Budget allocation strategy generation method and device, equipment and storage medium

Abstract

The invention discloses a budget allocation strategy generation method and device, equipment and a storage medium, and the method comprises the steps: carrying out the disassembly of a total budget amount according to a budget period and each time period, and obtaining a budget disassembly result; performing target disassembly according to the delivery target type and the industry to which the enterprise belongs to obtain a target disassembly result; combining the budget disassembling result and the target disassembling result to obtain a preliminary budget allocation framework; embedding historical delivery data of the same industry and the optimal practice of the industry into a rule engine; and based on the preliminary budget allocation framework, the static basic data and the dynamic real-time data, generating at least two types of alternative budget allocation strategies by adopting a mode of fusing a machine learning model and an embedded rule engine. According to the method, the client side data, the platform side data and the industry side data are systematically integrated, and the distribution strategy is generated through fusion of the rule engine and the machine learning model, so that adaptive fusion and real-time optimization of the strategy are realized, and the generation efficiency is remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to a method, apparatus, device, and storage medium for generating budget allocation strategies. Background Technology

[0002] With the continued advancement of global business expansion, the demand for overseas market development is growing stronger. Overseas multimedia advertising, with its wide reach and diverse communication formats, has become a core means for brands to enter overseas markets. Currently, the global overseas advertising market is showing steady growth. Platforms covering multiple scenarios such as search, social media, short video, e-commerce, and youth-oriented social media advertising, including search, social media, e-commerce, and short video platforms, have become the mainstream channels for overseas advertising by businesses.

[0003] In overseas advertising scenarios, the core objective for businesses is to maximize campaign effectiveness within a limited budget. This requires budget allocation strategies to comprehensively consider factors such as the company's total budget, campaign objectives, product and industry characteristics, as well as user profiles, conversion costs, policy rules, and traffic fluctuation cycles on different overseas platforms. However, current overseas advertising is characterized by multiple platforms, multiple objectives, and high dynamism, making traditional budget allocation methods inadequate for meeting actual needs and revealing numerous shortcomings.

[0004] Existing budget allocation schemes mainly fall into two categories. The first category is the manual experience-based budget allocation scheme, which heavily relies on operations personnel or ad optimization specialists with overseas advertising experience to complete the budget allocation. The specific process is as follows: first, operations personnel collect basic customer information and static data from overseas platforms; then, they break down the budget based on their personal experience; and finally, during the campaign, they periodically review the data manually and adjust the budget allocation manually. This scheme is not only inefficient, but the allocation results are also greatly limited by personal experience, making it difficult to cope with complex and ever-changing campaign scenarios. In addition, manual adjustments are lagging and easily lead to wasted budget.

[0005] The second type is basic algorithm-based budget allocation tools. Some companies or platforms offer these tools, which generate budget allocation strategies using single-dimensional data and fixed rules. Their data input only includes two core data points: the client's total budget and the historical traffic share or average conversion cost of each platform. The algorithm logic uses a linear allocation model, and the output is a fixed "Platform-Budget Share" table. Furthermore, the strategy cannot be automatically adjusted during the campaign period and requires manual intervention. These tools suffer from limited data dimensions and rigid algorithm logic, failing to fully adapt to the diverse needs of businesses and the dynamic changes of platforms. Some agencies may even recommend platforms within their agency qualifications based on their own KPI considerations, deviating from the core need of maximizing client advertising effectiveness. Summary of the Invention

[0006] The purpose of this invention is to provide a method, apparatus, device, and storage medium for generating budget allocation strategies, aiming to address the problems of low efficiency and poor adaptability in existing budget allocation schemes.

[0007] In a first aspect, embodiments of the present invention provide a method for generating a budget allocation strategy, comprising: Collect customer-side data, platform-side data, and industry-side data; wherein, the customer-side data includes total budget amount, budget period, target type of advertising, and the industry to which the enterprise belongs; the platform-side data includes static basic data and dynamic real-time data; the industry-side data includes historical advertising data of the same industry and industry best practices; The total budget amount is broken down into different time periods according to the budget cycle to obtain the budget breakdown results; Based on the target type and the industry to which the enterprise belongs, the target is broken down to obtain the target breakdown results; The budget breakdown results and the target breakdown results are combined to obtain a preliminary budget allocation framework; The historical campaign data and best practices of the same industry are embedded into the rules engine; Based on the aforementioned preliminary budget allocation framework, static basic data, and dynamic real-time data, at least two types of alternative budget allocation strategies are generated by fusing machine learning models with an embedded rule engine. The types of alternative budget allocation strategies include conservative, balanced, and aggressive.

[0008] In a second aspect, embodiments of the present invention provide a budget allocation strategy generation apparatus, comprising: The data collection unit is used to collect customer-side data, platform-side data, and industry-side data. The customer-side data includes total budget amount, budget period, target type, and the industry to which the company belongs. The platform-side data includes static basic data and dynamic real-time data. The industry-side data includes historical campaign data from the same industry and industry best practices. The budget breakdown unit is used to break down the total budget amount into different time periods according to the budget period to obtain the budget breakdown result; The target decomposition unit is used to decompose the target according to the target type and the industry to which the enterprise belongs, and obtain the target decomposition result; A combination unit is used to combine the budget breakdown results and the target breakdown results to obtain a preliminary budget allocation framework; The embedding unit is used to embed the historical deployment data and industry best practices of the same industry into the rule engine; The generation unit is used to generate at least two types of alternative budget allocation strategies based on the preliminary budget allocation framework, static basic data and dynamic real-time data by using a machine learning model and an embedded rule engine. The types of alternative budget allocation strategies include conservative, balanced and aggressive.

[0009] Thirdly, embodiments of the present invention provide a computer device including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the budget allocation strategy generation method described in the first aspect.

[0010] Fourthly, embodiments of the present invention also provide a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program that, when executed by a processor, implements the budget allocation strategy generation method described in the first aspect.

[0011] This invention discloses a method, apparatus, device, and storage medium for generating budget allocation strategies. The method includes: collecting customer-side data, platform-side data, and industry-side data; wherein, the customer-side data includes total budget amount, budget period, target type, and the industry to which the enterprise belongs; the platform-side data includes static basic data and dynamic real-time data; the industry-side data includes historical campaign data and industry best practices; the total budget amount is broken down into different time periods according to the budget period to obtain budget breakdown results; the target type and the industry to which the enterprise belongs are broken down into targets to obtain target breakdown results; the budget breakdown results and the target breakdown results are combined to obtain a preliminary budget allocation framework; the historical campaign data and industry best practices are embedded into a rule engine; based on the preliminary budget allocation framework, static basic data, and dynamic real-time data, at least two types of alternative budget allocation strategies are generated by fusing machine learning models with the embedded rule engine, wherein the types of alternative budget allocation strategies include conservative, balanced, and aggressive. This invention systematically integrates customer-side data, platform-side data, and industry-side data. Through the fusion of a rule engine and a machine learning model, it generates at least two alternative budget allocation schemes, achieving adaptive fusion and real-time optimization of strategies. This scheme avoids decision-making biases caused by reliance on human experience and overcomes the limitation of basic algorithms relying solely on single-dimensional data. It enables budget allocation strategies to synchronously match changes in the market environment and personalized customer needs, significantly improving the scientific rigor and dynamic adaptability of strategy generation, while also increasing the efficiency of budget allocation scheme generation. Embodiments of this invention also provide a budget allocation strategy generation device, a computer-readable storage medium, and a computer device, all possessing the aforementioned beneficial effects, which will not be elaborated further here. Attached Figure Description

[0012] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0013] Figure 1 A flowchart illustrating the method for generating budget allocation strategies; Figure 2 A schematic block diagram of a device for generating budget allocation strategies. Detailed Implementation

[0014] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0015] It should be understood that, when used in this specification and the appended claims, the terms “comprising” and “including” indicate the presence of the described features, integrals, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more of its features, integrals, steps, operations, elements, components and / or collections thereof.

[0016] It should also be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the invention. As used in this specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0017] It should also be further understood that the term "and / or" as used in this specification and the appended claims refers to any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0018] Please see Figure 1 This embodiment provides a method for generating a budget allocation strategy, including: S101: Collect customer-side data, platform-side data, and industry-side data; wherein, the customer-side data includes total budget amount, budget period, target type of advertising, and the industry to which the enterprise belongs; the platform-side data includes static basic data and dynamic real-time data; the industry-side data includes historical advertising data of the same industry and industry best practices; In this embodiment, when a fast-moving consumer goods (FMCG) company seeking to expand overseas initiates an overseas advertising budget allocation request, the system automatically activates the data collection module. During the client-side data collection phase, the system obtains the total budget amount and preset budget period through a dedicated portal connected to the company, clarifying its target type of simultaneous brand exposure and sales conversion, and recording the company's FMCG industry attributes. On the platform-side data collection side, the system simultaneously captures static basic data from target advertising platforms such as search advertising platforms and social advertising platforms, including rules such as platform entry thresholds and ad format restrictions, as well as dynamic real-time data covering dynamic indicators such as current platform traffic volume and user click-through rates. For industry-side data, the system retrieves historical advertising data from companies in the same FMCG category by accessing the built-in industry feature library, analyzing advertising strategies and performance feedback under different scenarios, and integrating best practices in overseas advertising within the industry to form a complete multi-dimensional data source, providing comprehensive data support for subsequent budget analysis.

[0019] Taking the overseas advertising needs of a 3C company during its new product launch period in the North American market as an example, we conducted multi-dimensional data collection. In the client-side data collection phase, we obtained its total budget, monthly budget cycle, and budget adjustment threshold allowing for a certain range of fluctuations. We clarified the target type of advertising, focusing on conversion rate and ROI (Return on Investment, a core indicator for overseas advertising, calculated as: ROI = (Direct revenue from advertising - Advertising cost) ÷ Advertising cost × 100%, reflecting the return on each unit of budget). We determined the weighting of these two factors and the target KPI for cost per conversion. Simultaneously, we recorded information about the company's 3C industry, the product lifecycle during the new product launch period, and its core target market in North America. In the platform-side data collection phase, we collected static basic data on target platform advertising formats such as search ads on search advertising platforms and feed ads on short video advertising platforms, as well as the minimum advertising thresholds and billing methods such as CPC and CPM for each platform. Through API integration with the backends of each platform, we captured dynamic real-time data such as traffic volume, user age, gender, and geographic profiles, current conversion costs, competitor advertising share, and updates to platform review rules. In terms of industry-side data collection, we extracted historical data on the average ROI of search advertising platforms in the North American market for the 3C industry from the constructed industry feature library, as well as industry best practices for exposure during the new product launch period, to provide comprehensive data support for the generation of subsequent budget allocation strategies.

[0020] It should be noted that the acquisition of implementation data in this embodiment complies with laws and regulations, social ethics, and public interests, and does not infringe on the privacy of others. This implementation data includes customer-side data, platform-side data, industry-side data, and other data to be acquired subsequently.

[0021] S102: Break down the total budget amount into different time periods according to the budget cycle to obtain the budget breakdown results; Specifically, the total budget amount is broken down into different time periods according to the budget cycle, resulting in the following budget breakdown: Obtain the total budget amount, budget period, platform traffic fluctuation characteristics, platform historical traffic data, and industry-specific time period traffic distribution data from the industry feature library; The budget period is divided into multiple consecutive time periods based on the length of the budget period and the characteristics of platform traffic fluctuations. The traffic weight coefficient for each time period is calculated based on the platform's historical traffic data and the same industry time period traffic distribution data in the industry feature library. The total budget amount is allocated according to the proportion of the traffic weight coefficient to obtain the budget breakdown result.

[0022] This example uses a maternal and infant product company focusing on the European market as an example. The company plans to promote its new infant food products through overseas advertising, with the core objective of increasing market penetration and sales in key European countries. To ensure precise utilization of the limited budget, the system initiates a budget time-segmentation process, as detailed below: During the data acquisition phase, the system first connects to the enterprise's customer management backend to collect the total quarterly budget amount, specifying a three-month budget cycle. Simultaneously, based on the enterprise's selected search advertising platform and social advertising platform—two core advertising platforms—the system retrieves the traffic fluctuation characteristics of these platforms over the past six months via API interfaces. This reveals that in the European market, weekday traffic is concentrated between 9:00-11:00 AM, 3:00-5:00 PM, and 7:00-9:00 PM, with parents exhibiting the highest interaction frequency during the evening hours. On weekends, traffic gradually increases after 10:00 AM, peaking between 2:00-4:00 PM. Furthermore, the system extracts historical traffic data from both platforms over the past six months, including key metrics such as exposure, user session duration, and click-through rate for each time period. It also retrieves traffic distribution data for the maternal and infant industry in the European market from an industry feature database, focusing on the time-based advertising patterns of leading complementary food brands to identify the unique characteristics of traffic fluctuations before and after holidays, such as Children's Day and the European summer holidays, which experience periods of increased traffic.

[0023] In the time-segmentation phase, the system combines the three-month budget cycle length with the aforementioned platform traffic fluctuation characteristics to perform refined time-segmentation. First, the total cycle is divided into three monthly sub-cycles. Then, each monthly sub-cycle is further divided into two modules: weekdays and weekends. Weekdays are further subdivided into six consecutive time slots: morning (7-9 am), late morning (9-12 pm), noon (12-2 pm), afternoon (2-6 pm), evening (6-10 pm), and night (after 10 pm). Weekends are adjusted to four time slots: morning (9-12 pm), afternoon (12-6 pm), evening (6-10 pm), and night (after 10 pm). Special time slot labels are reserved for holidays to allow for flexible adjustments for special occasions. This segmentation closely aligns with the daily routines of the target users of maternal and infant products (parents), ensuring a high degree of match between time-segment coverage and user activity periods.

[0024] During the weighting coefficient calculation phase, the system employs a weighted fusion algorithm to process historical traffic data from the platform and traffic distribution data from the same industry across different time periods. For historical traffic data, data from the past three months is given higher weighting, with a focus on traffic volume and conversion efficiency at each time period. For traffic distribution data from the same industry across different time periods, successful campaign characteristics of leading brands are extracted as a benchmark to correct for deviations in the system's own data. The algorithm calculates the traffic weighting coefficients for each time period: weekday evenings (6-10 PM) have the highest coefficients due to higher user activity and conversion intentions; weekend afternoons (12-6 PM) have the second highest coefficients, aligning with the leisure browsing habits of family users; midday and evening periods have lower coefficients, corresponding to natural low points in user traffic. Simultaneously, for holiday periods expected to see traffic peaks, the corresponding base coefficients are adjusted upwards.

[0025] In the budget allocation phase, the system precisely breaks down the company's total quarterly budget according to the traffic weighting coefficients calculated above for each time period. For example, the highest proportion of the budget is allocated to weekday evenings to ensure strong advertising during peak activity times; the second highest proportion is allocated to weekend afternoons to cover the core browsing windows of home users; and lower proportions are allocated to midday and evening periods to avoid resource waste. This breakdown results in a structured budget breakdown with holiday adjustment plans. This result is then synchronized to subsequent target breakdown and strategy generation phases, providing precise time-based data support for platform matching and budget allocation, ensuring that advertising accurately captures peak user traffic and maximizes budget utilization efficiency.

[0026] In some embodiments, the traffic weighting coefficients for each time period are calculated based on the platform's historical traffic data and the same industry's time-period traffic distribution data in the industry feature library, including: The platform's historical traffic data is standardized, converting different levels of exposure, clicks, and active users into standardized values. Outlier removal from traffic distribution data for the same industry during different time periods; The analytic hierarchy process was used to determine the fusion weights between the standardized values ​​and the excluded industry-specific time-period traffic distribution data. Based on the fusion weight, the standardized value is weighted and summed with the traffic distribution data of the same industry in different time periods after elimination, so as to obtain the traffic weight coefficient of each time period.

[0027] This embodiment uses a maternal and infant product company focusing on the European market as the application target. Based on data from its selected search advertising platform and social advertising platform, it calculates the traffic weight coefficient for each time period. The specific process is as follows: The system first extracts historical traffic data from the two platforms over the past six months, covering three core metrics: impressions, clicks, and active users. Due to the significant differences in the magnitude of these three metrics (e.g., impressions in millions, active users in tens of thousands), a min-max standardization method is used to process the data. The value range of each metric is uniformly mapped to the [0,1] interval, and the influence of different magnitudes is eliminated through formulas to obtain the standardized values ​​for the three metrics for each time period. For example, the standardized value for impressions during weekday evening hours on the search advertising platform is 0.92, the standardized value for clicks is 0.88, and the standardized value for active users is 0.90.

[0028] Next, traffic distribution data for the maternal and infant industry in the European market was retrieved from the industry feature database, including traffic records for leading complementary food brands for the past year. The quartile method was used to detect outliers, calculating the first, second, and third quartiles to determine the threshold for outlier detection. Extreme data exceeding the threshold range (such as sudden increases in traffic due to special marketing activities or sudden drops in traffic due to system failures) were removed, retaining only valid data that conforms to the industry's normal fluctuation patterns to ensure the accuracy of subsequent calculations.

[0029] Then, a hierarchical analysis model was constructed, with "obtaining accurate traffic weight coefficients" as the target layer and "historical traffic data of the platform" and "traffic distribution data of the same industry during the same period" as the criterion layers. Five experts in the field of overseas advertising were invited to compare the importance of the two criterion layer indicators pairwise, forming a judgment matrix. The rationality of the judgment matrix was verified through a consistency test. After the test passed, the weights of each criterion layer were calculated: the fusion weight of historical traffic data of the platform was 0.6, and the fusion weight of traffic distribution data of the same industry during the same period was 0.4, which highlights the targeting of the company's own advertising data while taking into account the reference value of industry patterns.

[0030] Next, for each time period, the standardized historical traffic data of the platform (the average of the standardized values ​​of the three types of indicators) and the traffic distribution data of the same industry during the same time period after removing outliers are weighted and summed according to the fusion weights determined above. For example, the average platform data for weekday evenings on the search advertising platform is 0.90, and the industry data is 0.85, so the traffic weight coefficient for this time period is calculated to be 0.90×0.6+0.85×0.4=0.88; the average platform data for weekend afternoons is 0.75, and the industry data is 0.80, so the corresponding traffic weight coefficient is 0.75×0.6+0.80×0.4=0.77. The traffic weight coefficients for all time periods are calculated in this way to form a complete coefficient matrix, providing core data support for subsequent budget period breakdown.

[0031] S103: Based on the target type and the industry to which the enterprise belongs, the target is broken down to obtain the target breakdown results; Specifically, based on the target type and the industry of the enterprise, the target is broken down, and the target breakdown results include: The target types are broken down into multiple dimensions to clarify the attribute characteristics of each sub-target; Based on the industry to which the company belongs, extract the common patterns of target decomposition in the industry under the same advertising objectives from the industry-side data; By combining the attribute characteristics of sub-targets with the common patterns of target decomposition, an adaptation relationship between sub-targets and multimedia advertising platforms is established. The adaptation relationship between each sub-objective and its corresponding platform is integrated into the object decomposition result.

[0032] This embodiment focuses on the common launch objective of increasing market share for new products in the FMCG industry, and conducts standardized objective decomposition work. The specific process is as follows: The core objective of increasing market share for new products is broken down into three sub-objectives: brand awareness enhancement, user interest stimulation, and purchase conversion promotion. The characteristics of each sub-objective are clearly defined: the brand awareness enhancement sub-objective targets a broad potential consumer group, with the behavioral guidance focused on conveying the core selling points of the new product (such as ingredient advantages and usage scenarios), and its value is measured by exposure and brand search volume; the user interest stimulation sub-objective targets a group with potential demand for new FMCG products, with the behavioral guidance focused on attracting user interaction (such as liking, saving, and completing questionnaires), and its value is measured by interaction rate and content dwell time; the purchase conversion promotion sub-objective targets a high-intent consumer group, with the behavioral guidance focused on driving users to complete their purchase decisions, and its value is measured by conversion rate and repurchase rate.

[0033] Next, we extracted common patterns in the target breakdown of new product market share improvement campaigns within the FMCG industry from industry-side data. We found that the industry generally follows a tiered sub-target progression logic of awareness-interest-conversion. In the early stages, brand awareness sub-targets are the core; in the mid-stage, the focus shifts to cultivating user interest; and in the later stage, the focus is on purchase conversion. Regarding platform adaptation, brand awareness sub-targets tend to align with platforms with strong social and short-video attributes; user interest sub-targets can be promoted simultaneously on social and content platforms; and purchase conversion sub-targets prioritize search and e-commerce platforms.

[0034] Then, combining the characteristics of sub-targets with common industry patterns, an appropriate correlation is established. The brand awareness enhancement sub-target is adapted to short video advertising platforms and social advertising platforms. The former relies on short videos to efficiently convey the selling points of new products, while the latter covers a wide consumer group to achieve large-scale exposure. The user interest stimulation sub-target, in addition to continuing with the two major social platforms mentioned above, supplements with diversified advertising platforms, using content formats combining text and images with short videos to enhance interactive guidance. The purchase conversion promotion sub-target focuses on adapting to search advertising platforms and e-commerce advertising platforms. The former accurately reaches users with purchase intentions through search keywords, while the latter shortens the conversion path by relying on the shopping scenario of e-commerce platforms.

[0035] The sub-goals and their compatibility with the platform are then structurally integrated to form a goal breakdown. The results are presented in a structured list, clearly defining the compatible platforms, behavioral guidance strategies, and value measurement dimensions for each sub-goal. The integrated results clearly define the execution vehicles and evaluation standards for each stage, providing a unified goal orientation for subsequent budget allocation and strategy development.

[0036] S104: Combine the budget breakdown results and the target breakdown results to obtain a preliminary budget allocation framework; Specifically, the budget amounts for each time period provided by the budget breakdown results are aligned and matched in the spatiotemporal dimensions with the platform sub-objective requirements determined by the target breakdown results obtained by the strategy generation module. This process establishes a correlation matrix among time periods, platforms, and sub-objectives.

[0037] Next, based on the established association matrix, the available budget for each time period is initially allocated according to the priority and difficulty of the sub-goals to be achieved by each platform within that time period, through the preset allocation logic in the rule engine. For time periods that require rapid achievement of exposure goals, the budget proportion for social media platforms is increased accordingly; for time periods that focus on conversion goals, the budget is tilted towards search and e-commerce platforms.

[0038] Next, threshold verification and risk verification will be carried out in sequence to ensure the compliance, feasibility, and low risk of budget allocation. The specific process is as follows: We retrieved the minimum placement threshold standards for each compatible overseas advertising platform from the static basic data on the platform side, and clarified the minimum daily budget constraints for short video advertising platforms, social advertising platforms, diversified advertising platforms, search advertising platforms, and e-commerce advertising platforms; at the same time, we extracted the budget details data of "day-by-day-by-platform" in the preliminary budget allocation framework to form a daily budget list for the entire platform.

[0039] The system verifies each data point on the list. If a platform's daily budget is found to be below its minimum spending threshold, the system automatically initiates an adjustment mechanism, raising the platform's daily budget to the minimum threshold. Adjustments are first deducted from the platform with the lowest daily budget priority to ensure the total budget does not exceed the upper limit. For example, if a diversified advertising platform's daily budget is found to be below the minimum threshold, after being raised to the standard amount, the corresponding amount is deducted from low-priority platforms that are experiencing low traffic during that day.

[0040] After adjustments are made, re-verify that the daily budgets for all platforms meet the threshold requirements and that the total budget remains unchanged. Once verification is successful, update the daily budget data in the initial budget allocation framework and generate adjustment records, clearly specifying the adjustment targets, amounts, and funding sources.

[0041] Next, the policies and rules of various platforms are integrated from the static basic data and dynamic real-time data on the platform side, highlighting the restrictions on specific advertising content; at the same time, a compliance risk threshold is set (a compliance rate of less than 80% is considered high risk), and a three-dimensional risk assessment model is established based on the matching degree of policy terms, reference to historical compliance cases, and adaptation to real-time review rules.

[0042] Based on the evaluation model, risk assessments are then performed on the adjusted budget allocation details, with a focus on evaluating whether the content allocated to each platform complies with platform policies. For example, if a certain type of advertising content faces strict promotional restrictions on a particular platform, and the evaluation shows that the compliance rate of the campaign allocated to that platform's budget is only 70%, which is below the set threshold and is therefore deemed a high-risk allocation.

[0043] Subsequently, an adjustment process was initiated for the budgets of high-risk platforms, transferring that portion of the budget to low-risk platforms. The adjustments were based on the sub-objective compatibility within the initial budget allocation framework, prioritizing transfers to compliant platforms within the same sub-objective. For example, the budgets of the aforementioned high-risk platforms were transferred to another platform with more lenient policy restrictions and compatible with the same campaign sub-objective, ensuring that the overall compliance rate rose above the threshold after the adjustment.

[0044] Then, a comprehensive risk review was conducted on the adjusted budget allocation plan to confirm that the compliance rates of all platforms met the threshold requirements. The final budget data was synchronized to the initial budget allocation framework, and a risk assessment report and adjustment records were generated to provide a secure and compliant budget basis for subsequent strategy integration.

[0045] S105: Embed the aforementioned historical campaign data and industry best practices into the rule engine; This embodiment focuses on upgrading the rule engine of a general-purpose overseas advertising intelligent system. By standardizing, transforming, and systematically embedding historical advertising data and best practices from the same industry, it solves the problems of traditional rule engines having a single logic and lacking industry adaptability. It constructs a closed-loop rule system of "data support - rule execution - dynamic optimization", providing accurate and industry-aligned core rule support for subsequent budget allocation strategy generation and integration with machine learning models.

[0046] Specifically, it extracts general-purpose historical advertising data from the system's built-in industry feature library in batches, forming a multi-dimensional data source. This includes: Platform performance data: Quantitative indicators such as time-of-day traffic volume, conversion cost range, user interaction rate, and ad approval rate of mainstream platforms such as short video advertising platforms, social advertising platforms, and search advertising platforms; Budget allocation data: Historical range of budget allocation for each platform under different campaign objectives (exposure, interaction, conversion), time period budget tilt patterns, adjustment frequency and adjustment magnitude records; Performance-related data: The correlation between different formats such as short video ads and search ads and performance metrics such as exposure and conversion rate, as well as the impact coefficient of external factors such as traffic fluctuations and policy adjustments on the performance of the campaign; Compliance risk data: Cases of violations and penalties on various platforms, data on advertising based on policy restrictions, and operational details of high-compliance advertising cases.

[0047] After all data has been deduplicated and missing values ​​filled (using the industry mean substitution method), it is stored according to the "platform-target-time period" dimension.

[0048] Next, industry best practices were compiled into an unstructured collection of experiences, which was then screened and verified by six senior experts in overseas advertising, eliminating outdated and situationally limited content. Specifically, this includes: Campaign lifecycle strategy: platform selection preferences in the early exposure accumulation stage, content format combination schemes in the mid-interaction cultivation stage, and budget allocation logic in the late conversion sprint stage; Compliance Operating Guidelines: Prohibited advertising content, qualification review requirements, solutions for violations, and alternative templates for compliant content on various platforms; Emergency adjustment mechanism: Budget transfer strategies and methods for switching ad delivery formats in case of sudden events such as a 30% or greater drop in platform traffic within 24 hours or sudden policy changes; High-performance case studies: Key operational points (such as precise time-based matching and platform combination) and scenario adaptation conditions for high-ROI campaigns in the industry.

[0049] Subsequently, a five-dimensional tagging system based on "target audience - platform type - time period - budget allocation - performance metrics" was used to standardize and label historical data, transforming discrete data into usable data modules. For example, precise data modules were constructed such as "Exposure - Short video ad platform - Weekday evening - 30%-40% - Exposure increase 25%-35%" and "Conversion - Search ad platform - Weekend afternoon - 15%-20% - Conversion rate 8%-12%". The min-max normalization method is used to normalize the data, mapping different levels of metrics (such as millions of exposures and tens of thousands of active users) to the [0,1] range, eliminating differences in data magnitude and ensuring consistency when calling rules.

[0050] Next, the unstructured best practices are transformed into "condition-action" type rule statements, and logical checks are performed to avoid conflicts and ambiguities. A specific example is as follows: Strategy-based rule: "When the traffic weight coefficient of the current time period is ≥0.8 (preset peak threshold), the budget tilt mechanism of the core sub-target adaptation platform is triggered, and the budget proportion of the platform for the time period is adjusted to the high value range of the corresponding data module." Compliance rules: "If the content being delivered contains platform restrictions (such as false advertising), a compliant content template will be automatically matched, and the highest priority of compliance rules will be triggered to suspend the original content delivery plan." Emergency rule: "When the target platform's traffic drops by more than 30% in 24 hours, the budget transfer process will be automatically initiated, transferring 10%-15% of the platform's budget to an alternative platform with traffic fluctuation ≤5% under the same sub-target."

[0051] Then, a four-level rule base architecture is built within the rule engine to ensure the accuracy and orderliness of rule invocation: Level 1 Library: Classified into “Data Rules”, “Strategy Rules”, “Compliance Rules”, and “Emergency Rules”, with compliance rules having the highest priority, followed by emergency rules, and strategy and data rules serving as the basic support. Secondary database: further subdivided into "Exposure," "Interaction," and "Conversion" categories based on campaign objectives; Level 3 Library: Further subdivided into dedicated sub-libraries based on platform type, such as short video advertising platforms and social advertising platforms; Level 4 library: Divided by time period, forming precise rule units of "target-platform-time period".

[0052] Then, the structured data modules and rule statements are imported into the corresponding rule bases to establish a two-way association mapping. For example, the "Conversion - Search Advertising Platform - Weekend Afternoon" data module is bound to the "Conversion Stage Budget Allocation Rule" to ensure that the corresponding data range is automatically retrieved when the rule is executed; compliance rules are linked to the platform policy data module to achieve dynamic rule adaptation when policies are updated. When multiple rules are triggered simultaneously, they are executed in order of priority. For rules with the same priority, the expected ROI after execution of each rule is calculated by the effect prediction model, and the optimal solution is selected.

[0053] S106: Based on the aforementioned preliminary budget allocation framework, static basic data, and dynamic real-time data, at least two types of alternative budget allocation strategies are generated by fusing machine learning models with an embedded rule engine. The types of alternative budget allocation strategies include conservative, balanced, and aggressive.

[0054] Specifically, based on the initial budget allocation framework, static basic data, and dynamic real-time data, at least two types of alternative budget allocation strategies are generated by fusing machine learning models with an embedded rule engine. These alternative budget allocation strategies include conservative, balanced, and aggressive strategies. Based on the constraints of static basic data and the preliminary budget allocation framework, the rule engine is invoked to generate a baseline strategy; By using machine learning models to extract features and recognize patterns from dynamic real-time data, predictive allocation strategies are generated. By fusing the baseline strategy output by the rules engine with the predictive strategy output by the machine learning model, at least two classes of alternative budget allocation strategies are generated.

[0055] More specifically, once the embedding of historical campaign data and industry best practices into the rules engine is complete, and a preliminary budget allocation framework has been established, the system simultaneously acquires static basic data (minimum campaign threshold, billing method, ad format rules, etc.) and dynamic real-time data (real-time traffic volume, user profile changes, conversion cost fluctuations, competitor campaign intensity, platform policy updates, etc.) from various overseas advertising platforms. It then deploys a Gradient Boosting Tree (XGBoost) machine learning model for dynamic data processing and strategy prediction. The specific steps are as follows: The system first extracts the constraints from the initial budget allocation framework, including core parameters such as total budget amount, budget period, sub-goal priority ranking, and platform compatibility. Simultaneously, it loads static basic data from the platform side, clarifying the compliance boundaries for each platform (e.g., minimum daily budget of $100, cost cap for CPC billing). Based on this information, it invokes a rule engine embedded with industry data and best practices, triggering the built-in "compliance priority - basic performance guarantee" rule logic: prioritizing ensuring each platform's budget meets the minimum deployment threshold, allocating the basic budget to each sub-goal's compatible platform according to industry norms, and referencing historical budget allocation patterns from the same industry to ensure sufficient basic deployment during peak traffic periods. Finally, a baseline strategy is generated, centered on stability and compliance. Budget allocation favors platforms and time periods with proven historical performance; for example, 60% of the basic budget is allocated to mature platforms with stable traffic and low conversion cost fluctuations, while peak-time budgets account for no less than 40% of the total budget.

[0056] Next, the system inputs the real-time collected dynamic data into the XGBoost model. First, the data is cleaned to remove abnormal traffic data caused by platform failures and short-term data distortions caused by sudden marketing activities of competitors. Then, feature engineering is performed to extract core feature dimensions, including: traffic trend features (traffic growth rate of each time period in the past 24 hours, traffic forecast value for the next 7 days), conversion potential features (deviation rate of real-time conversion cost from historical average, matching degree of high-intent user profile), competitive environment features (change rate of competitor ad intensity, bidding heat of core keywords), and policy adaptation features (matching degree of the latest platform review rules with the ad content).

[0057] The model then learns the correlation patterns between historical dynamic data and campaign performance to identify key trend signals in the current data. For example, when it detects that the real-time traffic growth rate of an emerging platform (such as a short video advertising platform) exceeds the historical average by 30%, and the high-intent user profile matching rate reaches 85%, while the intensity of competitor campaigns is low, the model determines that the platform is a high-potential campaign carrier; when it identifies that the predicted conversion cost during weekend evenings is 20% lower than the weekday average, it determines that this period is a high-value period. Based on the above identification results, a predictive allocation strategy is generated: an additional 15%-20% budget allocation is given to high-potential platforms, the budget share for high-value periods is increased to 50% of the total budget, while the basic budget share for low-potential platforms (real-time conversion cost exceeding the threshold by 20%) is reduced.

[0058] The system employs a weighted fusion mechanism of "rule-based benchmark + model prediction," setting fusion weights according to different risk preferences to generate three types of alternative strategies: conservative, balanced, and aggressive. Conservative Strategy: This strategy centers on a benchmark strategy (70% weighting), incorporating a moderate amount of low-risk elements from a predictive strategy (30% weighting). Prioritizing campaign stability and compliance, this strategy only slightly increases budget allocations to high-value periods with high validation rates within the predictive strategy, without adding additional budgets to high-potential emerging platforms. For example, maintaining a 60% budget allocation for established platforms, increasing the peak-hour budget allocation from 40% to 45% of the benchmark strategy, without adjusting the base budget for low-potential platforms, setting an overall compliance target of ≥95%, and expecting ROI fluctuations to be controlled within ±5%.

[0059] Balanced Strategy: The baseline and predictive strategies are each weighted 50%, balancing stability and potential growth. While ensuring a base budget (45%) on mature platforms and core time slots, 10%-15% of the budget is shifted to high-potential platforms, increasing the peak-hour budget to 50%, and setting a dynamic adjustment threshold (automatically reducing the budget when the platform's real-time conversion cost exceeds expectations by 10%). This strategy is expected to achieve a compliance rate of ≥90%, while reserving 20% ​​budget flexibility to capture potential opportunities, with an expected ROI improvement of 10%-15% compared to the baseline strategy.

[0060] Aggressive Strategy: This strategy centers on a predictive approach (70% weighting), with a baseline strategy serving only as a foundation (30% weighting). It significantly reduces the base budget for mature platforms to 30%, allocating 25%-30% to high-potential platforms, increasing the budget allocation for high-value periods to 55%, and relaxing constraints on some non-core compliance metrics (e.g., allowing short-term budget fluctuations of ±15% for some platforms). This strategy focuses on uncovering high-return potential, with an expected ROI improvement of over 20% compared to the baseline strategy. However, the compliance risk threshold is set at ≥85%, and an emergency adjustment mechanism is included: automatically switching to a balanced strategy if the campaign performance falls short of 20% of the predicted value.

[0061] Through the aforementioned fusion process, the system generates three types of alternative strategies that cover different risk appetites and campaign objectives, solving the problem that traditional single strategies are difficult to adapt to diverse campaign scenarios. The conservative strategy is suitable for campaigns with low risk tolerance and a pursuit of stable results; the balanced strategy is suitable for general needs that balance stability and growth; and the aggressive strategy is suitable for campaigns seeking high returns and willing to bear certain risks. Subsequently, the system can optimize based on the client's chosen strategy type, combined with subsequent threshold verification and risk verification processes, ultimately outputting an executable budget allocation plan. Simultaneously, it provides strategy performance feedback data to the dynamic iteration module, continuously optimizing the fusion accuracy of the model and rule engine.

[0062] In some embodiments, the generation of predictive allocation strategies by performing feature extraction and pattern recognition on dynamic real-time data using machine learning models includes: Dynamic real-time data is cleaned, denoised, and standardized to form standardized time-series data; Based on normalized time-series data, multi-dimensional features representing platform traffic change trends, user behavior patterns, and market competition intensity are extracted through machine learning models to obtain feature vectors; By using machine learning models to perform deep learning and pattern recognition on feature vectors, a nonlinear mapping relationship between dynamic real-time data features and budget allocation effects is established, and a predictive budget allocation scheme is generated based on the nonlinear mapping relationship.

[0063] Specifically, the system inputs the collected raw, dynamic, real-time data into the preprocessing module, which uses the quartile method to identify and remove outliers. For example, it removes data such as sudden drops in traffic caused by platform server failures or short-term spikes in conversion costs caused by competitors' large-scale marketing campaigns; and it fills in missing data using linear interpolation to ensure data integrity.

[0064] Then, to smooth out the random fluctuation noise in the traffic data, a moving average method is used to preserve the core trend characteristics of the data. For example, a 3-hour moving average is calculated for the hourly traffic data to filter out the impact of short-term random fluctuations on the data trend.

[0065] Next, the min-max normalization method is used to uniformly map dynamic indicators of different magnitudes (such as traffic volume in tens of thousands and conversion cost in US dollars) to the [0,1] interval, eliminating the interference of magnitude differences on model training, and finally forming a normalized time series dataset with a time granularity of 1 hour.

[0066] The XGBoost model then performs deep feature mining on the normalized time-series data, extracting features from three core dimensions to construct a high-dimensional feature vector: Platform traffic change trend characteristics: including traffic growth rate for each time period in the past 24 hours, traffic forecast for the next 7 days, probability distribution of peak traffic periods, traffic fluctuation coefficient (deviation rate between real-time traffic and historical average), etc., comprehensively characterizing the dynamic change pattern of platform traffic.

[0067] User behavior pattern characteristics: covering real-time user profile matching degree (overlap rate with target user profile), user dwell time change rate, click conversion rate fluctuation trend, proportion of high-intent users (those who complete the collection / add to cart action), etc., accurately capturing the dynamic changes in user behavior.

[0068] Market competition intensity characteristics: Extract characteristics such as the rate of change in competitor advertising intensity (the growth rate of competitor ad exposure in the past 24 hours), the bidding heat of core keywords, the direction of competitor budget shifts, and the overall industry advertising saturation, to reflect the dynamic changes in the market competition environment.

[0069] The three types of features are combined in time series order to form a feature vector containing 128 dimensions, providing a data foundation for subsequent pattern recognition.

[0070] Next, the model constructs a training set using historical normalized time-series data, feature vectors, and corresponding campaign performance data (such as ROI and conversion rates). Through iterative training, it learns the non-linear mapping relationship between dynamic real-time data features and budget allocation effects. For example, the model learns that when "the traffic growth rate of a certain platform exceeds 30% and the advertising intensity of competitors is 20% lower than the average," adding 15% of the budget to that platform can improve ROI by 25%.

[0071] The model then performs pattern recognition on the current input feature vectors and, combined with the established nonlinear mapping relationship, predicts the performance under different budget allocation schemes. For example, if it identifies a high-potential pattern on a short video advertising platform with "traffic growth rate of 35%, user profile matching degree of 88%, and competitor advertising intensity decreased by 22%", it predicts that adding 20% ​​of the budget to this platform can increase conversion rate by 30%; if it identifies a low-performance pattern on a search advertising platform with "conversion cost exceeding historical average by 25% and user dwell time decreasing by 18%", it predicts that maintaining the original budget will lead to a 15% decrease in ROI.

[0072] Following the performance predictions, the model generates a predictive budget allocation strategy. This strategy focuses on potential discovery and risk mitigation, with specific adjustments including: increasing the budget allocation for high-potential platforms by 15%-20% from the baseline, increasing the budget allocation for high-value periods (predicted conversion costs 20% lower than the average) to 55% of the total budget, while reducing the budget for low-performance platforms and periods by 10%-15%, and transferring the reduced budget to high-potential platforms, thus forming a predictive allocation plan that is both targeted and flexible.

[0073] Through the above process, the machine learning model achieves accurate processing and in-depth analysis of dynamic real-time data. The generated predictive budget allocation strategy can promptly capture market opportunities and mitigate potential advertising risks. This strategy lays the foundation for subsequent integration with the rule engine's baseline strategy, helping the system generate multiple alternative budget allocation schemes covering different risk preferences, thereby improving the flexibility and controllability of overseas advertising campaigns.

[0074] In some embodiments, fusing the baseline policy output by the rules engine with the predictive policy output by the machine learning model to generate at least two classes of alternative budget allocation policies includes: Acquire the user's risk preference dimension and the diversity dimension of the platform portfolio; Align and map the baseline strategy with the predictive strategy across both platform and time dimensions; The aligned baseline policy and predictive policy are weighted and fused using a fusion algorithm to generate a fused policy set. Based on the risk preference and diversity dimensions, the strategies in the fusion strategy set are evaluated and clustered in multiple dimensions to obtain the evaluation and clustering results. Based on the evaluation clustering results, at least two alternative budget allocation strategies are selected from the set of fusion strategies.

[0075] Specifically, the system obtains user-inputted risk preference parameters (such as conservative, balanced, or aggressive) and platform diversity demand index through an interactive interface. This index is derived by calculating the dispersion of budget proportions across platforms. Then, the system aligns the baseline strategy generated by the rules engine with the strategy predicted by the machine learning model in the spatiotemporal dimensions: in the time dimension, the daily granular baseline strategy from the rules engine is expanded into a weekly granular strategy based on the platform's historical traffic fluctuation curve, while the hourly predicted strategies from the machine learning model are aggregated into a daily granular strategy; in the platform dimension, a mapping matrix is ​​established to determine mandatory and optional platforms, ensuring that the strategy covers core advertising channels while also considering the potential of emerging platforms. Next, a dynamic weighted fusion algorithm is used to perform weighted calculations based on the aligned strategies using the risk preference coefficient α (range 0-1) and the diversity coefficient β (1-α), generating a fusion strategy set including the baseline strategy, predicted strategies, and intermediate strategies. Finally, through a multi-dimensional evaluation model, cluster analysis is performed using the DBSCAN density clustering algorithm based on characteristic indicators such as budget concentration, risk index, ROI volatility, and compliance rate, dividing the strategy space into conservative, balanced, and aggressive strategy clusters. Ultimately, based on the match between users' risk preferences and diverse needs, at least two alternative strategies are selected from the clustering results: for risk-averse users, a conservative, balanced strategy with a budget concentration of less than 40% and a compliance rate of more than 95% is selected; for risk-seeking users, an aggressive strategy combination with a budget concentration of more than 60% and including high-potential emerging platforms is recommended. This process realizes a human-machine collaborative strategy generation mechanism, which retains the stability of accumulated industry experience while injecting dynamic data-driven innovation.

[0076] In some embodiments, based on a general-purpose overseas advertising intelligent delivery system, a meta-learning technology module and an NSGA-II multi-objective optimization algorithm module are added to the existing framework of merging baseline and predictive strategies, forming an upgraded system with dynamic weight adaptation and multi-objective balancing. The system has completed its basic data reserves, including global regional market volatility data for the past three years (such as traffic fluctuation coefficients, competitor advertising fluctuations, and policy adjustment frequency), full-category product lifecycle stage characteristic data (advertising behavior and effect correlation data for the introduction, growth, maturity, and decline stages), and multi-objective KPI quantitative data (brand awareness monitoring indicators, user loyalty scores, market share statistics, etc.). Simultaneously, the system connects to real-time data interfaces of mainstream platforms such as short video advertising platforms, social advertising platforms, and search advertising platforms, and has the ability to update core data every 15 minutes, providing ample data support for meta-learning model training and multi-objective optimization calculations.

[0077] Among them, the dynamic weight adaptive mechanism takes meta-learning technology as its core. By learning the correlation between historical market scenarios and weight allocation, it realizes the automatic and scenario-based adjustment of the fusion ratio between the rule engine and the machine learning model, breaking the limitations of the original fixed weight.

[0078] Specifically, the system first constructs a core model based on MAML (Model-Independent Meta-Learning). This model takes market scenario characteristics and product lifecycle characteristics as inputs and outputs optimal fusion weights. Through multiple rounds of meta-training and fine-tuning, it possesses the ability to quickly adapt weight prediction to new scenarios. At the feature selection level, market scenario characteristics cover three dimensions: a comprehensive market volatility index (calculated by weighting traffic volatility coefficient, competitor placement volatility, and policy adjustment frequency, with a value range of 0-1, where 0 represents extremely low volatility and 1 represents extremely high volatility), regional market competition intensity, and platform policy stability. Product lifecycle characteristics are determined by three indicators: user growth rate, market penetration rate, and repurchase rate, specifying the specific stage (Introduction stage: user growth rate ≥ 30%, market penetration rate < 10%; Growth stage: user growth rate 10%-30%, market penetration rate 10%-30%; Maturity stage: user growth rate < 10%, market penetration rate 30%-60%; Decline stage: user growth rate < 0%, market penetration rate ≥ 60%).

[0079] The training data employs a three-dimensional dataset of scenario-weights-effects, covering deployment cases across different market scenarios and product stages over the past three years. 80% of the dataset serves as the meta-training set, and 20% as the validation set. For example, for scenarios with market volatility of 0.2 (low volatility) and product maturity, the historically optimal fusion weights were 0.7 for the rule engine and 0.3 for the machine learning model, resulting in an 18% increase in ROI compared to fixed weights. For scenarios with market volatility of 0.8 (high volatility) and product growth, the historically optimal fusion weights were 0.3 for the rule engine and 0.7 for the machine learning model, resulting in a 25% increase in market share growth rate. The model learns the weight adjustment patterns under different scenarios through meta-training and then fine-tunes using the validation set, ultimately achieving an optimal weight prediction accuracy of over 85%.

[0080] After model deployment, the system collects market scenario characteristics and product lifecycle characteristics data in real time, automatically inputs them into the meta-learning model to generate the optimal fusion weights for the current scenario, and sets up a weight re-evaluation mechanism every 24 hours to dynamically optimize based on real-time deployment results. A specific scenario execution example is as follows: In the North American market and mature 3C product scenarios, the system monitored a comprehensive market volatility index of 0.25 (low volatility), a product user growth rate of 5%, and a market penetration rate of 45% (mature stage characteristics). The meta-learning model output fusion weights: rule engine 0.65, machine learning model 0.35. This weight allocation primarily relies on the mature delivery logic of the rule engine to ensure delivery stability, while appropriately incorporating real-time adjustment suggestions from the machine learning model to address potential minor market changes. In actual implementation, the rule engine dominates the budget allocation for core time slots on mature platforms such as search advertising platforms and e-commerce advertising platforms, while the machine learning model fine-tunes the proportion of information flow ads on social advertising platforms based on real-time conversion data. Ultimately, this achieves conversion cost fluctuation control within ±3%, reducing costs by 7% compared to a fixed-weight strategy.

[0081] In the Southeast Asian market and the growth stage of beauty products, the overall market volatility index was 0.75 (high volatility), the product user growth rate was 25%, and the market penetration rate was 15% (characteristics of the growth stage). The meta-learning model output fusion weights were: rule engine 0.3, machine learning model 0.7. This weighting emphasizes the machine learning model's ability to capture dynamic markets, while the rule engine only ensures basic compliance and the minimum deployment threshold. In practice, the machine learning model dominates the allocation of short video advertising budgets on short video advertising platforms, dynamically adjusting the deployment intensity based on real-time traffic peaks (such as local evening 8-10 pm), while also exploring incremental opportunities on emerging platforms; the rule engine strictly controls the budgets of each platform to ensure they do not fall below the minimum threshold, ensuring a compliance rate of ≥90%. After implementing this strategy, the market share increased by 12% monthly, a 5 percentage point increase compared to the fixed-weight strategy.

[0082] In addition, the system sets weight adjustment constraints, with the rule engine weight not less than 0.3 (to ensure basic compliance and stability) and the machine learning model weight not less than 0.2 (to ensure basic responsiveness to market dynamics), avoiding extreme risks caused by a single strategy dominating.

[0083] Next, the multi-objective collaborative optimization mechanism introduces NSGA-II (Non-dominated sorting genetic algorithm II), breaking the original single-objective optimization logic centered on ROI and exposure. It seeks the Pareto optimal solution among the three competitive objectives of brand awareness, user loyalty, and market share, achieving a balance between long-term value and short-term effects.

[0084] Specifically, the system first clarifies the quantitative indicators and weight allocation for three core objectives. Based on the core needs of advertising, the system sets the target weights as follows: market share (short-term growth target, weight 40%), brand awareness (medium-to-long-term building target, weight 30%), and user loyalty (long-term value target, weight 30%). The quantitative indicators for each objective are as follows: Market share: The percentage of industry advertising conversions obtained through third-party data platforms, updated weekly; Brand awareness: A weighted score of platform ad impressions, social media brand mentions, and search index growth rate, with a value range of 0-100; User loyalty: a weighted score of repurchase rate, user retention rate, and brand member registrations, with a value range of 0-100.

[0085] At the same time, set target constraints: monthly market share growth rate of no less than 5%, brand awareness score of no less than 60, and user loyalty score of no less than 55, to ensure that each target meets the basic standards and avoid neglecting one aspect for another.

[0086] The system then uses the NSGA-II algorithm to achieve multi-objective collaborative optimization, as follows: Based on the fusion strategy generated by the dynamic weight adaptive mechanism, 50 different platform budget allocation ratios, ad format combinations, and time period delivery plans are randomly generated to form the initial population. Each plan meets the total budget constraint and platform delivery threshold requirements.

[0087] Next, for each set of solutions, based on historical data and machine learning model predictions, the predicted scores for three objectives—market share, brand awareness, and user loyalty—are calculated. The overall fitness value is then calculated by combining the objective weights, and the constraints are checked to eliminate substandard solutions.

[0088] Then, the remaining options are non-dominated and ranked. Options that do not have a situation where "all objective scores of one option are better than those of another option" are grouped into the same non-dominated layer, forming the Pareto front set. For example, option A has a market share growth rate of 8%, brand awareness of 65, and user loyalty of 58; option B has a market share growth rate of 6%, brand awareness of 70, and user loyalty of 62. Both belong to the same non-dominated layer, but focus on different objectives.

[0089] Subsequently, the crowding degree of schemes in the same non-dominated layer is calculated to measure the degree of dispersion of schemes in the target space. Schemes with high crowding degree are preferentially retained to ensure the diversity of the Pareto front set.

[0090] Then, the roulette wheel selection method is used to select the solutions with high fitness and high crowding. Crossover (such as swapping the platform budget allocation ratio of two sets of solutions) and mutation (such as randomly adjusting the combination of advertising formats on a certain platform) are performed to generate the next generation population. This process is repeated 50 times to finally obtain a stable Pareto optimal solution set, which contains 10 differentiated optimization solutions.

[0091] Next, from the Pareto optimal solution set, and considering the user's target preferences, three core solutions are selected: Growth-First Plan: Focusing on increasing market share, the target weights are adjusted to 50% market share, 25% brand awareness, and 25% user loyalty. This plan allocates 40% of the budget to traffic-generating platforms such as short video advertising platforms, targeting young users and expanding market coverage through short video ads. The expected monthly market share growth rate is 10%, brand awareness is 62, and user loyalty is 56, which is suitable for the incremental demand during the product's growth stage.

[0092] Balanced Development Plan: Balancing the three objectives while maintaining the initial weighting. The plan adopts a combination of "mature platforms + emerging platforms," ​​with the budget allocated as follows: 30% for search advertising platforms, 25% for short video advertising platforms, 20% for social advertising platforms, 15% for e-commerce advertising platforms, and 10% for in-site advertising platforms. This plan balances short-term conversion with long-term brand building, with an expected monthly market share growth rate of 6%, brand awareness of 68%, and user loyalty of 60%, adapting to the stable needs of a mature product.

[0093] Value-driven solution: Focusing on improving user loyalty, the target weights are adjusted to 25% market share, 30% brand awareness, and 45% user loyalty. This solution increases the allocation of budget to member-exclusive ads on social advertising platforms and repurchase incentive ads on e-commerce advertising platforms, reaching 35% of the budget. The expected monthly market share growth rate is 4%, brand awareness is 72%, and user loyalty is 65%, aligning with the user value extraction needs of a mature product.

[0094] By combining the optimal weights generated by the dynamic weight adaptation mechanism with the Pareto optimal solution obtained from multi-objective collaborative optimization, a final strategy set is formed. The effectiveness is ensured through a process of small-scale testing, full deployment, and real-time monitoring. The system first conducts a 3-day small-scale test with 10% of the budget, comparing the actual effects of different solutions with predicted values, and eliminating solutions with a deviation rate exceeding 15%. After passing the test, the system is fully deployed, updating core data every 15 minutes to monitor the progress of each objective. If an objective fails to meet expectations (e.g., market share growth rate is below 3%), the meta-learning model is automatically triggered to re-evaluate the weights, and the NSGA-II algorithm is invoked to generate an adjustment plan, ensuring the dynamic adaptability of the strategy.

[0095] The dynamic weight adaptive mechanism, through meta-learning technology, enables scenario-based and automated adjustment of integrated weights, solving the problem that the original fixed weights could not adapt to complex market changes. This significantly improves the strategy's adaptability to market fluctuations and product lifecycles. The multi-objective collaborative optimization mechanism, through the NSGA-II algorithm, breaks through the limitations of single-objective optimization, achieving a scientific balance between market share, brand awareness, and user loyalty. This helps companies balance short-term growth and long-term value, increasing user lifetime value by 30% compared to a single ROI optimization strategy. The integration of these two mechanisms further improves the strategy system for intelligent overseas advertising, significantly enhancing the scientific nature, flexibility, and sustainability of advertising, and providing a more comprehensive solution for advertising at different stages and with different needs.

[0096] After strategy integration, the system automatically generates an intuitive visual output interface, providing users with clear decision support. The strategy charts include pie charts showing the budget distribution across platforms, clearly illustrating the allocation of the total budget to search advertising platforms, social advertising platforms, and short video advertising platforms; line charts dynamically present budget trends over different time periods within the budget cycle, helping users understand the correlation between traffic fluctuations and budget adjustments; and bar charts horizontally compare the expected performance differences of conservative, balanced, and aggressive strategies in key metrics such as ROI, impressions, and conversions, making the strengths and weaknesses of each strategy immediately apparent. The detailed tables list a structured mapping relationship of "Platform - Ad Format - Budget Amount - Expected KPI - Risk Level." For example, a search ad allocation of $6,000 on a search advertising platform has an expected ROI of 4.2, 150,000 impressions, and a medium risk level; a feed ad allocation of $4,000 on a social advertising platform has an expected conversion rate of 2,000 and a low risk level. Each strategy comes with a detailed explanation of its generation logic, such as "The conservative strategy, based on the attributes of the 3C industry and historical data from the North American market, prioritizes ensuring stable delivery of search ads on the search advertising platform." It also provides dynamic adjustment suggestions, such as "If the conversion cost on the short video advertising platform decreases by 15% compared to expectations, the budget can be increased by 5% to optimize ROI." These outputs not only ensure the interpretability of the strategies but also guide users to flexibly optimize during the campaign process through real-time data correlation, significantly improving the scientific nature and operational efficiency of budget allocation.

[0097] In one specific embodiment, a complete closed-loop monitoring and optimization system was constructed. This system automatically captures real-time campaign data from various advertising platforms hourly via API interfaces, including key metrics such as impressions, click-through rate, conversion cost, and ROI, and accurately compares these actual results with the expected KPIs set during strategy generation. When the system detects that the actual conversion cost on a platform consistently exceeds the expected target by 20%, or the impressions are below the expected value by 30%, it automatically triggers a multi-level early warning mechanism, promptly alerting relevant personnel through various methods such as dashboard highlighting, email notifications, and mobile push notifications.

[0098] Once an alert is triggered, the system immediately initiates an automatic adjustment process. Based on preset business rules, if conversion costs exceed expectations, the platform's budget is reduced by 10%, and the freed-up budget is transferred to platforms that have achieved the desired performance. The system then generates an initial adjustment plan. For complex situations, the system re-invokes the machine learning model, combining the latest market data and campaign performance to generate a more precise adjustment strategy. All these adjustment suggestions are accompanied by detailed modification instructions and expected performance assessments, and are automatically pushed to the client or operations team for review and confirmation.

[0099] More importantly, the system has established a comprehensive experience accumulation mechanism. At the end of each month, the system automatically selects high-performing strategy cases, such as successful strategies with a ROI consistently exceeding the industry average by 20%, and saves their core parameters and applicable conditions to the industry feature library. Simultaneously, the system identifies and records newly discovered campaign patterns, such as conversion peaks on short video advertising platforms in Southeast Asia during specific holidays, or the optimal campaign times for certain product types on specific platforms. This validated knowledge is systematically integrated into the rules engine, transforming it into reusable business rules to continuously optimize subsequent strategy generation logic, forming a self-learning, continuously evolving intelligent decision-making system.

[0100] This embodiment improves budget allocation efficiency and scalability through the aforementioned methods, reducing strategy generation time from "1-2 days" in existing technologies to "1-2 hours," supporting simultaneous service for 50+ clients, and increasing operational efficiency by over 80%. It eliminates the need for manual budget breakdown across platforms, reducing repetitive operations by over 70% and lowering the workload of operations staff. Simultaneously, multi-dimensional data fusion (client goals + platform dynamics + industry experience) improves the match between strategy and actual needs by 60%. Testing shows that clients using this tool have seen an average increase of 25%-40% in overseas advertising ROI and an average reduction of 15%-20% in conversion costs. Multiple strategy options cater to different risk appetites; for example, conservative strategies can control campaign performance fluctuations within 10%, while aggressive strategies can capture high-potential opportunities on new platforms.

[0101] Please see Figure 2 This embodiment provides a budget allocation strategy generation device 200, including: The data collection unit 201 is used to collect customer-side data, platform-side data, and industry-side data. The customer-side data includes total budget amount, budget period, target type, and the industry to which the company belongs. The platform-side data includes static basic data and dynamic real-time data. The industry-side data includes historical campaign data from the same industry and industry best practices. Budget breakdown unit 202 is used to break down the total budget amount into different time periods according to the budget cycle to obtain budget breakdown results; The target decomposition unit 203 is used to decompose the target according to the target type and the industry to which the enterprise belongs, and obtain the target decomposition result; The combination unit 204 is used to combine the budget breakdown results and the target breakdown results to obtain a preliminary budget allocation framework; Embedding unit 205 is used to embed the historical deployment data of the same industry and industry best practices into the rule engine; The generation unit 206 is used to generate at least two types of alternative budget allocation strategies based on the preliminary budget allocation framework, static basic data and dynamic real-time data by using a machine learning model and an embedded rule engine. The types of alternative budget allocation strategies include conservative, balanced and aggressive.

[0102] Furthermore, the budget breakdown unit 202 includes: The data acquisition subunit is used to acquire the total budget amount, budget period, platform traffic fluctuation characteristics, platform historical traffic data, and industry-specific time period traffic distribution data from the industry feature library; The time period division subunit is used to divide the budget period into multiple consecutive time periods based on the length of the budget period and the platform traffic fluctuation characteristics. The weight calculation subunit is used to calculate the traffic weight coefficient for each time period based on the platform's historical traffic data and the same industry time period traffic distribution data in the industry feature library; The proportional allocation subunit is used to allocate the total budget amount according to the proportion of the traffic weight coefficient to obtain the budget breakdown result.

[0103] Furthermore, the weight calculation subunit includes: The standardization processing subunit is used to standardize the historical traffic data of the platform, converting different levels of exposure, clicks and active users into standardized values. The outlier removal subunit is used to remove outliers from the same industry's time-period traffic distribution data. The weighted fusion subunit is used to determine the fusion weight of the standardized value and the excluded same-industry time-period traffic distribution data using the analytic hierarchy process. The weighted summation subunit is used to perform a weighted summation of the standardized value and the excluded traffic distribution data of the same industry time period based on the fusion weight, so as to obtain the traffic weight coefficient of each time period.

[0104] Furthermore, the target disassembly unit 203 includes: The type decomposition subunit is used to decompose the target type in multiple dimensions and clarify the attribute characteristics of each sub-target; The pattern decomposition subunit is used to extract common patterns of target decomposition in the same type of target under the same industry based on the industry to which the enterprise belongs; The association establishes sub-units, which are used to combine the attribute characteristics of the sub-targets with the common rules of the target decomposition to establish an adaptation association between the sub-targets and the multimedia advertising platform; The relationship integration subunit is used to integrate the adaptation relationships between each sub-target and the corresponding platform into the target decomposition result.

[0105] Furthermore, the generation unit 206 includes: The rule engine calling subunit is used to call the rule engine to generate a baseline strategy based on the constraints of static basic data and the preliminary budget allocation framework; The strategy generation subunit is used to extract features and recognize patterns from the dynamic real-time data using a machine learning model to generate a predictive allocation strategy. The strategy fusion subunit is used to fuse the baseline strategy output by the rule engine with the predictive strategy output by the machine learning model to generate at least two alternative budget allocation strategies.

[0106] Furthermore, the strategy generation subunit includes: The data processing subunit is used to clean, denoise, and standardize the dynamic real-time data to form standardized time-series data. The feature extraction subunit is used to extract multi-dimensional features representing the platform's traffic change trend, user behavior patterns, and market competition intensity based on the normalized time-series data and through the machine learning model, to obtain a feature vector; The pattern recognition subunit is used to perform deep learning and pattern recognition on the feature vector using the machine learning model, establish a nonlinear mapping relationship between dynamic real-time data features and budget allocation effects, and generate a predictive budget allocation scheme based on the nonlinear mapping relationship.

[0107] Furthermore, the strategy fusion subunit includes: The dimension acquisition subunit is used to acquire the user's risk preference dimension and the diversity dimension of the platform combination; The alignment subunit is used to align and map the baseline policy and the predictive policy in the platform and time dimensions. The weighted fusion subunit is used to perform weighted fusion of the aligned baseline policy and predictive policy using a fusion algorithm to generate a fusion policy set; The feature clustering subunit is used to perform multi-dimensional evaluation and feature clustering on each strategy in the fusion strategy set based on the risk preference dimension and the diversity dimension, so as to obtain the evaluation clustering result; The screening subunit is used to select at least two alternative budget allocation strategies from the set of fusion strategies based on the evaluation clustering results.

[0108] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the above-described apparatus and unit can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0109] The present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed, can implement the methods provided in the above embodiments. The storage medium may include various media capable of storing program code, such as a USB flash drive, a portable hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0110] The present invention also provides a computer device, which may include a memory and a processor. The memory stores a computer program, and when the processor calls the computer program in the memory, it can implement the methods provided in the above embodiments. Of course, the computer device may also include various network interfaces, power supplies, and other components.

[0111] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since it corresponds to the method disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to in the method section. It should be noted that those skilled in the art can make various improvements and modifications to this invention without departing from its principles, and these improvements and modifications also fall within the protection scope of the claims of this invention.

[0112] It should also be noted that, in this specification, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusivity.

[0113] The term "comprises" implies that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprises a..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

Claims

1. A method for generating a budget allocation strategy, characterized in that, The method comprises the following steps: Collecting customer-side data, platform-side data and industry-side data; wherein, the customer-side data comprises total budget amount, budget period, target type of delivery and industry to which the enterprise belongs; the platform-side data comprises static basic data and dynamic real-time data; the industry-side data comprises historical delivery data of the same industry and industry best practices; According to the budget period, the total budget amount is disassembled according to each time period to obtain a budget disassembly result; According to the target type of delivery and the industry to which the enterprise belongs, target disassembly is performed to obtain a target disassembly result; The budget disassembly result and the target disassembly result are combined to obtain a preliminary budget allocation framework; The historical delivery data of the same industry and the industry best practices are embedded into a rule engine; Based on the preliminary budget allocation framework, the static basic data and the dynamic real-time data, at least two types of alternative budget allocation strategies are generated by adopting a machine learning model and a rule engine embedded in a fusion manner, and the types of the alternative budget allocation strategies comprise conservative type, balanced type and aggressive type.

2. The budget allocation policy generation method of claim 1, wherein, The budget disassembly result comprises the following steps: Obtaining the total budget amount, the budget period, platform traffic fluctuation characteristics, platform historical traffic data and the same industry time period traffic distribution data in the industry characteristic library input by the customer; According to the length of the budget period and the platform traffic fluctuation characteristics, the budget period is divided into a plurality of continuous time periods; Based on the platform historical traffic data and the same industry time period traffic distribution data in the industry characteristic library, traffic weight coefficients of each time period are calculated; The total budget amount is allocated according to the proportion of the traffic weight coefficients to obtain a budget disassembly result.

3. The budget allocation policy generation method of claim 2, wherein, The traffic weight coefficients of each time period are calculated based on the platform historical traffic data and the same industry time period traffic distribution data in the industry characteristic library, which comprises the following steps: Standardizing the platform historical traffic data to convert exposure, click and active user numbers of different magnitudes into standardized values; Performing outlier rejection on the same industry time period traffic distribution data; Determining the fusion weight of the standardized values and the same industry time period traffic distribution data after outlier rejection by using the analytic hierarchy process; Based on the fusion weight, the standardized values and the same industry time period traffic distribution data after outlier rejection are weighted and summed to obtain the traffic weight coefficients of each time period.

4. The budget allocation policy generation method of claim 1, wherein, The target disassembly result comprises the following steps: Multi-dimensional disassembly is performed on the target type of delivery to clearly define the attribute characteristics of each sub-target; Based on the industry to which the enterprise belongs, the target disassembly commonality rule of the enterprise under the same type of delivery is extracted from the industry-side data; The attribute characteristics of the sub-targets and the target disassembly commonality rule are combined to establish the adaptive association between the sub-targets and the multimedia advertising platform; The adaptive association relationship between each sub-target and the corresponding platform is integrated into a target disassembly result.

5. The budget allocation policy generation method of claim 1, wherein, The method comprises the following steps: generating at least two types of alternative budget allocation strategies based on the preliminary budget allocation framework, static basic data and dynamic real-time data, wherein the types of the alternative budget allocation strategies include conservative, balanced and aggressive; generating a benchmark strategy by calling the rule engine based on the constraint conditions of the static basic data and the preliminary budget allocation framework; extracting features and identifying patterns of the dynamic real-time data by the machine learning model to generate a predictive allocation strategy; 6. The budget allocation policy generation method of claim 5, wherein, fusing the benchmark strategy output by the rule engine and the predictive strategy output by the machine learning model to generate at least two types of alternative budget allocation strategies. The method of extracting features and identifying patterns of the dynamic real-time data by the machine learning model to generate a predictive allocation strategy comprises: cleaning, denoising and standardizing the dynamic real-time data to form normalized time series data; extracting multi-dimensional features representing platform traffic trends, user behavior patterns and market competition intensity from the normalized time series data by the machine learning model to obtain a feature vector; 7. The budget allocation policy generation method of claim 5, wherein, performing deep learning and pattern recognition on the feature vector by the machine learning model to establish a nonlinear mapping relationship between dynamic real-time data features and budget allocation effects, and generating a predictive budget allocation scheme based on the nonlinear mapping relationship. The method of fusing the benchmark strategy output by the rule engine and the predictive strategy output by the machine learning model to generate at least two types of alternative budget allocation strategies comprises: obtaining the risk preference dimension of the user and the diversity dimension of the platform combination; aligning and mapping the benchmark strategy and the predictive strategy in the platform dimension and the time dimension; performing weighted fusion of the aligned benchmark strategy and predictive strategy by a fusion algorithm to generate a fusion strategy set; performing multi-dimensional evaluation and feature clustering on each strategy in the fusion strategy set based on the risk preference dimension and the diversity dimension to obtain an evaluation clustering result; 8. A budget allocation strategy generation apparatus characterized by comprising: selecting at least two types of alternative budget allocation strategies from the fusion strategy set according to the evaluation clustering result. The method comprises the following steps: a collection unit for collecting customer-side data, platform-side data and industry-side data; wherein the customer-side data includes total budget amount, budget period, target type and industry of the enterprise; the platform-side data includes static basic data and dynamic real-time data; the industry-side data includes historical advertising data of the same industry and industry best practices; a budget disassembly unit for disassembling the total budget amount according to each time period to obtain a budget disassembly result; a target disassembly unit for target disassembly according to the target type and the industry of the enterprise to obtain a target disassembly result; a combination unit for combining the budget disassembly result and the target disassembly result to obtain a preliminary budget allocation framework; an embedding unit for embedding the historical advertising data of the same industry and the industry best practices into a rule engine. The generating unit is configured to generate at least two types of candidate budget allocation strategies based on the preliminary budget allocation framework, static basic data and dynamic real-time data, by adopting a machine learning model and an embedded rule engine in a fusion manner, wherein the types of the candidate budget allocation strategies include a conservative type, a balanced type and an aggressive type.

9. A computer device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, The processor implements the budget allocation strategy generation method according to any one of claims 1 to 7 when executing the computer program.

10. A computer-readable storage medium, characterized in that, The computer readable storage medium stores a computer program, and the computer program causes the processor to execute the budget allocation strategy generation method according to any one of claims 1 to 7 when the computer program is executed by the processor.

Images