Intelligent delivery targeting penetration decision method for precision marketing

Abstract

The application discloses a smart delivery targeting penetration decision method for accurate marketing, and particularly relates to the technical field of smart delivery targeting penetration decision, and comprehensively identifies potential abnormalities through unified processing of a standardized index set and dynamic association construction of a multistage threshold value; with the aid of the constraint of a delivery strategy shock index on delivery behavior, smooth adjustment of dynamic bidding, material optimization and traffic distribution can be realized in a fluctuating environment, the controllability of budget consumption and the continuity of delivery effect are ensured, combined with multi-dimensional penetration strategies of competition, scene, crowd and material, high-value user groups are more accurately reached and material matching degree is improved, so that the return on investment and conversion rate remain stable in growth, shock and resource waste in the delivery process are reduced, and the cross-platform collaborative optimization capability and risk prevention and control level are improved.

Description

Technical Field

[0001] This invention relates to the field of intelligent targeted penetration decision-making technology, and more specifically, to an intelligent targeted penetration decision-making method for precision marketing. Background Technology

[0002] In existing intelligent advertising systems, predictive models are typically used to forecast trends in key metrics such as Return on Investment (ROI) and conversion rates, allowing the system to adjust budgets and bids in advance. However, under sudden traffic fluctuations or special marketing events, historical data-driven predictive models often become distorted, leading to overestimation of future ROI. When the system increases its spending based on inaccurate predictions, it rapidly escalates budget consumption and bid levels. Simultaneously, the risk control module provides protection based on multiple real-time monitored thresholds (such as excessively rapid consumption, substandard conversion results, and insufficient account balance). Under conditions of a budget surge, these thresholds are easily triggered simultaneously. Thus, the overly optimistic prediction layer and the overly restrictive risk control layer create a contradiction: the former pushes the system to expand spending, while the latter forcibly freezes or suspends plans. This combination directly leads to drastic fluctuations in the advertising strategy, manifesting as frequent switching between "increased spending—forced suspension" within a short period. This not only disrupts the continuity and convergence of the campaign but also causes drastic fluctuations in ROI and waste of resources, potentially even missing crucial traffic windows. Summary of the Invention

[0003] In order to overcome the above-mentioned deficiencies of the prior art, embodiments of the present invention provide an intelligent targeting and penetration decision-making method for precision marketing, so as to solve the problems mentioned in the background art.

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

[0005] The intelligent targeting and penetration decision-making method for precision marketing includes the following steps:

[0006] By acquiring campaign performance, traffic, creative performance, and account health metrics through cross-platform interfaces, behavioral log collection, and third-party user tag data, the acquired data is time-aligned, formatted, and missing values ​​are filled in to generate a standardized set of metrics.

[0007] Multi-level threshold checks are applied to the standardized indicator set to identify abnormal states, and dynamic correlations between indicators are constructed based on the abnormality markers to form an indicator correlation information set;

[0008] Based on the indicator correlation information set and historical time series data, combined with periodic characteristics and external factors, the short-term trend of the indicator is predicted, and information on prediction inaccuracies, sudden fluctuations and risk control trigger conflicts is obtained to generate the oscillation index of the investment strategy.

[0009] Based on the trend prediction of indicators, the correlation information of indicators and the oscillation index of the advertising strategy, the bids of each channel are dynamically adjusted, high-quality creatives are selected for reuse or new creatives are generated, and A / B testing plans are automatically configured, while the traffic ratio of each channel is adjusted.

[0010] By combining user behavior characteristics, platform traffic distribution, and interest tags, competitive penetration, scenario penetration, audience penetration, and creative penetration are executed, and the data on the performance of the campaign is fed back to update the standardized indicator set.

[0011] In a preferred embodiment, multi-level threshold checks are applied to the standardized indicator set to identify abnormal states, and dynamic correlations between indicators are constructed based on the abnormality markers to form an indicator correlation information set, as detailed below:

[0012] The standardized index set is decomposed into a multi-scale time series to generate a set of time series components.

[0013] Calculate adaptive thresholds for the time series component set, and output the threshold set and threshold confidence.

[0014] A multi-strategy detector is applied in parallel to the threshold set and a standardized metric set is used for detection, outputting the original anomaly label set;

[0015] Confidence fusion and multi-level classification are performed on the original anomaly label set to generate a multi-level anomaly label set;

[0016] Perform temporal aggregation and denoising on the multi-level anomaly marker set to generate a cleaned set of anomaly time periods and an anomaly co-occurrence statistics table;

[0017] For the cleaned set of abnormal time periods and the standardized index set aligned with time, calculate the conditional coherence and time-delay correlation coefficients, and output the initial candidate set of associations:

[0018] Perform causal directionality verification and counterfactual perturbation test on the initial candidate set of associations to generate a weighted list of directed association edges;

[0019] The weighted directed edge list is graph-based, dynamically weighted and smoothed, and anomaly propagation capabilities are calculated to output the final set of index association information.

[0020] In a preferred embodiment, based on the indicator association information set and historical time-series data, combined with periodic characteristics and external factors, the short-term trend of the indicator is predicted, as follows:

[0021] The indicator association information set, historical time series data, periodic features and external factors are merged according to a unified time benchmark, and data integrity verification, missing data marking and delay completion are performed to form an enhanced time series dataset with source traceability;

[0022] The enhanced time series dataset is decomposed into multiple scales to extract long-term trends, periodic components and short-term residuals respectively. Anomaly window masking and noise suppression are applied to the residuals, and the decomposed component set is output.

[0023] Using the decomposed component set and the indicator association information set as input, propagation-weighted cross-indicator lag features, short-term residual features, periodic embedding features and external event codes are constructed according to the association edge weights and lag candidate list; thus forming a feature matrix for modeling.

[0024] The feature matrix is ​​fed into the predictor set for parallel inference and rolling adaptation, and the original predicted value sequence and the error distribution of each predictor are output.

[0025] Using the original predicted value sequence and the error distribution of each predictor as input, backtesting residual calibration, short-term bias correction and confidence interval estimation are performed, and the calibrated short-term prediction package is output.

[0026] In a preferred embodiment, the prediction inaccuracy and sudden fluctuation information includes the following indicators: return on investment, conversion rate, consumption rate, click-through rate, and material performance deviation.

[0027] In a preferred embodiment, the risk control trigger conflict information includes the following indicators: budget consumption, insufficient balance, and event sequences triggered by multiple thresholds simultaneously.

[0028] In a preferred embodiment, the deployment strategy oscillation index is obtained by weighted summation of each indicator among prediction inaccuracy, sudden fluctuation information, and risk control trigger conflict information.

[0029] The technical effects and advantages of this invention are as follows:

[0030] 1. This invention introduces a campaign strategy oscillation index into the context of precision marketing, which explicitly quantifies the conflict information between the prediction layer and the risk control layer and integrates it into dynamic decision-making. This enables the system to balance foresight and robustness when facing sudden traffic fluctuations and special marketing events, thereby avoiding overexpansion due to inaccurate predictions and frequent freezes due to tightened risk.

[0031] 2. This invention, through the unified processing of standardized indicator sets and the dynamic correlation of multi-level thresholds, enables the system to comprehensively identify potential anomalies and enhance its adaptability to complex scenarios by incorporating periodic characteristics and external events in trend prediction. By leveraging the constraint of the campaign strategy oscillation index on campaign behavior, dynamic bidding, creative optimization, and traffic allocation can be smoothly adjusted in volatile environments, ensuring controllable budget consumption and continuous campaign performance. Simultaneously, by combining multi-dimensional penetration strategies based on competition, scenarios, audiences, and creatives, the system can more accurately reach high-value user groups and improve creative matching, allowing ROI and conversion rates to maintain stable growth. This significantly reduces oscillations and resource waste during the campaign process and enhances cross-platform collaborative optimization capabilities and risk control levels, thus achieving efficient utilization of marketing resources and continuous improvement in ROI even in uncertain environments. Attached Figure Description

[0032] To facilitate understanding by those skilled in the art, the present invention will be further described below with reference to the accompanying drawings;

[0033] Figure 1 This is a flowchart of a method according to an embodiment of the present invention. Detailed Implementation

[0034] 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.

[0035] Example: Figure 1 This invention presents an intelligent targeted and penetrating decision-making method for precision marketing, comprising the following steps:

[0036] By acquiring campaign performance, traffic, creative performance, and account health metrics through cross-platform interfaces, behavioral log collection, and third-party user tag data, the acquired data is time-aligned, formatted, and missing values ​​are filled in to generate a standardized set of metrics.

[0037] Multi-level threshold checks are applied to the standardized indicator set to identify abnormal states, and dynamic correlations between indicators are constructed based on the abnormality markers to form an indicator correlation information set;

[0038] Based on the indicator correlation information set and historical time series data, combined with periodic characteristics and external factors, the short-term trend of the indicator is predicted, and information on prediction inaccuracies, sudden fluctuations and risk control trigger conflicts is obtained to generate the oscillation index of the investment strategy.

[0039] Based on the trend prediction of indicators, the correlation information of indicators and the oscillation index of the advertising strategy, the bids of each channel are dynamically adjusted, high-quality creatives are selected for reuse or new creatives are generated, and A / B testing plans are automatically configured, while the traffic ratio of each channel is adjusted.

[0040] Combine user behavior characteristics, platform traffic distribution and interest tags to perform competitive penetration, scenario penetration, audience penetration and creative penetration, and feed back the data of the campaign performance to update the standardized indicator set;

[0041] We acquire metrics such as campaign performance, traffic, creative performance, and account health by using cross-platform interfaces, behavioral log collection, and third-party user tagging data. The acquired data is then time-aligned, formatted, and missing values ​​are filled in to generate a standardized set of metrics, as detailed below:

[0042] Retrieve raw event stream: Retrieve raw event streams through platform interface queries, real-time log collectors, and third-party data export interfaces, including campaign performance, traffic, creative performance, and account health metrics, and generate raw event streams with source timestamps;

[0043] The metrics for campaign performance include: return on investment, cost of conversion, conversion rate, and engagement rate.

[0044] The traffic metrics include: impressions, clicks, and consumption rate;

[0045] The performance metrics for the materials include: material competitiveness score, completion rate, and interaction rate;

[0046] The account health metrics include: budget utilization rate, balance adequacy, and planned investment quantity.

[0047] Buffer and tag the event stream: Write the raw event stream to a low-latency buffer layer and add acquisition metadata (data source identifier, reception time, original timestamp, access batch ID) to each event to generate a buffered event stream with metadata;

[0048] Unify time base and align buffered event streams: Normalize and reorder buffered event streams based on a unified time base (e.g., UTC or time zone specified by advertising campaign), handle cross-source clock drift and time zone differences, and generate time-aligned event sequences;

[0049] Mapping identifiers and deduplicating time-aligned event sequences: Perform identifier mapping and deduplication on time-aligned event sequences. Unify cross-platform identifiers through user ID mapping table, device fingerprint or probability matching, and merge duplicate events according to the same request ID or time window rules to generate a set of deduplicated and uniformly identified event records.

[0050] Associating the attribution window with the event record set for attribution: The event record set that has been deduplicated and uniformly identified is marked with event attribution according to business attribution rules (such as click attribution window, view attribution window, attribution priority), forming an event record set with attribution attributes;

[0051] Complete the event set for delayed attribution and merge late events: Apply a late data completion strategy (based on watermark / watermark judgment, incremental backfilling and confidence decay) to the event record set with attribution attributes, merge delayed conversion or cost events into the corresponding attribution records, and generate an event set with enhanced integrity.

[0052] Standardized fields and semantic mapping enhance the integrity of event sets: Perform field mapping, type conversion and enumeration normalization on the event sets with enhanced integrity according to a unified semantic dictionary (e.g., map "impression / cnt" on different platforms to a unified "exposure" field), and generate a semantically consistent intermediate indicator set;

[0053] Quality verification and intelligent correction of intermediate indicator set: Perform quality checks (range test, mutation detection, cross-source verification) on intermediate indicator set, and correct abnormal or missing items according to priority strategy: first, trace back to the source and re-pull; second, use the nearest window interpolation or regression based on similar channels to fill in; finally, label the confidence level and retain the correction log to generate a standardized indicator set with confidence level and traceability record.

[0054] Package and write standardized indicator sets into feature storage and publish them for downstream use: Write standardized indicator sets with confidence and source records into queryable feature storage (supporting time window retrieval and incremental updates), while updating the metadata directory and version number, and output standardized indicator sets that can be directly called.

[0055] Multi-level threshold checks are applied to the standardized indicator set to identify abnormal states, and dynamic correlations between indicators are constructed based on the anomaly markers to form an indicator correlation information set, as detailed below:

[0056] Multi-scale time-series decomposition of the standardized indicator set generates a set of time-series components (trend, seasonality, residuals): Multi-scale decomposition decomposes the standardized indicator set into long-term trends, periodic components and short-term residuals, providing clear objects for subsequent threshold setting and noise suppression.

[0057] Adaptive thresholds are calculated for the time series component set, and the threshold set and threshold confidence scores are output: short-term thresholds and long-term thresholds are generated based on quantiles, decomposed seasonal cycles and rolling baselines, and confidence scores are calculated for each threshold (the confidence scores are obtained by weighted summation of the standard deviation of historical indicators and sample size).

[0058] A multi-strategy detector is applied in parallel to the threshold set and a standardized index set is used for detection. The output is the original anomaly label set and the scores of each detector. The multi-strategy detectors, including robust Z-score (based on MAD), EWMA / CUSUM control chart, change point detection (such as mutation point detection), and prediction residual detection (prediction model residual exceeding the limit), are used in parallel to score the standardized index set under the threshold set constraints and produce the original anomaly labels (including detector source and score).

[0059] The original anomaly label set is subjected to confidence fusion and multi-level classification to generate a multi-level anomaly label set (including start / end time, severity, detector votes and confidence). The output of each detector is weighted and confidence fusion strategy to generate three levels of anomaly labels: mild, moderate and severe. The detector consistency, duration and confidence curve are recorded for each anomaly label.

[0060] Perform time-series aggregation and denoising on the multi-level anomaly marker set to generate a cleaned set of anomaly time periods and an anomaly co-occurrence statistics table: merge close anomaly events according to a preset merging window, remove isolated short-term noise events, count the number of co-occurrences and co-occurrence sequence patterns of different indicators in the same time window, and output the cleaned set of anomaly time periods and an anomaly co-occurrence statistics table.

[0061] For the cleaned set of abnormal time periods and the standardized index set aligned with time, calculate the conditional synergy and time-delay correlation coefficients, and output the initial candidate set of associations (including time delay, association direction candidates, and correlation strength):

[0062] The logic for obtaining the conditional coordination quantity is as follows:

[0063] Choose a probability density estimation strategy for continuous variables (e.g., isometric histogram estimation, kernel density estimation, or parameter estimation based on the Gaussian assumption), and estimate the marginal and joint probability densities for index A, index B, and condition set Z on the sample set, respectively.

[0064] Based on the estimated probability density, calculate the following entropy terms (described in words as "calculate entropy"):

[0065] Calculate the conditional entropy for the condition set Z;

[0066] Calculate the joint entropy of the joint distribution of index A and condition set Z;

[0067] Calculate the joint entropy of the joint distribution of index B and condition set Z;

[0068] Calculate the joint entropy of the ternary joint distribution of index A, index B, and condition set Z;

[0069] (The calculations for each of the above items are performed by discretizing the sample probability using the selected density estimation method or by approximating it with continuous entropy, and then summing / integrating according to the definition of entropy to obtain the numerical value.)

[0070] The above entropy quantities are numerically synthesized according to the "algebraic combination of conditional mutual information": that is, the numerical expression of the conditional synergy quantity is obtained by the sum and difference of the corresponding entropy terms (the entropy terms are used as intermediate quantities for algebraic combination, and the numerical value is the conditional synergy quantity).

[0071] The logic for obtaining the time-delay correlation coefficient is as follows:

[0072] Set the hysteresis scan range (the maximum duration of positive and negative hysteresis ranges, for example, ±N time steps), and determine the hysteresis step size;

[0073] For each pair of indicators A and B, and for each lag value τ, calculate the time-lag correlation values ​​according to the following steps:

[0074] a. Shift the time series of indicator A and the time series of indicator B relative to each other on the time axis by τ (if τ is positive, the B series shifts backward relative to A; if τ is negative, it shifts forward).

[0075] b. After shifting, only the overlapping time window portion of the two sequences is retained, and the mean is removed from the overlapping segments (the sample mean of each overlapping segment is subtracted respectively).

[0076] c. Calculate the sample covariance of the overlapping segment, which is the sample mean of the product of the two sequences in the overlapping segment after removing the mean.

[0077] d. Calculate the sample standard deviation of each overlapping segment;

[0078] e. Divide the covariance by the product of the two sample standard deviations to obtain the normalized correlation coefficient value under the lag τ;

[0079] For each lag τ, repeat the calculation steps of the time lag correlation value for all abnormal time periods and summarize the corresponding lag correlation coefficient according to the predetermined aggregation strategy (e.g., taking the average or weighted average, where the weight can be based on the severity of the abnormality or the segment length).

[0080] Perform causal directionality verification and counterfactual perturbation test on the initial candidate association set to generate a weighted list of directed association edges (including direction, lag, weight and confidence). Adopt causal determination methods (such as Granger-type determination or constraint test based on time series prediction residuals) for candidate associations and implement small-amplitude counterfactual perturbation simulation within a controlled historical window to verify the downstream response. Combine the results of the two methods to assign directionality, lag value, weight and confidence to each candidate association.

[0081] The weighted directed edge list is graph-based, dynamically weighted smoothed, and anomaly propagation capability is calculated. The final indicator association information set (including nodes, weighted directed edges, lag, positive and negative relationships, confidence, anomaly propagation score, and most recent trigger time) is output. The weighted edges are written into a time series graph database, and exponential smoothing is used to fuse historical weights with the current anomaly weighted update. The influence propagation simulation is performed on the graph to calculate the anomaly propagation score of each edge, and finally, an indicator association information set that can be used for decision-making is formed.

[0082] Based on the indicator correlation information set and historical time series data, combined with periodic characteristics and external factors, the short-term trend of the indicator is predicted as follows:

[0083] The indicator association information set, historical time series data, periodic features (hourly / daily / weekly codes) and external factors (holidays, promotional plans, market event identifiers, etc.) are merged according to a unified time benchmark, and data integrity verification, missing data marking and delay completion are performed to form an enhanced time series dataset with source traceability and confidence labeling.

[0084] The enhanced time series dataset is decomposed into multiple time series at different scales to extract long-term trends, periodic components and short-term residuals. Anomaly window masking and noise suppression are applied to the residuals, and the decomposed component set (trend / season / residual) and its time window label are output.

[0085] Using the decomposed component set and the index association information set as input, propagation-weighted cross-index lag features, short-term residual features, periodic embedding features and external event codes are constructed according to the association edge weights and lag candidate list; a feature matrix for modeling is formed with feature confidence scores attached.

[0086] The feature matrix is ​​fed into the predictor set for parallel inference and rolling adaptation: complementary short-term pattern predictors (local trend extrapolators, residual predictors, time series learners and tree regressors, etc.) are trained / updated online on the rolling window, and the weights of each predictor are dynamically adjusted based on the performance of the most recent validation window, outputting the original predicted value sequence and the error distribution of each predictor.

[0087] Using the original predicted value sequence and the error distribution of each predictor as input, backtesting residual calibration, short-term bias correction and confidence interval estimation are performed, and the calibrated short-term prediction package is output.

[0088] Acquire information on forecast inaccuracies, sudden fluctuations, and risk control-triggered conflicts to generate a volatility index for investment strategies.

[0089] The information on prediction inaccuracies and sudden fluctuations includes the following indicators: return on investment, conversion rate, consumption rate, click-through rate, and deviation in creative performance;

[0090] The return on investment (ROI) is used to measure the ratio of the economic benefits generated by advertising to the cost of investment. The final ROI is obtained by calculating the total revenue generated by advertising, calculating the corresponding advertising expenditure, and then normalizing the ratio of revenue to expenditure.

[0091] The conversion rate is used to measure the proportion of users who take a target action (such as purchase, registration, or download) after being reached by an advertisement; the conversion rate is obtained by counting the number of users who complete the target action within a fixed time period and dividing it by the total number of users reached by the advertisement.

[0092] The consumption rate is used to measure the speed and efficiency of advertising budget usage. It is obtained by statistically analyzing the actual amount of advertising budget spent within a fixed time period and comparing it with the historical average consumption.

[0093] The click-through rate (CTR) is used to measure users' interest in and interaction with an advertisement. It is obtained by counting the number of times an advertisement is clicked within a fixed time period and dividing the number of times the advertisement is displayed.

[0094] The performance deviation of the advertising material is used to measure the deviation of the overall performance of the advertising material in various indicators from the historical or expected baseline. It is obtained by first extracting the core performance indicators (completion rate, interaction rate, click rate, etc.) of each material, comparing the difference between the current core performance indicators of each material and the historical average, and then summing them up to obtain the performance deviation value of the material.

[0095] The risk control trigger conflict information includes the following indicators: budget consumption, insufficient balance, and event sequences triggered by multiple thresholds simultaneously;

[0096] The budget consumption is used to measure the progress and scale of advertising budget usage within the campaign period. By calculating the advertising expenses consumed within a fixed campaign period and comparing them with the budget allocated for that period, the budget consumption ratio is obtained and used as a representative indicator of budget consumption.

[0097] The term "insufficient balance" is used to measure the adequacy of the remaining budget in an account relative to future spending needs. By statistically analyzing the available balance of the account in real time and comparing it with the estimated future spending needs, when the balance is less than the needs, it is marked as insufficient balance, and the ratio of available balance to spending needs is calculated as a representative indicator of insufficient balance.

[0098] The event sequence of multiple thresholds being triggered simultaneously is used to measure the situation where multiple risk thresholds are triggered simultaneously in the same time window during the deployment process. It is achieved by statistically analyzing the threshold triggering events of two or more thresholds within the same window in consecutive time windows, forming an event sequence from the consecutively occurring triggering records, and using the total number of threshold triggering events in the event sequence as a representative indicator of the event sequence of multiple thresholds being triggered simultaneously.

[0099] The oscillation index of the deployment strategy is obtained by weighted summation of each indicator in the information of prediction inaccuracy, sudden fluctuations, and risk control triggering conflict information.

[0100] Based on the predicted trends of indicators, indicator correlation information, and the oscillation index of the advertising strategy, the bids for each channel are dynamically adjusted, high-quality creatives are selected for reuse or new creatives are generated, and A / B testing plans are automatically configured. At the same time, the traffic distribution of each channel is adjusted, as follows:

[0101] Receive calibrated short-term forecast packages, indicator correlation information sets, campaign strategy oscillation index, current account budget and campaign constraints, real-time performance snapshots of each channel and historical bidding backtracking data, and merge them to generate a decision context package with confidence level and version number;

[0102] Using the decision context package as input, candidate actions are generated for each advertising campaign and channel, including: differentiated bid adjustment instructions (increase / decrease percentage, target bid range and minimum modification granularity), creative operation instructions (reuse of high-quality creatives, new product launch suggestions, variant generation requests), traffic allocation instructions (channel share adjustment, time period reallocation), and candidate A / B experiment configurations (split ratio, grouping key, early stop rule); the output is a set of executable candidate actions with metadata (priority, expected impact, minimum execution unit).

[0103] Using candidate action sets, historical bidding backtracking data, real-time traffic prediction, and the oscillation index of the delivery strategy as input, a rapid simulation (including shadow auction replay, budget burn prediction, and conversion link response simulation) is performed in a low-cost shadow environment. The estimated revenue, budget consumption curve, and trigger risk threshold probability of each candidate action are obtained, and they are weighted and summed to obtain the candidate action score. The output is a candidate action score table with estimated effect and risk score.

[0104] Using the candidate action scoring table as input, actions are selected according to multi-objective strategies (e.g., prioritizing ROI stability, short-term conversion, and long-term value) and anti-vibration strategies (using hysteresis suppression, momentum buffering, and upper limit of change amplitude driven by the oscillation index of the deployment strategy). An execution plan is generated in stages and time sequence (start time, duration, rate limit), and a deployable action plan with rollback conditions and observation points is output.

[0105] Taking the action plan as input, the system issues price change instructions, material delivery commands, and traffic allocation commands according to the plan through the channel interface, and simultaneously creates micro-experiment (A / B) configurations: traffic splitting key, group ratio, early stop and statistical effectiveness threshold, obfuscation control and sample size requirements; at the same time, it records the version, reason and expected indicators of each change for auditing and traceability, and outputs the execution change event flow and experiment ID;

[0106] By combining user behavior characteristics, platform traffic distribution, and interest tags, competitive penetration, scenario penetration, audience penetration, and creative penetration are executed, and the performance data is fed back to update the standardized indicator set, as detailed below:

[0107] The processed user behavior characteristics (access frequency, nearest path, conversion probability, value estimation, etc.), platform traffic distribution (time period traffic curve, bidding depth, winning price distribution), interest tag set, and current creative library, budget constraints and decision signals (predicted trends, indicator correlation information, and campaign strategy oscillation index) are merged to generate a campaign execution context package with timestamp, confidence level and version number.

[0108] Using the delivery execution context package as input, candidate actions are generated for competition penetration, scene penetration, audience penetration, and creative penetration respectively:

[0109] Competitive penetration candidates (such as time-based premiums, automatic bid upper / lower limits, preemptive bid windows);

[0110] Candidates for scenario penetration (such as weighted delivery during a specified time period, or scenario-triggered centralized overload delivery plans);

[0111] Audience penetration candidate (such as interest tag combination package, similarity expansion rules and exclusion list);

[0112] Material penetration candidates (such as format priority mapping, creative variant priority and personalized templates);

[0113] Obtain a list of candidate actions with priority, expected impact estimate, and minimum execution unit;

[0114] The candidate action list is used as input to perform frequency limits, budget rate limits, privacy protection verification (edge ​​hashing and homomorphic matching rules), anti-fraud checks and compliance checks of the delivery strategy volatility index. Non-compliant / high-risk actions are eliminated or downgraded and a cleaned list of executable actions and security labels are generated.

[0115] Taking the list of executable actions as input, the creative template is mapped to a platform-compatible creative package according to the display specifications and format requirements of each platform. The personalized placeholders are replaced and variants (including downgraded and rollback versions) are generated. At the same time, the routing rules (target audience, priority time period, frequency limit, A / B grouping key) of each creative package are generated, and the creative packages to be distributed and the routing table are output.

[0116] The program takes the creative package to be distributed and the routing table as input, executes bidding instructions, route allocation and creative listing through the channel interface; during the execution process, it collects and records atomic-level event streams (including display / winning / click / conversion events, bid, win rate, actual consumption and timestamp), and outputs a real-time execution event stream containing complete metadata.

[0117] Using real-time execution event streams as input, short-term performance metrics (ROI, conversion rate, click-through rate, consumption rate, win rate, and creative completion / interaction rate) are calculated based on the experiment ID and the split key. Incremental and causal backtesting are performed using the retained control group / time grouping, risk-triggered event logs are supplemented, and a performance summary set with confidence and attribution window annotations is output.

[0118] Using the performance aggregate as input, we first perform late attribution merging and confidence adjustment (backfilling the time window and lowering the confidence for delayed conversions), then write the adjusted metrics into the standardized metrics store, and update user behavior characteristics, creative performance records, platform traffic samples, and risk trigger counts.

[0119] This invention introduces a campaign strategy oscillation index into precision marketing scenarios, explicitly quantifies the conflict information between the prediction layer and the risk control layer, and integrates it into dynamic decision-making. This enables the system to balance foresight and robustness when facing sudden traffic fluctuations and special marketing nodes, thereby avoiding overexpansion due to inaccurate predictions and frequent freezes due to tightened risk.

[0120] This invention, through the unified processing of standardized indicator sets and the dynamic correlation of multi-level thresholds, enables the system to comprehensively identify potential anomalies and enhance its adaptability to complex scenarios by incorporating periodic characteristics and external events in trend prediction. By leveraging the constraint of the campaign strategy oscillation index on campaign behavior, dynamic bidding, creative optimization, and traffic allocation can be smoothly adjusted in volatile environments, ensuring controllable budget consumption and continuous campaign performance. Simultaneously, by combining multi-dimensional penetration strategies based on competition, scenarios, audiences, and creatives, the system can more accurately reach high-value user groups and improve creative matching, allowing ROI and conversion rates to maintain stable growth. This significantly reduces oscillations and resource waste during the campaign process and enhances cross-platform collaborative optimization capabilities and risk control levels, thus achieving efficient utilization of marketing resources and continuous improvement in ROI even in uncertain environments.

[0121] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0122] It should be understood that in the various embodiments of this application, the order of the above-mentioned processes does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0123] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. An intelligent targeted and penetrating decision-making method for precision marketing, characterized by: Includes the following steps: By acquiring campaign performance, traffic, creative performance, and account health metrics through cross-platform interfaces, behavioral log collection, and third-party user tag data, the acquired data is time-aligned, formatted, and missing values ​​are filled in to generate a standardized set of metrics. Multi-level threshold checks are applied to the standardized indicator set to identify abnormal states, and dynamic correlations between indicators are constructed based on the abnormality markers to form an indicator correlation information set; Based on the indicator correlation information set and historical time series data, combined with periodic characteristics and external factors, the short-term trend of the indicator is predicted, and information on prediction inaccuracies, sudden fluctuations and risk control trigger conflicts is obtained to generate the oscillation index of the investment strategy. Based on the trend prediction of indicators, the correlation information of indicators and the oscillation index of the advertising strategy, the bids of each channel are dynamically adjusted, high-quality creatives are selected for reuse or new creatives are generated, and A / B testing plans are automatically configured, while the traffic ratio of each channel is adjusted. Combine user behavior characteristics, platform traffic distribution and interest tags to perform competitive penetration, scenario penetration, audience penetration and creative penetration, and feed back the data of the campaign performance to update the standardized indicator set; Multi-level threshold checks are applied to the standardized indicator set to identify abnormal states, and dynamic correlations between indicators are constructed based on the anomaly markers to form an indicator correlation information set, as detailed below: The standardized index set is decomposed into a multi-scale time series to generate a set of time series components. Calculate adaptive thresholds for the time series component set, and output the threshold set and threshold confidence. A multi-strategy detector is applied in parallel to the threshold set and a standardized metric set is used for detection, outputting the original anomaly label set; Confidence fusion and multi-level classification are performed on the original anomaly label set to generate a multi-level anomaly label set; Perform temporal aggregation and denoising on the multi-level anomaly marker set to generate a cleaned set of anomaly time periods and an anomaly co-occurrence statistics table; For the cleaned set of abnormal time periods and the standardized index set aligned with time, calculate the conditional coherence and time-delay correlation coefficients, and output the initial candidate set of associations: Perform causal directionality verification and counterfactual perturbation test on the initial candidate set of associations to generate a weighted list of directed association edges; The weighted directed edge list is graph-based, dynamically weighted and smoothed, and anomaly propagation capability is calculated to output the final set of index association information. The external factors include holidays, promotional plans, and market event indicators; The information on prediction inaccuracies and sudden fluctuations includes the following indicators: return on investment, conversion rate, consumption rate, click-through rate, and deviation in creative performance; The risk control trigger conflict information includes the following indicators: budget consumption, insufficient balance, and event sequences triggered by multiple thresholds simultaneously; The oscillation index of the deployment strategy is obtained by weighted summation of each indicator in the information of prediction inaccuracy, sudden fluctuations, and risk control-triggered conflict.

2. The intelligent targeted penetration decision-making method for precision marketing according to claim 1, characterized in that: Based on the indicator correlation information set and historical time series data, combined with periodic characteristics and external factors, the short-term trend of the indicator is predicted as follows: The indicator association information set, historical time series data, periodic features and external factors are merged according to a unified time benchmark, and data integrity verification, missing data marking and delay completion are performed to form an enhanced time series dataset with source traceability; The enhanced time series dataset is decomposed into multiple scales to extract long-term trends, periodic components and short-term residuals respectively. Anomaly window masking and noise suppression are applied to the residuals, and the decomposed component set is output. Using the decomposed component set and the index association information set as input, propagation-weighted cross-index lag features, short-term residual features, periodic embedding features and external event codes are constructed according to the association edge weights and lag candidate list to form a feature matrix for modeling. The feature matrix is ​​fed into the predictor set for parallel inference and rolling adaptation, and the original predicted value sequence and the error distribution of each predictor are output. Using the original predicted value sequence and the error distribution of each predictor as input, backtesting residual calibration, short-term bias correction and confidence interval estimation are performed, and the calibrated short-term prediction package is output.

Images