Closed-loop detection and autonomous repair method and system for data quality of data-in-data station

Abstract

The invention discloses a closed-loop detection and autonomous repair method and system for data quality in data, and relates to the technical field of closed-loop detection and repair, and the method comprises the steps: building a user portrait through collecting multi-source data, building an objective function based on access and purchase data, comparing the objective function value with a preset threshold value, and obtaining a data quality closed-loop detection result. Judging the accuracy of the user portrait; if the user portrait reaches the standard, executing precise advertisement putting; if the access time does not reach the standard, screening the high-intention users according to the access time to form a coarse screening portrait, performing optimization calculation on the coarse screening portrait by adopting a simulated fire raising algorithm, updating an objective function, and performing re-evaluation to form a closed-loop optimization system. According to the method, the user portraits are dynamically optimized through a closed-loop system, efficient optimization is realized in combination with a simulated fire raising algorithm, the calculation efficiency is improved by adopting layered screening, and the conversion, the scale and the cost are balanced by utilizing a multi-dimensional objective function, so that the marketing return rate is remarkably improved while the putting effect is ensured.

Description

Technical Field

[0001] This invention relates to the field of closed-loop detection and repair technology, specifically to a method and system for closed-loop detection and autonomous repair of data quality in a data platform. Background Technology

[0002] In the e-commerce sector, increased sales rely on efficient traffic conversion, which means proactively acquiring targeted traffic through advertising to improve conversion rates and ensure that every advertising investment is converted into actual orders to the maximum extent possible, ultimately driving sustained sales growth.

[0003] For example, the invention patent with publication number CN112783681B discloses a method and device for multi-level closed-loop repair of self-service big data in power enterprises. This method includes: data collection at time intervals; anomaly identification using a threshold method; and partitioned repair of the captured anomaly data. This transforms the anomaly data repair process into a multi-level closed-loop process, enabling the repair of various types of data. The beneficial effects of this invention are: optimizing the anomaly data identification module, setting data calculation thresholds, implementing anomaly data identification based on a smoothing estimation threshold algorithm, setting repair levels according to data categories during data repair based on the data partitioning calculation process, and combining traditional data repair techniques to complete the multi-level closed-loop data repair process, thereby improving data repair integrity and data carrying capacity.

[0004] For example, the invention patent with publication number CN111426896A discloses a scalable data flow closed-loop test platform and method. This test platform is based on a scalable CPCI bus architecture and adopts RS422 bus technology. It includes a power module, a computer module and a rear output board, and N interface plug-ins and rear output boards mounted on a 3U backplane. This invention realizes automated closed-loop testing of power controller data flow, improves the efficiency of fault detection during the R&D process, and provides an effective guarantee for reducing the mean time to repair (MTBT) of radar during delivery. Furthermore, it does not require replacing the power module and computer module, supports 1 to N interface plug-ins and rear output boards, and can simultaneously test 1 to M power controllers, demonstrating strong scalability. The interface signal type of the rear output board can be transmitted in various forms such as RS232, RS422, LVDS, and fiber optics as needed. Only the interface type of the rear output board needs to be changed to meet the testing requirements of different power controller models, demonstrating strong versatility.

[0005] However, due to the distortion of basic information such as user profiles and interest tags that e-commerce operations rely on, the accuracy of advertising has decreased and the user conversion rate has continued to decline. In order to make up for this shortcoming, e-commerce operations have resorted to the crude method of simply increasing the budget and expanding the exposure to forcibly obtain conversions. This has resulted in the cost invested being disproportionate to the conversions obtained, causing a large amount of resources to be wasted on non-target groups, which in turn leads to uncontrolled marketing costs and increasingly worse advertising results. Summary of the Invention

[0006] Technical problems to be solved To address the shortcomings of existing technologies, this invention provides a data platform data quality closed-loop detection and autonomous repair method and system, which solves the problem of decreased advertising accuracy caused by distorted user profiles, and the vicious cycle of resource waste and cost overrun caused by extensive budget increases in an attempt to compensate for insufficient conversion.

[0007] To achieve the above objectives, this invention provides the following technical solution: a data platform data quality closed-loop detection and autonomous repair method, comprising the following specific steps: Step 1: Collect multi-source data to construct a user profile, including user age, user occupation, user region, and user shopping time period; obtain access data through front-end data tracking points, including the number of visitors, the number of purchases, and the actual access duration; Step 2: Preset the campaign cost, comprehensively analyze and standardize the number of purchases, visitors, and campaign costs, and construct an objective function; Step 3: Compare the output value of the objective function with... The target conversion threshold is compared: if the output value is greater than or equal to the threshold, proceed to step four; otherwise, proceed to step five; Step four: deliver the advertisement to devices that match the user profile, and end or return to step one; Step five: compare the actual access time corresponding to each visitor with the preset ideal access time: if the actual access time is greater than or equal to the ideal access time, record this user profile, obtain the coarse-screened user profile, and proceed to step six; otherwise, continue the comparison; Step six: perform a comprehensive calculation on the objective function corresponding to the coarse-screened user profile using the simulated ignition algorithm, output the objective function, and return to step three.

[0008] Furthermore, the specific method for constructing the objective function is as follows: ;in, Describe the objective function. Indicates the number of buyers. Indicates the number of visitors. Indicates the cost of deployment. and All of them represent positive real numbers.

[0009] Furthermore, the specific method for comprehensively calculating the objective function corresponding to the coarse-screened user profile using the simulated heating algorithm is as follows: Define an initial temperature and a final high temperature, and set a threshold for the number of single-round screenings and a probability formula. Starting from the initial temperature of the first round, the temperature is increased, and in each round, user age, user occupation, user region, and user shopping time period are randomly selected from the coarse-screened user profile to obtain a new objective function. The new objective function is calculated with the objective function to obtain the target value. The objective function for this round is determined based on the target value. This process continues until the number of random selections exceeds the threshold for the number of single-round screenings, at which point the next round is performed. This process continues in each subsequent round, with the temperature increasing until the final high temperature is reached, at which point the final objective function is determined.

[0010] Furthermore, the specific method for obtaining the target value is as follows: calculate the difference between the new objective function and the original objective function to obtain the target value.

[0011] Furthermore, the specific method for determining the objective function based on the target value is as follows: compare the target value with zero. If the target value is greater than or equal to zero, then update the new objective function to the next objective function. If the target value is less than zero, then determine according to the probability formula. If the probability is accepted, then update the new objective function to the next objective function. If the probability is not accepted, then update the original objective function to the next objective function.

[0012] Furthermore, the specific method for obtaining the heating temperature is as follows: [The text abruptly ends here, likely due to an incomplete sentence or a formatting error.] The temperature of each heat cycle is calculated by adding the temperature of the first heat cycle to the number of heat cycles, thus obtaining the result of the first heat cycle. The temperature of the wheel rises.

[0013] Furthermore, the probability formula is set as follows: ;in, Represents probability. Indicates the target value. Indicates the first The temperature of the wheel rises.

[0014] Furthermore, the specific method for determining the probability formula is as follows: the range from 0 to the probability value is set as the unacceptable probability range, and the range from the probability value to 1 is set as the acceptable probability range. A random function is set through code, with a range from 0 to 1. If the function value of the random function is in the acceptable probability range, it indicates that the probability is accepted; if the function value of the random function is in the unacceptable probability range, it indicates that the probability is not accepted.

[0015] A data platform data quality closed-loop detection and autonomous repair system includes the following specific modules: a data acquisition module, an objective function construction module, a target analysis module, a precise delivery module, a coarse-screen user profile module, and an optimal objective function analysis module. The data acquisition module collects multi-source data to construct user profiles, including user age, occupation, location, and shopping time period; it also obtains access data through front-end data tracking, including the number of visitors, the number of purchasers, and the actual access duration. The objective function construction module presets the delivery cost, comprehensively analyzes and standardizes the number of purchasers, visitors, and delivery cost, and constructs the objective function. The target analysis module outputs the value of the objective function... The system compares the output value with a preset target conversion threshold: if the output value is greater than or equal to the threshold, the precise delivery module is executed; otherwise, the coarse user profile screening module is executed. The precise delivery module delivers the advertisement to devices that match the user profile and then ends or returns to the data collection module. The coarse user profile screening module compares the actual access duration corresponding to each visitor with the preset ideal access duration: if the actual access duration is greater than or equal to the ideal access duration, the user profile is recorded to obtain the coarse user profile, and the optimal objective function analysis module is executed; otherwise, the comparison continues. The optimal objective function analysis module performs a comprehensive calculation on the objective function corresponding to the coarse user profile using a simulated ignition algorithm, outputs the objective function, and returns it to the target analysis module.

[0016] Beneficial effects Compared with the prior art, the embodiments of the present invention have at least the following advantages or beneficial effects: 1. By constructing a closed-loop detection system with an objective function as the evaluation standard: when the objective function value does not meet the standard, the system does not directly carry out extensive advertising, but automatically triggers a repair process. First, it quickly filters out high-intent users based on actual access time to form a coarse user profile, and then performs refined calculations through subsequent algorithms. This autonomous closed loop of evaluation, filtering and optimization enables the user profile to continuously self-correct based on real-time feedback data, ensuring that it always truly reflects the target customer group, thereby significantly improving the accuracy and conversion efficiency of advertising.

[0017] 2. By iteratively increasing the temperature from low to high, this reverse temperature control strategy, combined with the gradually increasing probability of accepting different solutions with each round, allows the algorithm to quickly lock onto promising search regions in the early stages, while in the later stages, it can escape local optima with a higher probability of acceptance and conduct a wider exploration. This mechanism ensures that the algorithm has the efficient convergence of a greedy algorithm in the early stages, while retaining the ability to escape local optima in the later stages. Therefore, in the complex optimization problem of advertising placement with multiple variables and nonlinearity, it is more likely to find the globally optimal or near-optimal user profile configuration and maximize the placement effect.

[0018] 3. Faced with massive amounts of user data, direct global optimization is computationally complex and slow. This invention designs a two-stage processing flow: First, it uses the strong intention indicator of actual access duration to quickly and computationally cheaply screen the original user profile, effectively narrowing down the scope of the dataset to be optimized. On this basis, it applies a computationally intensive simulated ignition algorithm to the scaled-down, coarsely screened user profile for deep optimization. This layered strategy, while ensuring final accuracy, cleverly concentrates computing resources on the most promising subset of data, avoiding unnecessary computational overhead on a large amount of invalid data. This significantly improves the overall response speed and operating efficiency of the system, making this technical solution highly practical in real high-concurrency business scenarios.

[0019] 4. The objective function of the invention does not solely pursue conversion rate or click volume, but combines purchase scale, conversion rate, and campaign cost for calculation. This allows the system to automatically avoid one-sided strategies such as high conversion but uncontrollable costs or low cost but no conversion. It always optimizes towards the comprehensive optimal direction of maximizing conversion under controllable costs. Therefore, the system outputs not only an accurate user profile, but also a mathematically verified, high-quality campaign strategy that balances effectiveness and cost, fundamentally ensuring the continuous optimization of marketing investment return rate.

[0020] Of course, any product implementing this invention does not necessarily need to achieve all of the advantages described above at the same time. Attached Figure Description

[0021] Figure 1 This invention provides a flowchart of a data quality closed-loop detection and autonomous repair method for a data middle platform.

[0022] Figure 2 This invention relates to a structural diagram of a data platform data quality closed-loop detection and autonomous repair system. Detailed Implementation

[0023] 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 embodiments of the present invention, and not all embodiments. 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.

[0024] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or 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.

[0025] Example 1: like Figure 1 As shown, this embodiment of the invention provides a method for closed-loop detection and autonomous repair of data quality in a data platform, including the following specific steps: Step 1: Integrate front-end data, back-end data, and third-party data to form multi-source data. The front-end includes apps, mini-programs, and websites; the back-end includes orders and customer service; and the third-party includes advertising platforms and social media. Clean and correlate the multi-source data to build user profiles, including user age, user occupation, user region, and user shopping time period. User age includes various age groups, user occupation includes various recorded occupations, user region includes provinces in different countries, and user shopping time period includes various purchase time periods. By embedding tracking points on the front end, when a user visits a page, clicks a button, or completes a purchase, pre-embedded tracking code is triggered to collect user behavior data, including the number of visitors, the number of purchases, and the actual visit duration. The number of visitors is calculated by deduplicating device IDs. The number of purchases is recorded when a special purchase event is triggered when the user completes payment and is redirected to the thank-you page. The actual visit duration is obtained by calculating the difference between the timestamp when the user leaves the page or triggers the next action and the timestamp when they enter the page. The access data is also cleaned to remove redundant values ​​and improve the data quality.

[0026] Step 2: Preset the campaign cost as the expense to increase exposure. Conduct a comprehensive analysis based on the number of buyers, visitors, and campaign cost, and standardize the data to eliminate differences in units and convert values ​​of different orders of magnitude into a unified numerical range to construct the objective function.

[0027] Step 3: Compare the output value of the objective function with the preset target conversion threshold. If the output value of the objective function is greater than or equal to the preset target conversion threshold, it means that the user profile at this time can accurately reflect the real situation, and then proceed to Step 4; if the output value of the objective function is less than the preset target conversion threshold, it means that the user profile at this time cannot accurately reflect the real situation, and then proceed to Step 5.

[0028] Step 4: In the advertising execution stage, the optimized accurate user portrait is transmitted to the advertising tool through the API interface. Based on the received portrait data, the advertising tool completes the creation or update of the targeted audience package in the platform background, sets the corresponding bidding strategy and creative combination, and finally accurately delivers the advertising materials to the mobile devices or personal accounts that meet the portrait characteristics; This process not only significantly improves the effective exposure of the product among the target population, but also the system will return the real-time data generated from the new round of delivery to the data collection stage in Step 1, restart the multi-source data integration and user portrait update process, thus forming a continuously optimized and self-evolving closed-loop delivery management system. Advertising tools such as ByteDance Marketing Solutions, etc.

[0029] Step 5: Compare the actual access duration corresponding to each number of visitors with the preset ideal access duration. If the actual access duration is greater than or equal to the ideal access duration, record the user portrait of this customer, indicating that this customer has a purchase intention, which is used to narrow the breadth of the user portrait, helping to quickly identify potential customers with high purchase intention, so as to concentrate computing resources on the most valuable user subset. This not only greatly improves the processing efficiency of the subsequent optimization module, avoids computing waste on invalid users, but also ensures that the finally output coarsely screened user portrait has a higher conversion probability basis, laying a high-quality data foundation for accurate delivery, obtaining the coarsely screened user portrait, and performing Step 6, otherwise continue the comparison.

[0030] Step 6: Comprehensively calculate the objective function corresponding to the coarsely screened user portrait through the simulated annealing algorithm, breaking through the limitation that traditional optimization methods are prone to falling into local optima, strategically accepting some sub-optimal solutions in the solution space, systematically exploring various combination possibilities of the coarsely screened user portrait, so as to accurately identify those sub-user groups with the highest conversion potential and the best cost-effectiveness, and output the objective function and then return to Step 3.

[0031] Example 2, the difference from Example 1 is that: The specific construction method of the objective function is: ; Where, represents the objective function, represents the number of purchases, reflecting the purchase scale and thus the breadth of the delivery, represents the number of visitors, represents the delivery cost, and both represent positive real numbers, to avoid the objective function being meaningless when the number of visitors or the delivery cost is zero, This represents the percentage of visitors who make a purchase, i.e., the conversion rate. It reflects the purchase conversion rate and, consequently, the depth of the campaign. This represents the reciprocal of the cost of advertising; the lower the cost of advertising, the better. The larger the value, the more efficient the deployment.

[0032] The specific method for comprehensively calculating the objective function corresponding to the coarsely screened user profile using the simulated ignition algorithm is as follows: Define an initial temperature and a final high temperature. The initial temperature is set to zero, and the final high temperature is set based on historical experience. Set a threshold for the number of single-round screenings and a probability formula. Starting from the initial temperature of the first round, the temperature is increased. In each round, user age, occupation, region, and shopping time period are randomly selected from the coarse-screened user profile to change the user profile, which in turn changes the number of buyers, visitors, and campaign costs, thus changing the objective function. The new objective function is obtained by calculating the new objective function with the original objective function. The target value is determined based on the target value. This process continues until the number of random selections exceeds the threshold for the number of single-round screenings, at which point the next round begins. Each subsequent round increases the temperature until the final high temperature is reached, at which point the final objective function is determined.

[0033] The specific method for obtaining the target value is as follows: The objective value is obtained by calculating the difference between the new objective function and the original objective function.

[0034] The specific method for determining the objective function based on the target value is as follows: The target value is compared with zero. If the target value is greater than or equal to zero, it means that the output value of the new objective function is greater than or equal to the output value of the original objective function. This indicates that the more people buy, the higher the conversion rate, or the lower the cost of advertising. The new objective function is then updated to be the target function for the next time. If the target value is less than zero, it means that the output value of the new objective function is less than the output value of the original objective function. This indicates that the fewer people buy, the lower the conversion rate, or the higher the cost of advertising. The probability formula is used to determine the target function. If the probability is accepted, the new objective function is still updated to be the target function for the next time to avoid the global optimum. If the probability is not accepted, the original objective function is updated to be the target function for the next time.

[0035] The specific method for obtaining the heating temperature is as follows: The first The temperature of each heat cycle is calculated by adding the temperature of the first heat cycle to the number of heat cycles, thus obtaining the result of the first heat cycle. The wheel heating temperature; ; in, Indicates the first The wheel heating temperature, Indicates the first The wheel heating temperature, Indicates the number of rounds.

[0036] The probability formula is set up as follows: ; in, Represents probability. It is in exponential form and ranges from 0 to 1. This represents the target value, which is less than zero. Indicates the first If the temperature rises in a cycle and is greater than zero, then Less than zero, making Monotonically decreasing, that is In form, , Greater than zero, in In China, due to As the temperature gradually increases, the output value of the objective function gradually increases. However, the conversion rate in the objective function is bounded, and the deployment cost will not increase indefinitely, meaning it will converge. Therefore, the output value of the objective function will gradually converge, approaching the optimal solution, and the objective value will gradually approach zero. Meanwhile, the heating temperature gradually increases... It is gradually decreasing, meaning the probability value is increasing.

[0037] The specific method for determining based on the probability formula is as follows: Set the range from 0 to the probability value as the unacceptable probability range, and set the range from the probability value to 1 as the accepted probability range. Set a random function in code, in floating-point form, with a range from 0 to 1. If the function value of the random function is in the accepted probability range, it means the probability is accepted; if the function value of the random function is in the unacceptable probability range, it means the probability is unacceptable. As the probability value increases, the unacceptable probability range increases, while the accepted probability range decreases.

[0038] Example 3: like Figure 2 As shown: A data platform data quality closed-loop detection and autonomous repair system includes the following specific modules: Data acquisition module: Collects multi-source data to build user profiles, including user age, user occupation, user region, and user shopping time period; obtains access data through front-end data tracking points, including the number of visitors, the number of purchases, and the actual access duration. Objective function construction module: Presets the campaign cost, comprehensively analyzes and standardizes the number of buyers, visitors, and campaign costs, and constructs the objective function; The target analysis module compares the output value of the objective function with the preset target conversion threshold. If the output value is greater than or equal to the threshold, the precise delivery module is executed; otherwise, the coarse user profile screening module is executed. Precision delivery module: Delivers advertisements to devices that match the user profile, and then ends or returns to the data collection module; Coarse user profile screening module: compares the actual access time corresponding to each visitor with the preset ideal access time: if the actual access time is greater than or equal to the ideal access time, record this user profile to obtain the coarse user profile, and execute the optimal objective function analysis module; otherwise, continue the comparison. Optimal Objective Function Analysis Module: This module uses a simulated fire-starting algorithm to comprehensively calculate the objective function corresponding to the coarsely screened user profile, outputs the objective function, and returns it to the compliance analysis module.

[0039] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to the specific implementations described. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. A method for closed-loop detection and autonomous repair of data quality in a data platform, characterized in that: The specific steps include the following: Step 1: Collect multi-source data to build user profiles, including user age, user occupation, user region, and user shopping time period; obtain access data through front-end data tracking, including the number of visitors, the number of purchases, and the actual access duration. Step 2: Preset the campaign cost, conduct a comprehensive analysis of the number of buyers, visitors, and campaign costs, and standardize the data to construct the objective function; Step 3: Compare the output value of the objective function with the preset target transformation threshold: if the output value is greater than or equal to the threshold, proceed to Step 4; otherwise, proceed to Step 5. Step 4: Deliver the advertisement to devices that match the user profile, and end or return to Step 1; Step 5: Compare the actual access duration for each visitor with the preset ideal access duration: If the actual access duration is greater than or equal to the ideal access duration, record this user profile to obtain a rough user profile and proceed to Step 6; otherwise, continue the comparison. Step Six: Perform a comprehensive calculation on the objective function corresponding to the coarsely screened user profile using the simulated fire-starting algorithm, output the objective function, and return to Step Three.

2. The data quality closed-loop detection and autonomous repair method for a data middle platform according to claim 1, characterized in that: The specific method for constructing the objective function is as follows: ； in, Describe the objective function. Indicates the number of buyers. Indicates the number of visitors. Indicates the cost of deployment. and All of them represent positive real numbers.

3. The data quality closed-loop detection and autonomous repair method for a data middle platform according to claim 1, characterized in that: The specific method for comprehensively calculating the objective function corresponding to the coarsely screened user profile using the simulated ignition algorithm is as follows: Define an initial temperature and a final high temperature, and set a threshold for the number of single-round screenings and a probability formula. Starting from the initial temperature of the first round, the temperature is increased. In each round, user age, occupation, region, and shopping time period are randomly selected from the coarse-screened user profile to obtain a new objective function. The new objective function is calculated with the objective function to obtain the target value. The objective function for this round is determined based on the target value. This process continues until the number of random selections exceeds the threshold for the number of single-round screenings, at which point the next round begins. This process continues in each subsequent round, with the temperature increasing until the final high temperature is reached, at which point the final objective function is determined.

4. The data quality closed-loop detection and autonomous repair method for a data middle platform according to claim 3, characterized in that: The specific method for obtaining the target value is as follows: The objective value is obtained by calculating the difference between the new objective function and the original objective function.

5. The data quality closed-loop detection and autonomous repair method for a data middle platform according to claim 3, characterized in that: The specific method for determining the objective function based on the target value is as follows: The objective value is compared with zero. If the objective value is greater than or equal to zero, the new objective function is updated to the objective function for the next step. If the objective value is less than zero, a probability formula is used to determine whether the objective function is acceptable or not. If the probability formula is not accepted, the new objective function is updated to the objective function for the next step.

6. The data quality closed-loop detection and autonomous repair method for a data middle platform according to claim 3, characterized in that: The specific method for obtaining the heating temperature is as follows: The first The temperature of each heat cycle is calculated by adding the temperature of the first heat cycle to the number of heat cycles, thus obtaining the result of the first heat cycle. The temperature of the wheel rises.

7. The data quality closed-loop detection and autonomous repair method for a data middle platform according to claim 3, characterized in that: The probability formula is set as follows: ； in, Represents probability. Indicates the target value. Indicates the first The temperature of the wheel rises.

8. The data quality closed-loop detection and autonomous repair method for a data middle platform according to claim 7, characterized in that: The specific method for determining the probability formula is as follows: Set the range from 0 to the probability value as the unacceptable probability range, and set the range from the probability value to 1 as the accepted probability range. Set a random function in the code, with a range from 0 to 1. If the function value of the random function is in the accepted probability range, it means the probability is accepted; if the function value of the random function is in the unacceptable probability range, it means the probability is unacceptable.

9. A data platform data quality closed-loop detection and autonomous repair system, used to implement the data platform data quality closed-loop detection and autonomous repair method according to any one of claims 1-8, characterized in that, The data platform data quality closed-loop detection and autonomous repair system includes: a data acquisition module, an objective function construction module, a compliance analysis module, a precise delivery module, a coarse screening user profile module, and an optimal objective function analysis module; The data acquisition module collects multi-source data to build user profiles, including user age, user occupation, user region, and user shopping time period; it also obtains access data through front-end data tracking points, including the number of visitors, the number of purchases, and the actual access duration. The objective function construction module: presets the deployment cost, comprehensively analyzes and standardizes the number of buyers, the number of visitors, and the deployment cost, and constructs the objective function; The target analysis module compares the output value of the objective function with a preset target conversion threshold. If the output value is greater than or equal to the threshold, the precise delivery module is executed; otherwise, the coarse user profile screening module is executed. The precise delivery module delivers advertisements to devices that match the user profile and then ends or returns to the data collection module. The coarse user profile module compares the actual access duration corresponding to each visitor with the preset ideal access duration. If the actual access duration is greater than or equal to the ideal access duration, the user profile is recorded to obtain the coarse user profile, and the optimal objective function analysis module is executed; otherwise, the comparison continues. The optimal objective function analysis module performs a comprehensive calculation on the objective function corresponding to the coarsely screened user profile using a simulated fire-starting algorithm, outputs the objective function, and returns it to the target analysis module.

Images