A multi-dimensional indicator evaluation and automatic grading system for brand promotion tasks

Abstract

This invention discloses a multi-dimensional indicator evaluation and automatic grading system for brand promotion tasks, relating to the field of brand promotion technology. It includes a data processing module for acquiring task feature data from a promotion task database, including dynamic features such as target audience segmentation, channel selection, and budget allocation. The module removes duplicate records and missing values ​​through data cleaning techniques and uses standardization techniques to convert the feature data into a structured feature dataset with a mean of 0 and a variance of 1, resulting in a structured feature dataset. This multi-dimensional indicator evaluation and automatic grading system for brand promotion tasks enables intelligent grading, optimized scheduling, and dynamic adjustment of promotion tasks, improving task execution efficiency and resource utilization, and providing more accurate decision support for promotional activities.

Description

Technical Field

[0001] This invention relates to the field of brand promotion technology, specifically to a multi-dimensional indicator evaluation and automatic grading system for brand promotion tasks. Background Technology

[0002] Marketing task management is a key area for modern enterprises to enhance their market competitiveness. Its core lies in optimizing resource allocation and performance evaluation through scientific methods to achieve efficient marketing. Reasonable task division and quality assessment not only improve marketing efficiency but also provide decision-makers with clear strategic guidance.

[0003] However, current methods for managing promotional tasks have significant limitations. Most solutions rely on manual experience or simple rule-based classification, lacking systematic and intelligent support, resulting in inaccurate task grading and difficulty in intuitively reflecting task quality levels. This extensive management approach cannot meet the refined needs of complex market environments. Automatic task classification and quality representation in promotional task management face multiple challenges. The primary issue is the diversity and complexity of task characteristics. Promotional tasks involve multiple dimensions, such as target audience, distribution channels, and budget size. The dynamic changes in these characteristics make a single classification standard difficult to adapt. The resulting technical challenge is how to extract key information from multi-dimensional characteristics and achieve automated classification. Furthermore, due to the lack of a unified quality assessment system, the classified tasks struggle to generate intuitive representation results, preventing decision-makers from quickly understanding task priority and potential value. Summary of the Invention

[0004] The purpose of this invention is to provide a multi-dimensional indicator evaluation and automatic grading system for brand promotion tasks, thereby solving the problems existing in the prior art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a multi-dimensional indicator evaluation and automatic grading system for brand promotion tasks, comprising:

[0006] The data processing module is used to obtain task feature data from the promotion task database, including dynamic features such as target audience segmentation, channel selection, and budget allocation. It removes duplicate records and missing values ​​through data cleaning technology and uses standardization technology to convert the feature data into a structured feature dataset with a mean of 0 and a variance of 1, thus obtaining a structured feature dataset.

[0007] The status monitoring module is used to obtain task execution status logs based on structured feature datasets. If the click-through rate or conversion rate in the task execution status logs is continuously lower than a preset threshold, the module adjusts the weight coefficients of target audience preferences and channel matching based on historical data fusion using feature weight update technology, and generates an optimized feature vector set using an incremental update strategy.

[0008] The task grouping and classification module is used to group tasks based on the optimized feature vector set, repeated clustering analysis technology based on the target audience's preferences and channel matching degree, and reclassify them based on the adjusted weight coefficients using decision tree classification technology to generate updated task classification results.

[0009] The visualization module is used to generate new task priority distribution maps and potential value heat maps based on the updated task classification results. The distribution map reflects the adjusted priority ratio, and the heat map is drawn based on the updated target audience coverage and channel conversion rate. This determines the dynamically adjusted task priorities and potential value, and obtains the dynamically adjusted representation results.

[0010] Preferably, the status monitoring module obtains a task execution status log based on the structured feature dataset. This log includes key information on target audience preferences, channel matching degree, and budget constraints extracted using principal component analysis (PCA) on the structured feature dataset. The module then determines the weight coefficients of each feature dimension and generates a feature vector set that includes target audience preferences, channel matching degree, and budget constraints, thus obtaining the feature vector.

[0011] Preferably, the status monitoring module obtains the task execution status log based on the structured feature dataset. If the weight coefficients in the feature vector set are higher than a preset threshold, the task is initially grouped based on the target audience's preferences and channel matching degree using clustering analysis technology to generate a task group set based on multi-dimensional dynamic features, thus obtaining the task group set.

[0012] Preferably, the status monitoring module obtains task execution status logs based on the structured feature dataset, including obtaining feature vectors for each task group from the task group set, and automatically classifying tasks based on target audience preferences, channel matching degree, and budget constraints using decision tree classification technology, generating task classification results containing high, medium, and low priorities, thus obtaining task classification results.

[0013] Preferably, the status monitoring module obtains task execution status logs based on the structured feature dataset, including task classification results, calculates the quality score of each group of tasks through a quality assessment model, and the model input includes target audience coverage, channel conversion rate and budget utilization rate, generating a set of task quality scores.

[0014] Preferably, the status monitoring module obtains task execution status logs based on the structured feature dataset, including generating task priority distribution maps and potential value heatmaps using visualization technology based on the task quality score set. The distribution map shows the proportion of high, medium, and low priority tasks, and the heatmap is drawn based on the target audience coverage and channel conversion rate to determine the task priority ranking and value distribution, thus obtaining hierarchical characterization results.

[0015] Preferably, the status monitoring module obtains a task execution status log based on the structured feature dataset, including extracting task priority ranking from the hierarchical representation results, combining the potential value heatmap, optimizing the scheduling of tasks based on task quality scores and budget utilization using a ranking algorithm, and generating a task execution sequence.

[0016] Preferably, the status monitoring module obtains a task execution status log based on a structured feature dataset. This log includes real-time data collected every minute for the task execution sequence using real-time monitoring technology, including click-through rate, conversion rate, and budget consumption rate. An anomaly detection mechanism is used to identify traffic anomalies or a sudden drop in conversion rate, and the task execution status log is generated.

[0017] Preferably, the data cleaning technology in the data processing module includes a duplicate record detection algorithm and a missing value imputation algorithm, wherein the missing value imputation algorithm adopts a mean substitution method based on similar features.

[0018] Preferably, the standardization technique in the data processing module is the Z-score standardization method.

[0019] As can be seen from the above technical solution, the present invention has the following beneficial effects:

[0020] This brand promotion task multi-dimensional indicator evaluation and automatic classification system acquires the target audience, distribution channels, and budget characteristics of promotion tasks, extracts key information using principal component analysis, and generates a feature vector set. Cluster analysis and decision tree techniques are used to group and classify tasks, and quality scores are calculated. Based on the classification results, a task priority distribution map and a value heatmap are generated to determine the task execution sequence. During execution, the task status is monitored in real time, and feature weights are dynamically adjusted when anomalies occur, and tasks are reclassified and prioritized. This invention achieves intelligent classification, optimized scheduling, and dynamic adjustment of promotion tasks, improving task execution efficiency and resource utilization, and providing more accurate decision support for promotional activities. Attached Figure Description

[0021] Figure 1 This is a connection diagram of the system modules of the present invention. Detailed Implementation

[0022] 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.

[0023] like Figure 1 As shown, the present invention provides a technical solution: a multi-dimensional indicator evaluation and automatic grading system for brand promotion tasks, comprising:

[0024] The data processing module is used to obtain task feature data from the promotion task database, including dynamic features such as target audience segmentation, channel selection, and budget allocation. It removes duplicate records and missing values ​​through data cleaning technology and uses standardization technology to convert the feature data into a structured feature dataset with a mean of 0 and a variance of 1, thus obtaining a structured feature dataset.

[0025] The status monitoring module is used to obtain task execution status logs based on structured feature datasets. If the click-through rate or conversion rate in the task execution status logs is continuously lower than a preset threshold, the module adjusts the weight coefficients of target audience preferences and channel matching based on historical data fusion using feature weight update technology, and generates an optimized feature vector set using an incremental update strategy.

[0026] The task grouping and classification module is used to group tasks based on the optimized feature vector set, repeated clustering analysis technology based on the target audience's preferences and channel matching degree, and reclassify them based on the adjusted weight coefficients using decision tree classification technology to generate updated task classification results.

[0027] The visualization module is used to generate new task priority distribution maps and potential value heat maps based on the updated task classification results. The distribution map reflects the adjusted priority ratio, and the heat map is drawn based on the updated target audience coverage and channel conversion rate. This determines the dynamically adjusted task priorities and potential value, and obtains the dynamically adjusted representation results.

[0028] The system first cleanses and standardizes the dynamic feature data in the promotion task database through a data processing module, forming a unified structured feature dataset and resolving issues of noise and scale inconsistencies in the original data. The status monitoring module monitors task execution status in real time, identifies poorly performing tasks, and adjusts target audience preferences and channel matching based on historical performance data using feature weight update techniques, ensuring that the optimized feature vectors better reflect the current market situation. The task grouping and grading module utilizes repeated clustering and decision tree algorithms to automate grouping and grading, effectively reducing manual intervention and improving the scientific rigor and adaptability of the grading process. Finally, the visualization module transforms complex data analysis results into intuitive visual graphics, helping decision-makers quickly identify task priorities and potential value, thereby guiding subsequent resource allocation and strategy adjustments.

[0029] 1. Data standardization

[0030] The original feature data X is standardized using the z-score standardization formula:

[0031] ;

[0032] in, These are the standardized feature data values. By subtracting the mean and dividing by the standard deviation, the data is transformed into a distribution with a mean of 0 and a variance of 1, which helps to eliminate the influence of different feature data scales. μ is the mean of the feature data, and σ is the standard deviation of the feature data. The parameters μ and σ are statistically calculated based on the training dataset to ensure data balance.

[0033] 2. Status monitoring and threshold judgment

[0034] The system continuously monitors click-through rate (CTR) and conversion rate (CVR) to determine if they are below preset thresholds. , .

[0035] Judgment condition: CTR < or CVR < ;

[0036] in and Determined based on industry averages and business expectations for past tasks, for example =0.02, =0.005.

[0037] 3. Feature weight update

[0038] When the values ​​fall below the threshold, a dynamic update method based on historical data is used to adjust the weights.

[0039] ;

[0040] in, These are the updated feature weights. They represent the importance coefficient of the current feature in the model after adjustment using the dynamic update method. This value will replace the previous weights and be used in subsequent grouping and ranking calculations. Represents the original weights. This represents the performance score of the i-th task in the historical data. This represents the performance metric of the corresponding feature, where α is the update coefficient (usually between 0.7 and 0.9), and N is the number of historical tasks. The update coefficient α is obtained through empirical parameter tuning or cross-validation to ensure that the new weights take into account both historical performance and current adjustments.

[0041] 4. Incrementally update feature vectors

[0042] The updated feature vector is achieved by: ;

[0043] in, This represents the updated feature vector. It indicates the adjusted representation of the current task in the feature space, used for subsequent clustering and hierarchical calculations. This represents the original feature vector. This represents the feature vector before the update, reflecting the previous attribute features of the task. This is the incremental adjustment amount of the feature vector. Calculated based on the deviation of the most recent data, it reflects the trend of recent market or user behavior changes.

[0044] in It is calculated based on the deviation of the most recent data, dynamically reflecting market trends.

[0045] 5. Repeated clustering and hierarchical classification

[0046] Using K-means or DBSCAN clustering algorithms, grouping is performed based on the optimized feature vector set:

[0047] ;

[0048] Where C represents the clustering result (the set of all categories), the goal is to find an optimal C, argmin is the minimum operator, representing the partition that minimizes the objective function among all possible clustering partitions C, and K is the number of clusters, i.e., the number of categories. Let x be the set of the k-th category (cluster), containing all data points belonging to that category, and let x be a data point, i.e., a task feature vector in the optimization feature vector set. Let K be the centroid of the k-th category (cluster), which is the average position of all data points in that category. The initial value of K is determined based on the "elbow method".

[0049] 6. Decision tree hierarchical structure

[0050] Construct a decision tree based on adjusted weight coefficients, and select splitting features according to information gain. The information gain formula is:

[0051] ;

[0052] Where IG(D,A) represents information gain. It indicates how much the uncertainty of the dataset is reduced after selecting feature A to split the dataset D. The greater the information gain, the more suitable feature A is as a splitting attribute for a decision tree. D is the current dataset, containing data records used for training (task feature data). A is the feature, a candidate attribute used to split the dataset D, such as "channel matching degree" or "target audience preference". Entropy(D) is the entropy of the dataset D, measuring the uncertainty or purity in the dataset. V is the number of possible values ​​for feature A. Let A be the subset of data when feature A takes the vth value.

[0053] 7. Visual output

[0054] Based on the updated grouping and grading results, priority distribution charts (bar charts or stacked bar charts) and potential value heat maps (two-dimensional or three-dimensional color mapping maps) are drawn to provide intuitive analytical basis.

[0055] This system improves the quality and consistency of input data through automatic data cleaning and standardization; it enhances the system's adaptability by enabling real-time monitoring and dynamic weight updates to quickly reflect status feedback during task execution in the task model; it automatically groups and classifies tasks using advanced clustering and classification algorithms, significantly reducing labor costs and improving decision-making efficiency; and its visualization module further enhances the readability and operability of the results, allowing managers to more intuitively grasp the priority and potential value of tasks, thereby optimizing brand promotion strategies and improving promotion effectiveness and return on investment.

[0056] In a practical application, a major e-commerce platform planned a brand promotion campaign during the "Double Eleven" shopping festival, targeting users aged 18 to 35 who prefer technology and fashion products. The platform's main advertising channels included social media (such as Douyin and Weibo), search engine advertising, and in-app recommendations, with an initial budget of 3 million yuan. A dynamic allocation strategy was adopted to adapt to market feedback. The system first extracted feature data from the platform's "Double Eleven" promotional tasks over the past five years. After data cleaning and standardization, a structured input dataset was formed. During the initial operation phase, the system found that the conversion rate of some tasks was below the preset threshold of 0.5%. The status monitoring module immediately activated, dynamically adjusting the weight coefficients of target audience preferences and channel matching through a feature weight update method. Simultaneously, an incremental update strategy was applied to optimize the feature vectors, reflecting the latest market changes. Subsequently, the system used the K-means algorithm to cluster the optimized feature vectors, classifying tasks into high-value, medium-value, and low-value categories. Then, a decision tree was constructed based on the adjusted weight coefficients to reclassify each task. Ultimately, the visualization module generated updated task priority distribution maps and potential value heatmaps, clearly demonstrating the priority ratio of each task and its corresponding market potential. Based on these analysis results, the management team optimized resource allocation strategies and subsequently improved the overall conversion rate to 1.2%, achieving an approximately 30% increase in ROI, fully validating the system's efficiency and adaptability in actual brand promotion tasks.

[0057] The status monitoring module obtains task execution status logs based on the structured feature dataset. It uses principal component analysis to extract key information on target audience preferences, channel matching degree, and budget constraints from the structured feature dataset, determines the weight coefficients of each feature dimension, and generates a feature vector set including target audience preferences, channel matching degree, and budget constraints.

[0058] 1. Principal Component Analysis (PCA) Dimensionality Reduction

[0059] PCA is performed on the structured feature dataset X, and the covariance matrix is ​​calculated as follows:

[0060] ;

[0061] Perform eigenvalue decomposition on the covariance matrix: ;

[0062] Where Σ is the covariance matrix, describing the covariance relationship between features and used to measure the linear correlation between features. n is the number of data samples. X is the matrix representation of the structured feature dataset, where rows represent samples and columns represent features, and it is usually standardized (mean 0, variance 1). Let X be the transpose of matrix X (with rows and columns interchanged). Q is the eigenvector matrix, with column vectors being the eigenvectors of Σ, representing the directions of the principal components. This is an eigenvalue matrix containing the eigenvalues ​​corresponding to each principal component, reflecting the contribution of that principal component to the total variance (the larger the value, the more important it is). This is the transpose of the eigenvector matrix Q.

[0063] 2. Feature Dimension Selection

[0064] Select the top k principal components whose cumulative contribution rate reaches a preset threshold (e.g., 95%), eliminate redundant and noisy data, and ensure maximum information retention.

[0065] 3. Determining the weighting coefficients

[0066] Based on the selected principal component loadings, determine the weight coefficients of each feature. :

[0067] ;

[0068] k is the number of principal components selected or the number of features considered. is the weight coefficient of the j-th feature. represents the importance proportion of feature j in the overall feature vector, used for subsequent feature vector construction. This represents the loading of feature j along the principal component direction. It reflects the contribution or correlation of this feature in the corresponding principal component.

[0069] 4. Construct a feature vector set

[0070] The target audience preferences, channel fit, and budget constraints are combined according to determined weighting coefficients to form a feature vector:

[0071] ;

[0072] in, This is the feature vector for the i-th task. It contains a weighted combination of three features: target audience preferences, channel matching degree, and budget constraints, which are used for subsequent analysis (e.g., clustering, classification). These are the standardized feature values ​​of the target audience preferences, channel matching degree, and budget constraints for the i-th task, respectively. These are the corresponding weighting coefficients.

[0073] Ultimately, the resulting feature vector set will be used for subsequent state monitoring, grouping, and classification.

[0074] This method effectively reduces feature dimensionality and noise interference by introducing principal component analysis (PCA), thereby improving the efficiency and accuracy of feature extraction. A weighting strategy based on principal component loadings ensures a reasonable contribution of each feature dimension to the overall analysis results, overcoming the bias that may arise from manually setting weights. The construction of the feature vector set enables the system to transform complex multidimensional features into concise vector representations, facilitating subsequent clustering, classification, and visualization analysis, and improving the system's adaptability and the efficiency of generalization tasks.

[0075] The status monitoring module obtains task execution status logs based on the structured feature dataset. If the weight coefficients in the feature vector set are higher than a preset threshold, the task is initially grouped based on the target audience's preferences and channel matching degree through clustering analysis technology, generating a task group set based on multi-dimensional dynamic features.

[0076] 1. Weight threshold determination

[0077] First, the system monitors the weight coefficients of each feature in the feature vector set. If it exists:

[0078] ;

[0079] in, The weight coefficient for the j-th feature; This is a preset threshold (set based on business experience or historical data, for example, 0.4). When the weight exceeds this threshold, it indicates that the feature currently has a significant impact on task performance, triggering subsequent clustering.

[0080] 2. Extract feature dimensions for grouping

[0081] The target audience's preferences and channel matching were selected as the main features for initial grouping.

[0082] Constructing feature sub-vectors: ;

[0083] in, This is the sub-feature vector of the i-th task, used for initial task grouping, containing two weighted features: target audience preference and channel matching degree; The weight coefficients corresponding to the preferences of the target audience reflect the importance of this feature to the overall feature representation; The weight coefficient corresponding to the channel matching degree reflects the importance of this feature; The standardized feature value represents the preference of the target audience, indicating the feature performance of the i-th task in this dimension; Let be the standardized feature value of the channel matching degree, representing the feature performance of the i-th task in this dimension.

[0084] 3. Perform cluster analysis

[0085] Using K-means or DBSCAN algorithms, for all tasks Perform clustering.

[0086] 4. Generate task group sets

[0087] After clustering is completed, the task group set is output. Each of them It includes promotional tasks with similar performance, which facilitates subsequent tiering and optimization.

[0088] For the k-th task group, the tasks within the group are similar in characteristics such as target audience preferences and channel matching, which facilitates subsequent unified optimization and management.

[0089] By automating task grouping based on significant weighted features, the system can quickly identify task sets with similar target audience preferences and channel matching, avoiding subjective biases caused by manual segmentation and improving the efficiency and scientific rigor of task management. Cluster analysis reduces the complexity of the high-dimensional feature space while retaining the most critical information, making subsequent task optimization, grading, and resource allocation more accurate. Furthermore, a dynamic triggering mechanism (weight exceeding threshold) ensures the system can respond promptly to significant changes in market characteristics, enhancing its adaptability and real-time performance.

[0090] The status monitoring module obtains task execution status logs based on the structured feature dataset, including the feature vector of each task group obtained from the task group set. It automatically classifies tasks based on target audience preferences, channel matching degree and budget constraints using decision tree classification technology, and generates task classification results containing high, medium and low priorities.

[0091] 1. Feature Vector Extraction

[0092] For each task i, extract the complete weighted feature vector from the previously generated task group set:

[0093] ;

[0094] in, This is the weighted feature vector for the i-th task, containing three weighted features, which are used for subsequent classification or other analyses. Standardized feature values ​​representing the preferences of the target audience; Standardized feature values ​​for channel matching degree; Budget constraints (standardized eigenvalues); These are the corresponding feature weight coefficients.

[0095] 2. Generation of hierarchical rules

[0096] Based on the decision tree training results, the system automatically generates hierarchical rules. The final classification results include three categories: high priority, medium priority, and low priority, with priority determined based on the comprehensive performance of task features. For example:

[0097] High priority: High audience preference and channel matching, with moderate or low budget constraints.

[0098] Medium priority: At least one core feature performs at a medium level.

[0099] Low priority: Key features are low or budget is limited.

[0100] 3. Output task classification results

[0101] Each task is assigned a priority label, forming a task hierarchy result, which is used for subsequent resource allocation and optimization strategies.

[0102] This method utilizes an automated decision tree classification mechanism, avoiding subjective biases caused by manually set rules. It dynamically determines splitting features and classification rules based on actual data, ensuring the scientific rigor and adaptability of the classification. Introducing budget constraints as one of the classification features ensures that the classification results consider both market potential and resource feasibility. The classification output is concise and clear (high, medium, and low priority), facilitating rapid development of follow-up strategies by the management team and significantly improving task management efficiency and return on investment.

[0103] The status monitoring module obtains task execution status logs based on the structured feature dataset, including task classification results. It calculates the quality score for each group of tasks through a quality assessment model. The model input includes target audience coverage, channel conversion rate, and budget utilization rate, generating a set of task quality scores.

[0104] 1. Input indicator collection

[0105] For each group of tasks Extract the following three core metrics:

[0106] Target audience coverage ( ): The proportion of the target population covered to the expected population.

[0107] Channel conversion rate ( : The conversion effect of the task group on the selected channel (number of conversions / number of clicks).

[0108] Budget utilization rate ( ): The proportion of the budget already used to the planned budget.

[0109] 2. Standardized processing

[0110] To eliminate the scaling effect of different metrics, min-max normalization or z-score standardization is used:

[0111] or ;

[0112] Where x is the original index value, Let μ and σ be the minimum and maximum values ​​of the indicator, respectively, and μ and σ be the mean and standard deviation. This is a standardized or normalized indicator value. After transformation, this value is used for subsequent modeling or analysis to ensure that different features are comparable.

[0113] 3. Weighting coefficient setting

[0114] Set the weight for each indicator based on corporate strategy or experience data. , , ,For example:

[0115] =0.4 (Attention to population coverage);

[0116] =0.4 (conversion effect);

[0117] =0.2 (budget efficiency).

[0118] 4. Calculation of Task Quality Score

[0119] Quality score for each task group Calculation formula:

[0120] ;

[0121] in, These are the three standardized input metrics.

[0122] 5. Generate a set of task quality scores.

[0123] Final form: ;

[0124] Each This represents the overall quality score of the k-th task group.

[0125] By introducing a multi-dimensional quality assessment model, the system can comprehensively quantify the actual performance of each task, avoiding the misleading effects of a single indicator (such as conversion rate) and ensuring the fairness and scientific nature of the assessment results. Standardization and weighting mechanisms allow scores to be dynamically adjusted according to the company's actual goals, meeting the strategic needs of different stages. The generated task quality score set provides an objective and quantitative basis for subsequent resource optimization and allocation, improving overall promotion efficiency and return on investment.

[0126] The status monitoring module obtains task execution status logs based on the structured feature dataset. It also generates task priority distribution maps and potential value heatmaps using visualization technology based on the task quality score set. The distribution map shows the proportion of high, medium, and low priority tasks, while the heatmap is drawn based on the target audience coverage and channel conversion rate. This determines the task priority ranking and value distribution, resulting in a hierarchical characterization result.

[0127] First, the system extracts the score for each task group from the previously obtained task quality score set and categorizes these tasks into three levels—high priority, medium priority, and low priority—based on the generated priority labels. Then, the system counts the number of tasks belonging to each priority category and calculates the proportion of each category. For example, it calculates the proportion of high-priority tasks to all tasks, and the same calculation method is used for medium and low priority tasks. These proportions are used to generate a priority distribution chart, typically in the form of a bar chart or pie chart, clearly showing the percentage of different priority tasks in the overall task set.

[0128] Next, the system generates a potential value heatmap based on the target audience coverage and channel conversion rate for each task. In the chart, the horizontal axis represents target audience coverage, and the vertical axis represents channel conversion rate. The system plots each task's performance on these two dimensions as points, and uses color intensity or type to reflect the task's value based on its quality score. Higher scores are indicated by darker colors or colors closer to high-value hues.

[0129] Finally, the system integrates the results of the priority distribution map and the potential value heatmap to generate an overall hierarchical representation. This result provides visualized and quantitative data support for subsequent resource allocation, task optimization, and strategy adjustment, enabling management teams to more intuitively identify the market potential and execution priority of various tasks.

[0130] By integrating visualization technology, the system transforms complex, multi-dimensional task evaluation results into intuitive and easy-to-understand graphical representations, helping management teams quickly identify high-value task groups and allocate resources rationally. The priority distribution map provides information on the overall proportion of task levels, helping to determine the rationality of the overall layout of promotional resources; the potential value heatmap reflects the coverage and conversion potential of individual tasks, assisting in the development of personalized optimization strategies. This process significantly improves decision-making efficiency and the controllability of promotional activities.

[0131] The status monitoring module obtains task execution status logs based on the structured feature dataset, including extracting task priority ranking from the hierarchical representation results, combining the potential value heatmap, and optimizing task scheduling based on task quality score and budget utilization using a ranking algorithm to generate task execution sequences.

[0132] 1. Priority sorting extraction

[0133] The system first reads the priority label of each task from the hierarchical representation results, including high priority, medium priority, and low priority. All tasks are then initially sorted according to priority level, with high-priority tasks listed first, followed by medium-priority and low-priority tasks in that order.

[0134] 2. Combine with potential value heat map

[0135] Based on the initial ranking, the system further considers a potential value heatmap. This heatmap reflects the task's performance in terms of target audience coverage and channel conversion rate, providing visual and numerical information about the task's market potential. The system incorporates this information into subsequent ranking considerations to ensure that tasks with high market value receive higher priority in execution.

[0136] 3. Applications of sorting algorithms

[0137] The system performs a comprehensive ranking based on the preliminary ranking results and the potential value heatmap data. The ranking is mainly based on two indicators: task quality score and budget utilization rate.

[0138] First, the system assigns higher ranking weights to tasks with higher quality scores. When quality scores are similar or identical, it further compares budget utilization rates. Tasks with higher budget utilization rates are prioritized to improve the efficiency of fund use. If both quality scores and budget utilization rates are equal, the system refers to coverage and conversion rates in the potential value heatmap to select the better-performing task for priority.

[0139] 4. Generate task execution sequence

[0140] After sorting, the system generates a task execution sequence, arranging tasks in an optimized order. This sequence ensures that high-priority, high-market-value, and high-budget-efficiency tasks are executed first, with subsequent tasks arranged sequentially, forming a rational and efficient execution strategy.

[0141] This implementation method achieves comprehensive and intelligent task scheduling by integrating three dimensions: priority, market potential (potential value), and resource utilization efficiency. The sorting algorithm avoids the bias caused by a single indicator, ensuring that the task execution order aligns with business priorities while also considering return on investment and budget efficiency, significantly improving the overall effectiveness of promotional activities and the efficiency of fund utilization. Furthermore, the automated sorting and scheduling process reduces manual intervention, improving the system's response speed and flexibility.

[0142] The status monitoring module obtains task execution status logs based on the structured feature dataset. This includes real-time data collected every minute for task execution sequences using real-time monitoring technology, including click-through rate, conversion rate, and budget consumption rate. An anomaly detection mechanism identifies traffic anomalies or sudden drops in conversion rate, generating task execution status logs.

[0143] During system operation, for each task executed in a predetermined order, the status monitoring module automatically collects three real-time data points every minute: click-through rate (CTR), conversion rate, and budget expenditure rate. CTR is calculated by dividing the number of ad clicks by the number of impressions within that time period to obtain a percentage, used to measure ad attractiveness. Conversion rate is calculated by dividing the number of users who achieved the expected conversion behavior by the number of users who clicked the ad to obtain the percentage of promotional effectiveness. Budget expenditure rate is calculated by dividing the current budget expenditure by the pre-set total budget for that task to obtain the current budget usage percentage. Before entering the analysis process, the collected data undergoes data cleaning to remove duplicate data, fill in missing values, and remove outliers to ensure the accuracy of the analysis. Subsequently, the system performs anomaly detection on these three metrics for each task. The anomaly detection mechanism compares the currently collected data with the average level or set normal range of the task's historical data. If a significant fluctuation in CTR or conversion rate is detected within a short period, or if the budget expenditure rate shows unexpected abnormal changes, such as a conversion rate significantly lower than the previous average for two consecutive time periods, the system will identify this as an anomaly. Once an anomaly is detected, the system immediately generates a task execution status log containing the time, task identifier, anomaly metric name, current metric value, and anomaly type. This log will be continuously updated, serving as a basis for task adjustment, suspension, or optimization, and will also provide feedback to the scheduling module and administrators, supporting timely responses and mitigating potential risks.

[0144] Through real-time monitoring and anomaly detection, the system can instantly grasp the actual performance of task execution, respond quickly to anomalies, and effectively avoid budget waste and decreased effectiveness. Automatically generated task execution status logs provide detailed data support for subsequent data analysis, task adjustments, and system learning, improving the transparency and controllability of the overall promotional campaign. Furthermore, the minute-level data collection frequency balances response speed with system resource consumption, ensuring both high efficiency and cost-effectiveness in monitoring.

[0145] The data cleaning techniques in the data processing module include a duplicate record detection algorithm and a missing value imputation algorithm. The missing value imputation algorithm adopts a mean substitution method based on similar features.

[0146] In the system's data processing, duplicate record detection is first performed on the structured feature dataset. The system uses multiple key fields, such as task name, target audience identifier, channel code, and budget information, to perform combination matching, identifying records with completely identical content or identical key fields. When duplicate records are found, the system retains only one valid record and deletes the others, avoiding data redundancy that could introduce errors into subsequent analysis. After deduplication, the system scans all fields of the dataset, identifying records with missing values ​​and marking these missing data units. Next, for each missing data unit, the system searches the entire dataset for complete records with similar features based on other existing feature values ​​in that record. Similarity can be determined by calculating the distance between features, such as the difference between feature values, or by matching based on the similarity ratio of category features, ensuring that the selected complete records are highly similar to the missing records in key attributes such as target audience category, budget range, and channel type. After finding a set of similar records, the system extracts the data corresponding to the missing field from these records, adds up all the values ​​of these data, and divides by the number of records to obtain the average value for that field. The system then uses this average value to replace the missing value, thus completing data imputation. This method ensures that the imputed values ​​fully reflect the data distribution with similar attributes to the missing records, avoiding the bias that may arise from simply using the average of all data. After imputing all missing values, the system performs consistency verification, checking the data type, value range, and logical relationships of each field to ensure data integrity and correctness, providing a high-quality data foundation for subsequent data analysis and model training.

[0147] This data cleaning method significantly reduces data redundancy and improves data processing efficiency by automatically detecting and deleting duplicate records. It employs a mean-based imputation method based on similar features for missing value completion, which, compared to traditional global mean imputation or simple deletion of missing records, better maintains data representativeness and accuracy, reducing analytical errors caused by missing data handling. This approach enhances the reliability of subsequent feature extraction, model training, and task evaluation, laying a solid data foundation for scientific decision-making in brand promotion tasks.

[0148] The standardization technique used in the data processing module is the Z-score standardization method.

[0149] During the system's data processing, after data cleaning is completed and the features to be processed are identified, the system standardizes the data for each feature. First, the system sums the values ​​of that feature across all records, then divides this sum by the number of records to calculate the feature's mean. Next, the system calculates the standard deviation. Specifically, for each data point, the system subtracts the calculated mean from its value, obtaining the difference. Then, it squares all these differences, obtaining a set of squared differences. The system sums all these squared differences, divides this sum by the number of records minus one, and finally performs a square root operation on the result to obtain the feature's standard deviation. After calculating the mean and standard deviation, the system transforms each data point: subtracting the mean from the original value yields the deviation, which is then divided by the standard deviation to obtain the standardized value. Through this method, the data for all features is transformed into a distribution with a mean of zero and a standard deviation of one. This eliminates scale differences between different features, making the contributions of all features comparable. This standardization process is particularly suitable for subsequent data analysis and model training. It can prevent features with large value ranges from having excessive weight in the algorithm, thereby improving the accuracy and stability of subsequent operations such as principal component analysis, clustering, and classification.

[0150] The Z-score standardization method effectively eliminates scale differences between different features, allowing for fair comparison of feature contributions and preventing large numerical features from having excessive weight in the model. Furthermore, this method is highly adaptable, applicable to most machine learning and data analysis models, and particularly beneficial for subsequent processes such as principal component analysis, feature weight updates, clustering, and classification, improving the overall system's analytical accuracy and execution efficiency. In addition, Z-score standardization has relatively relaxed requirements on data distribution, maintaining good standardization results across various data feature distribution types.

[0151] 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 multi-dimensional indicator evaluation and automatic grading system for brand promotion tasks, comprising a data processing module for acquiring task feature data from a promotion task database, including dynamic features such as target audience segmentation, channel selection, and budget allocation; removing duplicate records and missing values ​​through data cleaning techniques; and converting the feature data into a structured feature dataset with a mean of 0 and a variance of 1 using standardization techniques; and a status monitoring module for obtaining a task execution status log based on the structured feature dataset. If the click-through rate or conversion rate in the task execution status log is continuously lower than a preset threshold, the system adjusts the weights of target audience preferences and channel matching based on historical data fusion using feature weight update techniques. The system employs an incremental update strategy to generate an optimized feature vector set. The task grouping and grading module uses repeated clustering analysis based on the optimized feature vector set, grouping tasks according to target audience preferences and channel matching. It then uses decision tree classification based on adjusted weight coefficients to re-grade tasks, generating updated task grading results. The visualization module uses visualization techniques to generate new task priority distribution maps and potential value heatmaps based on the updated task grading results. The distribution map reflects the adjusted priority proportions, and the heatmap is drawn based on the updated target audience coverage and channel conversion rates, determining the dynamically adjusted task priorities and potential value. The representation results are as follows: The status monitoring module obtains task execution status logs based on the structured feature dataset. This logs include key information on target audience preferences, channel matching degree, and budget constraints extracted using principal component analysis (PCA) on the structured feature dataset. The weight coefficients of each feature dimension are determined, generating a feature vector set including target audience preferences, channel matching degree, and budget constraints. If the weight coefficients in the feature vector set are higher than a preset threshold, clustering analysis is used to initially group tasks based on target audience preferences and channel matching degree, generating a task group set based on multi-dimensional dynamic features. The feature vectors of each task group are then obtained from the task group set and classified using decision tree technology. Based on target audience preferences, channel matching, and budget constraints, tasks are automatically categorized into high, medium, and low priorities, generating task categorization results. For each task categorization result, a quality assessment model is used to calculate a quality score. The model input includes target audience coverage, channel conversion rate, and budget utilization, generating a set of task quality scores. Based on this set of quality scores, visualization techniques are used to generate a task priority distribution map and a potential value heatmap. The distribution map shows the proportion of high, medium, and low priority tasks, while the heatmap is drawn based on target audience coverage and channel conversion rate, determining the task priority ranking and value distribution, thus obtaining the categorization representation results.Task priority ranking is extracted from the hierarchical representation results. Combined with a potential value heatmap, a ranking algorithm is used to optimize task scheduling based on task quality scores and budget utilization, generating a task execution sequence. For this sequence, real-time monitoring technology collects real-time data every minute, including click-through rate, conversion rate, and budget consumption rate. An anomaly detection mechanism identifies traffic anomalies or sudden drops in conversion rates, generating a task execution status log.

2. The multi-dimensional indicator evaluation and automatic grading system for brand promotion tasks according to claim 1, characterized in that: The data cleaning techniques in the data processing module include a duplicate record detection algorithm and a missing value imputation algorithm. The missing value imputation algorithm adopts a mean substitution method based on similar features.

3. The multi-dimensional indicator evaluation and automatic grading system for brand promotion tasks according to claim 1, characterized in that: The standardization technique used in the data processing module is the Z-score standardization method.

Images