A Similar Advertising Contextual Analysis Method and System Based on the COSMO Algorithm

Abstract

This invention relates to the field of advertising push technology, specifically disclosing a method and system for contextual analysis of similar advertisements based on the COSMO algorithm. The method includes vectorizing advertisements and scenarios based on the same shared semantic space to obtain advertisement vectors and semantic vectors; comparing advertisement vectors and semantic vectors to determine vector similarity; selecting advertisements whose vector similarity reaches a preset similarity threshold; and executing the delivery process. The similarity threshold is a dynamic threshold. In the process of matching advertisements with scenarios, this invention constructs a multi-dimensional state-aware similarity threshold adaptive adjustment architecture, enabling the advertisement matching standard to be dynamically adjusted according to the system's real-time operating capabilities, advertisement resource supply capabilities, and the historical maturity of the scenario. This process is entirely completed autonomously by intelligent devices, with extremely fast response speed and high timeliness.

Description

Technical Field

[0001] This invention relates to the field of advertising push technology, specifically a similar advertising scenario analysis method and system based on the COSMO algorithm. Background Technology

[0002] In the field of computational advertising and recommendation systems, contextualized advertising is a core technology of precision marketing. Its core objective is to push the right ad creative to the right user in the right context, thereby optimizing ad click-through rate, conversion rate and user experience. The key to achieving this goal lies in calculating the similarity between the advertisement and the current presentation scenario, and then filtering and matching based on a preset similarity threshold. Traditional methods generally adopt static or semi-static threshold setting strategies, where the threshold remains fixed over a long period or is only periodically adjusted based on a few business rules. In large-scale, high-concurrency online advertising systems, this setting strategy has very low timeliness. For example, during peak traffic periods, the number of queries per second increases sharply, the 95th percentile request response time is prolonged, CPU and memory utilization approaches the critical point, and the time consumption for real-time similarity calculation increases. If a high similarity threshold is maintained, it will lead to increased matching calculation latency, slower system response, and even service degradation or timeout errors, seriously affecting user experience and system stability. Therefore, how to provide an automated threshold update scheme based on smart devices to smoothly and quickly adjust the similarity threshold, thereby improving the overall robustness, resource utilization efficiency, and scenario adaptability of the advertising system, is the technical problem that this invention aims to solve. Summary of the Invention

[0003] The purpose of this invention is to provide a similar advertising scenario analysis method and system based on the COSMO algorithm to solve the problems mentioned in the background art.

[0004] To achieve the above objectives, the present invention provides the following technical solution:

[0005] A scenario-based analysis method for similar advertisements based on the COSMO algorithm, the method comprising:

[0006] Based on the same shared semantic space, ads and scenarios are vectorized to obtain ad vectors and semantic vectors;

[0007] Compare ad vectors and semantic vectors to determine vector similarity, select ads whose vector similarity reaches a preset similarity threshold, and execute the delivery process;

[0008] The similarity threshold is a dynamic threshold, and the determination process is as follows:

[0009] Acquire system operation status monitoring data, and calculate the system health factor based on the monitoring data;

[0010] Obtain advertising resource supply data, and calculate the advertising inventory supply coefficient based on the advertising resource supply data;

[0011] The secondary correction coefficient is calculated based on the system health factor and the advertising inventory supply coefficient.

[0012] Obtain historical exposure data for the target advertising scenario, and calculate the scenario's historical maturity coefficient based on the historical exposure data;

[0013] Based on the preset basic similarity threshold, the secondary correction coefficient, and the scene historical maturity coefficient, the target scene similarity threshold is calculated, and the current scene similarity threshold is adjusted to the target scene similarity threshold.

[0014] As a further aspect of the present invention: the step of acquiring system operation status monitoring data and calculating the system health factor based on the monitoring data includes:

[0015] Get the current number of queries per second, 95th percentile request response time, CPU utilization, memory utilization, and real-time similarity calculation time;

[0016] Calculate the system ingress pressure coefficient based on the current number of queries per second and the 95th percentile request response time;

[0017] Calculate the resource consumption coefficient based on CPU utilization, memory utilization, and real-time similarity calculation time;

[0018] Based on the combined factors of the system ingress pressure coefficient and the computational resource consumption coefficient, a system health factor is determined. The system health factor is negatively correlated with both the system ingress pressure coefficient and the computational resource consumption coefficient. The higher the health factor, the healthier the system.

[0019] As a further aspect of the present invention: the step of calculating the system inlet pressure coefficient includes:

[0020] Get the current number of queries per second and the 95th percentile request response time;

[0021] Perform max-min normalization on the current number of queries per second to obtain the current query per second index;

[0022] The target response time threshold is compared with the 95th percentile request response time, and the complement of the ratio is taken as the request response time index.

[0023] The system entry pressure coefficient is obtained by taking a weighted average of the current query count per second index and the request response time index. The system entry pressure coefficient is directly proportional to both the current query count per second index and the request response time index.

[0024] As a further aspect of the present invention: the step of calculating the resource consumption coefficient includes:

[0025] Obtain CPU utilization, memory utilization, and real-time similarity calculation time;

[0026] The system resource stress index is obtained by taking the larger value of CPU utilization and memory utilization and comparing it with the maximum utilization.

[0027] The similarity calculation time is calculated by comparing the real-time similarity calculation time with the maximum allowed calculation time threshold to obtain the similarity calculation time index.

[0028] The larger value between the system resource tension index and the similarity calculation time index is taken as the resource consumption coefficient. Calculate the resource consumption coefficient Output range 0-1, calculate resource consumption coefficient. The closer the value is to 1, the more strained the computing resources are.

[0029] As a further aspect of the present invention: the step of calculating the advertising inventory supply coefficient includes:

[0030] Get the number of available ad creatives and ad fill rate;

[0031] The available ad creative index is obtained by comparing the number of available ad creatives with the minimum number of creatives required.

[0032] The advertising inventory supply coefficient is calculated by weighted linear combination of the available advertising creativity index and the advertising fill rate. The value range of the advertising inventory supply coefficient is constrained to be between 0 and 1, and the larger the value, the more abundant the advertising supply. Both the available advertising creativity index and the advertising fill rate have a positive contribution to the supply coefficient.

[0033] As a further aspect of the present invention: the step of calculating the secondary correction coefficient includes:

[0034] Obtain the system health factor and advertising inventory supply coefficient;

[0035] Substitute the system health factor and the advertising inventory supply coefficient into the formula. Obtain the secondary correction coefficient Secondary correction coefficient The output range is 0-1, and the closer it is to 1, the more ideal the system state. As a system health factor, This is the advertising inventory supply coefficient.

[0036] As a further aspect of the present invention: the step of calculating the historical maturity coefficient of a scenario includes:

[0037] Obtain the historical exposure count of the scene;

[0038] The scene historical exposure count is processed by a nonlinear normalization function to obtain the scene historical exposure count index, which is positively correlated with the historical exposure count.

[0039] After setting the upper limit of the historical exposure count index of the scene to 1 using the min function, the historical maturity coefficient of the scene is obtained.

[0040] As a further aspect of the present invention: the step of calculating the target scene similarity threshold includes:

[0041] Obtain the basic similarity threshold, secondary correction coefficient, and scene historical maturity coefficient;

[0042] Substituting the basic similarity threshold, secondary correction coefficient, and scene historical maturity coefficient into the formula Obtain the target scene similarity threshold ,in, Based on the basic similarity threshold, To adjust the sensitivity coefficient, The weighting is affected by the secondary correction coefficient. This is the secondary correction coefficient. This represents the historical maturity coefficient of the scenario.

[0043] The present invention also provides a similar advertising scenario analysis system based on the COSMO algorithm. The system is used to implement the aforementioned similar advertising scenario analysis method based on the COSMO algorithm. The system includes:

[0044] The data vectorization module is used to vectorize advertisements and scenarios based on the same shared semantic space to obtain advertisement vectors and semantic vectors.

[0045] The vector comparison module is used to compare ad vectors and semantic vectors, determine vector similarity, select ads whose vector similarity reaches a preset similarity threshold, and execute the delivery process.

[0046] The similarity threshold is a dynamic threshold, and the determination process is as follows:

[0047] Acquire system operation status monitoring data, and calculate the system health factor based on the monitoring data;

[0048] Obtain advertising resource supply data, and calculate the advertising inventory supply coefficient based on the advertising resource supply data;

[0049] The secondary correction coefficient is calculated based on the system health factor and the advertising inventory supply coefficient.

[0050] Obtain historical exposure data for the target advertising scenario, and calculate the scenario's historical maturity coefficient based on the historical exposure data;

[0051] Based on the preset basic similarity threshold, the secondary correction coefficient, and the scene historical maturity coefficient, the target scene similarity threshold is calculated, and the current scene similarity threshold is adjusted to the target scene similarity threshold.

[0052] Compared with the prior art, the beneficial effects of the present invention are as follows: In the process of matching advertisements with scenarios, the present invention constructs a multi-dimensional state-aware similarity threshold adaptive adjustment architecture, which enables the advertisement matching standard to be dynamically adjusted according to the real-time operation capability of the system, the supply capability of advertising resources, and the historical maturity of the scenario. This process is completed autonomously by the intelligent device, with extremely fast response speed and strong timeliness. Attached Figure Description

[0053] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention.

[0054] Figure 1 This is a flowchart of a similar advertising scenario analysis method based on the COSMO algorithm.

[0055] Figure 2 This is a structural diagram of a similar advertising scenario analysis system based on the COSMO algorithm. Detailed Implementation

[0056] To make the technical problems to be solved, the technical solutions, and the beneficial effects of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the present invention and are not intended to limit the present invention.

[0057] Figure 1 The flowchart below illustrates a scenario-based analysis method for similar advertisements based on the COSMO algorithm. In this embodiment of the invention, a scenario-based analysis method for similar advertisements based on the COSMO algorithm includes:

[0058] Based on the same shared semantic space, ads and scenarios are vectorized to obtain ad vectors and semantic vectors;

[0059] Compare ad vectors and semantic vectors to determine vector similarity, select ads whose vector similarity reaches a preset similarity threshold, and execute the delivery process;

[0060] The process of matching advertisements with scenarios is actually part of existing technology. Existing technologies offer numerous schemes that match advertisements with scenarios and then select which scenarios to target. Their basic principle is to first map the advertisement and scenario into data, called vectors or feature sets. By comparing this data, the similarity between the advertisement content and the scenario can be obtained, which is equivalent to the matching degree. When the vector similarity reaches a preset similarity threshold, the advertisement is pushed to the scenario. A hidden premise is that the vectorization process of the advertisement and scenario needs to be based on the same shared semantic space so that their corresponding vectors can be compared. A relatively simple solution is to convert both the scenario and the advertisement into word sets, and then convert the words into word vectors. In this case, they can be compared.

[0061] The key point of this invention is that the similarity threshold used to determine whether a match is made is a dynamic threshold, and the process for determining the similarity threshold is as follows:

[0062] Acquire system operation status monitoring data, and calculate the system health factor based on the monitoring data;

[0063] Obtain advertising resource supply data, and calculate the advertising inventory supply coefficient based on the advertising resource supply data;

[0064] The secondary correction coefficient is calculated based on the system health factor and the advertising inventory supply coefficient.

[0065] Obtain historical exposure data for the target advertising scenario, and calculate the scenario's historical maturity coefficient based on the historical exposure data;

[0066] Based on the preset basic similarity threshold, the secondary correction coefficient, and the scene historical maturity coefficient, the target scene similarity threshold is calculated, and the current scene similarity threshold is adjusted to the target scene similarity threshold.

[0067] Specifically, the above content is explained as follows: Based on the current number of queries per second, 95th percentile request response time, CPU utilization, memory utilization, and real-time similarity calculation time, a system health factor is calculated. This factor is used to quantitatively evaluate the overall operating status of the advertising delivery system. Its value is usually between 0 and 1. The closer the value is to 1, the more ideal the system's operating status, the more abundant the resources, and the faster the response; the closer the value is to 0, the more pressure the system faces or the more strained the resources.

[0068] Based on the number of available ad creatives and the ad fill rate, an ad inventory supply coefficient is calculated. This coefficient measures the sufficiency of ad creative resources available for placement in the current advertising system. Its value is typically between 0 and 1. A value closer to 1 indicates sufficient ad creative inventory and a large number of ads available for matching; a value closer to 0 indicates tight ad creative inventory, which may lead to fewer ad placement opportunities.

[0069] A secondary correction coefficient is calculated based on the system health factor and the advertising inventory supply coefficient. This coefficient is an intermediate adjustment parameter derived from the comprehensive calculation of the system health factor and the advertising inventory supply coefficient. Its function is to make preliminary dynamic adjustments to the basic similarity threshold to reflect the overall status of the system's current performance and advertising supply.

[0070] Based on the historical exposure count of a scenario, a historical maturity coefficient is calculated. This coefficient is used to assess the historical data accumulation and stability of a specific advertising scenario. New scenarios have lower maturity coefficients due to less data, while mature scenarios with extensive exposure and data accumulation have higher maturity coefficients. This coefficient reflects the different needs of scenarios for matching accuracy and generalization ability.

[0071] Based on the basic similarity threshold, secondary correction coefficient, and scene historical maturity coefficient, the target scene similarity threshold is calculated and adjusted to the target scene similarity threshold. The target scene similarity threshold is an ideal standard for ad-scene matching, dynamically calculated by the system based on factors such as real-time operating status, ad supply, and scene historical maturity. The system will filter and match ads according to this target threshold.

[0072] The above-mentioned approach achieves adaptive and smooth adjustment of the similarity threshold by comprehensively considering multiple dynamic dimensions such as real-time system performance, advertising inventory supply level, and historical maturity of the scenario. This overcomes the limitations of static threshold or single-dimensional adjustment in existing technologies, effectively solves the balance problem between system stability, advertising supply efficiency, and scenario adaptability, and significantly improves the overall efficiency, robustness, and scenario adaptability of the advertising delivery system.

[0073] Regarding step S100, the step of acquiring system operation status monitoring data and calculating the system health factor based on the monitoring data includes:

[0074] Get the current number of queries per second, 95th percentile request response time, CPU utilization, memory utilization, and real-time similarity calculation time;

[0075] Calculate the system ingress pressure coefficient based on the current number of queries per second and the 95th percentile request response time;

[0076] Calculate the resource consumption coefficient based on CPU utilization, memory utilization, and real-time similarity calculation time;

[0077] Based on the combined factors of the system ingress pressure coefficient and the computational resource consumption coefficient, a system health factor is determined. The system health factor is negatively correlated with both the system ingress pressure coefficient and the computational resource consumption coefficient. The higher the health factor, the healthier the system.

[0078] The step of calculating the system inlet pressure coefficient includes:

[0079] Get the current number of queries per second and the 95th percentile request response time;

[0080] Perform max-min normalization on the current number of queries per second to obtain the current query per second index;

[0081] The target response time threshold is compared with the 95th percentile request response time, and the complement of the ratio is taken as the request response time index.

[0082] The system entry pressure coefficient is obtained by taking a weighted average of the current query count per second index and the request response time index. The system entry pressure coefficient is directly proportional to both the current query count per second index and the request response time index.

[0083] Specifically, the steps for calculating the resource consumption coefficient include:

[0084] Obtain CPU utilization, memory utilization, and real-time similarity calculation time;

[0085] The system resource stress index is obtained by taking the larger value of CPU utilization and memory utilization and comparing it with the maximum utilization.

[0086] The similarity calculation time is calculated by comparing the real-time similarity calculation time with the maximum allowed calculation time threshold to obtain the similarity calculation time index.

[0087] The larger value between the system resource tension index and the similarity calculation time index is taken as the resource consumption coefficient. Calculate the resource consumption coefficient Output range 0-1, calculate resource consumption coefficient. The closer the value is to 1, the more strained the computing resources are.

[0088] In one example of the technical solution of this invention, the process of determining the system health factor is described. This factor is used to quantify the real-time operational health of the system. The specific steps are as follows: obtaining the current number of queries per second, 95th percentile request response time, CPU utilization, memory utilization, and real-time similarity calculation time. Obtaining these parameters is fundamental to evaluating the system's operational status. These parameters can be collected in real-time or periodically using system monitoring tools, log analysis systems, or performance counters. For example, monitoring systems such as Prometheus and Grafana can be deployed to collect these metrics, or they can be obtained through APIs provided by the operating system.

[0089] Based on the current number of queries per second and the 95th percentile request response time, a system entry pressure coefficient is calculated. This coefficient aims to quantify the external request load pressure currently borne by the system and provide a more comprehensive reflection of the system entry's health status. The coefficient can be calculated using various methods, such as weighted averaging or nonlinear mapping.

[0090] The current number of queries per second (QPS) is normalized using a max-min normalization process to obtain the QPS index. The original QPS data is then converted into a dimensionless value between 0 and 1 for easier calculation with other metrics. Normalization eliminates differences in the dimensions and numerical ranges of different metrics.

[0091] The target response time threshold is compared with the 95th percentile request response time, and the complement of this ratio (i.e., one minus the ratio) is taken as the request response time index; this is used to measure how well the system's response speed is relative to the preset performance target. The design of taking the complement of the ratio ensures that the index value is higher as the actual response time is closer to or lower than the target threshold.

[0092] The system ingress pressure coefficient is obtained by weighted averaging and summing the current query count per second index and request response time index. The system ingress pressure coefficient is directly proportional to both the current query count per second index and request response time index. Specifically, the calculation method is as follows: substitute the current query count per second index and request response time index into the formula... Obtain the system inlet pressure coefficient System inlet pressure coefficient Output range 0-1, system inlet pressure coefficient The closer the value is to 1, the greater the system inlet pressure. The weight that affects QPS is usually between 0 and 1, and can be set according to the actual business scenario. This represents the current query count per second index. This is the request response time index; the formula allows for flexible adjustment of the relative importance of QPS and response time when assessing system ingress pressure.

[0093] Based on CPU utilization, memory utilization, and real-time similarity calculation time, the computational resource consumption coefficient is calculated to assess the stress on internal computational resources (such as CPU and memory) and the execution efficiency of core algorithms.

[0094] The system resource stress index is obtained by comparing the larger of CPU utilization and memory utilization with the maximum utilization. This index is used to quantify the pressure level of the system's core computing resources, ensuring that even if one of the resources reaches a high level, the stress level of the system resources can be reflected in a timely manner.

[0095] The ratio of real-time similarity calculation time to the maximum allowed calculation time threshold is used to obtain the similarity calculation time index, which directly reflects the performance of the core similarity calculation module.

[0096] The larger value between the system resource tension index and the similarity calculation time index is taken as the resource consumption coefficient. Calculate the resource consumption coefficient Output range 0-1, calculate resource consumption coefficient. The closer the value is to 1, the more strained the computing resources are; this ensures that the computing resource consumption coefficient can sensitively capture the performance bottleneck of any critical resource or core computing link.

[0097] Based on a comprehensive analysis of the system ingress pressure coefficient and the computational resource consumption coefficient, a system health factor is determined. The system health factor is negatively correlated with both the system ingress pressure coefficient and the computational resource consumption coefficient; a higher health factor indicates a healthier system. The specific calculation method is as follows: substitute the system ingress pressure coefficient and the computational resource consumption coefficient into the formula... Obtain system health factors System health factor Output range 0-1, system health factor The closer the value is to 1, the healthier the system. This is the system inlet pressure coefficient. This is used to calculate the resource consumption coefficient. This calculation method allows the system health factor to comprehensively reflect the overall operational status of the system, from external request processing to internal resource consumption.

[0098] This invention obtains an accurate system health factor by performing multi-dimensional and refined monitoring and evaluation of the system's operating status. Specifically, the solution first focuses on two main aspects: external load and internal resource consumption. Regarding external load, the solution collects real-time data on the current number of queries per second and the 95th percentile request response time, and normalizes these data to obtain the current number of queries per second index and request response time index. These two indices are then combined using a weighted average to calculate the system entry pressure coefficient. This coefficient quantifies the external request pressure the system is under, reflecting the performance bottleneck at the system entry point. Regarding internal resource consumption, the solution monitors CPU utilization, memory utilization, and real-time similarity calculation time to calculate the system resource stress index and similarity calculation time index, respectively. The larger of these two indices is selected as the computational resource consumption coefficient to ensure sensitive capture of internal resource pressure. Finally, the system entry pressure coefficient and computational resource consumption coefficient are averaged and their complements are taken to generate a comprehensive system health factor. This health factor comprehensively reflects the overall operating status of the system from external request processing to internal resource consumption; a value closer to 1 indicates a healthier system. Through this hierarchical, multi-indicator fusion calculation mechanism, the present invention can overcome the limitations of traditional single-indicator evaluation, providing a more accurate and reliable basis for the subsequent calculation of secondary correction coefficients. This enables the adjustment of the target scene similarity threshold to more accurately respond to the actual operating health of the system, avoiding performance degradation or service interruption due to system overload or resource shortage, and ensuring the stability and effectiveness of the advertising scenario analysis method.

[0099] Through the above technical solution, this invention provides a comprehensive and refined system health assessment mechanism. This mechanism, by comprehensively considering system ingress pressure and internal computing resource consumption, avoids the one-sidedness of single-indicator assessments, enabling the system health factor to more accurately reflect the system's current true operating status. When the system faces high concurrency requests or internal resource constraints, this health factor can promptly and sensitively capture and quantify these changes. This precise system health assessment provides a solid foundation for the subsequent calculation of the secondary correction coefficient, thereby ensuring that the adjustment of the target scenario similarity threshold can be more intelligent and adaptive. For example, when the system load is high, the health factor decreases, thus affecting the secondary correction coefficient and prompting a corresponding adjustment of the similarity threshold to avoid system overload and ensure service stability and user experience. Conversely, when the system is idle or resources are sufficient, the health factor is higher, allowing for more proactive threshold adjustments to improve the efficiency and effectiveness of ad matching. Therefore, this solution significantly improves the robustness and adaptability of similar ad scenario analysis methods, effectively balancing system performance and ad effectiveness.

[0100] Regarding step S200, the step of calculating the advertising inventory supply coefficient includes:

[0101] Get the number of available ad creatives and ad fill rate;

[0102] The available ad creative index is obtained by comparing the number of available ad creatives with the minimum number of creatives required.

[0103] The advertising inventory supply coefficient is calculated by weighted linear combination of the available advertising creativity index and the advertising fill rate. The value range of the advertising inventory supply coefficient is constrained to be between 0 and 1, and the larger the value, the more abundant the advertising supply. Both the available advertising creativity index and the advertising fill rate have a positive contribution to the supply coefficient.

[0104] The above solution provides a quantitative approach to assess advertising inventory supply capacity, obtaining the number of available ad creatives and ad fill rate. The number of available ad creatives refers to the total number of approved and usable ad creatives (such as images, videos, and copy combinations) currently in the advertising platform or advertiser's account. This number directly reflects the advertising system's potential for ad matching and delivery in a specific scenario. This number can be obtained by querying real-time data in the ad creative management system or dynamically updated by statistically analyzing the number of added and deleted creatives within a specific time period. Ad fill rate refers to the percentage of ad requests received by the advertising system during ad delivery that are ultimately successfully matched and displayed. It is a key indicator for measuring the advertising system's monetization capability and ad placement efficiency. This fill rate can be obtained by monitoring ad request logs and ad display logs in real time, calculating the ratio of successfully displayed ads to the total number of requests, or by obtaining this statistical data through the API interface provided by the advertising platform.

[0105] The available ad creative index is obtained by comparing the number of available ad creatives with the minimum number of creatives required. The minimum number of creatives required refers to the minimum number of available ad creatives set to ensure that the advertising system can provide sufficiently diverse ad content, avoid ad fatigue, and meet basic advertising needs. Falling below this number may lead to insufficient ad supply or a decline in ad quality. This requirement can be preset to a fixed value based on historical data analysis, industry standards, or operational strategies, or dynamically adjusted according to the complexity of the advertising scenario and the diversity of the user group.

[0106] The advertising inventory supply coefficient is calculated by weighted linearly combining the available ad creativity index and the ad fill rate. The value of the advertising inventory supply coefficient is constrained to be between 0 and 1, with a higher value indicating a more abundant ad supply. Both the available ad creativity index and the ad fill rate contribute positively to the supply coefficient. The specific calculation method is as follows: substitute the available ad creativity index and the ad fill rate into the formula... Obtain the advertising inventory supply coefficient Advertising inventory supply coefficient Output range 0-1, advertising inventory supply coefficient The closer the value is to 1, the more abundant the advertising supply. The weight of the number of creative ideas. This is a value between 0 and 1 used to balance the relative importance of the available ad creative index and ad fill rate when calculating the ad inventory supply coefficient. A higher weight means that the number of creatives has a greater impact on the ad supply coefficient. This weight can be set through expert experience or optimized through historical data analysis using machine learning models (such as regression analysis). The available advertising creativity index, This refers to the ad fill rate.

[0107] Through the above technical solution, this invention comprehensively considers two key indicators: the number of available ad creatives and the ad fill rate, and introduces minimum creative quantity requirements and creative quantity weights, thereby providing a more comprehensive and accurate ad inventory supply coefficient. This multi-dimensional, weighted evaluation method effectively solves the problem of one-sidedness that may result from evaluating ad inventory supply based on a single indicator, and avoids the deviation of the secondary correction coefficient caused by inaccurate evaluation. Therefore, when calculating the secondary correction coefficient, more realistic ad inventory supply information can be obtained, making the calculation of the secondary correction coefficient more accurate, thereby ensuring that the adjustment of the target scene similarity threshold is more scientific and reasonable, and ultimately improving the efficiency and effectiveness of ad placement.

[0108] Regarding step S300, the step of calculating the secondary correction coefficient includes:

[0109] Obtain the system health factor and advertising inventory supply coefficient;

[0110] Substitute the system health factor and the advertising inventory supply coefficient into the formula. Obtain the secondary correction coefficient Secondary correction coefficient The output range is 0-1, and the closer it is to 1, the more ideal the system state. As a system health factor, This is the advertising inventory supply coefficient.

[0111] The above content explains the process of determining the secondary correction coefficient, obtaining the system health factor and the ad inventory supply coefficient. The system health factor is a quantitative assessment of the overall operating status and performance of the system. This factor reflects the system's ability to handle current load and maintain responsiveness. The system health factor can be obtained in various ways. For example, it can be calculated comprehensively based on real-time system monitoring data, such as CPU utilization, memory utilization, network latency, error rate, and queries per second; or it can be predicted and output using machine learning models by learning from and analyzing historical system performance data. The ad inventory supply coefficient is a quantitative assessment of the availability and sufficiency of ad resources. This coefficient indicates whether there are sufficient and suitable ad creatives and ad slots to meet ad fill demand. The ad inventory supply coefficient can be obtained in various ways. For example, it can be calculated comprehensively based on the number of currently available ad creatives, ad fill rate, ad quality score, and predictions of future ad inventory; or it can be obtained by analyzing the diversity and relevance of existing ad creatives to assess their matching degree with current ad demand.

[0112] Substitute the system health factor and the advertising inventory supply coefficient into the formula. Obtain the secondary correction coefficient Secondary correction coefficient The output range is 0-1, and the closer it is to 1, the more ideal the system state. As a system health factor, This is the advertising inventory supply coefficient. The formula is designed to comprehensively consider the system's operational status and the supply of advertising resources. The first part of the formula... Indicates system health factor and advertising inventory supply coefficient The average value reflects the average state of the entire system. The second part of the formula... Then, as a penalty, when the system health factor... and advertising inventory supply coefficient When there are significant differences between them, this penalty term will reduce the secondary correction coefficient. The value of the secondary correction coefficient is ensured only when both system health and ad supply are in good condition and in balance. Only then can it approach the ideal value of 1. Conversely, if one of them performs poorly or there is a significant imbalance between them, the secondary correction coefficient... This will be adjusted accordingly, thus avoiding overly aggressive or unreasonable adjustments to the target scene similarity threshold. In this way, the secondary correction coefficient... It can more accurately and robustly reflect the overall operating status and resource matching degree of the current advertising system, providing a reliable basis for subsequent target scene similarity threshold calculation.

[0113] Through the above technical solution, this invention can comprehensively assess the system's health status and advertising inventory supply, and generate a secondary correction coefficient using a balancing formula. This method avoids the one-sidedness of judging based on a single indicator and effectively solves the technical problem of how to accurately and robustly generate a secondary correction coefficient in complex and ever-changing environments. This secondary correction coefficient can more accurately reflect the system's true carrying capacity and resource matching degree, thus providing a more reliable and refined basis for the adaptive adjustment of the target scene similarity threshold, making the adjustment of the similarity threshold more reasonable and efficient, thereby optimizing the advertising effect and improving the overall stability and resource utilization of the system.

[0114] Regarding step S400, the step of calculating the historical maturity coefficient of the scenario includes:

[0115] Obtain historical exposure counts for a specific advertising scenario; obtaining historical exposure counts for a specific scenario refers to the system retrieving the total number of times a particular advertising scenario was displayed within a certain period from historical data. This historical exposure count is fundamental data for assessing the maturity of the scenario and can be obtained from the advertising campaign log database or by aggregating exposure events through a real-time data stream processing system.

[0116] The scene's historical exposure count is processed using a nonlinear normalization function to obtain a scene historical exposure count index, which is positively correlated with the historical exposure count. The specific calculation method is as follows: substitute the scene's historical exposure count into the formula. Calculate and obtain the historical exposure index of the scene. ,in, For the historical exposure count of the scene, This step sets the threshold for the number of exposures required for scene maturity. The aim is to transform the raw, potentially highly variable scene exposure data into a standardized, physically meaningful maturity indicator. The formula uses a logarithmic function for normalization, effectively mitigating the excessive influence of extreme exposure data on maturity assessment, resulting in a more stable and reasonable assessment. This represents the actual number of historical exposures, while This is a preset threshold for the number of exposures required for a scenario to reach a mature state. This threshold can be determined based on business experience or historical data analysis. The calculation process can be completed in a dedicated calculation module or data processing framework.

[0117] After limiting the historical exposure count index of the scene to an upper limit of 1 using the min function, the historical maturity coefficient of the scene is obtained. This step is to ensure that the calculated historical maturity coefficient of the scene is always within a controllable and meaningful range (usually between 0 and 1). By limiting the upper limit of the index to 1, abnormally high maturity coefficients due to excessively high scene exposure counts can be avoided, thus ensuring the stability and accuracy of subsequent similarity threshold adjustments. This can be achieved by calculating... Then apply directly This is achieved through operations.

[0118] The aforementioned technical solution provides a method for quantitatively assessing the historical maturity of a scenario. By introducing a calculation formula for the historical exposure frequency index of a scenario and combining it with the exposure frequency threshold required for scenario maturity, it can transform raw, potentially highly variable scenario exposure data into a standardized, physically meaningful maturity indicator. In particular, the use of a logarithmic function for normalization effectively mitigates the excessive influence of extreme exposure data on maturity assessment, making the maturity assessment more stable and reasonable. At the same time, by using a min function to limit the upper limit of the index, it ensures that the historical maturity coefficient of the scenario remains within a controllable and meaningful range (0 to 1), avoiding calculation deviations caused by data anomalies. This precise and robust historical maturity coefficient of the scenario can more accurately reflect the actual maturity level of the scenario, thus providing a reliable input for the calculation of the aforementioned target scenario similarity threshold. This allows for more refined and adaptive adjustment of the similarity threshold, thereby optimizing the scenario-based analysis and delivery effects of similar advertisements.

[0119] Regarding step S500, the step of calculating the target scene similarity threshold includes:

[0120] The system acquires a basic similarity threshold, a secondary correction coefficient, and a scenario historical maturity coefficient. The basic similarity threshold is a preset similarity threshold used as an adjustment benchmark. It can be an initial value determined based on historical experience data, expert knowledge, or system default configurations. For example, it can be set as the average similarity requirement for a general scenario or set according to the overall advertising strategy. The secondary correction coefficient reflects a comprehensive assessment of the system's health and ad inventory supply. It can be calculated in real-time by a dedicated system monitoring module, for example, by analyzing indicators such as server load, response time, ad creative quantity, and fill rate. This coefficient can be obtained through real-time calculation by an independent evaluation module or from a pre-trained model. The scenario historical maturity coefficient characterizes the historical performance and maturity of a specific advertising scenario. It can be calculated based on data such as historical exposures, click-through rates, and conversion rates for that scenario, for example, by periodically updating it through a scenario historical data analysis module. This coefficient can be obtained through statistical analysis of historical data or by predicting the long-term performance of the scenario using a machine learning model.

[0121] Substituting the basic similarity threshold, secondary correction coefficient, and scene historical maturity coefficient into the formula Obtain the target scene similarity threshold ,in, Based on the basic similarity threshold, To adjust the sensitivity coefficient, The weighting is affected by the secondary correction coefficient. This is the secondary correction coefficient. This represents the historical maturity coefficient of the scenario.

[0122] The specific explanation of the above calculation process is as follows: Substitute the basic similarity threshold, the secondary correction coefficient, and the scene historical maturity coefficient into the formula. Obtain the target scene similarity threshold ,in, The basic similarity threshold is, To adjust the sensitivity coefficient, used to control the degree of response of the target similarity threshold to changes in the correction term, for example, when When the value is large, the threshold will change more drastically with the correction term; when When the value is small, the threshold changes more gradually. This coefficient is usually determined through system tuning or A / B testing to achieve the best adjustment effect; The secondary correction coefficient is weighted to balance the relative importance of the secondary correction coefficient and the scenario historical maturity coefficient in the target threshold calculation. For example, when system health and ad supply (reflected by the secondary correction coefficient) are considered more important, the weight can be increased. The value of can be reduced; conversely, if the historical performance of a scene (reflected by the scene's historical maturity coefficient) receives more attention, then the value of can be reduced. The value of this weight can be configured based on business needs and strategies, or dynamically adjusted through algorithm optimization. This is the secondary correction coefficient. The calculation step, which substitutes the aforementioned coefficients into the formula, aims to use a mathematical model to comprehensively calculate the three key coefficients obtained above, thereby obtaining a target scene similarity threshold that accurately reflects the current system state, advertising supply, and historical scene performance. The target scene similarity threshold is the final similarity threshold obtained after comprehensive consideration and precise calculation, used to guide ad matching and delivery; it is a dynamically adjusted value that can adaptively change based on real-time system state and historical data.

[0123] Through the above technical solution, this invention provides a quantitative and controllable method for calculating the similarity threshold of a target scene. By introducing a comprehensive mathematical formula, it can accurately integrate the basic similarity threshold, the secondary correction coefficient reflecting the real-time state of the system and the advertising supply capacity, and the scene historical maturity coefficient characterizing the long-term performance of the scene. This solves the technical problem of how to scientifically and dynamically determine the similarity threshold in complex advertising scenarios. Specifically, this solution makes the adjustment of the similarity threshold no longer dependent on a single indicator or empirical judgment, but based on the synergistic effect of multi-dimensional information. The introduction of the influence weights of the adjustment sensitivity coefficient and the secondary correction coefficient further enhances the flexibility and adaptability of threshold adjustment, enabling the system to intelligently balance the importance of real-time performance and historical experience according to actual operating conditions and business strategies. This ensures that the calculated target scene similarity threshold can more accurately reflect the current environment, thereby optimizing the advertising matching effect, improving the accuracy and efficiency of advertising delivery, and avoiding the waste of system resources or insufficient advertising supply caused by improper threshold settings.

[0124] Figure 2 The diagram shows the structural composition of a similar advertising scenario analysis system based on the COSMO algorithm. In this embodiment of the invention, a similar advertising scenario analysis system based on the COSMO algorithm, system 10, includes:

[0125] The data vectorization module 11 is used to vectorize advertisements and scenarios based on the same shared semantic space to obtain advertisement vectors and semantic vectors.

[0126] Vector comparison module 12 is used to compare ad vectors and semantic vectors, determine vector similarity, select ads whose vector similarity reaches a preset similarity threshold, and execute the delivery process.

[0127] The similarity threshold is a dynamic threshold, and the determination process is as follows:

[0128] Acquire system operation status monitoring data, and calculate the system health factor based on the monitoring data;

[0129] Obtain advertising resource supply data, and calculate the advertising inventory supply coefficient based on the advertising resource supply data;

[0130] The secondary correction coefficient is calculated based on the system health factor and the advertising inventory supply coefficient.

[0131] Obtain historical exposure data for the target advertising scenario, and calculate the scenario's historical maturity coefficient based on the historical exposure data;

[0132] Based on the preset basic similarity threshold, the secondary correction coefficient, and the scene historical maturity coefficient, the target scene similarity threshold is calculated, and the current scene similarity threshold is adjusted to the target scene similarity threshold.

[0133] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for contextual analysis of similar advertisements based on the COSMO algorithm, characterized in that, The method includes: Based on the same shared semantic space, ads and scenarios are vectorized to obtain ad vectors and semantic vectors; Compare ad vectors and semantic vectors to determine vector similarity, select ads whose vector similarity reaches a preset similarity threshold, and execute the delivery process; The similarity threshold is a dynamic threshold, and the determination process is as follows: Acquire system operation status monitoring data, and calculate the system health factor based on the monitoring data; Obtain advertising resource supply data, and calculate the advertising inventory supply coefficient based on the advertising resource supply data; The secondary correction coefficient is calculated based on the system health factor and the advertising inventory supply coefficient. Obtain historical exposure data for the target advertising scenario, and calculate the scenario's historical maturity coefficient based on the historical exposure data; Based on the preset basic similarity threshold, the secondary correction coefficient, and the scene historical maturity coefficient, the target scene similarity threshold is calculated, and the current scene similarity threshold is adjusted to the target scene similarity threshold.

2. The similar advertising scenario analysis method based on the COSMO algorithm according to claim 1, characterized in that, The steps of acquiring system operation status monitoring data and calculating the system health factor based on the monitoring data include: Get the current number of queries per second, 95th percentile request response time, CPU utilization, memory utilization, and real-time similarity calculation time; Calculate the system ingress pressure coefficient based on the current number of queries per second and the 95th percentile request response time; Calculate the resource consumption coefficient based on CPU utilization, memory utilization, and real-time similarity calculation time; Based on the combined factors of the system inlet pressure coefficient and the computational resource consumption coefficient, a system health factor is determined. The system health factor is negatively correlated with both the system inlet pressure coefficient and the computational resource consumption coefficient. The higher the health factor, the healthier the system.

3. The similar advertising scenario analysis method based on the COSMO algorithm according to claim 2, characterized in that, The steps for calculating the system inlet pressure coefficient include: Get the current number of queries per second and the 95th percentile request response time; Perform max-min normalization on the current number of queries per second to obtain the current query per second index; The target response time threshold is compared with the 95th percentile request response time, and the complement of the ratio is taken as the request response time index. The system entry pressure coefficient is obtained by weighted average summation of the current query count per second index and request response time index. The system entry pressure coefficient is directly proportional to both the current query count per second index and request response time index.

4. The similar advertising scenario analysis method based on the COSMO algorithm according to claim 3, characterized in that, The steps for calculating the resource consumption coefficient include: Obtain CPU utilization, memory utilization, and real-time similarity calculation time; The system resource stress index is obtained by taking the larger value of CPU utilization and memory utilization and comparing it with the maximum utilization. The similarity calculation time is calculated by comparing the real-time similarity calculation time with the maximum allowed calculation time threshold to obtain the similarity calculation time index. The larger value between the system resource tension index and the similarity calculation time index is taken as the resource consumption coefficient. Calculate the resource consumption coefficient Output range 0-1, calculate resource consumption coefficient. The closer the value is to 1, the more strained the computing resources are.

5. The similar advertising scenario analysis method based on the COSMO algorithm according to claim 1, characterized in that, The steps for calculating the advertising inventory supply coefficient include: Get the number of available ad creatives and ad fill rate; The available ad creative index is obtained by comparing the number of available ad creatives with the minimum number of creatives required. The advertising inventory supply coefficient is calculated by weighted linear combination of the available advertising creativity index and the advertising fill rate. The value range of the advertising inventory supply coefficient is constrained to be between 0 and 1, and the larger the value, the more abundant the advertising supply. Both the available advertising creativity index and the advertising fill rate have a positive contribution to the supply coefficient.

6. The similar advertising scenario analysis method based on the COSMO algorithm according to claim 1, characterized in that, The steps for calculating the secondary correction coefficient include: Obtain the system health factor and the advertising inventory supply coefficient; Substitute the system health factor and the advertising inventory supply coefficient into the formula. Obtain the secondary correction coefficient Secondary correction coefficient The output range is 0-1, and the closer it is to 1, the more ideal the system state. As a system health factor, This is the advertising inventory supply coefficient.

7. The similar advertising scenario analysis method based on the COSMO algorithm according to claim 1, characterized in that, The steps for calculating the historical maturity coefficient of a scenario include: Obtain the historical exposure count of the scene; The scene historical exposure count is processed by a nonlinear normalization function to obtain the scene historical exposure count index, which is positively correlated with the historical exposure count. After setting the upper limit of the historical exposure count index of the scene to 1 using the min function, the historical maturity coefficient of the scene is obtained.

8. The similar advertising scenario analysis method based on the COSMO algorithm according to claim 1, characterized in that, The steps for calculating the target scene similarity threshold include: Obtain the basic similarity threshold, secondary correction coefficient, and scene historical maturity coefficient; Substituting the basic similarity threshold, secondary correction coefficient, and scene historical maturity coefficient into the formula Obtain the target scene similarity threshold ,in, Based on the basic similarity threshold, To adjust the sensitivity coefficient, The weighting is affected by the secondary correction coefficient. This is the secondary correction coefficient. This represents the historical maturity coefficient of the scenario.

9. A similar advertising scenario analysis system based on the COSMO algorithm, the system being used to implement the similar advertising scenario analysis method based on the COSMO algorithm as described in any one of claims 1 to 8, characterized in that, The system includes: The data vectorization module is used to vectorize advertisements and scenarios based on the same shared semantic space to obtain advertisement vectors and semantic vectors. The vector comparison module is used to compare ad vectors and semantic vectors, determine vector similarity, select ads whose vector similarity reaches a preset similarity threshold, and execute the delivery process. The similarity threshold is a dynamic threshold, and the determination process is as follows: Acquire system operation status monitoring data, and calculate the system health factor based on the monitoring data; Obtain advertising resource supply data, and calculate the advertising inventory supply coefficient based on the advertising resource supply data; The secondary correction coefficient is calculated based on the system health factor and the advertising inventory supply coefficient. Obtain historical exposure data for the target advertising scenario, and calculate the scenario's historical maturity coefficient based on the historical exposure data; Based on the preset basic similarity threshold, the secondary correction coefficient, and the scene historical maturity coefficient, the target scene similarity threshold is calculated, and the current scene similarity threshold is adjusted to the target scene similarity threshold.

Images