Advertisement material life cycle management system based on smart contract

Abstract

The invention relates to the technical field of advertisement material management, and discloses an advertisement material life cycle management system based on a smart contract. The system comprises a data analysis module, and the data analysis module is used for collecting a creation data set and a historical performance data set of advertisement materials; constructing an advertisement material performance prediction model based on the historical performance data set, and predicting to obtain advertisement material performance parameters; defining a management rule set of the life cycle of the advertisement material through the smart contract, wherein the management rule set comprises an auditing rule, a putting rule and an off-shelf rule; according to the advertisement material performance parameters, inputting the created data set into a rule matching engine of the smart contract, triggering execution of the management rule set, and obtaining an advertisement material state classification result; and in combination with the state classification result, performing adaptive allocation of advertisement resources to obtain an advertisement material life cycle management result. According to the method, automatic and intelligent management of the full life cycle of the advertisement materials from creation to off-shelf is realized.

Description

Technical Field

[0001] This invention relates to the field of advertising material management technology, specifically to an advertising material lifecycle management system based on smart contracts. Background Technology

[0002] In the current advertising industry, the lifecycle management of ad creatives involves multiple stages, including creation, review, placement, optimization, and removal. Its management efficiency and quality directly impact the final outcome of an advertising campaign. Traditional ad creative management methods rely heavily on manual operation and centralized system coordination, which has revealed numerous problems in actual operation. During the creative content creation phase, relevant personnel need to manually enter basic information about the content, such as content type, size, and theme. This data collection is often incomplete or untimely, resulting in a lack of comprehensive foundational data support for subsequent stages. Furthermore, in terms of performance evaluation, the collection and analysis of historical performance data largely relies on manual processing, making it difficult to cover multi-dimensional metrics such as impressions, click-through rate, and conversion rate. Moreover, the data updates are often delayed, failing to provide effective reference for subsequent content processing. In the review process, the application of review standards under the traditional model is easily influenced by the subjective factors of reviewers. Different reviewers may have different judgments on the same type of material, resulting in inconsistent review results. At the same time, the review process is cumbersome, requiring approval from multiple departments and levels, which often leads to excessively long review cycles and delays in the release of high-quality materials. In the release phase, there is the problem of rigid release strategies. Most systems execute according to preset release plans, making it difficult to dynamically adjust the release intensity and channel allocation based on the real-time performance of the materials. This results in some high-potential materials failing to receive sufficient exposure resources, while inefficient materials still occupy a large amount of release budget. The decision-making process for removing ad creatives also relies on manual judgment, lacking clear data-driven basis. This often results in inefficient creatives not being removed in a timely manner, continuing to consume resources, or high-quality creatives being removed prematurely due to misjudgment. Furthermore, in traditional management models, the enforcement of rules at each stage depends on the centralized platform's access control. The modification and execution of rules lack transparency, making it susceptible to human intervention in data or rule tampering, thus reducing the credibility of the management process. With the rapid development of the advertising industry, the number and types of ad creatives have surged, and traditional management methods can no longer meet the needs for efficient, accurate, and transparent management. There is an urgent need for an automated, intelligent, and reliable ad creative lifecycle management solution. Summary of the Invention

[0003] The purpose of this invention is to provide an advertising creative lifecycle management system based on smart contracts to solve the problems mentioned in the background art.

[0004] To achieve the above objectives, the present invention provides an advertising creative lifecycle management system based on smart contracts. The system includes a data analysis module, which is used to perform: Collect the creation data set and historical performance data set of advertising creatives; Based on the aforementioned historical performance data set, an advertising creative performance prediction model is constructed to predict and obtain advertising creative performance parameters; The set of management rules for the lifecycle of advertising creatives is defined through smart contracts, including review rules, delivery rules, and removal rules; Based on the performance parameters of the advertising creative, the data set for creating the advertising creative is input into the rule matching engine of the smart contract, triggering the execution of the management rule set and obtaining the status classification result of the advertising creative; Based on the advertising creative status classification results, adaptive allocation of advertising resources is performed to obtain advertising creative lifecycle management results.

[0005] Preferably, the collection of advertising creative creation data sets and historical performance data sets includes: Collect the creation time parameters, material type parameters, and content feature parameters of the advertising materials, and integrate them to obtain a set of advertising material creation data; Collect exposure, click, and conversion rate parameters of advertising creatives during historical campaigns, and integrate them to obtain a set of historical performance data for advertising creatives; The data set of advertising creative creation and the data set of advertising creative historical performance are stored in a decentralized database.

[0006] Preferably, based on the historical performance data set, an advertising creative performance prediction model is constructed to predict the advertising creative performance parameters, including: Extract time-series features from the historical performance data set of the advertising creative; Based on event-driven algorithms, a performance prediction model for advertising creatives is constructed. The time series features are used as training data to train and validate the advertising creative performance prediction model; Input the current ad creative creation data set into the trained ad creative performance prediction model, and obtain the ad creative performance parameters from the prediction output.

[0007] Preferably, the set of management rules for defining the lifecycle of advertising creatives through smart contracts includes review rules, placement rules, and removal rules, including: Define review rules in smart contracts to verify whether the content of advertising materials meets compliance standards; Define delivery rules in the smart contract to trigger delivery conditions based on the performance parameters of the advertising creative; Define removal rules in the smart contract to trigger removal conditions based on the status classification results of the advertising material; The review rules, release rules, and delisting rules are compiled into smart contract executable code.

[0008] Preferably, based on the performance parameters of the advertising creative, the data set for creating the advertising creative is input into the rule matching engine of the smart contract, triggering the execution of the management rule set, and obtaining the advertising creative status classification results, including: Based on the performance parameters of the advertising creative, multiple candidate status levels are obtained through screening. Select multiple rule matching engines corresponding to the multiple candidate state levels; The advertising creative creation data set is input into the multiple rule matching engines respectively. Each rule matching engine executes the corresponding rule in the management rule set and outputs multiple rule matching results. By integrating the matching results of the multiple rules, the status classification result of the advertising material is calculated.

[0009] Preferably, selecting multiple rule matching engines corresponding to the multiple candidate state levels includes: The number of matching engines is calculated based on the error range between the performance parameters of the advertising creative and the multiple candidate status levels. Based on decentralized logic, multiple rule matching engines are selected corresponding to multiple candidate state levels; Randomly select a number of rule matching paths from the plurality of rule matching engines; The data set created by the advertising creative is input into the multiple rule matching paths, and multiple rule matching results are output.

[0010] Preferably, based on the advertising creative status classification results, adaptive allocation of advertising resources is performed to obtain advertising creative lifecycle management results, including: Extract the status level parameter from the status classification results of the advertising materials; Calculate the advertising resource allocation ratio based on resource optimization algorithms; Based on the aforementioned advertising resource allocation ratio, the resource allocation instruction of the smart contract is triggered; Execute the resource allocation instructions to obtain the advertising creative lifecycle management results.

[0011] Preferably, the calculation of advertising resource allocation ratios based on resource optimization algorithms includes: Analyze the trend of status changes in the status classification results of the advertising materials; Based on trend matching algorithms, identify the characteristics of advertising resource demand; Based on the advertising resource demand characteristics, calculate multiple candidate values ​​for resource allocation ratios; Select the candidate values ​​of the resource allocation ratio that meet the constraints as the advertising resource allocation ratio.

[0012] Preferably, the resource allocation instruction triggered by the smart contract based on the advertising resource allocation ratio includes: Map the advertising resource allocation ratio to the resource allocation rules of the smart contract; Generate a set of resource allocation instructions, including bandwidth adjustment instructions and priority update instructions; The resource allocation instruction set is executed through a smart contract event trigger.

[0013] Preferably, executing the resource allocation instruction to obtain the advertising creative lifecycle management results includes: Deploy the resource allocation instruction set in the advertising delivery network; Real-time monitoring of ad creative performance feedback data; Based on the performance feedback data of the advertising creative, adjust the set of resource allocation instructions; Output the adjusted ad creative lifecycle management results.

[0014] Compared with the prior art, the beneficial effects of the present invention are: This method provides a novel solution for ad creative lifecycle management by integrating data collection, intelligent prediction, smart contract rule definition, and adaptive resource allocation. At the data collection level, it incorporates both the ad creative creation dataset and historical performance dataset, covering multi-dimensional information from the creative's initial basic attributes to its past performance. This ensures that subsequent analysis and decision-making are based on a more comprehensive and richer data foundation, avoiding management biases caused by incomplete data. Building a performance prediction model for ad creatives and training it based on historical performance data enables proactive assessment of the potential performance parameters of ad creatives. This predictive capability allows the management process to move beyond passively summarizing historical data and instead identify the potential value of creatives in advance, providing directional guidance for subsequent review, deployment, and other stages, thus reducing resource waste caused by blind operations. By leveraging smart contracts to define sets of management rules for review, deployment, and delisting, traditional rules that rely on manual or centralized systems are transformed into coded, immutable, and automated execution logic. This process eliminates human interference in rule execution, ensuring consistency and stability throughout the rule's lifecycle. Simultaneously, the transparency of smart contracts allows all participants to clearly understand the rule content and execution process, enhancing the credibility and trustworthiness of the management process. The rule matching engine automatically matches the created dataset with the rules in the smart contract based on the predicted performance parameters and triggers execution, thus automating the classification of ad creative status. This automated process significantly reduces the need for manual intervention, minimizes errors that may occur during manual operation, and significantly improves the responsiveness of the management process, enabling creatives to quickly enter the appropriate lifecycle stage. Adaptive allocation of ad resources, based on status classification results, dynamically adjusts resource allocation according to the real-time status and predicted performance of creatives. For high-performing creatives, resources are automatically increased to amplify their impact; for underperforming creatives, resource support is promptly adjusted or terminated. This dynamic adjustment mechanism allows ad resources to flow to more valuable creatives, optimizing overall resource allocation efficiency and enhancing the utilization value of ad creatives throughout their lifecycle. This method achieves organic connection and efficient collaboration among all stages of advertising creative lifecycle management through the deep integration of data-driven approaches, intelligent prediction, and contractual rule execution. It effectively solves the problems of low efficiency, strong subjectivity, and insufficient transparency in traditional management methods, and promotes the development of advertising creative management towards intelligence, automation, and standardization. Attached Figure Description

[0015] Figure 1 This is a schematic diagram illustrating the working principle of the smart contract-based advertising material lifecycle management system described in this invention. Figure 2 Flowchart for building an ad creative performance prediction model; Figure 3 A flowchart defining the rules for smart contract management; Figure 4 A flowchart for selecting a decentralized rule matching engine. Detailed Implementation

[0016] 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.

[0017] Please see Figure 1 This invention provides a smart contract-based advertising creative lifecycle management system, the system including a data analysis module, which is used to perform: By combining blockchain technology with machine learning models, the entire process of managing advertising creatives, from creation to delivery, is automated. It collects creation data sets and historical performance data sets for advertising creatives, including parameters such as creative type, content characteristics, impressions, clicks, and conversion rates, and stores this data in a decentralized database. Based on historical performance data, an advertising creative performance prediction model is built, trained using time-series features to predict the performance parameters of new creatives. Smart contracts define review rules, delivery rules, and removal rules for automated management of the advertising creative lifecycle. A rule matching engine triggers corresponding management rules based on predicted performance parameters and creation data sets, generating advertising creative status classification results. Adaptive allocation of advertising resources is executed based on the status classification results, dynamically adjusting delivery strategies and optimizing advertising effectiveness.

[0018] Example 1: See Figure 2 The process of collecting the ad creative creation dataset and historical performance dataset involves the systematic integration of multi-dimensional parameters. The creation dataset includes the precise creation timestamp of the ad creative, the creative type classification identifier, and content feature vectors. The creative type classification identifier covers coded tags for formats such as static images, dynamic videos, interactive rich media, or plain text. The content feature vectors are extracted using computer vision algorithms, including but not limited to topic tag clustering results, keyword density distribution, color space histogram statistics, and texture complexity indicators. The historical performance dataset is extracted from the ad delivery system logs, including the cumulative number of impressions recorded by the impression counter, the number of user interactions in the click logs, and the actual number of conversion events required for conversion rate calculation. This raw data is asynchronously obtained from the ad delivery platform's open interface through distributed crawler nodes, or by capturing real-time status change messages of the delivery system through event listeners.

[0019] The collected data undergoes a preprocessing process. Timestamps are created, converted to a standard time format, and time zone identifiers are added. Material type classification identifiers are normalized and mapped to a pre-defined type encoding table. Content feature vectors undergo dimensionality reduction to eliminate redundant features, and principal component analysis is used to retain key feature components. Historical performance data, including exposure and click-through rates, undergoes data cleaning to remove outliers caused by system errors. Conversion rate data is recalculated and verified based on actual conversion definitions. The preprocessed dataset is transmitted to a decentralized storage network via an encrypted channel. A sharding strategy distributes the data across multiple network nodes, with each data shard appended with a timestamp hash value as a tamper-proof verification measure.

[0020] The process of building an ad creative performance prediction model begins with the extraction of time-series features from historical performance data sets. These time-series features include periodic fluctuation patterns in impressions, identified through Fourier transform to pinpoint daily peak periods. Click-through rate trends are smoothed using moving averages to smooth short-term fluctuations and extract long-term growth or decline trend lines. Conversion rate decay characteristics are quantified by calculating the slope parameter of historical decay curves. The prediction model built using an event-driven algorithm employs a recurrent neural network architecture. The input layer receives a three-dimensional tensor containing time-dimensional features. The hidden layer is designed with a bidirectional gated recurrent cell structure; forward propagation captures historical dependencies, and backpropagation learns future trend patterns. The output layer uses a fully connected network mapped to the performance parameter space.

[0021] The model training phase employs a time-series cross-validation strategy. The historical performance dataset is divided into consecutive time segments, with the first 80% of the data used as the training set and the last 20% as the validation set. During training, an adaptive moment estimation algorithm is used to optimize the loss function, defined as the mean squared error between the predicted performance parameters and the actual recorded values. After each training round, the error metric on the validation set is calculated, and an early stopping mechanism is triggered if the error fails to improve after three consecutive rounds of validation. Hyperparameter tuning uses a Bayesian optimization method to search for the optimal combination of learning rate and hidden layer units. After training, the model performs feature engineering on the new advertising creative creation dataset, converting the creative type encoding into a one-hot vector, and performing dimensionality reduction on the content feature vectors in the same way as the training data. The processed feature tensors are input into the prediction model, and the output layer generates three-dimensional vectors of predicted exposure, click-through rate, and conversion rate for a specified future time period.

[0022] The prediction results are accompanied by a confidence level assessment metric. The confidence level is obtained by calculating the historical prediction error distribution of the model on the validation set, and the confidence interval of the predicted value is output using quantile regression. When the lower limit of the confidence interval for the predicted click-through rate is higher than a preset threshold, it is marked as a high-confidence prediction; when the confidence interval crosses multiple state level thresholds, it is marked as a prediction requiring manual review. The final output set of ad creative performance parameters includes three attributes: predicted value, confidence interval, and confidence level, providing multi-dimensional decision-making basis for subsequent rule matching.

[0023] Decentralized database read and write operations follow a specific protocol. When data is written, a Merkle tree structure is generated to ensure data integrity, and each data block is appended with a blockchain transaction hash as proof of existence. Data queries employ a distributed hash table routing mechanism, locating data storage nodes through content addressing. Database version management uses multi-version concurrency control to ensure the temporal consistency of historical performance data. Data access permissions are managed through an access control list controlled by smart contracts, authorizing only nodes to read and write specific data partitions. Database performance monitoring records query latency and throughput metrics in real time, automatically triggering data sharding and rebalancing operations when performance degrades.

[0024] The entire implementation process forms a closed-loop data processing chain. When new advertising materials enter the system, the data collection process is automatically triggered, and the collected data is updated to a decentralized database in real time. Database update events trigger the model prediction process, and once the prediction results are generated, the rule matching process is immediately triggered. All operation records are permanently stored in a blockchain transaction log, forming an auditable operation trajectory. The system status monitoring panel displays the data collection progress, model training metrics, and prediction task queue depth in real time, providing visual support for operation and maintenance management. An exception handling mechanism captures abnormal events such as data collection timeouts and model prediction failures, automatically initiating a retry process or escalating to a manual intervention process.

[0025] Example 2: See Figure 3 The set of management rules defined in the smart contract is implemented through a modular code structure. The review rules module contains multi-layered content verification logic: the first layer performs basic format verification, checking the compatibility of the material file format and the integrity of the metadata; the second layer deploys computer vision algorithms to perform object recognition on image and video materials, marking potentially prohibited items or sensitive scenes; the third layer uses natural language processing technology to analyze the sentiment of keywords and the frequency of sensitive words in the text description. The compliance standard library stores preset rule thresholds, such as a list of symbols prohibited from being displayed in specific regions and a fingerprint database of copyrighted material features. When the material content triggers any rule threshold, the system automatically generates a review event log, recording the violation item number and the trigger time.

[0026] The ad delivery rules module establishes a dynamic triggering mechanism. This module receives ad creative performance parameters as input variables, including predicted impressions range, click-through rate (CTR) confidence interval, and conversion rate prediction. The rules engine matches these performance parameters with a preset threshold matrix: when the predicted CTR is 20% higher than the industry benchmark and the lower limit of the confidence interval exceeds the minimum guaranteed value, a full-domain delivery instruction is triggered; when the predicted conversion rate is in the medium range but the predicted impressions are high, a targeted audience delivery mode is triggered. The threshold matrix is ​​dynamically adjusted based on ad placement characteristics, with more stringent delivery standards applied to prime ad placements. A delivery strategy package is created simultaneously when a delivery instruction is generated, including budget allocation ratios, time period selection parameters, and geographic delivery priority weights.

[0027] The ad removal rules module is designed with state-driven logic. This module continuously receives data streams of ad creative status classification results. When the status indicator changes from "Active" to "Inefficient," the ad removal evaluation process is initiated. The evaluation process checks three dimensions of indicators: the deviation between the current real-time click-through rate and the predicted value, the slope of the conversion rate decline over a continuous period, and the trend of competitive ranking changes for similar creatives. If all three conditions are met simultaneously: the real-time click-through rate is lower than the lower limit of the predicted value's confidence interval, the hourly conversion rate shows negative growth for three consecutive hours, and the competitive ranking falls out of the top 50%, an immediate ad removal instruction is generated. The ad removal operation adopts a gradual execution strategy: first, new user exposure is paused, while subsequent conversion tracking of already reached users is maintained; after 24 hours, the campaign is completely terminated and resources are released.

[0028] The rule matching engine is deployed using a distributed architecture. Each candidate state level (high efficiency / medium efficiency / low efficiency) corresponds to an independent rule execution container, which encapsulates a complete rule logic unit. When ad creative performance parameters are input, the pre-classification algorithm calculates the Euclidean distance between the parameter and the center of each state level, selecting the two closest candidate levels as primary and backup processing channels. The system dynamically allocates the number of rule engine instances for each candidate level, based on parameter deviation: a single engine is allocated for deviations less than 5%, dual engines are allocated for parallel verification for deviations between 5% and 15%, and a three-engine consensus mechanism is triggered for deviations greater than 15%.

[0029] When the ad creative creation dataset is input into the rule matching engine, field mapping technology is used to transform the data structure. The creation time parameter is converted into the local time slot encoding corresponding to the delivery time zone; the creative type parameter is expanded into a multi-dimensional feature vector, such as sub-features for video creatives including duration segments, resolution levels, and subtitle presence; content feature parameters are generated as fixed-length feature fingerprints using a hash algorithm. The transformed data structure perfectly matches the input interface of the rule condition library, eliminating rule execution deviations caused by data format differences.

[0030] The rule matching process employs a layered triggering mechanism. The first layer performs basic rule matching, such as file format checks and metadata verification in review rules. The second layer performs composite condition matching, such as the logic and judgment of the "high click-through rate + medium conversion rate" combination condition in the ad placement rules. The third layer performs cross-rule correlation matching, such as simultaneously checking the three most recent review records when an ad removal rule is triggered. Each rule matching engine outputs a structured result object, including the rule number, trigger status, matching score, and evidence data index. The evidence data index points to a snapshot of the original performance parameters stored on the blockchain, which can be retrieved during result verification.

[0031] The integration of multi-engine outputs employs a confidence-weighted strategy. The matching score of each rule-matching engine is converted into a confidence weight, calculated based on the engine's historical decision accuracy: engines with a recent decision accuracy above 90% have a weight of 1.0, those between 80% and 90% have a weight of 0.8, and those below 80% have a weight of 0.6. The state classification suggestions from each engine are weighted and voted on; the classification result whose total weight exceeds a set threshold becomes the final state. When the output results of the primary and backup candidate level channels conflict, a conflict resolution protocol is initiated: the original prediction parameters stored on the blockchain are retrieved for secondary verification. If the verification result supports the primary channel, the primary result is adopted; otherwise, a manual arbitration process is initiated.

[0032] The smart contract rule code is updated using a version control mechanism. Each rule modification generates a new smart contract version, while the old version is retained in the historical version repository. New ad creatives are processed using the latest rule version by default, and already deployed creatives maintain their original rule version until their lifecycle ends. Rule change records are stored via blockchain transactions, including the changed content, effective time, and the operator's digital signature. The rule rollback function allows automatic switching to the most recent stable version when an anomaly in a new rule is detected.

[0033] The rule execution monitoring system tracks the status of each stage in real time. The dashboard displays the currently processed material number, the rule matching stage, the number of rules triggered, and the estimated completion time. The anomaly detection module identifies abnormal events such as rule matching timeouts and engine unresponsiveness, triggering automatic restarts or engine switching processes. The execution trajectory recording function saves the complete rule trigger sequence, forming a traceable decision path diagram that displays the complete logical chain from initial parameter input to final state classification.

[0034] The resource allocation pre-calculation mechanism starts early during the rule matching phase. When the rule matching engine outputs intermediate results, the resource scheduler begins simulating the resource requirements corresponding to different state classification results. For example, when a possible efficient state classification trend is detected, the bandwidth resources that need to be reserved are pre-calculated; when potential inefficiency risks are identified, a resource reclamation plan is prepared. The pre-calculated results are temporarily stored in a cache and take effect immediately after the final state classification is determined, shortening the resource allocation latency.

[0035] The rule feedback loop continuously optimizes the rule base. The system collects actual deployment effect data after rule decisions are made and compares it with the rule prediction effect. When a specific rule continuously deviates, it is marked as a rule to be optimized; when the rule matching accuracy is consistently lower than the set standard, a rule revision proposal is automatically submitted. After the proposal is approved by the governance committee, a new smart contract rule version is generated and deployed to the test network. After verification in the sandbox environment, it is pushed to the production environment.

[0036] Example 3: See Figure 4The selection process of the rule matching engine corresponding to the candidate state level adopts a distributed decision-making mechanism. The matching degree between the ad creative performance parameters and each candidate state level is determined by a similarity measure in a multi-dimensional feature space. Let the ad creative performance parameter vector be... ,in This represents the predicted exposure value. This represents the predicted click-through rate. This represents the predicted conversion rate. The center point vector of the candidate state level is defined as... ,in This indicates the state level number (high efficiency / medium efficiency / low efficiency correspond to k=1,2,3 respectively). This indicates the standard exposure benchmark value for this level. This represents the standard click-through rate benchmark value. This represents the benchmark value for standard conversion rate. The similarity metric uses a modified weighted Manhattan distance calculation:

[0037] in: These are the weighting coefficients for each performance parameter, determined based on historical data statistics. The highest value is selected to emphasize the crucial role of click-through rate. Distance calculation results. The smaller the value, the higher the match between the ad creative's performance and the status level. The system sets a dynamic threshold. , when the minimum Less than The primary candidate level is determined directly when the minimum level is reached; In When considering intervals, select the two closest levels as primary and backup candidates; when all When this occurs, it is determined to be an abnormal state and a manual review process is triggered.

[0038] The distributed deployment of the rule matching engine is based on the blockchain network node topology. Each candidate state level corresponds to a set of rule engine nodes, and the number of nodes is dynamically adjusted according to network load. The node selection algorithm considers three parameters: the node's historical response latency. Recent task success rate and resource surplus rate Node overall score The calculation method is as follows:

[0039] in: For adjustment coefficients, This represents the maximum allowable response latency for the network. The system periodically scores and sorts all available nodes, maintaining a priority node queue for each candidate status level. When a rule matching engine needs to be assigned, available nodes are selected sequentially from the head of the priority queue for the corresponding level until the required number of engines for the current task is met.

[0040] The number of matching engines is determined using an elastic scaling strategy. Base number of engines Set to 3, based on the priority of the ad creative. Deviation from performance parameters Make dynamic adjustments:

[0041] Where: priority Determined by the advertiser's credit rating and contract terms, the value ranges from [1,5]; deviation. Performance parameter vector With the nearest state center point The Euclidean distance normalized value. The system sets the maximum number of engines. When the calculated value exceeds the upper limit, the upper limit value is used. This design ensures that high-priority or abnormally fluctuating ad creatives receive more thorough rule validation.

[0042] The rule matching path is constructed using a directed acyclic graph (DAG) model. Each rule matching engine is abstracted as a graph node, and the data transmission channels between nodes are abstracted as directed edges. The path generation algorithm starts from the input node and sequentially connects the review rule node, the release rule node, and the delisting rule node, forming three independent processing chains. Parallel processing nodes are inserted into the chain structure. When multi-engine consensus is required, branch paths are created to process the same rule synchronously. The path weights are dynamically adjusted based on the network latency and throughput between nodes, and the routing algorithm prioritizes the path with the lowest weight to transmit data.

[0043] The transmission of ad creative creation data sets along the rule matching path employs a fragmented encryption mechanism. The original data set is divided into several data blocks, each with an appended Message Authentication Code (MAC) verification value. During transmission, each data block undergoes multiple encryption transformations: the initial encryption uses the sending node's private key, intermediate nodes add layers of encryption during forwarding, and the final receiving node decrypts using a pre-shared key. This multi-layered encryption scheme prevents data risks caused by single-point key leakage. Upon arrival at the rule matching engine, the MAC value is verified to ensure integrity; data blocks that fail verification are discarded and retransmission is requested.

[0044] Conflict detection during multi-engine parallel processing employs vector clock technology. Each rule matching engine maintains a local logical clock. A clock vector is attached when processing messages. ,in This represents the total number of engines participating in the processing. When an engine receives the processing result, it compares the clock vector to determine the order of events: if the vector... All less than The result of the judgment is as follows. Prior to If it exists and The result of the judgment is as follows. and A conflict exists. The conflict resolution committee is randomly elected from the current online nodes, and the final state classification result is corrected after reaching a consensus through the Byzantine fault-tolerant algorithm.

[0045] The rule matching engine's fault tolerance mechanism is designed with a three-layer protection system. The first layer implements heartbeat detection; the master node sends a liveness signal to the standby node every 5 seconds. If no signal is received within the timeout period, the standby node takes over. The second layer employs checkpoint recovery technology; after processing every 100 rule matching requests, the engine saves a state snapshot to distributed storage, and recovers from the most recent checkpoint upon restart. The third layer deploys a difference synchronization protocol; when the engine's offline time exceeds a threshold, it synchronizes the difference data by comparing operation logs, ensuring state consistency upon re-uploading.

[0046] Resource allocation and rule matching are coordinated via an event bus. When the rule matching engine generates intermediate results, it publishes state change events to the event bus. The resource scheduler subscribes to relevant event topics; upon receiving a "potentially efficient state" event, it pre-allocates bandwidth resources; upon receiving a "potentially delisted" event, it initiates a resource reclamation preparation process. The event bus uses a publish / subscribe model, supports many-to-many communication, and attaches digital signatures to event messages to prevent tampering. Event processing results are fed back to the rule matching engine, forming a closed-loop control system.

[0047] The performance monitoring system collects key metrics in real time during the rule matching process. This includes the CPU utilization of each engine node. Memory usage and network I / O load Composition of three-dimensional monitoring vectors The monitoring agent samples data every minute, using a Kalman filter to eliminate instantaneous fluctuation noise and generate a smoothed trend curve. When any dimension exceeds the warning threshold, a load balancing strategy is triggered: some matching tasks are migrated to low-load nodes, or virtual engine instances are dynamically added. Monitoring data is persistently stored in a time-series database, supporting multi-dimensional analysis and querying by node, time range, rule type, and other dimensions.

[0048] The rule-matching engine employs a traffic distribution strategy for its canary rollout. When a new version of the engine is launched, an initial 5% of the matching traffic is allocated for verification. This traffic allocation is adjusted based on the success rate of the new version. Dynamic adjustment: when At that time, increase the flow rate by 10% per hour; when When, maintain the current ratio; when If necessary, roll back to the old version and mark it as an anomaly. The canary release status is managed through blockchain smart contracts, and all traffic allocation records and version switching operations are stored on the blockchain, forming an immutable release audit trail.

[0049] The exception handling process covers the entire rule matching process. Input data exception detection includes format validation, range checks, and logical verification; exception data is transferred to an isolation area for manual processing. Rule execution exceptions capture syntax errors, runtime errors, and logical conflicts, triggering automatic engine restarts or rule hotfixes. Output result exception identification includes out-of-bounds results, contradictions, and low confidence levels; exception results trigger a multi-engine review process. All exception events generate standardized reports, including exception codes, occurrence times, impact scope, and remediation suggestions, which are pushed to the operations center via a distributed message queue.

[0050] Example 4: The implementation process of adaptive allocation of advertising resources is based on the dynamic analysis of advertising creative status classification results. The system first extracts the core parameters from the status classification results, including status level identifiers, confidence scores, and trend change markers. The status level identifiers adopt a three-level classification system: "Level A" represents creatives with high expected performance, "Level B" represents creatives with medium performance, and "Level C" represents creatives with low performance or those that need to be removed. The confidence score ranges from 0-100%, reflecting the degree of certainty in the status classification. The trend change markers record the status transition path within the last three analysis periods, such as "A→A→B" indicating that the creative performance has shifted from stable and high performance to medium performance. The input parameters of the resource optimization algorithm include multi-dimensional constraints and resource allocation constraints for typical time periods, as shown in Table 1.

[0051] Table 1: The resource allocation constraints for typical time periods are as follows.

[0052]

[0053] The algorithm first constructs a resource demand feature matrix during execution. For "Level A" content, the demand characteristics are high bandwidth requests, priority time slot preemption, and wide geographical coverage; "Level B" content exhibits medium bandwidth demand, balanced time slot distribution, and targeted geographical delivery; "Level C" content only retains minimum bandwidth guarantees and long-tail time slot delivery. Each cell in the matrix records the demand intensity for a specific resource dimension, with intensity values ​​automatically mapped according to the status level: Level A corresponds to values ​​of 0.8-1.0, Level B 0.4-0.7, and Level C 0.1-0.3.

[0054] The generation of candidate resource allocation ratios employs an iterative trial-and-error method. The system initializes three baseline ratio schemes: an aggressive scheme (70% for Level A, 25% for Level B, and 5% for Level C), a balanced scheme (55%, 35%, and 10%), and a conservative scheme (40%, 45%, and 15%). Each scheme is simulated in a virtual environment, and key indicators are collected during the simulation: bandwidth utilization fluctuation curve, time-period coverage saturation, and budget consumption rate. After the simulation, the algorithm calculates the constraint satisfaction score for each scheme. The scoring criteria include: peak-hour utilization must not fall below the constraint lower limit, coverage in key cities must meet the standard, and mobile terminal guarantee rate must strictly adhere to the red line value.

[0055] Candidate solutions that meet the constraints proceed to the final selection stage. The selection process considers two types of indicators: static indicators, including the solution's fit with the current network state, and dynamic indicators, focusing on the solution's adaptability to traffic changes over the next two hours. Fit analysis examines the required resource adjustments for each solution, prioritizing those with minimal differences from the existing network configuration. Adaptability assessment involves backtesting in similar historical scenarios, selecting solutions that have performed stably under similar conditions in the past. The final output advertising resource allocation ratios are rounded to two decimal places; for example, a possible output might be: Level A 61.25%, Level B 33.50%, and Level C 5.25%.

[0056] The resource allocation rules of smart contracts are mapped using a state machine model. The system maintains a current resource allocation state matrix, where rows represent resource types (bandwidth, time period, budget, etc.) and columns represent state levels (A, B, C). When a new allocation ratio is reached, the state machine calculates the difference between the target matrix and the current matrix, generating an incremental adjustment instruction sequence. For example, if it detects that the bandwidth ratio of level A needs to be increased by 8%, a "bandwidth reallocation" instruction is generated, with specific parameters including source level (B, C), target level (A), and transfer amount (8%). The instruction sequence is ordered according to dependencies to ensure that adjustments to basic resources precede changes to derived resources.

[0057] The resource allocation instruction set is constructed following the principle of atomicity. Each instruction contains complete execution elements: operation type (increase / decrease / transfer), resource category, numerical change, and effective time window. Instruction encoding uses a binary protocol: the first 2 bytes identify the operation type, the middle 4 bytes represent the resource category code, the following 8 bytes store the numerical parameters, and the last 4 bytes record the time parameters. For example, a bandwidth adjustment instruction is encoded as: 0x0001 (increase), 0x0000000A (bandwidth category), 0x00000000000000200 (512Mbps), 0x00015180 (24 hours). This compact encoding format is suitable for efficient transmission in blockchain transactions.

[0058] The implementation of priority update instructions relies on a tag exchange mechanism. The system maintains dynamic priority tags for each ad creative, with tag values ​​determined by both the status level and real-time performance. Initial tags are set based on status level: Level A creatives have tag values ​​of 300-400, Level B 200-299, and Level C 100-199. During campaign execution, tag values ​​are dynamically fine-tuned based on actual click-through rates (CTR): a CTR increase of more than 5% per hour increases the tag value by 20 points, while a decrease of 30 points is applied for two consecutive hours. The resource scheduler constructs a priority queue based on tag values, with creatives possessing higher tag values ​​receiving priority access to scarce resources.

[0059] The execution flow of smart contract event triggers is designed as an asynchronous pipeline pattern. The event listening module continuously monitors the resource allocation instruction queue and generates an event token when a new instruction is detected. The token contains the instruction hash, trigger condition, and execution context, and is transmitted to the execution node through a distributed message queue. After verifying the token signature, the execution node reads the corresponding smart contract code snippet from the blockchain and runs it in a sandbox environment. The execution result generates an event receipt, which contains resource change records and a state snapshot hash value. After all participating nodes confirm the validity of the receipt through a consensus mechanism, they update their local resource state.

[0060] The deployment of instructions for the advertising delivery network adopts a gradual strategy. The first phase involves trialing the new instructions on 5% of the node cluster to monitor for any anomalies in cluster performance metrics. During the trial run, data such as bandwidth utilization, request response latency, and error rate are collected and compared with historical baselines to verify security. Once no anomalies are confirmed, the second phase of deployment begins, gradually expanding to 30%, 70%, and eventually the entire network. Each phase is separated by a 15-minute observation period; any anomalies are immediately rolled back to the previous stable version. Deployment progress is managed through blockchain smart contracts; each node sends a signed confirmation transaction after confirming deployment completion, creating a verifiable deployment trajectory.

[0061] The real-time monitoring system constructs a multi-dimensional feedback data stream. The data acquisition layer captures the actual exposure of ad creatives, click coordinate heatmaps, and conversion event timestamps. The transmission layer sends the raw data to the stream processing engine via a dedicated data channel. The engine calculates minute-level performance metrics in real time: exposure completion rate (actual exposure / planned exposure), click concentration (percentage of clicks in the top 10% of the region), and conversion latency (average interval from click to conversion). These metrics are compared with predicted values ​​to generate deviation alarm signals. The signal levels are categorized based on the magnitude of the deviation: Attention (5-10%), Warning (10-20%), and Severe (>20%).

[0062] The dynamic adjustment of resource instructions employs a sliding window algorithm. The system maintains an indicator observation window for the most recent 15 minutes, with data points within the window weighted by time to calculate trend values. When three consecutive time points show a trend deviation in the same indicator (e.g., a continuous decline in click concentration), the instruction adjustment process is triggered. The adjustment magnitude is proportional to the trend slope; for example, if click concentration decreases by 5% per hour, the bandwidth allocation for that material in hotspot areas is reduced by 5%. Adjustment instructions are executed preferentially through a fast track, skipping the observation period of regular deployment, but are marked with a special flag for subsequent audit tracking.

[0063] The output of lifecycle management results includes structured reports and visualizations. The main body of the report comprises three parts: a resource allocation execution summary recording the difference between the actual allocation ratio and the planned value for each time period; a state transition trajectory showing the path of the creative's state changes from creation to the current state; and an anomaly handling log summarizing all automated intervention operations. The visualization uses a heatmap to display the resource density distribution across different regions and time periods, overlaid with actual click data to form a performance comparison chart. The final output is stored on the blockchain, including digital signatures and timestamps, serving as the basis for smart contract execution for ad settlement.

[0064] Example 5: The execution of resource allocation instructions in a smart contract begins with the parsing process of the instruction code. Bandwidth adjustment instructions include the target bandwidth value, effective timestamp, and scope identifier. The parser separates the opcode and parameter segments from the instruction. The opcode is mapped to a predefined operation type library, and the parameter segments are deserialized into structured data objects. Priority update instruction parsing involves tag version checking and conflict detection. When an ineffective older version of the same instruction is detected, the parameters of the old and new instructions are automatically merged to generate a comprehensive update scheme. A digital fingerprint is attached to the parsed instruction object. This fingerprint is generated by hashing the instruction content and serves as a benchmark for subsequent execution verification.

[0065] The resource allocation instruction set is deployed in the advertising network using a phased propagation strategy. In the initial phase, core network nodes are selected as instruction distribution centers, with each central node responsible for relaying instructions within a specific geographical area. The distribution protocol employs a tree-like topology, with the root node broadcasting instructions to its direct child nodes. After receiving and verifying the instructions, the child nodes forward them to their next-level nodes. Random delays are set for each level of forwarding to avoid momentary network congestion. When verifying the validity of an instruction, nodes perform a triple check: verification of the instruction signature against the system public key, consistency verification of the digital fingerprint and content, and confirmation of the validity of the effective time window. Any failure to perform these checks will result in the instruction being marked as suspicious and its propagation being suspended, awaiting manual review by the control center.

[0066] The real-time monitoring system constructs a multi-source data acquisition pipeline. An exposure monitor is deployed at the ad display interface, capturing the device fingerprint and spatiotemporal stamp of each ad display using lightweight tracking. A click tracking module is embedded in the page interaction layer, recording the screen coordinates, press duration, and subsequent scrolling trajectory of user clicks. A conversion rate analyzer connects to the business system's event bus, capturing the complete conversion chain from ad click to order completion. The acquisition terminals are equipped with local caches to temporarily store data during network interruptions and re-upload data via a breakpoint resume protocol after connection restoration. Once the data stream enters the processing cluster, it first undergoes time alignment to unify records from different sources onto a standardized timeline.

[0067] Real-time analysis of performance feedback data employs window aggregation technology. The system defines configurable sliding time windows, with a base window length of five minutes. Within each window period, the calculation module statistically analyzes the exposure completion rate metric, i.e., the ratio of actual exposures to planned exposures; it also plots a click heatmap to identify areas of high click density and low click density; and it calculates the median conversion latency to measure the user's response speed from click to conversion. A snapshot report is generated when the window closes, including statistical values ​​for each metric, outlier markers, and a data integrity score. Continuous window reports form a time series, which is then input into the trend detection engine.

[0068] The dynamic adjustment of resource allocation instructions is based on a deviation response mechanism. When the exposure completion rate is below 95% of the planned value for three consecutive windows, a bandwidth increase process is triggered: the bandwidth value is increased compensatorily according to the missing proportion, while the ad duration for the same period is extended. When the click heatmap shows that the click density in the core area has decreased by more than 10 percentage points, priority reallocation is initiated: the ad weight in the peripheral areas is reduced, and the released resources are concentrated in historically high click areas. When the median conversion latency exceeds twice the industry benchmark, display strategy optimization is performed: the display frequency of conversion guidance elements is increased, and the conversion path steps are shortened. Each adjustment generates an incremental adjustment instruction, maintaining version association with the original instruction.

[0069] The execution of instruction adjustments adopts a version overlay model. New adjustment instructions do not directly overwrite existing instructions, but are appended as supplementary instructions. The system maintains an instruction version chain, with each new version recording the hash of the parent version and the differences. Execution nodes merge instruction parameters from multiple versions to generate the currently effective comprehensive instruction set. A version rollback mechanism allows for quick restoration to any historical version if the effect of an adjustment deteriorates. Version management data is stored in a specific partition of the blockchain, and each version change generates an independent blockchain transaction, forming a complete version evolution graph.

[0070] The anomaly handling subsystem monitors the entire command execution process. The network transmission anomaly detector identifies command propagation delays, packet loss, and verification failures. The node execution anomaly monitor captures command interpretation errors, resource allocation conflicts, and permission verification failures. The effect feedback anomaly analyzer detects data interruptions, statistical anomalies, and logical contradictions. Anomalies are categorized according to preset levels: Level 1 anomalies trigger automatic repair processes, such as command retransmission or node switching; Level 2 anomalies initiate a local rollback mechanism to restore to the previous stable state; Level 3 anomalies escalate to the central console, generating an alarm report with debugging information. All anomaly handling records are appended with precise timestamps and node location information.

[0071] The lifecycle management results are output in both machine-readable and human-readable versions. The machine-readable version uses binary encoding and includes a final state matrix of resource allocation, a hash chain of instruction execution trajectories, and a time-series summary of performance metrics. The human-readable version generates a structured document, displaying the initial resource plan, dynamic adjustment records, final allocation ratios, and anomaly statistics in chapters. Visualization components convert key metrics into trend charts, render regional resource distribution as heat maps, and plot state transition paths as directed relationship graphs. The output document set is stored through a distributed file system, and index information is written to the blockchain to ensure the verifiability and immutability of the results.

[0072] The audit trail system records the complete operation chain. Starting from the generation of the initial instruction by the smart contract, it records operation logs for each stage: instruction parsing, network propagation, node execution, effect monitoring, and dynamic adjustment. Log entries include operation type, executing entity, target object, timestamp, and result code. Log storage adopts a hierarchical structure: recent logs are stored in a high-speed cache cluster for real-time querying, while historical logs are compressed and archived to a cold storage system. The audit interface supports multi-dimensional retrieval by material number, time range, node location, etc., and can reconstruct system state snapshots at any point in time. The tracking data is periodically generated into a Merkle tree and anchored to a public blockchain, providing verifiable proof of existence for third parties.

[0073] The resource release and recycling process is automatically triggered at the end of the product's lifecycle. When the ad removal rules confirm the termination of the creative's lifecycle, the resource recycling instruction is executed in reverse along the original allocation path. Bandwidth resources are returned to the public resource pool proportionally, the priority tag is set to the lowest recycling level, and the ad slots are marked as idle. The recycling process performs a resource leakage check: comparing the theoretical release amount with the actual recycling amount; if the difference exceeds a threshold, a resource tracking process is triggered to locate the unreleased resource occupancy points. Finally, a resource settlement report is generated, detailing the resource occupancy duration, consumption, and cost allocation data for each stage, serving as the original basis for ad billing.

[0074] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0075] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A smart contract-based advertising creative lifecycle management system, characterized in that, The system includes a data analysis module, which is used to perform: Collect the creation data set and historical performance data set of advertising creatives; Based on the aforementioned historical performance data set, an advertising creative performance prediction model is constructed to predict and obtain advertising creative performance parameters; The set of management rules for the lifecycle of advertising creatives is defined through smart contracts, including review rules, delivery rules, and removal rules; Based on the performance parameters of the advertising creative, the data set for creating the advertising creative is input into the rule matching engine of the smart contract, triggering the execution of the management rule set and obtaining the status classification result of the advertising creative; Based on the advertising creative status classification results, adaptive allocation of advertising resources is performed to obtain advertising creative lifecycle management results.

2. The smart contract-based advertising creative lifecycle management system according to claim 1, characterized in that, The collection of ad creative creation data sets and historical performance data sets includes: Collect the creation time parameters, material type parameters, and content feature parameters of the advertising materials, and integrate them to obtain a set of advertising material creation data; Collect exposure, click, and conversion rate parameters of advertising creatives during historical campaigns, and integrate them to obtain a set of historical performance data for advertising creatives; The data set of advertising creative creation and the data set of advertising creative historical performance are stored in a decentralized database.

3. The smart contract-based advertising creative lifecycle management system according to claim 2, characterized in that, Based on the aforementioned historical performance data set, an advertising creative performance prediction model is constructed to predict the following advertising creative performance parameters: Extract time-series features from the historical performance data set of the advertising creative; Based on event-driven algorithms, a performance prediction model for advertising creatives is constructed. The time series features are used as training data to train and validate the advertising creative performance prediction model; Input the current ad creative creation data set into the trained ad creative performance prediction model, and obtain the ad creative performance parameters from the prediction output.

4. The smart contract-based advertising creative lifecycle management system according to claim 1, characterized in that, The set of management rules for the lifecycle of advertising creatives defined through smart contracts includes review rules, placement rules, and removal rules. Define review rules in smart contracts to verify whether the content of advertising materials meets compliance standards; Define delivery rules in the smart contract to trigger delivery conditions based on the performance parameters of the advertising creative; Define removal rules in the smart contract to trigger removal conditions based on the status classification results of the advertising material; The review rules, release rules, and delisting rules are compiled into smart contract executable code.

5. The smart contract-based advertising creative lifecycle management system according to claim 4, characterized in that, Based on the performance parameters of the advertising creative, the data set for creating the advertising creative is input into the rule matching engine of the smart contract, triggering the execution of the management rule set, and obtaining the advertising creative status classification results, including: Based on the performance parameters of the advertising creative, multiple candidate status levels are obtained through screening. Select multiple rule matching engines corresponding to the multiple candidate state levels; The advertising creative creation data set is input into the multiple rule matching engines respectively. Each rule matching engine executes the corresponding rule in the management rule set and outputs multiple rule matching results. By integrating the matching results of the multiple rules, the status classification result of the advertising material is calculated.

6. The smart contract-based advertising creative lifecycle management system according to claim 5, characterized in that, Selecting multiple rule matching engines corresponding to the multiple candidate state levels includes: The number of matching engines is calculated based on the error range between the performance parameters of the advertising creative and the multiple candidate status levels. Based on decentralized logic, multiple rule matching engines are selected corresponding to multiple candidate state levels; Randomly select a number of rule matching paths from the plurality of rule matching engines; The data set created by the advertising creative is input into the multiple rule matching paths, and multiple rule matching results are output.

7. The smart contract-based advertising creative lifecycle management system according to claim 1, characterized in that, Based on the ad creative status classification results, adaptive allocation of ad resources is performed to obtain ad creative lifecycle management results, including: Extract the status level parameter from the status classification results of the advertising materials; Calculate the advertising resource allocation ratio based on resource optimization algorithms; Based on the aforementioned advertising resource allocation ratio, the resource allocation instruction of the smart contract is triggered; Execute the resource allocation instructions to obtain the advertising creative lifecycle management results.

8. The smart contract-based advertising creative lifecycle management system according to claim 7, characterized in that, Based on resource optimization algorithms, the calculation of advertising resource allocation ratios includes: Analyze the trend of status changes in the status classification results of the advertising materials; Based on trend matching algorithms, identify the characteristics of advertising resource demand; Based on the advertising resource demand characteristics, calculate multiple candidate values ​​for resource allocation ratios; Select the candidate values ​​of the resource allocation ratio that meet the constraints as the advertising resource allocation ratio.

9. The smart contract-based advertising creative lifecycle management system according to claim 8, characterized in that, Based on the aforementioned advertising resource allocation ratio, the resource allocation instructions triggered by the smart contract include: Map the advertising resource allocation ratio to the resource allocation rules of the smart contract; Generate a set of resource allocation instructions, including bandwidth adjustment instructions and priority update instructions; The resource allocation instruction set is executed through a smart contract event trigger.

10. The smart contract-based advertising creative lifecycle management system according to claim 9, characterized in that, Executing the resource allocation command to obtain the advertising creative lifecycle management results includes: Deploy the resource allocation instruction set in the advertising delivery network; Real-time monitoring of ad creative performance feedback data; Based on the performance feedback data of the advertising creative, adjust the set of resource allocation instructions; Output the adjusted ad creative lifecycle management results.

Images