Cross-platform advertisement placement optimization method and system

Abstract

The application provides a cross-platform advertisement optimization method and system, and belongs to the technical field of advertisement optimization. The method first analyzes advertisement optimization requirements to generate an advertisement optimization task definition containing a target function and a smoothing parameter, then collects multi-platform original reports and business backlink data to process and generate a unified index view; subsequently, candidate strategy combinations corresponding to each advertisement platform are generated and coded into strategy vectors in combination with the advertisement optimization task definition, and platform-level and strategy-level distribution of the total budget is completed; then, the candidate strategy combinations are converted into advertisement optimization parameters, and an advertisement optimization plan is created and updated; finally, multi-dimensional effect attribution is completed by aggregating advertisement optimization data, real performance indicators of the candidate strategy combinations are calculated, a strategy knowledge base is iteratively updated, and a self-iterative optimization closed loop is formed. The application can realize automatic and refined optimization of cross-platform advertisement optimization, improve advertisement optimization efficiency and income, and reduce customer acquisition cost.

Description

Technical Field

[0001] This application relates to the field of advertising delivery technology, and in particular to a cross-platform advertising delivery optimization method and system. Background Technology

[0002] Currently, when companies conduct cross-platform advertising, they generally rely on manual operation to complete the entire process. The business side can only manually set the overall budget and advertising goals, and roughly allocate the total budget to each advertising platform based on experience. The advertising team needs to independently build advertising plans on each platform, manually configure parameters such as bidding, frequency control, and creative combinations. Data from each platform is only displayed in its own backend, without a unified viewing perspective.

[0003] Subsequent analysis of campaign performance was limited to manually exporting reports from various platforms daily or weekly, followed by adjustments to budgets and bidding strategies based on experience. This approach not only resulted in infrequent adjustments and missed market optimization windows, but also made objective comparisons of cross-platform metrics impossible due to differences in statistical methods, conversion feedback delays, and deduplication rules across different advertising platforms. Furthermore, the vast potential for strategy combinations, comprised of multiple parameters such as bidding models and campaign pacing, made it difficult for manual, systematic multi-dimensional A / B testing and strategy exploration; only localized, experience-based adjustments were feasible.

[0004] In existing technologies, single-platform automated ad delivery tools can only achieve intelligent bidding and delivery within the platform, and cannot complete cross-platform budget coordination and performance attribution. Traditional cross-channel budget allocation methods can only make static linear allocations based on historical data, lacking online adaptive capabilities and the exploration-utilization closed loop of strategy exploration. They cannot solve the ad delivery decision-making difficulties caused by cross-platform metric differences and delayed feedback, resulting in low overall ad delivery efficiency, high customer acquisition costs, and difficulty in meeting the needs of large-scale and refined cross-platform advertising. Summary of the Invention

[0005] To address the aforementioned issues, the main objective of this application is to propose a cross-platform advertising optimization method and system. This system aims to automate and refine the entire cross-platform advertising process, thereby resolving technical problems in traditional advertising such as incomparable metrics, low adjustment frequency, insufficient strategy exploration, and difficulty in cross-platform collaboration. Ultimately, this will improve advertising efficiency and return on investment, and reduce customer acquisition costs.

[0006] To achieve the above objectives, this application proposes a cross-platform advertising optimization method, the method comprising:

[0007] The algorithm receives advertising placement requests and parses them into structured objectives and constraints that the algorithm can execute, generating a placement task definition. The placement task definition includes at least an objective function, a set of constraints, an indicator definition, an evaluation time window, an update frequency, and a smoothing parameter. Collect raw report data and post-business link data from multiple advertising platforms, process the collected data, and generate a unified indicator view; Based on the defined delivery task, the unified indicator view, the controllable parameter capabilities and strategy knowledge base of each advertising platform, candidate strategy combinations corresponding to each advertising platform are generated. Then, the candidate strategy combinations of each advertising platform are encoded into corresponding strategy vectors, and a one-to-one mapping relationship is established between each strategy vector and the platform parameter set of the corresponding advertising platform. The candidate strategy combinations include a baseline delivery strategy, a fallback delivery strategy, and an exploratory delivery strategy. The campaign is defined as a constraint and optimization objective. Based on the unified indicator view and the campaign results of the previous campaign cycle, the total budget is allocated at the platform level under the conditions of satisfying the total budget, the upper and lower limits of the proportion of advertising platforms, and the smoothing constraint, so as to obtain the target budget corresponding to each advertising platform. For each advertising platform, within the target budget range corresponding to the advertising platform, the platform budget is allocated to each candidate strategy in the corresponding candidate strategy combination, and the budget amount and traffic allocation ratio corresponding to each candidate strategy are dynamically adjusted by an online allocation algorithm to determine the target budget amount and target traffic allocation ratio corresponding to each candidate strategy under each advertising platform. Based on the one-to-one mapping relationship between the strategy vector and the platform parameter set of the corresponding advertising platform, the corresponding delivery plan is created and updated according to the target budget amount and target traffic allocation ratio of each candidate strategy under each advertising platform. Aggregate the delivery data and post-business link data of various advertising platforms and candidate strategy combinations, and generate an effect analysis report after data normalization and multi-dimensional effect attribution analysis; Based on the attribution analysis results, the true performance index of each candidate strategy combination is calculated, and the strategy knowledge base is iteratively updated based on the true performance index of each candidate strategy combination. The iteratively updated strategy knowledge base drives the next round of candidate strategy generation and budget allocation, forming a self-iterative optimization closed loop.

[0008] To achieve the above objectives, this application proposes a multi-agent collaborative advertising delivery optimization system for executing the method described in any embodiment of this application. The system includes a business and configuration layer, a cross-platform agent layer, a data and feature layer, and an external advertising platform layer. The business and configuration layer is used to receive advertising placement requests and send the advertising placement requests to the cross-platform pitcher intelligent agent layer; The data and feature layer is used to collect raw report data and business post-link data from multiple advertising platforms, process the collected data to generate a unified indicator view, and provide the unified indicator view to the cross-platform pitcher agent layer. The cross-platform pitcher agent layer is communicatively connected to the business and configuration layer, the data and feature layer, and the external advertising platform layer, respectively. The cross-platform pitcher agent layer is used to perform the following steps: The algorithm receives the advertising delivery request and parses it into structured objectives and constraints that the algorithm can execute, generating a delivery task definition. The delivery task definition includes at least an objective function, a set of constraints, a definition of indicators, an evaluation time window, an update frequency, and a smoothing parameter. Based on the defined campaign, the unified metric view, the controllable parameters of each advertising platform, and the strategy knowledge base, candidate strategy combinations are generated for each advertising platform. These candidate strategy combinations are then encoded into corresponding strategy vectors, and a one-to-one mapping relationship is established between each strategy vector and the platform parameter set of the corresponding advertising platform. The candidate strategy combinations include a baseline campaign strategy, a fallback campaign strategy, and an exploratory campaign strategy. Using the defined campaign as a constraint and optimization objective, and based on the unified metric view and the campaign results of the previous campaign cycle, the total budget is allocated at the platform level under the conditions of satisfying the total budget, the upper and lower limits of the advertising platform's share, and smoothing constraints, resulting in the target budget for each advertising platform. For each advertising platform, within the target budget range corresponding to the advertising platform, the platform budget is allocated to each candidate strategy in the corresponding candidate strategy combination, and the budget amount and traffic allocation ratio corresponding to each candidate strategy are dynamically adjusted by an online allocation algorithm to determine the target budget amount and target traffic allocation ratio corresponding to each candidate strategy under each advertising platform. Based on the one-to-one mapping relationship between the strategy vector and the platform parameter set of the corresponding advertising platform, the corresponding delivery plan is created and updated according to the target budget amount and target traffic allocation ratio of each candidate strategy under each advertising platform. Aggregate the delivery data and post-business link data of various advertising platforms and candidate strategy combinations, and generate an effect analysis report after data normalization and multi-dimensional effect attribution analysis; Based on the attribution analysis results, the true performance index of each candidate strategy combination is calculated, and the strategy knowledge base is iteratively updated based on the true performance index of each candidate strategy combination. The iteratively updated strategy knowledge base drives the next round of candidate strategy generation and budget allocation, forming a self-iterative optimization closed loop. The external advertising platform layer includes multiple cross-channel advertising platforms. The external advertising platform layer connects with the advertising agent intelligent agent layer through the advertising platform interface. The external advertising platform layer is used to receive the placement configuration and send back real-time placement data.

[0009] In the technical solution provided in this application embodiment, by parsing the received advertising placement requirements into structured objectives and constraints that can be executed by the algorithm, a placement task definition containing core elements such as objective function, constraint set, and smoothing parameters is generated. This transforms non-standardized placement requirements into standardized, algorithmic execution basis, which can solve the core pain points of subjective requirement parsing, inconsistent execution standards, and lack of quantitative definition of placement constraints in traditional cross-platform advertising placement from the source. It establishes a unified core criterion for subsequent algorithmic decision-making and execution throughout the entire process, realizing the accurate transformation of placement requirements from human understanding to machine execution. On this basis, by collecting and standardizing the original report data and business back-end data from multiple advertising platforms, a unified cross-platform indicator view is generated. The indicator caliber, data format, and statistical rules of different platforms are unified and standardized, which can completely solve the problems of incomparable data, inconsistent dimensions, and data distortion caused by backhaul delays in traditional cross-platform placement. It breaks down the data silos of various advertising platforms and provides comprehensive, standardized, and comparable accurate data support for subsequent budget allocation and strategy optimization, greatly improving the scientificity and rationality of placement decisions. Meanwhile, based on the definition of the campaign, a unified indicator view, the platform's controllable parameter capabilities, and the strategy knowledge base, a candidate strategy combination covering benchmark, fallback, and exploratory campaign strategies is generated. Through a one-to-one mapping between strategy vector encoding and the platform parameter set, the strategy combination can be accurately adapted to the execution parameters of each advertising platform. It not only relies on benchmark strategies to ensure the stability and basic effect of cross-platform campaigns, but also uses fallback strategies to avoid the risks caused by campaign anomalies, and uses exploratory strategies to explore the optimization space of campaigns. This solves the problems of traditional campaigns, such as strategy design relying on human experience, difficulty in cross-platform strategy adaptation, and insufficient strategy innovation. At the same time, it lays a standardized strategy foundation for the refined allocation of budgets. Furthermore, by defining the campaign task as a constraint and optimization goal, and combining a unified metrics view with the results of the previous campaign, the total budget is allocated at the platform level. Then, within the target budget range of each platform, the strategy-level budget and traffic ratio are dynamically adjusted through an online allocation algorithm. This enables two-tiered, refined, and dynamic configuration of the budget across both platform and strategy dimensions. It satisfies core campaign requirements such as total budget, platform share, and smoothing constraints, while dynamically optimizing resource allocation based on the real-time performance of each platform and strategy. This solves the problems of manual, static budget allocation and the disconnect between resource configuration and campaign performance in traditional campaigns, significantly improving the efficiency of advertising budget utilization and return on investment. Moreover, based on the mapping relationship between strategy vectors and platform parameters, the optimized candidate strategy combinations are automatically converted into executable campaign parameters for each advertising platform, completing the creation and updating of campaign plans. This replaces the tedious manual configuration of campaign plans for each platform, enabling automated and standardized implementation of cross-platform campaign plans. This solves the problems of low efficiency in cross-platform operations, error-prone manual configuration, and delayed plan updates in traditional campaigns, significantly reducing the manual operation costs and execution threshold for cross-platform campaigns.Furthermore, by aggregating campaign data and post-campaign data from various platforms and candidate strategy combinations, and conducting multi-dimensional performance attribution analysis after data normalization, the contribution of each platform and strategy combination can be accurately quantified. This solves the problems of ambiguous performance attribution and inability to accurately identify high-performing and inefficient strategies in traditional cross-platform campaigns, providing precise data support for campaign performance review and strategy optimization. Finally, based on the attribution analysis results, the true performance indicators of each candidate strategy combination are calculated, and the strategy knowledge base is iteratively updated accordingly. The iterated knowledge base then drives the next round of candidate strategy generation and budget allocation, forming a self-iterative optimization closed loop for the entire cross-platform campaign process: "requirements analysis - data unification - strategy generation - budget allocation - campaign execution - performance attribution - strategy iteration." This not only enables campaign strategies to continuously evolve with campaign data and steadily improve campaign performance, but also systematically accumulates high-quality cross-platform campaign experience into the knowledge base, enabling the reuse and continuous optimization of campaign strategies. This solves the problems of traditional campaigns where optimization actions lack data support, high-quality experience cannot be systematically accumulated, and campaign performance is difficult to continuously improve. The overall solution enables algorithmic, intelligent, and automated management of the entire cross-platform advertising campaign process. It combines high efficiency in campaign execution, precise budget allocation, scientific strategy optimization, and sustainable campaign results. It effectively reduces customer acquisition costs for cross-platform advertising, improves overall campaign efficiency and return on investment, and is suitable for the large-scale and refined cross-platform advertising needs of industries such as banking, consumer finance, and e-commerce.

[0010] It should be understood that the above general description and the following detailed description are merely exemplary and do not limit this application. Attached Figure Description

[0011] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0012] Figure 1 This is a flowchart of a cross-platform advertising optimization method provided in an embodiment of this application.

[0013] Figure 2 This application provides a flowchart of steps for parsing advertising demand into structured goals and constraints that the algorithm can execute, and generating a definition of the advertising task.

[0014] Figure 3 This is a flowchart illustrating the steps of collecting raw report data and business post-link data from multiple advertising platforms, processing the collected data, and generating a unified indicator view, as provided in one embodiment of this application.

[0015] Figure 4This application provides a flowchart of steps for generating candidate strategy combinations corresponding to each advertising platform based on the definition of the delivery task, a unified indicator view, the controllable parameter capabilities of each advertising platform, and a strategy knowledge base.

[0016] Figure 5 This application provides a flowchart of steps for allocating the total budget at the platform level under the conditions of total budget, upper and lower limits of advertising platform proportion, and smoothing constraints, based on the definition of the campaign task as a constraint and optimization goal, a unified indicator view and the campaign results of the previous campaign cycle as data basis, and obtaining the target budget corresponding to each advertising platform.

[0017] Figure 6 This application provides a flowchart of the steps for allocating the platform budget to each candidate strategy in the corresponding candidate strategy combination within the target budget range of the advertising platform, and dynamically adjusting the budget amount and traffic allocation ratio of each candidate strategy through an online allocation algorithm to determine the target budget amount and target traffic allocation ratio of each candidate strategy under each advertising platform.

[0018] Figure 7 This is a flowchart illustrating the steps of creating and updating a corresponding campaign plan based on a one-to-one mapping relationship between a strategy vector and the platform parameter set of the corresponding advertising platform, according to the target budget and target traffic allocation ratio of each candidate strategy under each advertising platform.

[0019] Figure 8 This is a flowchart illustrating the steps of generating an effect analysis report by aggregating the delivery data and post-business link data of various advertising platforms and candidate strategy combinations, as provided in one embodiment of this application, after data normalization and multi-dimensional effect attribution analysis.

[0020] Figure 9 This application provides a flowchart of steps for calculating the true performance index of each candidate strategy combination based on the attribution analysis results, iteratively updating the strategy knowledge base based on the true performance index of each candidate strategy combination, and using the iterated strategy knowledge base to drive the next round of candidate strategy generation and budget allocation, forming a self-iterative optimization closed loop.

[0021] Figure 10 This is a structural block diagram of a cross-platform advertising optimization system provided in an embodiment of this application. Detailed Implementation

[0022] To make the objectives, implementation methods, and advantages of this application clearer, exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, these exemplary embodiments are provided to make the description of this application more comprehensive and complete, and to fully convey the concept of the exemplary embodiments to those skilled in the art. It should be noted that the brief descriptions of terminology in this application are merely for the convenience of understanding the embodiments described below, and are not intended to limit the embodiments of this application. Unless otherwise stated, these terms should be understood in their ordinary and common meaning.

[0023] In the description of this application, it should be understood that the terms "first," "second," and "third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined as "first," "second," or "third" may explicitly or implicitly include one or more features.

[0024] The flowcharts shown in the accompanying drawings are merely illustrative and do not necessarily include all content and operations / steps, nor do they necessarily have to be performed in the described order. For example, some operations / steps can be broken down, while others can be combined or partially combined; therefore, the actual execution order may change depending on the specific circumstances.

[0025] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0026] Terminology Explanation: Large Language Model: This is an artificial intelligence model based on a deep learning architecture, pre-trained on a large-scale text corpus, capable of understanding natural language, generating text, logical inference, and content creation. In this solution, it is mainly used to automatically generate and optimize advertising copy and creative content based on audience profiles, targeting objectives, and advertising platform specifications, thereby improving creative production efficiency and standardization.

[0027] Intelligent agents are intelligent functional entities with independent perception, autonomous decision-making, automatic execution, and continuous learning capabilities. In this solution, intelligent agents are given clear business responsibilities and can independently complete corresponding tasks in the entire advertising delivery process based on standardized task cards and shared data space information. They can also iteratively optimize behavioral strategies based on performance feedback. Multiple intelligent agents collaborate and share data to form an automated, closed-loop advertising delivery optimization system.

[0028] Advertising task definition: This defines advertising delivery requirements as a structured expression that can be executed by the algorithm. It includes at least the objective function, constraint set, indicator definition, evaluation time window, update frequency, and smoothing parameters. It is the unified core principle for algorithmic decision-making and execution throughout the entire cross-platform advertising delivery process.

[0029] Unified Metrics View: A standardized data view generated after processing the original report data and business back-end data from multiple advertising platforms through dimensional alignment, cleaning and deduplication, metric normalization, and delayed backhaul modeling. It covers multi-dimensional observed data, estimated data, and risk-related indicators from various platforms, achieving unified comparability of cross-platform data.

[0030] Strategy parameter space: Based on the definition of the delivery task and the controllable parameter capabilities of each advertising platform, the types and value ranges of configurable delivery parameters for each platform are determined. This serves as the boundary basis for generating candidate strategy combinations. Parameters include bidding mode, bidding level, frequency control strategy, and delivery time period.

[0031] Candidate strategy combination: A set of feasible campaign strategies generated for each advertising platform, including at least a baseline campaign strategy, a fallback campaign strategy, and an exploratory campaign strategy, which consists of controllable parameters such as bidding mode, budget pacing, targeting frequency control, and creative combination.

[0032] Benchmark delivery strategy: Extracted from strategies with stable historical performance, this combination of strategies serves as a control group for strategy search and effect comparison, ensuring the basic effectiveness and stability of cross-platform delivery.

[0033] Backup strategy: In the event of abnormal fluctuations in campaign volume, risk control triggering, or platform interface anomalies, a combination of conservative strategies is used to ensure campaign continuity. This can achieve automatic downgrading or temporary takeover of campaigns, avoiding the risk of campaign interruption.

[0034] Exploratory targeting strategies: New strategies generated within the strategy parameter space of each platform through fine-grained search and local parameter perturbation are used to explore the optimization space for targeting and break through the limitations of strategies based on human experience.

[0035] Strategy Vector: A structured representation of candidate strategy combinations, containing core information such as platform identifier, bidding parameters, budget cadence, directional frequency control, and creative features, which can be recognized, compared, and updated by the algorithm module.

[0036] Platform parameter set: A set of natively identifiable and configurable delivery parameters for each advertising platform. A one-to-one mapping relationship is established between the strategy vector and this set, realizing the transformation of algorithm-level strategy combinations into platform-executable parameters.

[0037] Smoothing parameters: These parameters limit the adjustment range and frequency of budget and bidding control variables, and are used to reduce the impact of frequent and drastic parameter adjustments on the advertising platform's learning process and campaign stability.

[0038] Smoothing constraints: These are the delivery constraints established based on smoothing parameters. Their core purpose is to limit the adjustment range and frequency of control variables such as budget and bid. They serve as an important constraint basis for platform-level budget allocation.

[0039] Platform-level budget allocation: Based on the definition of the campaign task as a constraint, combined with a unified indicator view and the results of the previous campaign, the total budget is allocated among various advertising platforms to obtain the target budget for each platform. This is the first layer of the two-tier budget allocation.

[0040] Strategy-level budget / traffic allocation: Within the target budget range of each advertising platform, the platform budget is allocated to each corresponding candidate strategy, and the budget amount and traffic allocation ratio of each strategy are dynamically adjusted. This is the second layer of the two-tier budget allocation.

[0041] Online allocation algorithm: An algorithm for dynamic adjustment of budget / traffic at the strategy level. It can combine real-time delivery data from a unified metrics view to balance the "exploration" and "utilization" of strategies, and achieve dynamic optimization of budget and traffic allocation.

[0042] Expected marginal revenue: The estimated revenue generated by each advertising platform's unit budget is calculated based on a unified indicator view, the results of the previous round of campaigns, and the optimization goals defined in the campaign task definition. It is the core basis for platform-level budget allocation.

[0043] Post-campaign data: Business-side data after the conversion stage of advertising campaigns, including lead quality, loan disbursement data, and poor performance data, is an important data source for evaluating the true effectiveness of campaigns and conducting multi-dimensional attribution.

[0044] Metric normalization: This process, which converts metrics such as CTR, CVR, CPA, and ROI to a unified and comparable dimension to address the differences in statistical methods, postback delays, and deduplication rules across different advertising platforms, is the core step in generating a unified metric view.

[0045] Delayed data feedback modeling: This modeling process analyzes the time interval distribution of "exposure / click → conversion" in ad campaigns and performs prediction and completion of incomplete conversion data feedback. It can generate derived metrics such as predicted real-time ROI and predicted CPA, solving the problem of delayed conversion data feedback.

[0046] Multi-dimensional performance attribution analysis: By aggregating campaign data and post-business data according to dimensions such as advertising platform and candidate strategy combination, the analysis process quantifies the contribution of each dimension to the campaign performance, which can accurately identify high-quality / low-efficiency strategies and provide data support for strategy optimization.

[0047] Actual performance metrics: Based on the results of multi-dimensional effect attribution analysis, and in accordance with the indicator definitions and calculation rules defined in the campaign, the actual performance metrics of each candidate strategy combination during the campaign period are calculated and serve as the core basis for evaluating the true effectiveness of the strategy.

[0048] Strategy Knowledge Base: A database / data structure that stores the historical performance, risk characteristics, portability tags, and strategy adaptation rules of candidate strategy combinations in different scenario contexts, providing prior experience support for candidate strategy generation and budget allocation.

[0049] Self-iterative optimization closed loop: The strategy knowledge base is updated iteratively with the real performance indicators obtained from attribution analysis, and then the updated knowledge base drives the generation of the next round of candidate strategies and budget allocation, forming a cross-cycle optimization system of "deployment execution - effect attribution - strategy iteration - optimized deployment", which achieves continuous improvement in deployment effect.

[0050] Based on this, this application proposes a cross-platform advertising optimization method and system, aiming to achieve automated and refined intelligent optimization of the entire cross-platform advertising process, solve the technical problems of incomparable indicators, low adjustment frequency, insufficient strategy exploration, and difficulty in cross-platform collaboration in traditional advertising, improve advertising efficiency and return on investment, and reduce customer acquisition costs.

[0051] Reference Figure 1 , Figure 1 This is a flowchart of a cross-platform advertising optimization method provided in an embodiment of this application, including but not limited to steps S110 to S180.

[0052] Step S110: Receive advertising delivery requirements and parse them into structured objectives and constraints that the algorithm can execute, and generate a delivery task definition. The delivery task definition includes at least an objective function, a set of constraints, a definition of indicators, an evaluation time window, an update frequency, and smoothing parameters.

[0053] This step is the core starting point for cross-platform ad delivery optimization. It primarily involves the algorithmic and structured parsing of ad delivery requirements and the generation of ad delivery task definitions. Specifically, it receives ad delivery requirements from the business side, including natural language descriptions and / or structured fields. First, it extracts entities from the requirements, converting fuzzy expressions into algorithmically calculable task parameters. Then, it standardizes the indicator definitions for these parameters, establishing a unified set of indicators. Next, it verifies and completes the constraints corresponding to the standardized task parameters, forming a complete set of constraints. Based on this set of constraints and the task parameters, it constructs an objective function adapted to the ad delivery requirement. Subsequently, it combines the objective function, the constraint set, and the ad delivery requirement itself to determine the evaluation window for ad performance, the update frequency of ad delivery parameters, and smoothing parameters to limit the adjustment range and frequency of budget and bidding control variables. Finally, it integrates the objective function, constraint set, indicator definitions, evaluation window, update frequency, and smoothing parameters to generate a structured ad delivery task definition. This provides a unified and standardized core basis for subsequent algorithmic decision-making and execution throughout the entire cross-platform ad delivery process.

[0054] Reference Figure 2 , Figure 2 This application provides a flowchart of steps for parsing advertising demand into structured goals and constraints that the algorithm can execute, and generating a delivery task definition, including but not limited to steps S210 to S250.

[0055] Step S210: Extract entities from advertising placement requirements, convert the fuzzy expressions in advertising placement requirements into task parameters that the algorithm can calculate, and perform standardization processing on the indicator caliber of the task parameters to determine a unified indicator caliber definition.

[0056] This step involves entity extraction and metric standardization of advertising placement requirements. Advertising placement requirements input from the business side typically include natural language descriptions, structured fields, or a combination of both. These often contain numerous vague and non-quantifiable expressions that cannot be directly recognized and calculated by the algorithm. This step first extracts core business entities from the advertising placement requirements, identifying key elements such as budget, placement goals, placement cycle, and risk control requirements. The vague expressions are then converted into quantifiable task parameters that the algorithm can directly calculate. Subsequently, all extracted task parameters undergo unified metric standardization, clarifying the definitions, statistical rules, calculation methods, and judgment criteria for various core metrics, thus establishing a unified metric definition throughout the entire process.

[0057] For example, if the business side inputs the advertising requirements as "The total budget for this month's credit product advertising is 1 million yuan, with a focus on increasing the number of high-quality leads, a target unit effective lead cost reduction of 20% compared to historical levels, and a default rate not exceeding a preset threshold," this step will extract core entities such as "total budget of 1 million yuan," "advertising period of the current month," "20% reduction in effective lead cost," and "default rate control." It will then convert "increasing the number of high-quality leads" and "20% cost reduction" into quantifiable task parameters such as "maximizing the number of effective leads" and "effective lead CPA (Cost Per Action) ≤ historical value × 80%." Simultaneously, a standardized definition of "effective lead" will be established, such as defining "effective lead" as "leads with a quality score ≥ 80 points and passing risk control compliance verification." The calculation rule for "effective lead CPA" will be defined as "total advertising cost ÷ number of effective leads," thus avoiding decision-making and execution errors caused by misunderstandings of indicator definitions in subsequent stages of advertising.

[0058] Step S220: Perform consistency verification and completion on the constraints corresponding to the task parameters after standardization of indicator caliber to form a complete set of constraints.

[0059] This step involves consistency verification and completion of constraints. Based on the quantitative task parameters standardized in step S210, various corresponding deployment constraints are identified. These constraints cover all types, including business-side preset constraints, risk control and compliance constraints, platform operation constraints, and deployment stability constraints. This step performs logical consistency verification on all identified constraints, identifies and corrects conflicting, numerically contradictory, and ambiguous content, and completes any missing necessary constraints. If any unclear or business-side confirmation is found during the verification process, a follow-up confirmation can be initiated. Ultimately, a complete, consistent, and implementable set of constraints is formed, defining clear and explicit constraint boundaries for subsequent objective function construction and full-process deployment execution.

[0060] For example, if the parameters extracted in step S210 include "Platform A's budget share is not less than 50%" and "The sum of the budget shares of all platforms is 100%", and the initial constraints lack "The lower limit of the budget share of Platforms B and C", this step will verify the logical consistency between "Platform A's share ≥ 50%" and the shares of other platforms, and at the same time complete the necessary constraints "Platform B's budget share ≥ 20%" and "Platform C's budget share ≥ 10%". If the initial constraints contain conflicting content such as "Unlimited adjustment of budget per hour" and "Frequency fluctuation of ad placement not exceeding 10%", they will be corrected to determine the reasonable constraint "The adjustment range of budget per hour shall not exceed 5% of the current budget", ultimately forming a complete set of constraints including total budget constraints, platform share constraints, risk control indicator constraints, and budget adjustment constraints.

[0061] Step S230: Based on the set of constraints and task parameters, construct an objective function that adapts to the advertising delivery requirements.

[0062] This step involves constructing the objective function to adapt to the advertising delivery requirements. The objective function is the core guiding principle for the algorithm to optimize delivery, allocate budgets, and adjust strategies. This step takes the complete set of constraints formed in step S220 as a premise, combines the quantified task parameters obtained in step S210, and constructs an adapted mathematical objective function based on the core requirements of this advertising delivery. This transforms the delivery objectives on the business side into a mathematical model that the algorithm can solve and optimize. The form of the objective function can be flexibly set according to different delivery requirements, and all objective functions must satisfy the boundary requirements of the constraint set.

[0063] For example, if the core requirement of this campaign is "to maximize the number of valid leads for credit products within a total budget of 1 million yuan and a default rate of ≤3%", then the objective function is "maxf(x) = number of valid leads", with the constraints being "total campaign cost ≤ 1 million yuan, default rate ≤ 3%, and budget proportions for each platform within a preset range". If the core requirement of this campaign is "to minimize the effective order CPA for e-commerce products, with no less than 1000 effective orders", then the objective function is "minf(x) = effective order CPA", with the constraints being "effective order quantity ≥ 1000 orders, total budget ≤ preset value". This objective function provides a clear optimization direction for subsequent algorithmic decisions.

[0064] Step S240: Based on the objective function, the set of constraints, and the advertising needs, determine the evaluation time window, the update frequency of the advertising parameters, and the smoothing parameters. The smoothing parameters are parameters that limit the adjustment range and frequency of budget and bidding control variables.

[0065] This step involves determining the evaluation time window, update frequency, and smoothing parameters. Combining the objective function constructed in step S230, the complete set of constraints formed in step S220, and the actual business scenario, channel characteristics, and execution requirements of this advertising campaign, this step identifies three core parameters to ensure the rhythm and stability of the campaign execution: First, the evaluation time window, which clarifies the statistical, calculation, and evaluation cycle of the campaign's effectiveness, used to determine the effect of strategy and budget adjustments. Second, the update frequency of the campaign parameters, determining the frequency of automatic algorithmic adjustments for core campaign parameters such as budget, bid, and traffic allocation, adapting to the high-frequency automated optimization needs of cross-platform campaigns. Third, the smoothing parameter, which quantitatively limits the adjustment magnitude and frequency of control variables such as budget and bid, used to avoid frequent and drastic parameter adjustments affecting the advertising platform's model learning process and overall campaign stability.

[0066] For example, to address the demand for credit product advertising across information feeds, short videos, and search platforms, and considering the objective function of "minimizing effective lead CPA" and constraints, the evaluation time window is set at "1 hour," meaning the advertising performance of each platform and strategy is analyzed and evaluated every hour. The advertising parameter update frequency is also set at "1 hour," meaning the algorithm adjusts the budget and bid every hour based on the performance evaluation results. A smoothing parameter is defined as follows: "single-round budget adjustment ≤ 5% of the current budget, single-round bid adjustment ≤ 3% of the current bid, and the interval between two adjustments is no less than 30 minutes." This smoothing parameter limits the adjustment range and frequency, preventing drastic fluctuations in advertising data.

[0067] Step S250: Integrate the objective function, constraint set, indicator definition, evaluation time window, update frequency and smoothing parameter to generate a structured deployment task definition.

[0068] This step is the integration and generation step for the structured ad placement task definition. As the final step in the ad placement demand analysis process, this step integrates the indicator definitions determined in step S210, the constraint set formed in step S220, the objective function constructed in step S230, and the evaluation time window, ad placement parameter update frequency, and smoothing parameters determined in step S240. All core elements are integrated into a standardized, structured ad placement task definition that can be directly invoked by the algorithm. This ad placement task definition serves as the unified input basis for all subsequent stages, including cross-platform data collection and processing, candidate strategy combination generation, platform-level and strategy-level budget allocation, ad placement plan creation and updating, and performance attribution analysis. All ad placement optimization algorithm modules operate based on this ad placement task definition, achieving standardized, algorithmic, and refined control of the entire cross-platform ad placement process.

[0069] For example, the integrated campaign definition will explicitly include: metric criteria (definition of effective leads, CPA calculation rules, etc.), objective function (min effective lead CPA), constraint set (total budget of 1 million yuan, platform A's share ≥50%, defect rate ≤3%, etc.), evaluation time window (1 hour), update frequency (1 hour), and smoothing parameters (budget adjustment ≤5%, bid adjustment ≤3%). All elements are stored in a structured form for subsequent algorithm modules to directly read and execute.

[0070] In this embodiment, the steps of parsing advertising placement requirements into structured goals and constraints and generating placement task definitions are implemented through a progressive process design. This includes entity extraction, standardization of definitions, constraint verification and completion, objective function construction, core parameter determination, and final integration. First, the vague and non-quantifiable natural language placement requirements from the business side are transformed into quantifiable task parameters that the algorithm can directly calculate. Simultaneously, through standardized processing of unified indicator definitions, decision-making and execution errors caused by misunderstandings of indicators and inconsistent statistical rules in subsequent cross-platform placement stages can be avoided from the outset, addressing the technical pain points of subjective and low-standardization in traditional placement. Second, by verifying and completing the consistency of constraints, constraint conflicts are effectively corrected, and necessary constraints are supplemented, forming a complete and implementable set of constraints. This provides clear and consistent execution boundaries for subsequent placement optimization, avoiding placement control issues caused by ambiguity in constraints. Furthermore, by transforming business campaign objectives into a suitable mathematical objective function, and combining this with the scientific setting of evaluation time windows, update frequencies, and smoothing parameters, a clear optimization guide and rhythm, as well as stability assurance, are provided for the algorithm to conduct cross-platform budget allocation and strategy adjustments. This enables high-frequency automated adjustment of campaign parameters while avoiding the impact of frequent and drastic parameter adjustments on the platform's learning process and campaign stability through smoothing parameters. The resulting structured campaign task definition integrates the core elements required for the entire algorithm execution process, becoming the unified input basis for all subsequent cross-platform data processing, strategy generation, campaign execution, and performance attribution. This achieves standardized and algorithmic control of the entire cross-platform advertising campaign process, laying a solid foundation for subsequent intelligent optimization across the entire chain. It significantly improves the accuracy of campaign demand analysis, the standardization of campaign execution, and the scientific nature of algorithmic decision-making, effectively reducing the subjectivity and operational costs of manual analysis.

[0071] Step S120: Collect raw report data and business back-end data from multiple advertising platforms, process the collected data, and generate a unified indicator view.

[0072] This step is the core data processing stage for cross-platform advertising optimization, and it provides crucial data support for subsequent algorithmic budget allocation and strategy optimization. This step begins by comprehensively collecting raw report data from various advertising platforms, including news feeds, short videos, and search engines, covering the entire advertising execution process, such as impressions, clicks, conversions, and campaign costs. Simultaneously, it collects back-end data from the business side, including core business data following campaign conversion, such as lead quality, loan disbursement data, and data on poor performance. Subsequently, the collected multi-source heterogeneous data underwent full-process standardization processing. First, the data was aligned according to preset dimensions such as platform, strategy combination, and time window. Then, data cleaning and consistency processing were performed to unify the time zone, time granularity, and currency and tax standards. Outlier removal, missing value completion, and cross-platform deduplication based on user and device identifiers were completed. Next, cross-platform indicator normalization processing was performed on the cleaned data to transform core indicators such as CTR (Click-Through Rate), CVR (Conversion Rate), CPA (Cost Per Action), and ROI (Return on Investment) from various platforms to a unified and comparable dimension, forming standardized indicator data. Then, based on historical campaign data, delayed data feedback modeling is performed to analyze the distribution of time intervals from exposure and click to conversion. Incomplete conversion data is estimated and supplemented to generate a derived indicator set including estimated real-time ROI and estimated CPA. Finally, standardized indicator data and derived indicator sets are integrated to generate a unified indicator view covering multi-dimensional observed data, estimated data, and risk-related indicators across various advertising platforms and strategy combinations, along with indicator confidence levels. This completely breaks down data silos between advertising platforms and solves the problems of incomparable data and decision distortion caused by delayed data feedback in traditional cross-platform campaigns. It provides a unified, standardized, and precise data basis for subsequent platform-level and strategy-level budget allocation and performance attribution analysis.

[0073] Reference Figure 3 , Figure 3 This application provides a flowchart of steps for collecting raw report data and business post-link data from multiple advertising platforms, processing the collected data, and generating a unified indicator view, including but not limited to steps S310 to S350.

[0074] Step S310: Collect raw report data and business back-end data from multiple advertising platforms, and perform unified alignment processing on the collected data according to preset dimensions.

[0075] This step involves multi-source data collection and unified alignment processing across preset dimensions. First, it comprehensively collects raw report data from various advertising platforms, including full-link campaign execution data such as impressions, clicks, cost per click, conversions, and campaign spending from platforms like news feeds, short videos, and search. Simultaneously, it collects post-campaign data from the business side, covering core business data following campaign conversion, such as lead quality scores, number of valid leads, number of loans disbursed, default rate, and order conversion rate. Then, according to preset dimensions such as advertising platform, campaign strategy combination, time window, and campaign account, all collected multi-source heterogeneous data undergoes unified alignment processing, merging data of different formats and statistical dimensions into a unified dimension system, laying the foundation for subsequent data standardization processing.

[0076] For example, the spending statistics of information flow platforms by "day" and short video platforms by "hour" are uniformly aligned to the "hour" level time window dimension. At the same time, the conversion data of each platform are associated with the corresponding strategy combination identifier, so as to achieve unified matching of data from different platforms and different formats at the dimension level and avoid dimension confusion in subsequent data processing.

[0077] Step S320 involves cleaning and standardizing the dimension-aligned data to achieve data standardization, outlier handling, and cross-platform deduplication.

[0078] This step involves data cleaning and standardization. Based on the dataset aligned in step S310, this step performs comprehensive cleaning and standardization, focusing on three key aspects: First, data standardization: The time zones, currency units, tax calculation rules, and statistical methods across different platforms are standardized. For example, the campaign costs across platforms are converted to RMB excluding tax, and the time data is converted to East 8 time zone. Second, outlier and missing value handling: Pre-defined outlier detection rules are used to remove abnormal data such as a tenfold or greater increase in campaign spending or negative conversion rates. Interpolation, averaging, or fitting based on historical similar data are used to complete missing campaign and conversion data. Third, cross-platform deduplication of conversions: Based on core identifiers such as unique user identifiers, device identifiers, and lead IDs, duplicate conversions by the same user across different advertising platforms are deduplicated to avoid misjudgments of campaign performance due to repeated conversion data.

[0079] For example, if the same user generates credit product leads through both the information flow platform and the search platform, this step will use the user's mobile phone number as a unique identifier to deduplicate the duplicate conversion records, retaining only the relevant data from the first conversion, ensuring the authenticity and accuracy of the conversion data, and ultimately obtaining a well-organized dataset with unified dimensions, consistent standards, and complete data.

[0080] Step S330: Perform cross-platform indicator normalization processing on the cleaned and standardized data to transform the indicators of each advertising platform to a unified dimension and form standardized indicator data.

[0081] This step is a cross-platform metric normalization process. Since different advertising platforms have different statistical rules and definition standards for core advertising metrics, direct comparison is meaningless. This step performs cross-platform metric normalization on the standardized dataset obtained in step S320. Based on preset normalization rules, all core advertising metrics such as CTR, CVR, CPA, and ROI from each advertising platform are converted to a unified calculation and evaluation dimension, eliminating statistical differences between platforms and forming standardized metric data.

[0082] For example, news feed platforms include conversions generated within 3 days of a click in CVR statistics, while search platforms include conversions generated within 24 hours of a click. This step unifies the conversion statistics period for both to "effective conversions generated within 48 hours of a click," and then recalculates the CVR metrics for each platform. Simultaneously, the CPA metric for each platform is uniformly defined as "the average cost of acquiring a valid lead." Through metric normalization, direct comparison and analysis of core metrics across different advertising platforms can be achieved.

[0083] Step S340: Based on historical data, perform delayed backhaul modeling and transformation prediction and completion on standardized indicator data to generate a derived indicator set containing prediction indicators.

[0084] This step involves delayed data feedback modeling and conversion prediction completion. There is a significant delay in the feedback of conversion data and post-business data from advertising campaigns. For example, after a lead in a loan product is converted, loan disbursement and default data may be delayed by several days. Making decisions based solely on the already received data can easily lead to distorted decisions. This step addresses this issue by performing delayed data feedback modeling on standardized metrics based on historical campaign data. First, it analyzes the distribution patterns of time intervals in each stage of the historical data—"exposure / click → conversion → post-business results"—to construct a delayed data feedback prediction model. Then, it uses this model to predict and complete the conversion data and post-business data that have not yet been fully fed back, generating a derived metric set that includes predicted real-time ROI, predicted effective lead CPA, predicted loan disbursement rate, and predicted default rate. Confidence levels are assigned to each predicted metric to achieve real-time and accurate evaluation of campaign performance.

[0085] For example, based on historical data, it is found that after a loan product lead is converted, 70% of the loan disbursement data will be returned within 3 days, and 20% will be returned on the 4th-5th day. This step will use this pattern to estimate and complete the remaining loan disbursement data for leads that currently only return loan disbursement data within 3 days, and then calculate a more realistic real-time ROI, providing data support for the algorithm's real-time decision-making.

[0086] Step S350: Integrate standardized indicator data and derived indicator sets to generate a unified indicator view that covers multi-dimensional observed data, estimated data and risk-related indicators from various advertising platforms.

[0087] This step is the unified indicator view integration and generation step. As the final step in this data processing flow, this step integrates the standardized indicator data obtained in step S330 with the derived indicator set generated in step S340, and combines risk-related indicators to construct a unified indicator view. This view covers all data from various advertising platforms and strategy combinations under different time windows, specifically including observed actual delivery data, estimated data supplemented by predictions, and risk-related indicators such as risk-weighted ROI, indicator volatility coefficient, and default rate warning value. All data are accompanied by corresponding statistical dimensions and indicator confidence levels.

[0088] For example, the generated unified metrics view clearly displays the observed impressions, clicks, and actual cost of a certain exploratory strategy combination on the information flow platform within a one-hour time window, as well as the estimated number of valid leads, estimated valid lead CPA, and estimated defect rate, while also indicating that the confidence level of each estimated metric is 85%. This unified metrics view provides comprehensive, standardized, and comparable unified data support for all subsequent algorithmic decision-making stages, such as platform-level budget allocation, strategy-level budget traffic adjustment, and multi-dimensional performance attribution analysis, enabling integrated management and application of cross-platform data.

[0089] In this embodiment, the steps of collecting original report data and post-business link data from multiple advertising platforms and processing them to generate a unified indicator view involve a progressive standardized processing flow from multi-source data collection and alignment, cleaning and consistency, indicator normalization, delayed backhaul modeling and completion to final integration. Firstly, this achieves comprehensive collection and dimensional unification of cross-platform campaign execution data and post-business link data, solving the problems of scattered data sources and chaotic dimensions in traditional cross-platform campaigns, laying a solid foundation for subsequent data processing. Secondly, by unifying data definitions, handling outliers and missing values, and deduplicating duplicate data across platforms, it effectively corrects data distortion issues, ensuring data authenticity, completeness, and consistency, and avoiding misjudgments of campaign performance due to duplicate statistics and data anomalies. Thirdly, through cross-platform indicator normalization, it eliminates differences in statistical rules and definitions of core campaign indicators across different advertising platforms, enabling direct comparison and analysis of indicators from various platforms, completely breaking down cross-platform data silos and solving the core pain point of incomparable indicators in traditional campaigns. Meanwhile, based on historical data-driven delayed data transmission modeling and conversion prediction completion, the system accurately addresses the decision-making lag and distortion issues caused by delayed data transmission in advertising campaigns. The generated predictive derived indicators provide precise data support for the algorithm's high-frequency real-time decision-making. The final unified indicator view integrates standardized measured data, accurate predictive data, and risk-related indicators, along with indicator confidence levels and statistical dimensions. This provides a comprehensive, standardized, and comparable unified data basis for all algorithmic decision-making stages, including platform-level and strategy-level budget allocation, multi-dimensional effect attribution analysis, and strategy optimization. This significantly improves the scientific rigor and accuracy of cross-platform campaign decisions, while laying a data foundation for comprehensive and refined evaluation of campaign performance. It effectively avoids resource misallocation caused by data issues, improving the overall data utilization efficiency and optimization effect of advertising campaigns.

[0090] Step S130: Based on the task definition, unified indicator view, controllable parameter capabilities of each advertising platform, and strategy knowledge base, generate candidate strategy combinations corresponding to each advertising platform. Then, encode the candidate strategy combinations of each advertising platform into corresponding strategy vectors and establish a one-to-one mapping relationship between each strategy vector and the platform parameter set of the corresponding advertising platform. The candidate strategy combinations include baseline delivery strategy, fallback delivery strategy, and exploratory delivery strategy.

[0091] This step is the core of cross-platform advertising optimization strategy construction and coding, providing a standardized strategic foundation for subsequent budget allocation and campaign execution. First, using campaign task definitions as constraint boundaries and a unified metric view as data basis, and combining the controllable parameter capabilities of each advertising platform with historical strategy performance and adaptation rules accumulated in the strategy knowledge base, a dedicated strategy parameter space is defined for each advertising platform. Then, based on this space, candidate strategy combinations that combine stability, security, and exploratory potential are generated. The baseline campaign strategy selects high-quality strategies with stable historical performance from the strategy knowledge base; the fallback campaign strategy is configured with conservative parameters that comply with risk control and platform rules to ensure campaign continuity; and the exploratory campaign strategy explores new optimization space through fine-grained search and local parameter perturbation. Simultaneously, all generated strategies undergo constraint pruning, feasibility verification, and deduplication to ensure the compliance and implementability of the candidate strategy combinations. Subsequently, the candidate strategy combinations of each advertising platform are structured and encoded to generate strategy vectors containing core information such as platform identifiers, bidding parameters, budget cadence, targeting frequency control, and creative features. Finally, a one-to-one mapping relationship is established between each strategy vector and the corresponding advertising platform's set of platform parameters. This achieves accurate conversion of algorithm-level strategy vectors into natively identifiable and configurable delivery parameters of the advertising platform. This not only solves the problems of traditional campaign strategy design relying on human experience and the difficulty of cross-platform strategy adaptation, but also provides a standardized basis for strategy identification and execution for subsequent algorithmic budget allocation and traffic adjustment, ensuring the consistency and accuracy of cross-platform strategy delivery.

[0092] Reference Figure 4 , Figure 4 This application provides a flowchart of steps for generating candidate strategy combinations corresponding to each advertising platform based on the definition of the delivery task, a unified indicator view, the controllable parameter capabilities of each advertising platform, and a strategy knowledge base, including but not limited to steps S410 to S440.

[0093] Step S410: Based on the definition of the delivery task and the controllable parameter capabilities of each advertising platform, determine the corresponding strategy parameter space for each advertising platform.

[0094] This step involves defining the strategy parameter space for each advertising platform. The strategy parameter space is the set of configurable delivery parameters and their value boundaries for each advertising platform. It forms the fundamental constraints for generating candidate strategy combinations. This step is guided by the constraint set and optimization objectives defined in the delivery task definition. Combined with the actual controllable parameter capabilities of each advertising platform, it delineates the platform-specific strategy parameter space, clarifies the configurable delivery parameter types, parameter value ranges, and parameter combination rules for each platform, and eliminates parameter configuration methods not supported by each platform. This ensures that all subsequently generated strategies comply with the platform's execution requirements. The strategy parameters cover core delivery configuration items such as bidding mode, bid level range, budget rhythm, delivery time period, frequency control strategy, and creative combination matching rules.

[0095] For example, for information flow advertising platforms, which support two bidding modes, OCPC (Optimized Cost Per Click) and CPC, this step defines the bidding mode parameter range as {OCPC, CPC}. Simultaneously, considering the constraint in the campaign definition that "the daily budget for a single strategy should not be less than 500 yuan", the daily budget parameter range for this platform is defined as 500 yuan to 30% of the platform-level target budget. Search advertising platforms support the CPA bidding mode, and its bidding mode parameter range is defined as {CPA}. Based on the platform's capabilities, corresponding upper and lower limits for bidding levels are set, ultimately generating a unique strategy parameter space for each platform that conforms to both the task constraints and the platform's capabilities.

[0096] Step S420: Based on the strategy knowledge base, unified indicator view, and strategy parameter space of each advertising platform, generate the benchmark delivery strategy and the fallback delivery strategy corresponding to each advertising platform.

[0097] This step generates the baseline and fallback delivery strategies for each advertising platform. Based on historical strategy performance, risk characteristics, and scenario adaptation rules accumulated in the strategy knowledge base, and combined with real-time delivery data from each platform in the unified metrics view, as well as the strategy parameter space defined in step S410, this step generates the baseline and fallback delivery strategies for each advertising platform. These two types of strategies each have their own function, ensuring the stability and risk resistance of cross-platform delivery. The baseline delivery strategy selects a high-quality strategy combination from the strategy knowledge base that matches the current delivery scenario, has stable historical performance, and meets delivery targets. This serves as the basic execution strategy and performance benchmark for this delivery. The fallback delivery strategy is configured with a conservative parameter combination that complies with risk control compliance rules and platform requirements. It is used to take over delivery in case of abnormal fluctuations, risk control triggers, or platform interface anomalies, ensuring delivery continuity.

[0098] For example, for short video platform advertising of credit products, the strategy combination of "short video platform - credit customer acquisition - historical CPA target and volatility coefficient <10%" is retrieved from the strategy knowledge base. Combined with the real-time traffic data of the short video platform in the current unified indicator view, the bidding level is fine-tuned and determined as the benchmark advertising strategy. At the same time, a conservative parameter combination of "CPC bidding mode + conservative bidding level + universal audience targeting across the platform + daily budget of 500 yuan" is configured as the backup advertising strategy for the platform to ensure that the advertising is not interrupted in extreme cases.

[0099] Step S430: Combining the optimization objectives defined in the campaign task with the unified indicator view, conduct strategy exploration within the strategy parameter space of each advertising platform to generate exploratory campaign strategies corresponding to each advertising platform.

[0100] This step involves generating exploratory campaign strategies for each advertising platform. Exploratory campaign strategies are used to uncover new optimization opportunities and overcome the limitations of strategies based on human experience. This step is guided by the optimization objectives defined in the campaign task, and combines real-time campaign potential data from each platform in a unified metrics view. Within the strategy parameter space defined in step S410, strategies are explored through fine-grained parameter searches, local parameter perturbations around the benchmark campaign strategy, and orthogonal experiments with different parameter combinations. This generates exploratory campaign strategies for each advertising platform, ensuring that the parameter configurations of these strategies do not exceed the boundaries of the strategy parameter space, while also considering both campaign diversity and targeted exploration.

[0101] For example, if the optimization objective defined in the campaign is "minimizing the CPA of effective leads for credit products," and the unified metrics view shows that the traffic quality of the information flow platform is higher during the evening hours, this step will focus on the platform's baseline campaign strategy and conduct local parameter perturbations within the strategy parameter space: adjusting the bid level within ±10% of the baseline value, limiting the campaign time to 18:00-23:00, and adjusting the frequency control strategy from "3 times a day" to "2 times a day," generating multiple sets of exploratory campaign strategies with different parameter combinations. Simultaneously, fine-grained combinations of the platform's creative combinations and audience targeting strategies will be explored to generate more diverse exploratory strategies and uncover better parameter configuration combinations.

[0102] Step S440 involves constraining, pruning, verifying feasibility, and deduplicating the baseline, fallback, and exploratory placement strategies for each advertising platform to obtain candidate strategy combinations for each platform.

[0103] This step involves strategy constraint pruning, feasibility verification, and deduplication. It comprehensively refines the baseline and fallback strategies generated in step S420, as well as the exploratory strategies generated in step S430, ultimately yielding candidate strategy combinations that can be implemented on various advertising platforms. The core tasks are threefold: First, constraint pruning: eliminating all strategy combinations that violate risk control and compliance constraints, platform operation constraints, and budget constraints defined in the campaign definition, such as strategies targeting sensitive customer groups or those with single-strategy budgets exceeding parameter ranges. Second, feasibility verification: verifying whether the parameter configurations of the remaining strategies match the actual execution capabilities of the corresponding advertising platforms, and identifying any parameter conflicts or invalid configurations, such as verifying whether the bidding model and creative combination of the strategy are supported by the platform. Third, deduplication: merging strategy combinations with substantially equivalent parameter configurations, eliminating duplicate strategies, and reasonably controlling the number of candidate strategies to avoid excessive budget and traffic dispersion due to too many strategies.

[0104] For example, in the strategy set of the short video platform, the exploratory strategy of "single strategy daily budget of 400 yuan" that violates the constraint of "not less than 500 yuan" is removed. The parameter conflict strategy of "OCPC bidding mode + no conversion target setting" is verified and deleted. The equivalent strategy of "the bidding level, the time period and the frequency control strategy are completely the same and only the creative tags are different" is merged. Finally, the candidate strategy combination of the short video platform is obtained, consisting of 1 set of benchmark placement strategies, 1 set of backup placement strategies and 8 sets of exploratory placement strategies. Each strategy meets the constraint requirements, has the feasibility of implementation and has no duplication.

[0105] In this embodiment, a progressive process—first determining the strategy parameter space, then generating baseline and fallback strategies, followed by strategy exploration, and finally constraint pruning and deduplication—enables precise definition of the types and value boundaries of configurable delivery parameters for each platform. This ensures that subsequent strategy generation is within the constraints of the delivery task and the platform's capabilities, preventing the generation of invalid strategies from the outset. Secondly, the baseline delivery strategy generated based on the strategy knowledge base and unified indicator view can quickly reuse historical best practices and be fine-tuned with real-time data, guaranteeing the basic effectiveness and stability of the delivery. The fallback delivery strategy provides automatic degradation protection in case of abnormal fluctuations or interface failures, effectively mitigating the risk of delivery interruption. Furthermore, the exploratory strategy generation within the strategy parameter space, through local perturbations and fine-grained searches, continuously uncovers new optimization space, overcoming the limitations of manual experience and enhancing strategy innovation capabilities and long-term benefits. Ultimately, through constraint pruning, feasibility verification, and deduplication, illegal, conflicting, and duplicate strategies are eliminated, resulting in a compliant, implementable, and clearly structured combination of candidate strategies. This provides a standardized and executable strategy foundation for subsequent budget allocation and campaign execution, and achieves a comprehensive optimization effect of "stable campaign execution + risk mitigation + continuous exploration" while ensuring campaign continuity and stability. This significantly improves the generation efficiency, implementation adaptability, and overall optimization effect of cross-platform campaign strategies.

[0106] Step S140: Define the campaign task as the constraint and optimization goal, and use the unified indicator view and the campaign results of the previous campaign cycle as data basis. Under the conditions of satisfying the total budget, the upper and lower limits of the proportion of advertising platforms, and the smoothing constraint, the total budget is allocated at the platform level to obtain the target budget corresponding to each advertising platform.

[0107] This step is the platform-level budget allocation step in cross-platform advertising optimization, belonging to the first layer of a two-tier budget allocation system. This step uses the objective function, constraint set, and smoothing parameters set in the campaign definition as optimization criteria and boundary limits. It is supported by standardized and comparable cross-platform indicator data provided by the unified indicator view and the actual campaign results of the previous campaign cycle. Under the premise of strictly meeting multiple constraints such as the total budget limit, the upper and lower limits of the budget proportion of each advertising platform, and smoothing constraints, the total budget is allocated and optimized across platforms through a preset resource allocation algorithm. It comprehensively considers the real-time campaign effect, estimated revenue, risk indicators, and historical performance of each advertising platform, and rationally allocates the total budget to different advertising platforms such as information flow, short video, and search. Finally, it obtains the target budget corresponding to each advertising platform in this campaign cycle, realizing the tilting of budget resources towards high-return, low-risk, and high-potential platforms. This ensures that the overall campaign meets business constraints and stability requirements, maximizes the overall campaign optimization goals, and improves the overall budget utilization efficiency and return on investment.

[0108] Reference Figure 5 , Figure 5 This application provides a flowchart of steps for determining the target budget for each advertising platform by defining the campaign task as a constraint and optimization goal, using a unified indicator view and the campaign results of the previous campaign cycle as data, and allocating the total budget at the platform level under the conditions of satisfying the total budget, the upper and lower limits of the proportion of advertising platforms, and smoothing constraints. This includes steps S510 to S540.

[0109] Step S510: Based on the unified indicator view and the results of the previous campaign, and combined with the optimization objectives defined in the campaign task, calculate the expected marginal revenue per unit budget of each advertising platform.

[0110] This step involves calculating the expected marginal return per unit budget for each advertising platform. Based on standardized and comparable real-time performance metrics from a unified metrics view, as well as actual spending, conversion, and revenue data from the previous campaign cycle, and combined with optimization objectives defined in the campaign task (such as maximizing effective leads, minimizing CPA, and maximizing risk-weighted ROI), a pre-set revenue prediction model is used to calculate the incremental conversion value and incremental business revenue that each additional unit of budget investment will bring to each advertising platform, thus obtaining the expected marginal return per unit budget for each platform.

[0111] For example, in a credit customer acquisition scenario, if the optimization goal is to maximize the number of effective leads, and the unified metrics view shows that the current effective lead CPA of the information flow platform is 80 yuan, and the previous round of spending of 100,000 yuan brought 1,250 effective leads, while the search platform has a CPA of 100 yuan, and the previous round of spending of 80,000 yuan brought 800 effective leads, then the expected marginal lead revenue per unit budget of the information flow platform is higher than that of the search platform, providing a basis for budget allocation.

[0112] Step S520 involves adjusting the expected marginal revenue of each advertising platform for risk and stability.

[0113] This step involves adjusting the risk and stability of expected marginal returns. Simply allocating budgets based on expected marginal returns can easily overlook factors such as platform volatility, data latency, and risk control risks, leading to unstable campaigns. Therefore, this step uses risk indicators and historical volatility coefficients from a unified indicator view, combined with smoothing constraints and risk control requirements in the campaign definition, to perform a weighted adjustment of the original expected marginal returns for each platform. Specifically, it introduces adjustment factors such as default rate weight, indicator confidence level, campaign stability coefficient, and platform traffic volatility coefficient. Returns for high-risk, high-volatility, and low-confidence platforms are adjusted downwards, while returns for low-risk, stable, and reliable platforms are appropriately smoothed, making the return indicators more closely reflect the actual achievable campaign results.

[0114] For example, a short video platform has high expected marginal revenue, but the unified indicator view shows that its lead defect rate is 5%, the indicator confidence level is only 70%, and the historical campaign volatility coefficient exceeds 15%. Therefore, its marginal revenue is reduced by a coefficient of 0.8. On the other hand, a search platform has low risk, high confidence level, and low volatility. It is adjusted by a coefficient of 0.95 to make the budget allocation more stable.

[0115] Step S530: Using the task definition as a constraint, under the conditions of satisfying the total budget, the upper and lower limits of the advertising platform's proportion and the smoothing constraint, perform a constrained budget allocation calculation on the modified expected marginal revenue.

[0116] This step involves a constrained budget allocation optimization calculation. The overall constraint framework is defined by the campaign task. Under strict constraints such as total budget limits, upper and lower limits on the budget proportions of each advertising platform, and smoothing adjustment constraints, the optimization objective is the expected marginal return after risk and stability adjustments. Linear programming, integer programming, or heuristic allocation algorithms are used to find the optimal solution within the constraint boundaries, maximizing overall campaign revenue.

[0117] For example, the total budget for this round is 1 million yuan. The task definition constraints are as follows: information flow platforms account for 40%–60%, search platforms 20%–30%, and short video platforms 10%–20%. The budget adjustment for each platform in this round shall not exceed ±10% compared to the previous round. Within the above constraints, the algorithm calculates the allocation based on the adjusted marginal revenue to avoid allocation results that exceed the constraints or fluctuate drastically, ensuring that the allocation scheme is legal, executable, and controllable.

[0118] Step S540: Convert the budget allocation calculation results into a budget plan that each advertising platform can execute, so as to determine the target budget corresponding to each advertising platform.

[0119] This step involves budget plan conversion and target budget determination. The theoretically optimized allocation values ​​obtained in step S530 are converted into executable budget plans that comply with the billing rules, campaign granularity, and account management requirements of each advertising platform. These plans are then broken down according to campaign time windows (hours, days, weeks), and the values ​​are rounded, boundary-calibrated, and conflict-checked to ultimately determine the target budgets that can be directly executed on each advertising platform within the current campaign period.

[0120] For example, after optimization and calculation, the theoretical allocation amount for the information flow platform is 512,300 yuan. Combined with the minimum unit of the platform's daily budget, which is rounded down to 512,000 yuan, and after verifying that its proportion and smoothing range meet the constraints, it is determined as the target budget for this round of the platform. Similarly, the target budgets for the search platform and the short video platform are determined to form a complete and executable platform-level budget plan, providing an upper limit for the total amount of internal strategy-level budget allocation for each platform in the future.

[0121] In this embodiment, a progressive process is employed: first, the expected marginal revenue per unit budget of each advertising platform is calculated; then, risk and stability adjustments are made; finally, constrained optimization calculations are performed; and finally, an executable budget plan is generated. This process accurately quantifies the incremental revenue potential of each platform based on a unified indicator view and the results of the previous round of campaigns, providing objective data support for the reasonable allocation of budget resources. Simultaneously, by introducing correction factors such as failure rate, indicator confidence level, and traffic fluctuations to calibrate the marginal revenue against risk, the problem of simply pursuing revenue while ignoring campaign stability and business risks can be effectively avoided, making budget allocation more aligned with actual campaign performance. Furthermore, by conducting constrained optimization calculations under strict adherence to the total budget, platform share upper and lower limits, and smoothing constraints, the budget allocation plan can be guaranteed not to exceed business and risk control boundaries. The algorithm maximizes overall campaign revenue, and the smoothing constraints limit drastic fluctuations in platform budgets, ensuring the continuity and stability of campaigns across advertising platforms. Ultimately, the theoretical calculation results are converted into target budgets that can be directly executed on each platform, which can improve the adaptability and feasibility of the budget plan. This enables the scientific, robust, and refined allocation of cross-platform budget resources, significantly improving the overall budget utilization efficiency and return on investment. At the same time, it provides a reliable total amount basis for subsequent strategic budget allocation within the platform.

[0122] Step S150: For each advertising platform, within the target budget range corresponding to the advertising platform, allocate the platform budget to each candidate strategy in the corresponding candidate strategy combination, and dynamically adjust the budget amount and traffic allocation ratio corresponding to each candidate strategy through an online allocation algorithm to determine the target budget amount and target traffic allocation ratio corresponding to each candidate strategy under each advertising platform.

[0123] This step is a strategy-level refined allocation and dynamic adjustment step within the platform in the two-tier budget allocation system. For each advertising platform, under the constraint of the total target budget determined in step S140, this step first combines the real-time performance, estimated revenue, and risk indicators of each candidate strategy in the unified indicator view to initially allocate the overall platform budget to the baseline delivery strategy, the backup delivery strategy, and various exploratory delivery strategies. Then, through an online allocation algorithm, the delivery data of each candidate strategy is read in real time. Guided by the optimization goals defined in the delivery task, the budget amount and traffic allocation ratio of each strategy are dynamically adjusted within the target budget range. Resources are continuously tilted towards candidate strategies with excellent performance, higher estimated revenue, and controllable risk, while a minimum guarantee amount is reserved for the backup strategy to maintain delivery stability. Finally, the target budget amount and target traffic allocation ratio corresponding to each candidate strategy under each advertising platform are determined, achieving refined, dynamic, and revenue-maximizing allocation of budget and traffic at the strategy level, effectively improving the resource utilization efficiency and overall delivery effect within a single platform.

[0124] Reference Figure 6 , Figure 6 This application provides an embodiment of the process of allocating the platform budget to each candidate strategy in the corresponding candidate strategy combination within the target budget range of the advertising platform, and dynamically adjusting the budget amount and traffic allocation ratio of each candidate strategy through an online allocation algorithm to determine the target budget amount and target traffic allocation ratio of each candidate strategy under each advertising platform, including but not limited to steps S610 to S630.

[0125] Step S610: For each advertising platform, within the target budget range corresponding to the advertising platform, allocate initial budget amount and initial traffic allocation ratio for each candidate strategy in the candidate strategy combination corresponding to the advertising platform, and define the delivery role of each candidate strategy.

[0126] This step involves defining the initial budget, traffic allocation, and delivery roles for each candidate strategy. For each advertising platform, within the platform's predetermined target budget, this step allocates initial budget amounts and initial traffic proportions to the baseline delivery strategy, fallback delivery strategy, and exploratory delivery strategy based on the type and functional positioning of each strategy in the candidate strategy portfolio. It also clarifies the role and priority of each strategy during the delivery process, ensuring a robust delivery structure. Specifically, the baseline delivery strategy, as the primary delivery unit, receives a higher proportion of initial budget and traffic. The fallback delivery strategy, as a safety net unit, receives only a small, fixed budget to meet minimum continuous delivery requirements. The exploratory delivery strategy, as an incremental development unit, receives a moderate trial budget to control exploration risks.

[0127] For example, a certain information flow platform has a target budget of 500,000 yuan for this round. Its candidate strategy combination includes 1 set of baseline strategies, 1 set of fallback strategies, and 8 sets of exploratory strategies. In this step, 60% (300,000 yuan) of the initial budget and 60% of the initial traffic ratio are allocated to the baseline strategy, 2% (10,000 yuan) of the initial budget and 2% of the initial traffic ratio are allocated to the fallback strategy, and the remaining 38% (190,000 yuan) of the budget and 38% of the traffic are evenly distributed to the 8 sets of exploratory strategies. At the same time, the baseline strategy is defined as the main placement role, the fallback strategy as the fallback guarantee role, and the exploratory strategy as the effect testing role.

[0128] Step S620: Based on the preset online allocation algorithm, guided by the optimization objectives defined in the delivery task, and combined with the delivery data in the unified indicator view, dynamically adjust the budget amount and traffic allocation ratio of each candidate strategy based on the preset delivery constraints.

[0129] This step is a dynamic optimization step based on an online allocation algorithm. Guided by the objective function defined in the campaign task (e.g., minimizing effective lead CPA, maximizing effective lead volume), and based on standardized data such as real-time exposure, clicks, conversions, costs, and risks for each candidate strategy in a unified metrics view, this step uses an online allocation algorithm to continuously and dynamically adjust the budget and traffic allocation ratios for each candidate strategy, without exceeding preset constraints such as the platform's target budget, the upper and lower limits of the single strategy budget, and the smooth adjustment range. Its core logic is: increase the budget and traffic share for strategies with excellent real-time performance and high estimated marginal returns; reduce resource allocation for strategies with poor performance and high risk; and maintain the minimum spending for fallback strategies, balancing the relationship between strategy "utilization" and "exploration."

[0130] For example, in a credit allocation scenario, a unified metrics view shows that the effective lead CPA of a certain exploratory strategy is significantly lower than that of the benchmark strategy, and the default rate meets the constraints. The online allocation algorithm then transfers part of the budget and traffic of other inefficient exploratory strategies to this high-quality strategy under the smoothing constraint that the single adjustment range does not exceed 10%. If a certain strategy experiences a sudden increase in cost or abnormal fluctuations, its budget ratio is automatically reduced to ensure that the overall allocation always converges in the optimal direction.

[0131] Step S630: After dynamic adjustment, determine the target budget amount and target traffic allocation ratio for each candidate strategy under each advertising platform to form a strategy-level budget and traffic allocation plan for each platform.

[0132] This step involves determining the strategy-level target budget and traffic allocation plan. After multiple rounds of online dynamic adjustments and reaching a stable deployment state, this step locks in the budget amount and traffic allocation ratio for each candidate strategy. It then determines the final target budget amount and target traffic allocation ratio for each group of candidate strategies under each advertising platform, integrating them to form a strategy-level budget and traffic allocation plan that can be directly deployed to the advertising platform for execution. This plan satisfies the platform's total budget constraints while achieving optimal resource allocation at the strategy level, and simultaneously retains a safety margin for fallback strategies and optimization space for exploratory strategies.

[0133] For example, after dynamic adjustment, the target budget of a high-quality exploratory strategy in the aforementioned information flow platform is increased to 15% of the total budget, the target budget of an inefficient strategy is reduced to 2%, the benchmark strategy is maintained at around 55%, and the fallback strategy remains unchanged at 2%. The resulting strategy-level allocation scheme can be directly used for the execution of each strategy on the advertising platform, providing a precise basis for resource allocation for the creation of subsequent campaign plans.

[0134] In this embodiment, a progressive process is adopted: first, initial budgets and traffic are allocated according to strategy type, and roles are clearly defined; then, online dynamic optimization is performed based on optimization goals and real-time delivery data, ultimately forming an executable allocation plan. This process firstly builds a robust resource structure in the initial stage of deployment. A high proportion of resources is allocated to the main baseline strategy to ensure basic deployment effectiveness; a minimum amount is reserved for backup strategies to ensure deployment continuity; and experimental resources are allocated to exploratory strategies to control trial-and-error risks, balancing deployment stability, security, and optimization potential. Secondly, by using an online allocation algorithm to dynamically adjust based on real-time data from a unified metric view, budgets and traffic can be continuously tilted towards high-yield, low-risk, high-quality strategies while strictly adhering to various constraints, while reducing the resource consumption of inefficient strategies. This effectively balances the full utilization of existing high-quality strategies with the exploration and discovery of new strategies, avoiding resource waste and fluctuations in deployment effectiveness. The resulting strategy-level allocation scheme is precisely adapted to the execution requirements of each advertising platform. It can meet the total budget constraints while achieving refined and maximized resource allocation within a single platform, significantly improving resource utilization efficiency and overall campaign performance at the strategy level. This lays a reliable resource allocation foundation for subsequent automated campaign execution and continuous iterative optimization.

[0135] Step S160: Based on the one-to-one mapping relationship between the strategy vector and the platform parameter set of the corresponding advertising platform, create and update the corresponding delivery plan according to the target budget amount and target traffic allocation ratio of each candidate strategy under each advertising platform.

[0136] This step automates the creation and updating of campaign plans. Based on the one-to-one mapping between the previously generated strategy vectors and the native executable parameter sets of each advertising platform, it accurately converts the algorithm-level strategy configurations into campaign parameter configurations that can be directly recognized and executed by each advertising platform. It strictly adheres to the target budget and target traffic allocation ratio corresponding to each candidate strategy on each platform, completing the creation and real-time updating of the campaign plan. Specifically, it first converts the structured strategy information contained in the strategy vectors, such as bidding mode, bid level, delivery time period, frequency control, and targeting rules, into a set of platform parameters that can be called by the corresponding advertising platform interface, based on the mapping relationship. Then, the determined target budget and target traffic allocation ratio are synchronously written into the campaign plan configuration. Simultaneously, based on the update frequency and smoothing parameter requirements defined in the campaign task, the campaign plan is updated in real-time during the campaign process to address strategy changes and budget adjustments. This ensures that the execution content on the platform is completely consistent with the algorithm optimization results, achieving full automation and standardization from strategy generation and resource allocation to platform execution. This avoids errors and delays caused by manual configuration, guaranteeing the accuracy, timeliness, and stability of cross-platform campaign execution.

[0137] Reference Figure 7 , Figure 7 This application provides a flowchart of steps for creating and updating a corresponding campaign plan based on a one-to-one mapping relationship between a strategy vector and the platform parameter set of the corresponding advertising platform, according to the target budget and target traffic allocation ratio of each candidate strategy under each advertising platform, including but not limited to steps S710 to S740.

[0138] Step S710: Based on the one-to-one mapping relationship between the strategy vector and the platform parameter set of the corresponding advertising platform, the candidate strategy combinations under each advertising platform are converted into delivery parameters that can be natively recognized by the corresponding advertising platform.

[0139] This step involves converting strategy vectors into native delivery parameters for the advertising platform. Based on a pre-established one-to-one mapping between strategy vectors and the parameter sets of each advertising platform, this step parses and converts the strategy vectors corresponding to candidate strategies generated by the algorithm into native delivery parameters that the respective advertising platform can directly recognize and execute. The strategy vectors contain structured strategy information such as bidding mode, bid level, delivery time period, audience targeting, frequency control, and creative combinations. The platform parameter sets consist of parameter names, data types, value ranges, and enumerated options specified by the open interfaces of each advertising platform. This mapping relationship achieves precise matching between strategy semantics and platform execution fields.

[0140] For example, the bidding mode recorded in the strategy vector is OCPC, the target conversion cost is 80 yuan, the delivery time is 18:00–23:00, and the frequency cap is 3 times per day. These are converted into the parameters corresponding to the information flow platform interface through the mapping relationship: bid_type=2, target_cpa=80, time_period=”18-23”, frequency_cap=3, so that the strategy content conforms to the platform execution specifications.

[0141] Step S720: For each candidate strategy under each advertising platform, match its corresponding target budget amount and target traffic allocation ratio, and integrate the delivery parameters with the corresponding budget and traffic parameters to form independent delivery configuration information for each candidate strategy.

[0142] This step integrates the delivery parameters and budget / traffic parameters to form an independent delivery configuration. For each candidate strategy under each advertising platform, this step matches the target budget and target traffic allocation ratio determined in step S150, and combines and encapsulates the converted platform native delivery parameters with the corresponding budget and traffic allocation parameters to form independent, complete, and directly deployable delivery configuration information for each candidate strategy.

[0143] For example, the conversion parameters of the benchmark delivery strategy under a certain information flow platform are matched with a target budget of 275,000 yuan and a target traffic allocation ratio of 55%, the exploratory premium strategy is matched with a budget of 75,000 yuan and a traffic ratio of 15%, and the fallback strategy is matched with a budget of 10,000 yuan and a traffic ratio of 2%. Each of these forms an independent delivery configuration item to ensure that the resource allocation corresponds one-to-one with the strategy content.

[0144] Step S730: Based on the delivery rules of each advertising platform, collect and standardize the independent delivery configuration information of all candidate strategies under the same platform to generate the initial delivery plan for each advertising platform.

[0145] This step involves aggregating and standardizing the placement configurations within the same platform to generate the initial placement plan. Based on the placement management rules, account structure, and plan hierarchy requirements of each advertising platform, this step aggregates, sorts, and standardizes the independent placement configuration information corresponding to all candidate strategies under the same platform. It unifies field order, data format, naming rules, and plan grouping methods, eliminating format conflicts and parameter anomalies to form an initial placement plan that conforms to the platform's interface specifications.

[0146] For example, the deployment configurations of 10 strategies under the information flow platform are uniformly aggregated under the same deployment account, sorted by "baseline strategy - exploration strategy - fallback strategy", and the plan name prefix and budget unit are unified to generate a complete initial deployment plan message, ensuring that the platform can parse it normally.

[0147] Step S740: Call the interfaces of each advertising platform to synchronize the corresponding initial campaign plan to the advertising platform and complete the creation of the campaign plan. If there is an existing campaign plan, update the corresponding campaign plan based on the new campaign configuration information.

[0148] This step involves synchronizing, creating, and updating the campaign's API. By calling the official APIs of various advertising platforms, this step encrypts and transmits the standardized initial campaign plan to the advertising platform's server. If there are no existing campaign plans for the corresponding strategy, the platform will directly create one. If a campaign plan with the same identifier already exists, the platform will update the original plan's parameters, budget, and traffic allocation based on the latest campaign configuration information, enabling real-time iteration of the campaign plan.

[0149] For example, the initial campaign plan is synchronized to the platform via the information flow platform's advertising API. The budget and traffic ratio of the baseline strategy plan that is already running online are updated. New plans are created for the new exploration strategy. The minimum budget for the fallback strategy remains unchanged. Finally, the creation or update of the full campaign plan is completed, achieving real-time consistency between algorithm decision-making and platform execution.

[0150] In this embodiment, a progressive process is employed: converting strategy vectors to platform native parameters, integrating delivery parameters and budget / traffic parameters, standardizing and aggregating configurations within the same platform, and synchronously updating platform interfaces. First, this achieves precise matching between the semantics of the algorithm strategy and the execution fields of various advertising platforms. The abstract strategy vector is standardized and converted into delivery parameters recognizable by the platform, resolving issues of cross-platform strategy format incompatibility and errors in manual configuration. Second, by encapsulating and integrating strategy content with target budgets and traffic allocation ratios, precise correspondence between resource allocation schemes and strategy execution configurations is ensured, avoiding delivery deviations caused by a disconnect between strategy and budget. Third, by aggregating and standardizing configurations within the same platform, delivery plans conform to the interface specifications of each platform, improving the stability and success rate of plan transmission and parsing. Finally, automated creation and real-time updates of delivery plans are achieved through the platform API interface. This completes the entire process from strategy decision-making to platform execution without manual intervention, significantly improving the efficiency of delivery plan deployment and adjustment, reducing manual operation costs and delays, and ensuring a high degree of consistency between algorithm optimization results and actual delivery execution. This provides reliable execution assurance for automated, refined, and real-time cross-platform delivery.

[0151] Step S170: Aggregate the delivery data and post-business link data of various advertising platforms and candidate strategy combinations, and generate an effect analysis report after data normalization and multi-dimensional effect attribution analysis.

[0152] This step is the closed-loop analysis step for campaign performance. It first aggregates and collects real-time campaign data (impressions, clicks, spending, conversions, etc.) and post-business data (effective leads, loan disbursements, default rates, order revenue, etc.) from various advertising platforms and candidate strategy combinations within those platforms. Then, it performs data standardization processing, including cleaning, alignment, normalization, and delay completion, according to a unified metric view. Subsequently, it conducts performance attribution analysis based on multiple dimensions such as campaign channels, strategy types, time periods, and audience targeting. This accurately quantifies the actual contribution and risk performance of each platform and strategy to core optimization objectives (such as CPA, ROI, and effective lead volume). Finally, it integrates these into a complete performance analysis report that includes campaign costs, conversion efficiency, revenue levels, risk indicators, and strategy ranking. This provides objective and comprehensive data support and decision-making basis for subsequent strategy knowledge base updates, budget allocation adjustments, and campaign plan iterations.

[0153] Reference Figure 8 , Figure 8 This is a flowchart of the steps for generating an effect analysis report after aggregating the delivery data and post-business link data of various advertising platforms and various candidate strategy combinations, and performing data normalization and multi-dimensional effect attribution analysis, as provided in an embodiment of this application, including but not limited to steps S810 to S840.

[0154] Step S810: Aggregate all delivery data by advertising platform and candidate strategy combination. The total delivery data includes delivery execution data and post-delivery data for each advertising platform and candidate strategy combination.

[0155] This step involves aggregating all data across multiple dimensions. It collects and summarizes all data generated during the campaign process, categorized by advertising platform and candidate strategy combination. Campaign execution data includes platform statistics such as impressions, clicks, spending, click-through rate, and initial conversions. Post-campaign data includes business outcome data such as valid lead assessment, lead quality score, loan amount, default rate, order gross profit, and risk level, achieving seamless data flow from ad reach to revenue generation.

[0156] For example, the overall delivery data of the three major platforms, namely information flow, short video and search, are collected at the same time, as well as the exposure, click and consumption data corresponding to the benchmark delivery strategy, the backup delivery strategy and multiple sets of exploratory delivery strategies under each platform. The subsequent review results, loan disbursement status and bad performance of the leads brought by each strategy are also associated to form a complete set of raw data.

[0157] Step S820: Perform data normalization processing on the aggregated full-volume delivery data to unify data dimensions, remove outliers, and fill in missing values, forming a standardized attribution dataset.

[0158] This step involves data normalization and the formation of a standardized attribution dataset. It standardizes the aggregated, multi-source, heterogeneous, and high-noise campaign data. Specifically, this includes aligning data from different time granularities and statistical dimensions, removing abnormal clicks, abnormal consumption, and fraudulent conversions, filling in missing data caused by interface delays or statistical deficiencies using interpolation or historical data fitting, and deduplicating cross-platform duplicate conversions and billing data. The final result is a standardized attribution dataset with unified dimensions, consistent definitions, and reliable quality, ensuring the accuracy of subsequent attribution analysis.

[0159] For example, data from some platforms that are statistically analyzed daily and those that are statistically analyzed hourly are aligned to the hourly level. Abnormal records with consumption increases exceeding three times the normal range are removed. For business back-end data that is not returned in time, the same strategy is used to supplement the average value of the same time period. At the same time, duplicate conversion records across platforms are removed according to the unique device identifier, resulting in standardized data that can be directly used for effect calculation.

[0160] Step S830: Based on the metric definition of the campaign task, conduct multi-dimensional effect attribution analysis on the standardized attribution dataset to quantify the contribution of each advertising platform and each candidate strategy combination to the campaign effect.

[0161] This step involves multi-dimensional performance attribution analysis and contribution quantification. Based on the unified metric definitions and optimization objectives in the campaign task definition, and using a standardized attribution dataset, this step conducts performance attribution analysis from multiple dimensions, including advertising platform, strategy type, campaign time period, and bidding model. Incremental contribution methods and weighted allocation methods are employed to calculate the actual contribution and return on investment efficiency of each advertising platform and each candidate strategy combination to core metrics such as effective leads, CPA, ROI, and risk level, thereby identifying high-value channels and high-quality strategies.

[0162] For example, attribution analysis reveals that the search platform contributes the most to the overall ROI. A certain exploratory strategy has a 15% lower effective lead CPA than the benchmark strategy with manageable risk, while an inefficient exploratory strategy has high consumption and insufficient conversion contribution. This clarifies the direction for subsequent resource allocation and strategy elimination.

[0163] Step S840: Integrate the results of multi-dimensional effect attribution analysis, combine them with the core indicators of the campaign to complete data visualization and conclusion summary, and generate a structured effect analysis report.

[0164] This step is for generating a structured performance analysis report. It integrates the results of the multi-dimensional attribution analysis above, and uses data visualization and conclusion summarization to present the overall campaign performance, platform performance, strategy strengths and weaknesses, and risk status around the core performance indicators. The result is a structured performance analysis report that includes an overview of the indicators, platform comparisons, strategy rankings, a summary of problems, and optimization suggestions.

[0165] For example, the report uses charts to show the matching of budget share and performance contribution of each platform, the ranking of CPA and ROI of each candidate strategy, the overall campaign target achievement rate, and extracts the characteristics of high-quality strategies and points out high-risk and inefficient strategies. This provides a direct and actionable basis for decision-making in the next round of campaigns, including updating the strategy knowledge base and optimizing platform-level and strategy-level budget allocation.

[0166] In this embodiment, a complete closed loop of "data aggregation—data standardization—performance attribution—report generation" is achieved. This not only enables comprehensive integration and standardized processing of data throughout the entire campaign process, ensuring data coverage of the entire campaign execution and business results chain, but also accurately quantifies the actual value of each platform and strategy through multi-dimensional attribution analysis, identifying high-quality channels and efficient strategies. Furthermore, based on standardized data and clear conclusions, it provides a scientific and actionable decision-making basis for strategy optimization, budget adjustment, and parameter configuration in the next round of campaigns. Its core benefits lie in ensuring the accuracy of the analytical foundation through full data aggregation and standardization, and clearly presenting the performance differences of each platform and strategy through multi-dimensional attribution and report output. This helps to accurately identify high-value channels and high-quality strategies, eliminate inefficient resources, and form a closed-loop management system of "campaign—analysis—optimization." This avoids resource waste, promotes continuous iteration of campaign strategies, improves the scientific nature, efficiency, and executability of the overall campaign, and ensures that each round of campaigns optimizes resource allocation based on objective data to maximize campaign value.

[0167] Step S180: Based on the attribution analysis results, calculate the true performance index of each candidate strategy combination, and iteratively update the strategy knowledge base based on the true performance index of each candidate strategy combination. The iteratively updated strategy knowledge base drives the next round of candidate strategy generation and budget allocation, forming a self-iterative optimization closed loop.

[0168] This step, as the core iterative link in the entire campaign optimization loop, focuses on the continuous updating of the strategy knowledge base and the formation of an optimized closed loop. Based on the results of the preliminary multi-dimensional attribution analysis, this step first accurately calculates the true performance indicators of each candidate strategy combination, focusing on core performance data such as effective conversion volume, CPA, and ROI for each strategy, clarifying the actual campaign performance of each candidate strategy combination and obtaining the true performance indicators for each strategy. Subsequently, based on these true performance indicators, the strategy knowledge base is iteratively updated, incorporating high-performing strategies (such as efficient and low-cost exploratory strategies and stable and reliable benchmark strategies) and optimizing parameter adaptation rules, eliminating inefficient and ineffective strategies, and supplementing new high-quality strategy features and adaptation experience. Finally, driven by the iteratively updated strategy knowledge base, it provides data support and rule references for the next round of candidate strategy generation and budget allocation, forming a self-iterative optimization closed loop of "strategy execution—effect analysis—knowledge base iteration—new strategy generation," continuously improving the adaptability and effectiveness of campaign strategies and achieving long-term iterative optimization.

[0169] Reference Figure 9 , Figure 9This application provides a flowchart of steps for calculating the true performance index of each candidate strategy combination based on the attribution analysis results, and iteratively updating the strategy knowledge base based on the true performance index of each candidate strategy combination, so as to drive the next round of candidate strategy generation and budget allocation with the iterated strategy knowledge base, forming a self-iterative optimization closed loop, including but not limited to steps S910 to S950.

[0170] Step S910: Based on the results of the multi-dimensional effect attribution analysis, calculate the actual performance indicators of each candidate strategy combination within the corresponding campaign period according to the indicator definitions and calculation rules defined in the campaign task.

[0171] Based on the obtained multi-dimensional effect attribution analysis results, this step strictly follows the unified indicator caliber, calculation formula and statistical period in the definition of the campaign task, and calculates the real campaign performance indicators for each candidate strategy combination under each advertising platform. The indicators include, but are not limited to, effective lead volume, CTR, CVR, CPA, ROI, lead failure rate, campaign stability coefficient, etc., to ensure that the performance results objectively reflect the actual business value of the strategy.

[0172] For example, in a credit customer acquisition scenario, for an exploratory strategy on an information flow platform, if the cost of spending 30,000 yuan and the number of valid leads generated in this round of deployment are calculated according to a unified standard, the actual CPA is 75 yuan, the ROI is 1.8, and the default rate is 2.1%, thus forming a set of actual performance indicators for this strategy.

[0173] Step S920: Associate the actual performance indicators of each candidate strategy combination with its deployment scenario context and parameter configuration information to form a complete strategy performance record.

[0174] This step associates and binds the actual performance indicators calculated in step S910 with the context of the corresponding candidate strategy combination and the full set of parameter configuration information to form a structured and traceable complete strategy performance record. The context includes the advertising platform, product category, target audience, and advertising time period, while the parameter configuration includes the bidding mode, bidding level, frequency control rules, and budget range.

[0175] For example, the performance indicators such as CPA and ROI of the above-mentioned information flow exploratory strategy are associated with scenarios and parameter information such as "information flow platform, credit customer acquisition, 20-45 year old customer group, evening placement, OCPC bidding" to form a complete strategy performance record.

[0176] Step S930: Based on the preset performance evaluation criteria, classify and determine the performance records of each candidate strategy combination, and mark high-quality, inefficient and strategy combinations to be explored.

[0177] This step, based on preset performance evaluation thresholds and grading standards, comprehensively scores and grades the performance records of each candidate strategy combination, clearly marking high-quality strategies, ordinary strategies, inefficient strategies, and strategies with further exploration value. The grading dimensions take into account effectiveness, cost, risk, and stability.

[0178] For example, a strategy with a CPA below 80 yuan, ROI ≥ 1.5, and a defect rate < 3% is considered a high-quality strategy, while a strategy with a CPA above 120 yuan or a defect rate > 5% is considered an inefficient strategy, and the rest are strategies to be explored. Based on this, the aforementioned exploratory strategies are judged to be high-quality strategies, while a strategy with high consumption and a defect rate of 6% is considered an inefficient strategy.

[0179] Step S940: Update the graded strategy performance records to the strategy knowledge base, iteratively replace and supplement the strategy data in the strategy knowledge base, and extract strategy adaptation rules and optimization experience.

[0180] This step involves batch writing the graded strategy performance records into the strategy knowledge base, updating and replacing the existing strategy data in the base: removing long-term inefficient strategy entries, adding high-quality strategy samples for this round, supplementing feature information of strategies to be explored, and extracting adaptation rules based on a large number of performance records, such as optimization experience like "using OCPC bidding + evening time slots in credit customer acquisition on information flow platforms can significantly reduce CPA", thus achieving continuous accumulation and evolution of the strategy knowledge base.

[0181] For example, exploratory strategies that are deemed high-quality in this round are added to the knowledge base, old strategies that have performed poorly for two consecutive rounds are deleted, and the optimal parameter range for the corresponding scenario is extracted.

[0182] Step S950 uses the iteratively updated strategy knowledge base as the core basis to provide prior support for the generation of the next round of candidate strategy combinations and the adjustment of strategy-level budget flow on each platform, forming a self-iterative optimization closed loop across cycles.

[0183] This step uses the iteratively updated strategy knowledge base as the core prior data for the next round of deployment. When generating the baseline and fallback strategies, calculating marginal returns, and allocating the initial strategy budget, it directly reuses high-quality strategy features and adaptation rules to avoid inefficient parameter combinations. This achieves a cross-cycle self-iterative optimization closed loop that results in more accurate candidate strategy generation, more reasonable budget allocation, and continuous improvement in deployment performance.

[0184] For example, the next round of deployment will be fine-tuned based on the high-quality exploratory strategies of this round, and more budget will be allocated to similar high-value strategies, forming a fully automated closed loop of "deployment - evaluation - iteration - optimization".

[0185] In this embodiment, by first accurately calculating the actual performance indicators of each candidate strategy combination, then associating them with scenario and parameter information to form a complete performance record, and then classifying and judging the merits of the strategies, finally updating the strategy knowledge base and driving the next round of deployment optimization, a complete "deployment-evaluation-iteration-optimization" closed loop is formed. Its core beneficial effect lies in its ability to objectively quantify the actual business value of each strategy, accurately distinguish between high-quality, inefficient, and unexplored strategies, and achieve the accumulation of experience with high-quality strategies and the elimination of inefficient strategies through iterative updates to the strategy knowledge base. Simultaneously, it extracts adaptation rules to provide scientific prior support for subsequent deployments. This avoids the repeated investment and resource waste of inefficient strategies, and allows the generation of the next round of candidate strategies to be more aligned with business scenarios and budget allocation to be more targeted, continuously improving deployment effectiveness and resource utilization efficiency. It achieves fully automated self-iterative optimization across cycles, driving the continuous evolution of the intelligent deployment system and ensuring long-term stable improvement in deployment effectiveness.

[0186] In some embodiments, refer to Figure 10 , Figure 10 This is a structural block diagram of a cross-platform advertising optimization system provided in one embodiment of this application. This application also proposes a cross-platform advertising optimization system for use with the cross-platform advertising optimization method provided in any embodiment of this application. The system includes a business and configuration layer 1010, a cross-platform advertising agent layer 1020, a data and feature layer 1030, and an external advertising platform layer 1040. Wherein: The business and configuration layer 1010 is used to receive advertising placement requests and distribute them to the cross-platform pitcher intelligent agent layer 1020.

[0187] The data and feature layer 1030 is used to collect raw report data and business post-link data from multiple advertising platforms, process the collected data to generate a unified indicator view, and provide the unified indicator view to the cross-platform pitcher agent layer.

[0188] The cross-platform pitcher intelligent agent layer 1020 is communicatively connected to the business and configuration layer 1010, the data and feature layer 1030 and the external advertising platform layer 1040 respectively. The cross-platform pitcher intelligent agent layer 1020 is used to execute the aforementioned steps S110, S130-S180.

[0189] The external advertising platform layer 1040 includes multiple cross-channel advertising platforms. The external advertising platform layer connects with the advertising agent intelligent agent layer 1020 through the advertising platform interface. The external advertising platform layer 1040 is used to receive the placement configuration and send back real-time placement data.

[0190] The cross-platform advertising optimization system provided in this application embodiment can implement all the method steps implemented in the cross-platform advertising optimization method embodiment and achieve the same technical effect. Here, the parts that are the same as those in the method embodiment and the beneficial effects will not be described in detail.

[0191] The preferred embodiments of the present application have been described above with reference to the accompanying drawings, but this does not limit the scope of the claims of the present application. Any modifications, equivalent substitutions, and improvements made by those skilled in the art without departing from the scope and substance of the embodiments of the present application shall be within the scope of the claims of the present application.

Claims

1. A cross-platform advertising optimization method, characterized in that, The method includes: The algorithm receives advertising placement requests and parses them into structured objectives and constraints that the algorithm can execute, generating a placement task definition. The placement task definition includes at least an objective function, a set of constraints, an indicator definition, an evaluation time window, an update frequency, and a smoothing parameter. Collect raw report data and post-business link data from multiple advertising platforms, process the collected data, and generate a unified indicator view; Based on the defined delivery task, the unified indicator view, the controllable parameter capabilities and strategy knowledge base of each advertising platform, candidate strategy combinations corresponding to each advertising platform are generated. Then, the candidate strategy combinations of each advertising platform are encoded into corresponding strategy vectors, and a one-to-one mapping relationship is established between each strategy vector and the platform parameter set of the corresponding advertising platform. The candidate strategy combinations include a baseline delivery strategy, a fallback delivery strategy, and an exploratory delivery strategy. The campaign is defined as a constraint and optimization objective. Based on the unified indicator view and the campaign results of the previous campaign cycle, the total budget is allocated at the platform level under the conditions of satisfying the total budget, the upper and lower limits of the proportion of advertising platforms, and the smoothing constraint, so as to obtain the target budget corresponding to each advertising platform. For each advertising platform, within the target budget range corresponding to the advertising platform, the platform budget is allocated to each candidate strategy in the corresponding candidate strategy combination, and the budget amount and traffic allocation ratio corresponding to each candidate strategy are dynamically adjusted by an online allocation algorithm to determine the target budget amount and target traffic allocation ratio corresponding to each candidate strategy under each advertising platform. Based on the one-to-one mapping relationship between the strategy vector and the platform parameter set of the corresponding advertising platform, the corresponding delivery plan is created and updated according to the target budget amount and target traffic allocation ratio of each candidate strategy under each advertising platform. Aggregate the delivery data and post-business link data of various advertising platforms and candidate strategy combinations, and generate an effect analysis report after data normalization and multi-dimensional effect attribution analysis; Based on the attribution analysis results, the true performance index of each candidate strategy combination is calculated, and the strategy knowledge base is iteratively updated based on the true performance index of each candidate strategy combination. The iteratively updated strategy knowledge base drives the next round of candidate strategy generation and budget allocation, forming a self-iterative optimization closed loop.

2. The method according to claim 1, characterized in that, The advertising delivery requirements are parsed into structured objectives and constraints that the algorithm can execute, generating a delivery task definition, including: The advertising placement requirements are extracted into entities, the fuzzy expressions in the advertising placement requirements are converted into task parameters that the algorithm can calculate, and the task parameters are standardized in terms of indicator caliber to determine a unified indicator caliber definition. The constraints corresponding to the task parameters after standardization of indicator definitions are checked for consistency and completed to form a complete set of constraints. Based on the set of constraints and the task parameters, construct an objective function that adapts to the advertising delivery requirements; Based on the objective function, the set of constraints, and the advertising delivery requirements, the evaluation time window, the update frequency of the delivery parameters, and the smoothing parameter are determined. The smoothing parameter is a parameter that limits the adjustment range and frequency of budget and bidding control variables. By integrating the objective function, the set of constraints, the definition of the indicator, the evaluation time window, the update frequency, and the smoothing parameter, a structured deployment task definition is generated.

3. The method according to claim 1, characterized in that, The process involves collecting raw report data and post-business link data from multiple advertising platforms, processing the collected data, and generating a unified metric view, including: Collect raw report data and business post-link data from multiple advertising platforms, and perform unified alignment processing on the collected data according to preset dimensions; The data after dimension alignment is cleaned and made consistent to achieve data standardization, outlier handling, and cross-platform deduplication. After cleaning and standardization, cross-platform metric normalization is performed on the data to transform the metrics of various advertising platforms to a unified dimension and form standardized metric data. Based on historical data, the standardized indicator data is modeled and transformed for prediction and completion using delayed backhaul, generating a derived indicator set containing prediction-type indicators. By integrating the standardized indicator data and the derived indicator set, a unified indicator view is generated that covers multi-dimensional observed data, estimated data, and risk-related indicators from various advertising platforms.

4. The method according to claim 1, characterized in that, Based on the defined delivery task, the unified metric view, the controllable parameter capabilities of each advertising platform, and the strategy knowledge base, candidate strategy combinations corresponding to each advertising platform are generated, including: Based on the defined delivery task and the controllable parameter capabilities of each advertising platform, determine the corresponding strategy parameter space for each advertising platform; Based on the strategy knowledge base, unified indicator view, and strategy parameter space of each advertising platform, a benchmark delivery strategy and a fallback delivery strategy corresponding to each advertising platform are generated. Combining the optimization objectives defined in the aforementioned campaign task with the unified metric view, strategies are explored within the strategy parameter space of each advertising platform to generate exploratory campaign strategies corresponding to each advertising platform. Constraint pruning, feasibility verification, and deduplication are performed on the baseline, fallback, and exploratory placement strategies of each advertising platform to obtain candidate strategy combinations for each platform.

5. The method according to claim 1, characterized in that, The process involves defining the campaign task as a constraint and optimization objective, using the unified metric view and the campaign results from the previous campaign cycle as data, and allocating the total budget at the platform level under the conditions of satisfying the total budget, upper and lower limits of the advertising platform's share, and smoothing constraints. This yields the target budget corresponding to each advertising platform, including: Based on the unified indicator view and the results of the previous campaign, combined with the optimization objectives defined in the campaign task, the expected marginal revenue per unit budget of each advertising platform is calculated. Risk and stability adjustments were made to the expected marginal revenue of each advertising platform. Using the aforementioned task definition as a constraint, under the conditions of satisfying the total budget, the upper and lower limits of the advertising platform's share, and the smoothing constraint, a constrained budget allocation calculation is performed on the modified expected marginal revenue. The budget allocation calculation results are converted into budget plans that each advertising platform can execute, in order to determine the target budget for each advertising platform.

6. The method according to claim 1, characterized in that, For each advertising platform, within the target budget range corresponding to the advertising platform, the platform budget is allocated to each candidate strategy in the corresponding candidate strategy combination, and the budget amount and traffic allocation ratio corresponding to each candidate strategy are dynamically adjusted through an online allocation algorithm to determine the target budget amount and target traffic allocation ratio corresponding to each candidate strategy under each advertising platform, including: For each advertising platform, within the target budget range corresponding to the advertising platform, an initial budget amount and an initial traffic allocation ratio are allocated for each candidate strategy in the candidate strategy combination corresponding to the advertising platform, and the delivery role of each candidate strategy is defined. Based on a preset online allocation algorithm, guided by the optimization objectives defined in the delivery task, and combined with the delivery data from a unified metrics view, the budget and traffic allocation ratio of each candidate strategy are dynamically adjusted based on preset delivery constraints. After dynamic adjustments, the target budget amount and target traffic allocation ratio for each candidate strategy under each advertising platform are determined, forming a strategy-level budget and traffic allocation plan for each platform.

7. The method according to claim 1, characterized in that, The process of creating and updating corresponding campaign plans based on the one-to-one mapping relationship between the strategy vector and the platform parameter set of the corresponding advertising platform, according to the target budget and target traffic allocation ratio corresponding to each candidate strategy under each advertising platform, includes: Based on the one-to-one mapping relationship between the strategy vector and the platform parameter set of the corresponding advertising platform, the candidate strategy combinations under each advertising platform are converted into delivery parameters that the corresponding advertising platform can natively recognize. For each candidate strategy under each advertising platform, match its corresponding target budget amount and target traffic allocation ratio, and integrate the delivery parameters with the corresponding budget and traffic parameters to form independent delivery configuration information for each candidate strategy; Based on the delivery rules of each advertising platform, the independent delivery configuration information of all candidate strategies under the same platform is collected and standardized to generate the initial delivery plan for each advertising platform. Call the interfaces of each advertising platform to synchronize the corresponding initial campaign plan to the advertising platform and complete the creation of the campaign plan. If there is an existing campaign plan, update the corresponding campaign plan based on the new campaign configuration information.

8. The method according to claim 1, characterized in that, The aggregated data from various advertising platforms and candidate strategy combinations, along with post-business data, are processed through data normalization and multi-dimensional performance attribution analysis to generate a performance analysis report, including: Aggregate all delivery data by advertising platform and candidate strategy combination. The total delivery data includes delivery execution data and post-business link data for each advertising platform and each candidate strategy combination. The aggregated full data of the campaign is processed to standardize the data dimensions, remove outliers and fill in missing values, forming a standardized attribution dataset. Based on the metrics defined in the aforementioned campaign task, a multi-dimensional effect attribution analysis is conducted on the standardized attribution dataset to quantify the contribution of each advertising platform and each candidate strategy combination to the campaign effect. By integrating multi-dimensional performance attribution analysis results and combining them with key performance indicators, data visualization and conclusion summarization are completed, generating a structured performance analysis report.

9. The method according to claim 1, characterized in that, Based on the attribution analysis results, the true performance indicators of each candidate strategy combination are calculated, and the strategy knowledge base is iteratively updated based on the true performance indicators of each candidate strategy combination. The iterated strategy knowledge base drives the next round of candidate strategy generation and budget allocation, forming a self-iterative optimization closed loop, including: Based on the results of multi-dimensional effect attribution analysis, the actual performance indicators of each candidate strategy combination in the corresponding campaign period are calculated according to the indicator caliber and calculation rules defined in the campaign task. The actual performance indicators of each candidate strategy combination are associated with its deployment scenario context and parameter configuration information to form a complete strategy performance record. Based on the preset performance evaluation criteria, the performance records of each candidate strategy combination are graded and marked as high-quality, inefficient, and strategy combinations to be explored. The graded strategy performance records are updated to the strategy knowledge base, the strategy data in the strategy knowledge base is iteratively replaced, supplemented and improved, and strategy adaptation rules and optimization experience are extracted. Using the iteratively updated strategy knowledge base as the core basis, it provides prior support for the generation of the next round of candidate strategy combinations and the adjustment of strategy-level budget flow on each platform, forming a self-iterative optimization closed loop across cycles.

10. A cross-platform advertising optimization system, characterized in that, The system is used to perform the method according to any one of claims 1-9, the system comprising a business and configuration layer, a cross-platform pitcher agent layer, a data and feature layer, and an external advertising platform layer; The business and configuration layer is used to receive advertising placement requests and send the advertising placement requests to the cross-platform pitcher intelligent agent layer; The data and feature layer is used to collect raw report data and business post-link data from multiple advertising platforms, process the collected data to generate a unified indicator view, and provide the unified indicator view to the cross-platform pitcher agent layer. The cross-platform pitcher agent layer is communicatively connected to the business and configuration layer, the data and feature layer, and the external advertising platform layer, respectively. The cross-platform pitcher agent layer is used to perform the following steps: The algorithm receives the advertising delivery request and parses it into structured objectives and constraints that the algorithm can execute, generating a delivery task definition. The delivery task definition includes at least an objective function, a set of constraints, a definition of indicators, an evaluation time window, an update frequency, and a smoothing parameter. Based on the defined campaign, the unified metric view, the controllable parameters of each advertising platform, and the strategy knowledge base, candidate strategy combinations are generated for each advertising platform. These candidate strategy combinations are then encoded into corresponding strategy vectors, and a one-to-one mapping relationship is established between each strategy vector and the platform parameter set of the corresponding advertising platform. The candidate strategy combinations include a baseline campaign strategy, a fallback campaign strategy, and an exploratory campaign strategy. Using the defined campaign as a constraint and optimization objective, and based on the unified metric view and the campaign results of the previous campaign cycle, the total budget is allocated at the platform level under the conditions of satisfying the total budget, the upper and lower limits of the advertising platform's share, and smoothing constraints, resulting in the target budget for each advertising platform. For each advertising platform, within the target budget range corresponding to the advertising platform, the platform budget is allocated to each candidate strategy in the corresponding candidate strategy combination, and the budget amount and traffic allocation ratio corresponding to each candidate strategy are dynamically adjusted by an online allocation algorithm to determine the target budget amount and target traffic allocation ratio corresponding to each candidate strategy under each advertising platform. Based on the one-to-one mapping relationship between the strategy vector and the platform parameter set of the corresponding advertising platform, the corresponding delivery plan is created and updated according to the target budget amount and target traffic allocation ratio of each candidate strategy under each advertising platform. Aggregate the delivery data and post-business link data of various advertising platforms and candidate strategy combinations, and generate an effect analysis report after data normalization and multi-dimensional effect attribution analysis; Based on the attribution analysis results, the true performance index of each candidate strategy combination is calculated, and the strategy knowledge base is iteratively updated based on the true performance index of each candidate strategy combination. The iteratively updated strategy knowledge base drives the next round of candidate strategy generation and budget allocation, forming a self-iterative optimization closed loop. The external advertising platform layer includes multiple cross-channel advertising platforms. The external advertising platform layer connects with the advertising agent intelligent agent layer through the advertising platform interface. The external advertising platform layer is used to receive the placement configuration and send back real-time placement data.

Images