the comparative analysis, the integrated prediction model is iteratively updated. The iterative update includes at least one of the following operations: adjusting model parameters, updating training samples, and reselecting key features.

9. An electronic device, characterized in that, It includes a processor, a memory, a program or instructions stored in the memory and executable on the processor, wherein the program or instructions, when executed by the processor, implement the steps of the method as described in any one of claims 1 to 8.

10. A computer program product, characterized in that, The computer program product includes a computer program stored on a non-transitory computer-readable storage medium, the computer program including program instructions that, when executed by a computer, implement the steps of the method as described in any one of claims 1 to 8.

Images