Marketing processing method and device based on algorithm elastic combination and computer equipment

Abstract

The invention relates to a marketing processing method and device based on elastic algorithm combination, computer equipment and a storage medium. The method comprises the steps that a model library comprising a plurality of sub-models with different function dimensions is constructed, and each sub-model is packaged into an independent function module with a standardized data interface; in response to a processing task request from a service side, obtaining a task intention and context information; based on the task intention and the context information, matching and scheduling at least two sub-models from a model library to form a candidate sub-model set adaptive to the current processing task; connecting the sub-models in the candidate sub-model set to form a combined model execution link for processing the task request; and executing the combined model execution link, processing the input business data, and generating and outputting a comprehensive processing result corresponding to the processing task request. By adopting the method, more accurate, dynamic and explainable portraits of customers in a diversified marketing scene can be screened, and intelligent touch can be realized.

Description

Technical Field

[0001] This application relates to the field of data classification technology, and in particular to a marketing processing method, apparatus, computer equipment, computer-readable storage medium, and computer program product based on algorithmic flexible combination. Background Technology

[0002] In the fintech sector, data-driven intelligent models are often used to assess a company's risk and potential. Traditional solutions mainly employ a static, fixed clustering-classification pipeline: first, clustering algorithms (such as K-means) are used to unsupervisedly group the company's multi-dimensional features, forming a predetermined number of clusters; then, experts manually label each cluster with a uniform risk label; finally, using the company's features as input and the manually labeled clustering labels as targets, classification models such as decision trees are trained to classify and predict new companies.

[0003] This technology suffers from the following drawbacks: First, the model architecture is rigid and lacks flexibility. The entire marketing model pipeline is fixed, making it difficult to quickly adapt to changes in market strategies, the addition of new marketing channels, or changes in recommendation algorithms. Second, the process is fragile and lacks closed-loop optimization. The link from customer segmentation to response prediction is static, and the actual effect data of marketing campaigns (such as conversion rates) cannot be effectively fed back to the model front end to optimize segmentation and prediction, causing the model strategy to gradually deviate from actual market feedback. Third, the interpretability and action guidance of the results are insufficient. The model typically only outputs a customer category or response probability, making it difficult to provide fine-grained customer characteristic analysis that leads to the result, preventing marketers from deeply understanding customer motivations and adjusting personalized communication strategies. Finally, system expansion and maintenance costs are high. The marketing model needs continuous iteration, but the existing tightly coupled architecture makes it extremely difficult to introduce new data sources, algorithms, or business rules.

[0004] Therefore, there is an urgent need for a marketing processing method, device, computer equipment, computer-readable storage medium, and computer program product based on algorithmic flexible combination, which can achieve more accurate, dynamic, and interpretable profiling, screening, and intelligent outreach to customers in diversified marketing scenarios. Summary of the Invention

[0005] Therefore, it is necessary to provide a marketing processing method, device, computer equipment, computer-readable storage medium, and computer program product that can achieve more accurate, dynamic, and interpretable customer profiling, screening, and intelligent outreach in diversified marketing scenarios, based on an algorithm-based flexible combination, to address the aforementioned technical problems.

[0006] Firstly, this application provides a marketing processing method based on algorithmic flexible combination, including:

[0007] Construct a model library containing multiple sub-models with different functional dimensions, wherein each sub-model is encapsulated as an independent functional module with a standardized data interface;

[0008] In response to processing task requests from the business side, obtain task intent and context information;

[0009] Based on the task intent and context information, at least two sub-models are matched and scheduled from the model library to form a set of candidate sub-models adapted to the current processing task.

[0010] The sub-models in the candidate sub-model set are connected to form a combined model execution link for the processing task request;

[0011] The combined model execution chain is executed to process the input business data, generate and output a comprehensive processing result corresponding to the processing task request.

[0012] In one embodiment, the construction of a model library comprising multiple sub-models with distinct functional dimensions includes:

[0013] The system cleans and formats the accessed structured and unstructured source data, performs data transformation and business indicator calculation on the structured source data, and uses a natural language processing model to extract text features from the unstructured source data to form feature data.

[0014] The feature data is iteratively divided to form multiple data clusters with similar features;

[0015] For each data cluster, a classification algorithm is used to train the model, generating a classification sub-model specifically for that data cluster;

[0016] The classification sub-model and other functional dimension sub-models are standardized, encapsulated, and registered in the model library.

[0017] In one embodiment, the iterative partitioning of the feature data to form multiple data clusters with similar features includes:

[0018] The feature data is vectorized, and the target number of clusters is set.

[0019] Initialize the center point of each data cluster, calculate the distance between each data point and each center point iteratively, and divide each data point into the corresponding cluster according to the nearest neighbor principle;

[0020] Based on the partitioning results, the mean of all data points in each data cluster is recalculated to update the position of the center point of that data cluster;

[0021] Output the final data clustering results, where data points within the same cluster have similar features.

[0022] In one embodiment, the step of matching and scheduling at least two sub-models from the model library based on the task intent and context information to form a set of candidate sub-models adapted to the current processing task includes:

[0023] The task intent and context information are structured and parsed to extract the task type and key parameters;

[0024] Based on the task type, a preset model matching rule library is queried to determine a set of candidate sub-model types applicable to the current task type;

[0025] Based on the key parameters, calculate the parameter matching degree of each sub-model in the model library that belongs to the candidate sub-model type set;

[0026] Based on the order of parameter matching degree, at least two sub-models are selected from the model library, and the selected sub-models are marked as scheduled to be added to the candidate sub-model set.

[0027] In one embodiment, connecting the sub-models in the candidate sub-model set to form a combined model execution link for the processing task request includes:

[0028] Obtain the input data pattern and output data pattern defined for each sub-model in the candidate sub-model set;

[0029] Based on the data processing logic required by the processing task request, determine the data flow direction and execution order among the various sub-models in the candidate sub-model set;

[0030] Based on the data flow and execution order, a data format conversion interface is configured, and the sub-models are connected sequentially through a standardized data interface to construct a combined model execution link.

[0031] In one embodiment, after generating and outputting the comprehensive processing result corresponding to the processing task request, the method further includes:

[0032] Obtain runtime performance metrics of the combined model execution chain and business feedback results on the comprehensive processing results;

[0033] Based on the runtime performance metrics and the business feedback results, the processing efficiency of the combined model execution link is evaluated;

[0034] Based on the evaluation results, the model matching rule base and the combined model execution link are adjusted and optimized;

[0035] Under a preset business scenario, performance tests are conducted on different combinations of sub-models formed after adjustment and optimization, and the candidate sub-model set is updated.

[0036] Secondly, this application also provides a marketing processing device based on algorithmic flexible combination, comprising:

[0037] The building module is used to build a model library containing multiple sub-models with different functional dimensions, wherein each sub-model is encapsulated as an independent functional module with a standardized data interface.

[0038] The acquisition module is used to respond to processing task requests from the business side and acquire task intent and context information;

[0039] The matching module is used to match and schedule at least two sub-models from the model library based on the task intent and context information to form a set of candidate sub-models that are adapted to the current processing task.

[0040] The building module is also used to connect the sub-models in the candidate sub-model set to form a combined model execution link for the processing task request;

[0041] The application module is used to execute the combined model execution link, process the input business data, and generate and output the comprehensive processing result corresponding to the processing task request.

[0042] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to perform the following steps:

[0043] Construct a model library containing multiple sub-models with different functional dimensions, wherein each sub-model is encapsulated as an independent functional module with a standardized data interface;

[0044] In response to processing task requests from the business side, obtain task intent and context information;

[0045] Based on the task intent and context information, at least two sub-models are matched and scheduled from the model library to form a set of candidate sub-models adapted to the current processing task.

[0046] The sub-models in the candidate sub-model set are connected to form a combined model execution link for the processing task request;

[0047] The combined model execution chain is executed to process the input business data, generate and output a comprehensive processing result corresponding to the processing task request.

[0048] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, performs the following steps:

[0049] Construct a model library containing multiple sub-models with different functional dimensions, wherein each sub-model is encapsulated as an independent functional module with a standardized data interface;

[0050] In response to processing task requests from the business side, obtain task intent and context information;

[0051] Based on the task intent and context information, at least two sub-models are matched and scheduled from the model library to form a set of candidate sub-models adapted to the current processing task.

[0052] The sub-models in the candidate sub-model set are connected to form a combined model execution link for the processing task request;

[0053] The combined model execution chain is executed to process the input business data, generate and output a comprehensive processing result corresponding to the processing task request.

[0054] Fifthly, this application also provides a computer program product, including a computer program that, when executed by a processor, performs the following steps:

[0055] Construct a model library containing multiple sub-models with different functional dimensions, wherein each sub-model is encapsulated as an independent functional module with a standardized data interface;

[0056] In response to processing task requests from the business side, obtain task intent and context information;

[0057] Based on the task intent and context information, at least two sub-models are matched and scheduled from the model library to form a set of candidate sub-models adapted to the current processing task.

[0058] The sub-models in the candidate sub-model set are connected to form a combined model execution link for the processing task request;

[0059] The combined model execution chain is executed to process the input business data, generate and output a comprehensive processing result corresponding to the processing task request.

[0060] The aforementioned marketing processing methods, devices, computer equipment, computer-readable storage media, and computer program products based on algorithmic flexible combination construct a standardized and modular sub-model library. They intelligently match and schedule model components according to dynamically parsed business intent, assembling them on demand into customized execution links, ultimately producing comprehensive results. This process fundamentally breaks the rigid architecture of traditional static model pipelines, enabling the system to flexibly and quickly respond to diverse marketing scenario needs, significantly improving the model's adaptability and agile response to changes in market strategies, new marketing channels, and customer segmentation requirements. Simultaneously, the modular design and standardized interfaces greatly enhance the system's scalability and maintainability, supporting low-cost and efficient integration and testing of new marketing models and strategies. By integrating the intelligence and results of multi-dimensional sub-models, it provides more comprehensive, granular, and interpretable customer insights and marketing decision-making basis than a single model, thereby significantly improving the accuracy of customer profiling, the precision of marketing screening, and the reliability of outreach strategies. Furthermore, this architecture lays a closed-loop foundation for the continuous optimization of model combination strategies based on marketing performance feedback (such as conversion rate and customer response), and promotes the transformation of marketing decision-making systems from experience-driven and static to data-intelligent driven and autonomous. Attached Figure Description

[0061] To more clearly illustrate the technical solutions in the embodiments of this application or related technologies, the drawings used in the description of the embodiments of this application or related technologies will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0062] Figure 1 This is an application environment diagram of a marketing processing method based on algorithmic flexible combination in one embodiment;

[0063] Figure 2 This is a flowchart illustrating a marketing processing method based on algorithmic flexible combination in one embodiment;

[0064] Figure 3 This is a flowchart illustrating a marketing processing method based on algorithmic flexible combination in another embodiment;

[0065] Figure 4 This is a structural block diagram of a marketing processing device based on algorithmic flexible combination in one embodiment;

[0066] Figure 5 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0067] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0068] It should be noted that the terms "first," "second," etc., used in this application can be used to describe various elements, but these elements are not limited by these terms. These terms are only used to distinguish the first element from the second element. The terms "comprising" and "having," and any variations thereof, used in this application, are intended to cover non-exclusive inclusion. The term "multiple" used in this application refers to two or more. The term "and / or" used in this application refers to one of the embodiments, or any combination of multiple embodiments.

[0069] In one exemplary embodiment, such as Figure 1 As shown, a marketing processing method based on algorithmic flexible combination is provided, including the following steps S102 to S110. Wherein:

[0070] Step S102: Construct a model library containing multiple sub-models with different functional dimensions, wherein each sub-model is encapsulated as an independent functional module with standardized data interfaces.

[0071] Specifically, firstly, building a model library containing multiple sub-models with distinct functional dimensions refers to establishing a pre-trained and well-managed pool of model resources. The distinct functional dimensions mean that each sub-model in the library is designed for a different stage or analytical perspective in the fintech data processing workflow. For example, the library might include clustering models for initial enterprise grouping, classification models for in-depth risk assessment based on specific grouping results, and rule-based decision models that do not rely on machine learning but are based solely on industry experience and business rules. The clustering model's functional dimension lies in discovering the inherent structure of the data, the classification model's functional dimension lies in performing supervised decision-making tasks, and the rule-based model's functional dimension lies in providing deterministic logical judgments. These sub-models together constitute a complementary model set covering the entire data processing chain.

[0072] Furthermore, each sub-model is encapsulated as an independent functional module with a standardized data interface. This standardized data interface defines the format, type, and structure of the input data accepted by each sub-model, as well as the format, type, and structure of its output results, implemented through a unified application programming interface (API) specification. For example, for a sub-model used to evaluate a company's R&D capabilities, its standardized input interface could be defined as receiving a JSON object containing key fields such as company name, total R&D expenditure, number of R&D personnel, and number of patent applications; its standardized output interface would be defined as returning a JSON object containing R&D capability score and its corresponding level. Through this encapsulation, each sub-model becomes a black-box, independent service. Regardless of the specific algorithm used internally (e.g., decision tree, logistic regression, or random forest), it presents a consistent and standardized calling method externally.

[0073] Step S104: In response to the processing task request from the business side, obtain the task intent and context information.

[0074] Specifically, during system operation, the business side initiates specific processing task requests. These requests carry a composite information carrier of business objectives and constraints. Upon responding to the request, the system parses and retrieves the task intent and contextual information contained within. The task intent refers to the core business objective the request aims to achieve. For example, in a fintech scenario, the task intent might be to conduct an initial screening of loan eligibility for a batch of newly registered technology companies, or to reassess the quarterly risk status of invested companies. Contextual information refers to auxiliary, background, or binding information related to the execution of this task, defining the specific environment and conditions for task execution. Contextual information may include, but is not limited to: the specific batch of data or list of companies targeted by this task, the processing time required by the business department (e.g., results must be returned within 2 hours), the latest policy or rule version number to be followed in this assessment, and the expected output format (e.g., a detailed scoring report or a simple pass / fail conclusion).

[0075] Step S106: Based on the task intent and context information, match and schedule at least two sub-models from the model library to form a set of candidate sub-models that are adapted to the current processing task.

[0076] Specifically, the system maintains a model capability directory, which is essentially a metadata database. This directory records a series of attribute labels for each sub-model, such as: model function (e.g., customer segmentation, credit scoring, text sentiment analysis), applicable business scenarios (e.g., pre-loan review, post-loan monitoring, marketing response prediction), required input data format and dimensions, and performance characteristics (e.g., computational complexity, average accuracy). The system compares the task intent with these attribute labels. For example, when the task intent is identified as making an initial assessment of default risk for technology companies, the system will prioritize matching sub-models whose function labels include risk identification and company assessment, and whose scenario label includes pre-loan. Contextual information is used for further refined filtering. For example, if the context indicates that this assessment should focus on intellectual property risk, the system will additionally favor sub-models whose descriptions involve patent analysis or innovation indicators.

[0077] Furthermore, the scheduler considers constraints in the context (such as task completion deadlines) and the real-time state of the system (such as the load of the computing cluster and the length of the task queue). Based on this information, it allocates appropriate computing resources to candidate models (e.g., specifying that they run on a containerized instance) and may determine their initial execution order or priority. For example, if the system is currently under high load, the scheduler may allocate more computing resources to a computationally intensive deep learning model or queue it up, while prioritizing a lightweight rule-based model to perform some of the filtering work first.

[0078] Ultimately, at least two sub-models that successfully pass functional matching and obtain resource scheduling permission are organized into a candidate sub-model set. For example, for a task involving a comprehensive enterprise assessment, the final candidate set might include: a data standardization sub-model, an industry classification sub-model, a financial health analysis sub-model, and a technological innovation assessment sub-model.

[0079] Step S108: Connect the sub-models in the candidate sub-model set to form a combined model execution link for processing task requests.

[0080] Specifically, the system can pre-define various common task logic templates or automatically infer the connection order by analyzing the dependencies between the inputs and outputs of sub-models. Each sub-model has its input and output interface data patterns explicitly defined during encapsulation. When building the link, the system obtains the interface specifications of each sub-model in the candidate set and checks the compatibility of preceding and following sub-models sequentially, i.e., whether the output data pattern of the predecessor model meets the input data pattern requirements of the successor model. If fully compatible, they are directly connected through their interfaces; if not fully compatible, the system automatically inserts a data conversion adapter between them.

[0081] The resulting combined model execution chain is a directed acyclic graph structure, where nodes are sub-models (or data transformation adapters) and edges represent standardized data flows. This chain is instantiated as an executable object, encapsulating complete execution logic, model call sequences, and data flow paths. This chain is a one-time solution dynamically generated for this processing task request; it dissolves upon task completion.

[0082] Step S110: Execute the combined model execution link, process the input business data, and generate and output the comprehensive processing result corresponding to the processing task request.

[0083] Specifically, the input business data is imported into the starting node of the chain, and then flows sequentially through each sub-model or data transformation adapter in the chain according to the preset data flow path and control logic. Each node, upon receiving the input data, executes its encapsulated specific function and outputs the processed result to the next node. For example, in a corporate credit assessment chain, the raw corporate data is first processed by the standardization sub-model, outputting cleaned structured data; this data is then processed by the clustering sub-model, outputting the category label to which the company belongs; this label, along with the raw data, is passed to a dedicated credit scoring sub-model for that category, outputting a preliminary credit score; finally, this score, along with a risk signal output from another parallel anti-fraud sub-model, is sent to a decision fusion sub-model, which integrates information from multiple sources according to preset rules or meta-learning strategies to generate the final comprehensive processing result.

[0084] The generated comprehensive processing result is not the output of a single sub-model in the process chain, but rather the final product of the entire chain's collaborative work. Its content and form closely correspond to the initial processing task request. This comprehensiveness is reflected in its integration of professional judgments from multiple sub-models and various algorithmic perspectives, thus providing a more comprehensive, reliable, and granular decision-making basis than a single model.

[0085] The system encapsulates and transmits the comprehensive processing results according to the output format and delivery requirements specified in the task request context information. The execution process of the entire link is usually monitored, and its performance indicators (such as the time consumed at each stage and resource utilization) and the quality indicators of the final output (such as confidence level) are recorded, providing a data foundation for subsequent link performance evaluation and optimization iteration.

[0086] The aforementioned marketing processing method based on algorithmic flexible combination constructs a standardized and modular sub-model library. Model components are intelligently matched and scheduled according to dynamically parsed business intent, and then assembled into customized execution links on demand. Finally, running this link produces comprehensive results. This process fundamentally breaks the rigid architecture of traditional static model pipelines, enabling the system to flexibly and quickly respond to diverse marketing scenario needs, significantly improving the model's adaptability and agile response to changes in market strategies, new marketing channels, and customer segmentation requirements. Simultaneously, the modular design and standardized interfaces greatly enhance the system's scalability and maintainability, supporting low-cost and efficient integration and testing of new marketing models and strategies. By integrating the intelligence and results of multi-dimensional sub-models, it provides more comprehensive, granular, and interpretable customer insights and marketing decision-making basis than a single model, thereby significantly improving the accuracy of customer profiling, the precision of marketing screening, and the reliability of outreach strategies. Furthermore, this architecture lays a closed-loop foundation for the continuous optimization of model combination strategies based on marketing performance feedback (such as conversion rate and customer response), and promotes the transformation of marketing decision-making systems from experience-driven and static to data-intelligent driven and autonomous.

[0087] In one exemplary embodiment, such as Figure 2 As shown, a model library is constructed containing multiple sub-models with different functional dimensions, including:

[0088] Step S202 involves cleaning and formatting the accessed structured and unstructured source data, performing data transformation and business indicator calculation on the structured source data, and using a natural language processing model to extract text features from the unstructured source data to form feature data.

[0089] Step S204: Iteratively divide the feature data to form multiple data clusters with similar features;

[0090] Step S206: For each data cluster, a classification algorithm is used to train the model and generate a classification sub-model specifically for that data cluster.

[0091] Step S208: Standardize and encapsulate the classification sub-model and other functional dimension sub-models respectively, and register them to the model library.

[0092] Specifically, the construction of the model library first processes multi-source data. Structured and unstructured source data are accessed through a distributed acquisition framework, formatted, and stored in a data lake, with data separated and stored according to access frequency. The accessed structured data (such as financial data) and unstructured data (such as text reports) are cleaned. The cleaned structured data undergoes transformation and calculation to generate business indicators such as R&D intensity and patent growth rate. Simultaneously, natural language processing models are used to analyze unstructured text, extracting text features such as technical keywords and public opinion sentiment.

[0093] After extracting structured business metrics and unstructured text features, the two types of features are aligned and merged to form unified, multi-dimensional feature data, which can also be considered a preliminary set of labels. Subsequently, the feature data undergoes quality verification, such as checking data integrity, reasonable range, and logical consistency. After verification, all feature labels are stored in a unified feature label library for centralized storage and management.

[0094] To improve data utilization efficiency, a corresponding data model and index structure are built upon this feature tag library. The data model defines the business relationships and logical constraints between features, while the index is designed to accelerate the retrieval speed for specific feature combinations or conditions. This design effectively supports subsequent in-depth mining and analysis of complex relationships between enterprises and indicators. The system utilizes containerization technology to encapsulate data processing tasks, enabling computing resources to dynamically scale elastically according to the load, thereby improving resource utilization and coping with traffic fluctuations. Furthermore, a persistence mechanism for the computing process is established by persistently storing the intermediate states and results of key computing processes.

[0095] Based on this feature data, a clustering algorithm is used to iteratively divide the data into multiple clusters with similar features. Next, for each cluster, a classification algorithm is used to train a classification sub-model specifically for that cluster, using the feature data and historical risk labels of the companies within that cluster. The model library not only includes the aforementioned classification sub-models but also models for other functional dimensions, such as decision sub-models built based on business rules. Finally, all sub-models are standardized and encapsulated, i.e., a unified input / output data interface is defined and registered in the model library directory, completing the construction of the model library.

[0096] In this embodiment, a unified feature is constructed by fusing multi-source data, and a dedicated classification model is trained based on clustering. The resulting sub-models are highly targeted and accurate. Combined with standardized encapsulation and management, the resulting model library has highly modular, reusable, and easily extensible characteristics, providing a high-quality and diverse model component foundation for subsequent on-demand combination of dynamic model links, effectively improving the adaptability and maintainability of the overall system.

[0097] In one exemplary embodiment, such as Figure 3 As shown, the feature data is iteratively divided to form multiple data clusters with similar features, including:

[0098] Step S302: Vectorize the feature data and set the target number of clusters;

[0099] Step S304: Initialize the center point of each data cluster, calculate the distance between each data point and each center point iteratively, and divide each data point into the corresponding cluster according to the nearest neighbor principle;

[0100] Step S306: Based on the partitioning results, recalculate the mean of all data points in each data cluster to update the position of the center point of the data cluster;

[0101] Step S308: Output the final data cluster partitioning result, where data points within the same cluster have feature similarity.

[0102] Specifically, firstly, the feature data generated in the previous step is transformed into a mathematical form that the algorithm can process, i.e., vectorized representation. Simultaneously, based on prior business knowledge or algorithm evaluation, the desired number of groups into which the data will be divided is set, i.e., the target number of clusters. Subsequently, multiple points are randomly selected or initialized in the feature vector space using a specific strategy as the initial center points of each data cluster.

[0103] The iterative optimization phase then begins. In each iteration, two key operations are performed: First, the distance (usually Euclidean distance) from each data point in the feature space (i.e., the feature vector of a company sample) to the current cluster centers is calculated. Based on the nearest neighbor principle, each data point is assigned to the cluster represented by its nearest center. Second, after all data points have been partitioned, for each newly formed cluster, the average value of all data points within that cluster across each feature dimension is recalculated. This average value vector becomes the new center point location for that cluster.

[0104] The above steps are repeated until a stopping condition is met, such as the change in the center point position being less than a preset threshold, or the maximum number of iterations being reached. Finally, the algorithm outputs a stable data cluster partitioning result. Since the algorithm's optimization objective is to minimize the sum of distances from samples within a cluster to their center point, the resulting data points within the same cluster exhibit high similarity in their feature representations, while the feature differences between different clusters are relatively large.

[0105] After obtaining multiple data clusters with similar features through the aforementioned clustering, a tree-based classification algorithm, such as decision trees, will be used to train a model for each individual data cluster. When training the classification sub-model specific to each cluster, the features most capable of distinguishing different categories are recursively selected to partition the data based on quantification criteria such as information gain and Gini impurity. Through this top-down feature selection and data partitioning, a tree-structured classification model is ultimately constructed. This model can perform refined risk classification predictions for new enterprise samples falling into their corresponding data clusters.

[0106] In this embodiment, by setting clear objectives and iterative optimization, high-dimensional feature data is systematically divided into multiple homogeneous subgroups. The sample features within each data cluster are highly similar, enabling the model to more accurately capture and characterize the unique patterns of different enterprise groups, thereby fundamentally improving the discrimination accuracy and generalization ability of subsequent classification sub-models.

[0107] In an exemplary embodiment, based on task intent and context information, at least two sub-models are matched and scheduled from a model library to form a set of candidate sub-models adapted to the current processing task, including:

[0108] The task intent and context information are structured and parsed to extract the task type and key parameters;

[0109] Based on the task type, query the preset model matching rule library to determine the set of candidate sub-model types applicable to the current task type;

[0110] Based on key parameters, calculate the parameter matching degree of each sub-model belonging to the candidate sub-model type set in the model library;

[0111] Based on the parameter matching degree, at least two sub-models are selected from the model library, and the selected sub-models are marked as scheduled to be added to the candidate sub-model set.

[0112] Specifically, firstly, the system performs structured parsing of the acquired task intent and context information, transforming it into explicit instructions that the machine can process, extracting task types and key parameters, and completing the transformation from natural language or semi-structured business requests to standardized technical parameters.

[0113] Subsequently, based on the parsed task type, the system queries a pre-defined model matching rule base. This rule base defines the recommended combinations of sub-model types for different task types. Based on this, the system identifies a set of candidate sub-model types, narrowing down the scope for subsequent precise matching. Within this set of candidate sub-model types, the system calculates the specific degree of matching between each sub-model in the model base that matches the current task, i.e., the parameter matching degree, based on the parsed key parameters. The calculation of the parameter matching degree may consider factors such as the sub-model's design goals, the fit between the required input data and the currently provided data, and the match between the model's complexity and the task's timeliness requirements. For example, for the parameter of risk preference threshold, a scoring model optimized for an aggressive investment strategy will have a lower matching degree than a model designed for a "conservative" strategy.

[0114] Finally, the system sorts all candidate models based on the calculated parameter matching degree and selects at least two sub-models according to a preset strategy (such as selecting the Top N, or selecting all models with matching degrees exceeding a threshold). After the selection is completed, the system marks these sub-models as scheduled in the resource scheduler.

[0115] In this embodiment, a two-stage progressive screening mechanism—coarse screening by task type and fine screening by key parameters—enables the rapid and accurate identification of model components highly compatible with the current task from the Haiquanzi model. This ensures that the selected model aligns with the business objectives in terms of functionality and that its specific execution parameters are optimally matched with the task details. Simultaneously, by pre-marking the selected model as scheduled, execution resources are bound and reserved in advance, significantly improving the startup efficiency and overall resource utilization of subsequent model combinations and task execution, achieving intelligent and automated on-demand allocation of model resources.

[0116] In an exemplary embodiment, the sub-models in the candidate sub-model set are connected to form a combined model execution chain for processing task requests, including:

[0117] Obtain the input data pattern and output data pattern defined for each sub-model in the candidate sub-model set;

[0118] Based on the data processing logic required by the task request, determine the data flow and execution order among the various sub-models in the candidate sub-model set;

[0119] Based on the data flow and execution order, a data format conversion interface is configured, and sub-models are connected sequentially through a standardized data interface to build a combined model execution link.

[0120] Specifically, first, the system obtains the interface contract for each sub-model in the candidate sub-model set, that is, the defined input data pattern and output data pattern. For example, it requires input of a JSON object containing an array of company IDs and financial indicators, and output of a JSON object containing a risk score and confidence level.

[0121] Subsequently, the system determines the data flow and execution order among these sub-models based on the data processing logic inherent in the current task request. For example, for a risk assessment task, the logical order is to first standardize the data, then segment the enterprises, then call different risk scoring models based on the segmentation results, and finally summarize the results.

[0122] After determining the order, the system sequentially checks adjacent sub-models in the sequence: if the output data pattern of the predecessor model fully meets the input data pattern requirements of the successor model, the two are directly connected through their standardized data interfaces. If the patterns are incompatible (e.g., the predecessor outputs score A, while the successor requires score B), the system automatically configures a data format conversion interface between them. This interface is a lightweight adapter responsible for data conversion, field renaming, or simple calculations according to predefined mapping rules to bridge the interface differences. Ultimately, through the above process, all sub-models (and data conversion interfaces inserted where necessary) are connected in series or parallel through standardized data interfaces in a determined order, forming a directed acyclic data processing graph, i.e., the combined model execution chain.

[0123] In this embodiment, the interface compatibility issue between heterogeneous sub-models is resolved by dynamically configuring the data format conversion interface, ensuring smooth and accurate data flow between multiple models. This allows previously independent and loosely connected functional modules to be quickly integrated into a logically rigorous and collaborative whole, thereby transforming modular model capabilities into end-to-end solutions that can directly address specific business problems. This significantly improves the flexibility and automation level of building customized analysis chains within the system.

[0124] In an exemplary embodiment, after generating and outputting the comprehensive processing result corresponding to the processing task request, the method further includes:

[0125] Obtain runtime performance metrics of the combined model execution chain and business feedback results on the comprehensive processing results;

[0126] The processing efficiency of the combined model execution chain is evaluated based on runtime performance metrics and business feedback results.

[0127] Based on the evaluation results, adjustments and optimizations were made to the model matching rule base and the combined model execution chain;

[0128] Under the preset business scenario, performance tests are conducted on different combinations of sub-models formed after adjustment and optimization, and the candidate sub-model set is updated.

[0129] Specifically, firstly, after the combined model execution link runs, the system actively collects two types of data: one is runtime performance indicators, such as the computation time, resource consumption (CPU / memory), and throughput of the entire link and its internal sub-models; the other is business feedback results, namely the evaluation of the comprehensive processing results of this output by downstream business systems or experts, such as whether the risk warning is accurate, whether the customer segmentation is effective, and cases of misjudgment confirmed after manual review.

[0130] Subsequently, based on the collected performance metrics and business feedback, the system quantitatively evaluates the processing efficiency of the combined model execution chain. The evaluation may comprehensively consider multiple dimensions such as efficiency (e.g., whether it meets timeliness requirements), effectiveness (e.g., accuracy, recall), and cost (e.g., computing resource consumption), forming a comprehensive performance score or diagnostic report.

[0131] Next, the model matching rule base is adjusted, for example, by modifying the mapping rules between specific task types and sub-model types, or adjusting the calculation weights of parameter matching degree. The execution chain of the combined model itself is also adjusted, for example, by replacing a sub-model that is not performing well, adjusting the connection order between sub-models, or inserting new data processing steps into the data flow.

[0132] Finally, to accumulate knowledge and prevent overfitting, the system conducts comprehensive performance tests and comparative experiments on different combinations of sub-models after adjustments and optimizations under preset typical business scenarios. Through testing, the advantageous scenarios and performance boundaries of each combination strategy are clarified (for example, combination A is efficient in high-concurrency simple query scenarios, while combination B is accurate in complex deep analysis scenarios). Based on these test conclusions, the system does not simply update the candidate sub-model set for a particular task, but rather preserves the validated combination strategies, their applicable scenarios, and performance data as knowledge and updates them to a higher-level model combination strategy library or experience base.

[0133] In this embodiment, a dynamic adjustment mechanism based on evaluation results enables online optimization and correction of model matching rules and combined link structures, thereby continuously improving the accuracy and efficiency of task execution. This not only enhances the system's reliability and adaptability to diverse scenarios but also promotes the continuous evolution of the entire model system from static deployment, ensuring long-term scalability and stability.

[0134] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0135] Based on the same inventive concept, this application also provides an algorithm-based flexible combination marketing processing apparatus for implementing the above-mentioned algorithm-based flexible combination marketing processing method. The solution provided by this apparatus is similar to the implementation scheme described in the above method. Therefore, the specific limitations of one or more algorithm-based flexible combination marketing processing apparatus embodiments provided below can be found in the limitations of the algorithm-based flexible combination marketing processing method described above, and will not be repeated here.

[0136] In one exemplary embodiment, such as Figure 4 As shown, a marketing processing device based on algorithmic flexible combination is provided, including:

[0137] Module 402 is used to build a model library containing multiple sub-models with different functional dimensions, wherein each sub-model is encapsulated as an independent functional module with a standardized data interface.

[0138] The 404 module is used to respond to a processing task request from the business side and obtain the task intent and context information.

[0139] The matching module 406 is used to match and schedule at least two sub-models from the model library based on task intent and context information to form a set of candidate sub-models that are adapted to the current processing task.

[0140] Module 402 is also used to connect the sub-models in the candidate sub-model set to form a combined model execution link for processing task requests;

[0141] Application module 408 is used to execute the combined model execution link, process the input business data, and generate and output the comprehensive processing results corresponding to the processing task requests.

[0142] In an exemplary embodiment, the construction module 402 is specifically used to clean and format the accessed structured source data and unstructured source data, perform data transformation and business indicator calculation on the structured source data, and use a natural language processing model to extract text features from the unstructured source data to form feature data; iteratively divide the feature data to form multiple data clusters with similar features; for each data cluster, use a classification algorithm to train the model and generate a classification sub-model specifically for that data cluster; and standardize and encapsulate the classification sub-model and other functional dimension sub-models respectively and register them to the model library.

[0143] In an exemplary embodiment, the construction module 402 is further configured to vectorize the feature data and set the target number of clusters; initialize the center point of each data cluster; iteratively calculate the distance between each data point and each center point; divide each data point into the corresponding cluster according to the nearest neighbor principle; recalculate the mean of all data points in each data cluster according to the division result to update the center point position of the data cluster; and output the final data cluster division result, wherein the data points in the same cluster have feature similarity.

[0144] In an exemplary embodiment, the matching module 406 is specifically used to perform structured parsing of task intent and context information, extract task type and key parameters; query a preset model matching rule library according to the task type to determine a set of candidate sub-model types applicable to the current task type; calculate the parameter matching degree of each sub-model in the model library that belongs to the set of candidate sub-model types based on the key parameters; select at least two sub-models from the model library according to the order of parameter matching degree, and mark the selected sub-models as scheduled to add them to the set of candidate sub-models.

[0145] In an exemplary embodiment, the construction module 402 is further configured to obtain the input data pattern and output data pattern defined by each sub-model in the candidate sub-model set; determine the data flow direction and execution order between each sub-model in the candidate sub-model set according to the data processing logic required by the processing task request; configure a data format conversion interface based on the data flow direction and execution order, and connect the sub-models sequentially through a standardized data interface to construct a combined model execution link.

[0146] In an exemplary embodiment, the application module 408 is further configured to obtain runtime performance metrics of the combined model execution link and business feedback results on the comprehensive processing results; evaluate the processing efficiency of the combined model execution link based on the runtime performance metrics and business feedback results; adjust and optimize the model matching rule base and the combined model execution link according to the evaluation results; and perform performance tests on different sub-model combinations formed after adjustment and optimization under a preset business scenario, and update the candidate sub-model set.

[0147] Each module in the aforementioned marketing processing device based on algorithmic flexible combination can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the operations corresponding to each module.

[0148] In one exemplary embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 5 As shown, this computer device includes a processor, memory, input / output interfaces (I / O), and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The database stores business data. The I / O interfaces are used for exchanging information between the processor and external devices. The communication interface is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements a marketing processing method based on algorithmic flexible combination.

[0149] Those skilled in the art will understand that Figure 5 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0150] In one exemplary embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the method described above.

[0151] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps of the above-described method.

[0152] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps of the method described above.

[0153] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0154] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, artificial intelligence (AI) processors, etc., and are not limited to these.

[0155] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0156] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A marketing processing method based on algorithmic elastic combination, characterized in that, The method includes: Construct a model library containing multiple sub-models with different functional dimensions, wherein each sub-model is encapsulated as an independent functional module with a standardized data interface; In response to processing task requests from the business side, obtain task intent and context information; Based on the task intent and context information, at least two sub-models are matched and scheduled from the model library to form a set of candidate sub-models adapted to the current processing task. The sub-models in the candidate sub-model set are connected to form a combined model execution link for the processing task request; The combined model execution chain is executed to process the input business data, generate and output a comprehensive processing result corresponding to the processing task request.

2. The method of claim 1, wherein, The construction of a model library containing multiple sub-models with different functional dimensions includes: The system cleans and formats the accessed structured and unstructured source data, performs data transformation and business indicator calculation on the structured source data, and uses a natural language processing model to extract text features from the unstructured source data to form feature data. The feature data is iteratively divided to form multiple data clusters with similar features; For each data cluster, a classification algorithm is used to train the model, generating a classification sub-model specifically for that data cluster; The classification sub-model and other functional dimension sub-models are standardized, encapsulated, and registered in the model library.

3. The method of claim 2, wherein, The iterative partitioning of the feature data to form multiple data clusters with similar features includes: The feature data is vectorized, and the target number of clusters is set. Initialize the center point of each data cluster, calculate the distance between each data point and each center point iteratively, and divide each data point into the corresponding cluster according to the nearest neighbor principle; Based on the partitioning results, the mean of all data points in each data cluster is recalculated to update the position of the center point of that data cluster; Output the final data clustering results, where data points within the same cluster have similar features.

4. The method of claim 1, wherein, Based on the task intent and context information, at least two sub-models are matched and scheduled from the model library to form a set of candidate sub-models adapted to the current processing task, including: The task intent and context information are structured and parsed to extract the task type and key parameters; Based on the task type, a preset model matching rule library is queried to determine a set of candidate sub-model types applicable to the current task type; Based on the key parameters, calculate the parameter matching degree of each sub-model in the model library that belongs to the candidate sub-model type set; Based on the order of parameter matching degree, at least two sub-models are selected from the model library, and the selected sub-models are marked as scheduled to be added to the candidate sub-model set.

5. The method of claim 1, wherein, The step of connecting the sub-models in the candidate sub-model set to form a combined model execution link for the processing task request includes: Obtain the input data pattern and output data pattern defined for each sub-model in the candidate sub-model set; Based on the data processing logic required by the processing task request, determine the data flow direction and execution order among the various sub-models in the candidate sub-model set; Based on the data flow and execution order, a data format conversion interface is configured, and the sub-models are connected sequentially through a standardized data interface to construct a combined model execution link.

6. The method of claim 4, wherein, After generating and outputting the comprehensive processing result corresponding to the processing task request, the method further includes: Obtain runtime performance metrics of the combined model execution chain and business feedback results on the comprehensive processing results; Based on the runtime performance metrics and the business feedback results, the processing efficiency of the combined model execution link is evaluated; Based on the evaluation results, the model matching rule base and the combined model execution link are adjusted and optimized; Under a preset business scenario, performance tests are conducted on different combinations of sub-models formed after adjustment and optimization, and the candidate sub-model set is updated.

7. A marketing processing device based on algorithmic elastic combination, characterized by, The device includes: The building module is used to build a model library containing multiple sub-models with different functional dimensions, wherein each sub-model is encapsulated as an independent functional module with a standardized data interface. The acquisition module is used to respond to processing task requests from the business side and acquire task intent and context information; The matching module is used to match and schedule at least two sub-models from the model library based on the task intent and context information to form a set of candidate sub-models that are adapted to the current processing task. The building module is also used to connect the sub-models in the candidate sub-model set to form a combined model execution link for the processing task request; The application module is used to execute the combined model execution link, process the input business data, and generate and output the comprehensive processing result corresponding to the processing task request.

8. A computer device comprising a memory and a processor, the memory storing a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 6.

9. A computer readable storage medium having stored thereon a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.

Images