Monitoring system and method for marketing quantitative indexes

Abstract

The invention relates to the technical field of marketing management, in particular to a marketing quantitative index monitoring system and method, and the system comprises a data collection layer, a data processing layer, an analysis calculation layer and a visual interaction layer. The data acquisition layer acquires original marketing data in real time through a multi-modal input interface, and processes the original marketing data to obtain processed marketing data; the data processing layer performs real-time windowing aggregation on the processed marketing data to generate a real-time calculation index, performs offline statistics on the historically stored batch marketing data, and constructs a historical reference database; the analysis and calculation layer identifies business anomalies through a historical benchmark database and an unsupervised detection algorithm, generates decision suggestions based on the business anomalies, and triggers a cross-system cooperation instruction based on the decision suggestions; and the visual interaction layer renders the index weight, the business exception and the state of the cross-system cooperation instruction in real time. According to the invention, real-time intelligent monitoring of marketing quantitative indexes can be realized.

Description

Technical Field

[0001] This disclosure relates to the field of marketing management technology, and in particular to a monitoring system and method for quantitative marketing indicators. Background Technology

[0002] Marketing metrics are standards for quantifying the process, effectiveness, efficiency, and ultimate value of marketing activities using measurable and statistical data. These metrics transform abstract marketing behaviors into concrete values, helping companies objectively measure the results of their marketing efforts. Therefore, it is necessary to monitor marketing metrics in real time.

[0003] In related technologies, the monitoring system for quantitative marketing indicators suffers from high delays in indicator updates and reliance on manual inspections. Furthermore, the indicators monitored by the monitoring system are scattered across multiple independent systems, lacking inter-system linkage analysis and making it difficult to support high-concurrency event handling. Summary of the Invention

[0004] This disclosure provides a monitoring system and method for quantitative marketing indicators to solve the problems of poor real-time performance, isolated indicators, high dependence on manual intervention, and low prediction accuracy in existing technologies, thereby achieving second-level indicator monitoring, cross-system linkage, and intelligent early warning.

[0005] In the first aspect, this disclosure provides a monitoring system for quantitative marketing indicators, including: a data acquisition layer, a data processing layer, an analysis and calculation layer, and a visualization and interaction layer; The data acquisition layer is used to acquire raw marketing data in real time through a multimodal input interface, and process the raw marketing data through a distributed message queue to obtain processed marketing data. The data processing layer is used to perform real-time windowed aggregation of the processed marketing data, generate real-time calculation indicators, and perform offline statistics on historically stored batch marketing data to build a historical benchmark database. The analysis and calculation layer is used to dynamically adjust the indicator weights based on the real-time calculated indicators through a time-series prediction model; identify business anomalies through the historical benchmark database and unsupervised detection algorithms; generate decision suggestions based on the business anomalies; and trigger cross-system collaboration instructions based on the decision suggestions. The visualization and interaction layer is used to render the status of the indicator weights, business anomalies, and cross-system collaboration instructions in real time.

[0006] Secondly, this disclosure provides a method for monitoring quantitative marketing indicators, applied to the aforementioned monitoring system for quantitative marketing indicators, including: Raw marketing data is acquired in real time through a multimodal input interface, and processed through a distributed message queue to obtain processed marketing data. The processed marketing data is aggregated in real time using a windowed format to generate real-time calculated metrics, and offline statistics are performed on historically stored batch marketing data to build a historical benchmark database. Based on the real-time calculated indicators, the indicator weights are dynamically adjusted through a time-series prediction model; business anomalies are identified through the historical benchmark database and unsupervised detection algorithms; decision suggestions are generated based on the business anomalies, and cross-system collaboration instructions are triggered based on the decision suggestions; The system renders the status of the indicator weights, business anomalies, and cross-system collaboration instructions in real time, and uploads user feedback to the analysis and computing layer.

[0007] In some embodiments, processing the original marketing data through a distributed message queue to obtain processed marketing data includes: The raw marketing data is categorized by event type to obtain data types; wherein, the data types include promotional activity order data, user clickstream data, and advertising exposure data; Assign a message topic to the data type, and partition the original marketing data within the message topic according to the event core identifier to obtain a first partitioned dataset; wherein, the first partitioned dataset contains a first sub-dataset divided according to the event type and the core identifier; The quantity of the first subset is dynamically adjusted based on real-time data flow to obtain processed marketing data.

[0008] In some embodiments, processing the original marketing data through a distributed message queue to obtain processed marketing data includes: The original marketing data is partitioned according to marketing regions or user groups to obtain a second partitioned dataset; wherein, the second partitioned dataset contains a second sub-dataset divided according to the marketing regions or user groups; The quantity of the second subset is dynamically adjusted based on real-time data flow to obtain processed marketing data.

[0009] In some embodiments, the real-time windowed aggregation of the processed marketing data to generate real-time calculated metrics includes: Set event time window parameters, and perform streaming aggregation on the processed marketing data based on the event time window parameters; The processed marketing data is grouped and calculated according to the business dimension identifier, and the real-time calculated indicators containing timestamps and dimension identifiers are output.

[0010] In some embodiments, the offline statistical analysis of historically stored bulk marketing data to construct a historical benchmark database includes: Extract time-period metrics from historical bulk marketing data; Calculate the average and standard deviation of the time period indicators, and construct the historical benchmark database based on the average and standard deviation.

[0011] In some embodiments, dynamically adjusting the indicator weights based on the real-time calculated indicators using a time-series prediction model includes: The real-time calculation indicators are input into the time series prediction model to capture the dynamic dependencies between the indicators, and the importance of each indicator to the decision objective is evaluated through the attention mechanism in the time series prediction model. Based on the importance level, a weight allocation coefficient is generated, and the real-time calculation index is weighted and fused based on the weight allocation coefficient to obtain the target index value after weight adjustment.

[0012] In some embodiments, identifying business anomalies through the historical benchmark database and unsupervised detection algorithms includes: The deviation is obtained by comparing the real-time calculated index with the standard deviation of the historical benchmark database; When the deviation exceeds a preset threshold, the label of the real-time calculated indicator is marked as a candidate anomaly; Verify and score the real-time calculated metrics labeled as candidate anomalies, and output a business anomaly identifier containing the anomaly score.

[0013] In some embodiments, generating decision suggestions based on the business anomalies and triggering cross-system collaboration instructions based on the decision suggestions includes: The system matches predefined response rules with the business anomaly identifier and generates decision recommendations containing execution actions based on the real-time calculation metrics and the response rules. The decision recommendations are parsed into cross-system collaboration instructions, and external business systems are called through application programming interfaces to execute the cross-system collaboration instructions. The execution status of the cross-system collaboration instructions is acquired and output in real time.

[0014] In some embodiments, uploading user feedback to the analysis and computing layer includes: Receive feedback information from users regarding the service anomaly identifier, and calibrate the preset threshold based on the feedback information to obtain the calibrated threshold; The calibrated threshold is uploaded to the analysis and calculation layer.

[0015] According to a third aspect of this disclosure, an electronic device is provided. The electronic device includes a memory and a processor, wherein the memory stores a computer program, and the processor executes the program to implement the method described above.

[0016] According to a fourth aspect of this disclosure, a computer-readable storage medium is provided having a computer program stored thereon that, when executed by a processor, implements the methods described above.

[0017] According to a fifth aspect of this disclosure, a computer program product is provided. The computer program product includes a computer program that, when executed by a processor, implements the methods described above in this disclosure.

[0018] This disclosure provides a monitoring system and method for quantitative marketing indicators. The system comprises: a data acquisition layer, a data processing layer, an analysis and calculation layer, and a visualization and interaction layer. The data acquisition layer acquires raw marketing data in real time through a multimodal input interface and processes the raw marketing data using a distributed message queue to obtain processed marketing data. The data processing layer performs real-time windowed aggregation of the processed marketing data to generate real-time calculated indicators and performs offline statistics on historically stored batch marketing data to build a historical benchmark database. The analysis and calculation layer dynamically adjusts indicator weights based on real-time calculated indicators using a time-series prediction model; identifies business anomalies using the historical benchmark database and unsupervised detection algorithms; generates decision suggestions based on business anomalies; and triggers cross-system collaboration instructions based on these suggestions. The visualization and interaction layer renders the status of indicator weights, business anomalies, and cross-system collaboration instructions in real time. This monitoring system for quantitative marketing indicators, through its four-layer architecture, fully leverages the advantages of a distributed computing framework, solving problems such as poor real-time performance, isolated indicators, and high reliance on manual intervention in traditional marketing monitoring. It promotes a shift in marketing monitoring from passive response to proactive early warning and intelligent decision-making, providing efficient and intelligent support for enterprise marketing management.

[0019] 1. For the data acquisition layer, the technical effect is to support multimodal data input, covering structured and unstructured data, and to achieve high-speed and stable data transmission through distributed message queues. This breaks down the barriers of data sources and provides comprehensive and continuous data source support for subsequent processing, ensuring that various marketing-related data can be incorporated into the monitoring system in a timely manner. 2. For the data processing layer, the technical effect is to achieve rapid data processing by relying on the real-time stream computing engine, while building a historical benchmark library by combining offline batch processing tools, forming a collaborative mode of real-time and offline data processing. This mode not only ensures the immediate response to changes in marketing data, but also establishes a reliable analytical reference with the help of historical data, providing a solid data foundation for indicator calculation and anomaly judgment. 3. For the analysis and calculation layer, the technical effect is to dynamically adjust the indicator weights through machine learning models, so that the system can flexibly adapt to market fluctuations, make indicator analysis more in line with actual business scenarios, accurately identify abnormal situations using anomaly detection algorithms, and achieve cross-platform linkage by integrating with multiple systems. This realizes the intelligentization of the entire process from indicator calculation to anomaly warning and cross-system collaboration, and improves the ability to perceive and respond to market changes. 4. For the visualization and interaction layer, the technical effect is to present various indicators in a variety of visualization ways, support multi-dimensional drill-down analysis, and adapt to mobile devices to ensure that relevant personnel can obtain information in real time and conveniently. The visualization and interaction layer transforms complex data analysis results into intuitive and easy-to-understand content, providing a clear basis for decision-making and building a bridge for information sharing for cross-departmental collaboration. Attached Figure Description

[0020] The present disclosure will be described in more detail below based on embodiments and with reference to the accompanying drawings: Figure 1 A schematic diagram of the structure of a monitoring system for quantitative marketing indicators provided in this embodiment of the disclosure; Figure 2 A flowchart illustrating a method for monitoring quantitative marketing indicators provided in this embodiment of the disclosure; Figure 3 This is a schematic diagram of the interface of the data acquisition layer provided in an embodiment of the present disclosure; Figure 4 This is a schematic diagram of the interface of the data processing layer provided in an embodiment of the present disclosure; Figure 5 A schematic diagram of the interface of the analysis and calculation layer provided in this embodiment of the disclosure; Figure 6 A schematic diagram of the interface of the visual interaction layer provided in the embodiments of this disclosure; Figure 7 This is a schematic diagram of a marketing data display interface provided in an embodiment of the present disclosure; Figure 8 A structural block diagram of an electronic device provided as an exemplary embodiment of this disclosure; Figure 9 A structural block diagram of a computer system provided as an exemplary embodiment of this disclosure; Figure 10 A structural block diagram of a computer program product provided for an exemplary embodiment of this disclosure.

[0021] In the accompanying drawings, the same parts are referred to by the same reference numerals, and the drawings are not drawn to scale. Detailed Implementation

[0022] To enable those skilled in the art to better understand the technical solutions of this disclosure, and to fully understand and implement the process of how this disclosure applies technical means to solve technical problems and achieve corresponding technical effects, the technical solutions in the embodiments of this disclosure will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this disclosure, not all embodiments. The embodiments of this disclosure and the various features within them can be combined with each other without conflict, and the resulting technical solutions are all within the protection scope of this disclosure. All other embodiments obtained by those skilled in the art based on the embodiments of this disclosure without creative effort should fall within the protection scope of this disclosure.

[0023] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this disclosure are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this disclosure described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0024] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0025] Example 1 In one embodiment, such as Figure 1 As shown, a monitoring system for quantitative marketing indicators is provided, including: a data acquisition layer 1, a data processing layer 2, an analysis and calculation layer 3, and a visualization and interaction layer 4.

[0026] The system comprises four layers: Data Acquisition Layer 1, which acquires raw marketing data in real time through a multimodal input interface and processes it using a distributed message queue to obtain processed marketing data; Data Processing Layer 2, which performs real-time windowed aggregation of the processed marketing data to generate real-time calculated metrics and performs offline statistics on historically stored batch marketing data to build a historical benchmark database; Analysis and Calculation Layer 3, which dynamically adjusts metric weights based on real-time calculated metrics using a time-series prediction model; identifies business anomalies using the historical benchmark database and unsupervised detection algorithms; generates decision suggestions based on business anomalies and triggers cross-system collaboration instructions based on these suggestions; and Visualization and Interaction Layer 4, which renders the status of metric weights, business anomalies, and cross-system collaboration instructions in real time.

[0027] In one possible embodiment, the data acquisition layer 1 widely accesses raw marketing data from both inside and outside the enterprise through multimodal input interfaces such as API interfaces, direct database connections, and file uploads. The raw marketing data includes structured data such as CRM orders and financial records, as well as unstructured data such as customer service dialogue logs and social media sentiment. These heterogeneous data are uniformly accessed by a distributed message queue and, after standardized processing, form a coherent post-processed marketing data stream, providing comprehensive and continuous data input for subsequent stages.

[0028] Data processing layer 2 receives processed marketing data from data acquisition layer 1. Relying on a real-time stream computing engine, data processing layer 2 performs real-time windowed aggregation of the processed marketing data, dynamically generating real-time calculation indicators such as human efficiency and business conversion, ensuring an immediate reflection of market dynamics. At the same time, data processing layer 2 uses offline batch processing tools to perform in-depth statistics and analysis on massive amounts of historically accumulated marketing data, constructing a historical benchmark database containing historical fluctuation patterns and benchmark values ​​of various indicators, providing a reference for anomaly judgment and trend analysis.

[0029] Based on the real-time calculation indicators output by the data processing layer 2, the analysis and calculation layer 3 calls the time series prediction model to dynamically adjust the weight allocation of various indicators, enabling the indicator system to adapt to market fluctuations. For example, during promotional periods, the weight of sales conversion indicators is automatically strengthened. At the same time, the analysis and calculation layer 3 combines a historical benchmark database and uses unsupervised detection algorithms to identify business anomalies that deviate from the normal fluctuation range. For example, if the conversion rate of a certain channel suddenly drops, once the analysis and calculation layer 3 detects the anomaly, it automatically generates decision suggestions containing the direction of root cause analysis and triggers cross-system collaboration instructions with systems such as technical support and order management to promote rapid response to anomalies.

[0030] The visualization interaction layer 4 will analyze and calculate the dynamic indicator weights, business anomaly information, and cross-system collaboration progress output by the calculation layer 3, and present them in a variety of forms such as 3D dashboards. This will allow users to perform multi-dimensional drill-down analysis of the data, such as drilling down from overall indicators to specific regional or channel segmented data. At the same time, the visualization interaction layer 4 is compatible with mobile devices, ensuring that management and relevant personnel can receive anomaly warnings and collaboration notifications in real time, thus achieving closed-loop management from data monitoring to decision execution.

[0031] It should be noted that the visualization and interaction layer is the key interface for users to interact with the monitoring system of marketing quantitative indicators and gain insights. Its core is to present complex data and analysis results to marketers in an intuitive and easy-to-understand way, and to provide necessary interactive functions.

[0032] In one possible implementation, the visualization interaction layer does not directly connect to the original data source. Instead, it subscribes to aggregated data streams processed by a distributed stream processing framework, such as Apache Flink or Spark Structured Streaming. For example, after calculating real-time marketing metrics such as click-through rate, conversion rate, and user activity, the Flink job can publish these metrics to a Kafka topic in the form of micro-batch or event streams. The front-end service or back-end API of the visualization interaction layer then subscribes to these real-time metric data through the Kafka Consumer to achieve near real-time data retrieval. This approach avoids the visualization interaction layer directly undertaking complex computational tasks while ensuring the real-time nature of the data.

[0033] Marketing metrics typically need to be calculated from raw clickstreams, transaction records, user behavior logs, and other data. At the data processing layer, such as using Apache Flink, a series of stream processing jobs can be defined to calculate these metrics in real time. For example, for click-through rate (CTR), Flink can count ad impressions and clicks per unit time in real time and calculate CTR = Clicks / Impressions. For conversion rate (CR), it can track the funnel path from browsing to purchase and calculate CR = Conversions / Visits. These calculation results are structured and sent downstream.

[0034] The analysis and computation layer is equipped with an anomaly detection module, such as an LSTM or Isolation Forest model. After identifying abnormal behavior, the analysis and computation layer will output information such as the type of anomaly, time, related indicators, and degree of anomaly, along with the context data that triggered the anomaly. These results can be pushed to a specific Kafka topic or stored in a low-latency database, such as Apache Druid or ClickHouse, for real-time querying and display by the visualization and interaction layer.

[0035] It should be noted that although the monitoring system for quantitative marketing metrics emphasizes real-time monitoring, historical data is crucial for trend analysis, anomaly tracing, and model training. Therefore, the visualization and interaction layer will call the backend data storage service, such as a data warehouse based on HDFS or a NoSQL database, through a dedicated API to obtain historical marketing metric data and anomaly records within a specific time range. Spark SQL can serve as a powerful batch processing and interactive query tool to efficiently query historical data stored in HDFS or other data lakes and return the results to the visualization and interaction layer for display.

[0036] In one possible embodiment, the visualization interaction layer also needs to obtain system metadata, such as a list of available marketing metrics, configuration parameters of the anomaly detection model, threshold settings, user permissions, etc. This information is usually stored in a configuration management service or a relational database, and the visualization interaction layer queries it during initialization or based on user operations.

[0037] In one possible embodiment, the visualization interaction layer should provide the following key functions and interaction processes: After a user logs into the system, the system loads the default or last saved dashboard layout according to the user's permissions. The user can customize their own dashboard by dragging and dropping, selecting indicators, adjusting the time granularity, etc. The real-time and historical indicator data obtained from the backend will be rendered into line charts, bar charts, pie charts, heatmaps, etc. For example, a user can create a dashboard that displays line charts of "ad click-through rate", "website conversion rate" and "average order value" in real time over the past hour. Once any indicator becomes abnormal, such as a sudden sharp drop in click-through rate, the corresponding chart on the dashboard will change color or flash, accompanied by a warning icon. In addition, the system continuously receives real-time marketing indicator data streams and updates the charts. Users can select different time windows, such as the last 5 minutes, 1 hour or 24 hours. The charts will automatically refresh to display the trend of the corresponding time period. For example, for the "user registration volume" indicator, the visualization interaction layer can display the real-time number of registrations per minute and overlay a trend line of the daily average registration volume to help marketers quickly determine whether the current registration performance meets expectations. In one possible embodiment, when the anomaly detection module in the analysis and calculation layer triggers an alarm, the visualization interaction layer will immediately pop up a notification on the interface or display an alarm list in a designated area. When the user clicks on the alarm notification, they can jump to the anomaly details page to view the detailed information of the anomaly. The detailed information includes: the time of the anomaly, the specific indicator value, the anomaly score, the anomaly type, and the indicator trend graph before and after the anomaly. For example, if the system detects a sudden increase in the "cost of a certain advertising campaign" in a short period of time, an alarm notification will pop up. Clicking on the details page will allow the user to see that the cost curve of the advertising campaign has a sharp peak at a certain moment, far exceeding the historical average and fluctuation range. The details page will also list the possible factors that caused the anomaly, such as traffic fraud, bidding strategy adjustments, and other preliminary analysis results.

[0038] In one possible implementation, users can select a specific point in time or time period on the metrics chart and drill down to view more granular data, such as drilling down from daily data to hourly data, or from the overall marketing campaign to the performance of specific advertising campaigns, channels, or regions. For example, if the overall "product purchase conversion rate" is found to be declining, users can click on the declining area on the chart and select to drill down by "advertising channel". The system will immediately display the purchase conversion rate of different advertising channels, revealing that the conversion rate of "social media advertising channels" has plummeted, thus pinpointing the source of the problem.

[0039] In one possible implementation, users can select two different time periods, regions, or user groups for comparative analysis. The system will display the corresponding indicator charts and data side by side, making it easy for users to identify differences and changes. For example, if marketers want to compare the "new user acquisition cost" of this week with that of last week, the system can display the cost trend charts and cumulative costs for the two weeks side by side, and calculate the year-on-year and month-on-month change rates. Furthermore, users can generate customized marketing analysis reports based on the currently selected indicators and time ranges, and support exporting to formats such as PDF and CSV. For example, the system automatically generates a weekly report containing the weekly performance of key marketing indicators, a summary of abnormal events, and preliminary analysis suggestions, which is then sent to the relevant personnel. In one possible embodiment, the marketing quantitative indicator monitoring system provided in this disclosure has a collaborative mechanism. This collaborative mechanism is key to achieving "intelligence" and "automation" in the intelligent monitoring system. It connects monitoring, analysis, decision-making, and execution, forming a closed loop. The specific content of the collaborative mechanism includes: automated or semi-automated collaboration between multiple modules to respond to monitored anomalies or trends; and the interaction process between the data acquisition layer (e.g., Kafka) and the data processing layer (e.g., Flink / Spark). Kafka, as a high-throughput distributed message queue, is responsible for real-time data collection from various marketing data sources (e.g., website logs, advertising platform APIs, CRM systems). Once data enters Kafka, Flink or Spark Streaming can consume these data streams in real-time for preprocessing, cleaning, format conversion, and initial aggregation. Kafka's publish / subscribe model and persistence features ensure data integrity and high availability. Even if the downstream processing system fails, data can continue to be processed after recovery. Flink and Spark... Streaming's stream processing capabilities ensure that data is processed within milliseconds or seconds, providing a foundation for subsequent real-time metric calculations and anomaly detection. This decoupled design makes the data source and processing logic independent, improving the system's scalability and fault tolerance.

[0040] The interaction between the data processing layer and the analysis and computation layer: After processing the raw data stream in real time, Flink or Spark SQL calculates various marketing quantification metrics. These real-time metrics are then written to a low-latency, high-concurrency metric storage database for use by the visualization interaction layer and the anomaly detection module. Flink or Spark's window operations (such as scrolling windows and sliding windows) can define aggregation behaviors within a time period and accurately calculate real-time metrics. For example, it can be set to calculate the ad clicks of the past 5 minutes every minute. The real-time writing of the calculation results ensures the "freshness" of the data, providing the latest status for monitoring and anomaly detection. The analysis and computation layer continuously obtains real-time metric data streams from the data processing layer. Once new metric data arrives, a pre-trained anomaly detection model (such as LSTM or Isolation Forest) immediately analyzes it to determine whether there is any abnormal behavior. The anomaly detection model can be deployed as a standalone microservice or integrated into a stream processing pipeline. For example, in Flink, an LSTM model can be integrated into the data stream to make real-time predictions for each newly arrived metric data point and calculate the prediction error. If the error exceeds a preset threshold, it is marked as an anomaly. Forest isolates outliers by constructing random decision trees. Its advantage lies in the fact that it does not require prior knowledge of the data distribution and has high computational efficiency. The collaborative mechanism requires the anomaly detection algorithm to be able to quickly process the input data and return the results in order to maintain real-time performance.

[0041] When the analytics layer detects anomalies in marketing metrics, it automatically triggers alert notifications. These notifications can be sent to pre-defined recipients (such as marketing team leaders and analysts) through various channels (e.g., email, SMS, DingTalk / WeChat messages, and in-system notifications). The alert notification system can be configured with flexible alert rules. For example, an alert can be triggered when the "ad placement cost" increases by more than 20% within 10 minutes and the absolute value exceeds 500 yuan. The notification content should include a detailed description of the anomaly, the time of occurrence, and links to relevant metric charts, allowing recipients to quickly understand the situation and take action. Furthermore, an alert noise reduction mechanism can be introduced to avoid frequent interruptions from duplicate or low-priority alerts. If the system detects a sustained abnormal decline in the click-through rate of an ad for a particular keyword, in addition to sending an alert, it can automatically send instructions to the ad placement system to pause the placement of that keyword or recommend optimizing the ad creative for that keyword. This requires a pre-defined rule engine for triggering conditions and response actions. This collaboration elevates intelligent monitoring from "discovering problems" to "solving problems," greatly improving marketing efficiency.

[0042] The visualization and interaction layer not only displays data and alerts but also provides a user feedback mechanism. Users can mark anomaly detection results (such as "false alarm" or "confirmed anomaly") or accept or reject automated suggestions. This feedback data will be collected for continuous optimization and retraining of the anomaly detection model. User feedback data is an important input for improving the model's accuracy and adaptability. For example, a reinforcement learning model can adjust its decision-making strategy based on the results of user suggestion acceptance. By retraining the model regularly or on demand, the system can adapt to new marketing patterns and anomaly types, achieving continuous learning and evolution.

[0043] This disclosure provides a monitoring system and method for quantitative marketing indicators. The system comprises: a data acquisition layer, a data processing layer, an analysis and calculation layer, and a visualization and interaction layer. The data acquisition layer acquires raw marketing data in real time through a multimodal input interface and processes the raw marketing data using a distributed message queue to obtain processed marketing data. The data processing layer performs real-time windowed aggregation of the processed marketing data to generate real-time calculated indicators and performs offline statistics on historically stored batch marketing data to build a historical benchmark database. The analysis and calculation layer dynamically adjusts indicator weights based on real-time calculated indicators using a time-series prediction model; identifies business anomalies using the historical benchmark database and unsupervised detection algorithms; generates decision suggestions based on business anomalies; and triggers cross-system collaboration instructions based on these suggestions. The visualization and interaction layer renders the status of indicator weights, business anomalies, and cross-system collaboration instructions in real time. This monitoring system for quantitative marketing indicators, through its four-layer architecture, fully leverages the advantages of a distributed computing framework, solving problems such as poor real-time performance, isolated indicators, and high reliance on manual intervention in traditional marketing monitoring. It promotes a shift in marketing monitoring from passive response to proactive early warning and intelligent decision-making, providing efficient and intelligent support for enterprise marketing management.

[0044] 1. For the data acquisition layer, the technical effect is to support multimodal data input, covering structured and unstructured data, and to achieve high-speed and stable data transmission through distributed message queues. This breaks down the barriers of data sources and provides comprehensive and continuous data source support for subsequent processing, ensuring that various marketing-related data can be incorporated into the monitoring system in a timely manner. 2. For the data processing layer, the technical effect is to achieve rapid data processing by relying on the real-time stream computing engine, while building a historical benchmark library by combining offline batch processing tools, forming a collaborative mode of real-time and offline data processing. This mode not only ensures the immediate response to changes in marketing data, but also establishes a reliable analytical reference with the help of historical data, providing a solid data foundation for indicator calculation and anomaly judgment. 3. For the analysis and calculation layer, the technical effect is to dynamically adjust the indicator weights through machine learning models, so that the system can flexibly adapt to market fluctuations, make indicator analysis more in line with actual business scenarios, accurately identify abnormal situations using anomaly detection algorithms, and achieve cross-platform linkage by integrating with multiple systems. This realizes the intelligentization of the entire process from indicator calculation to anomaly warning and cross-system collaboration, and improves the ability to perceive and respond to market changes. 4. For the visualization and interaction layer, the technical effect is to present various indicators in a variety of visualization ways, support multi-dimensional drill-down analysis, and adapt to mobile devices to ensure that relevant personnel can obtain information in real time and conveniently. The visualization and interaction layer transforms complex data analysis results into intuitive and easy-to-understand content, providing a clear basis for decision-making and building a bridge for information sharing for cross-departmental collaboration.

[0045] Example 2 Based on the above embodiments, such as Figure 2 As shown, a method for monitoring quantitative marketing indicators is provided. This method is applied to a monitoring system for quantitative marketing indicators and includes: Step 201: Obtain raw marketing data in real time through a multimodal input interface, and process the raw marketing data through a distributed message queue to obtain processed marketing data.

[0046] Here, the executing entity can obtain raw marketing data in real time through a multimodal input interface, and process the raw marketing data through a distributed message queue to obtain processed marketing data.

[0047] In one possible embodiment, the executing entity obtains raw marketing data from inside and outside the enterprise in real time through multimodal input interfaces such as API interfaces, direct database connections, and file uploads. This data includes structured data such as CRM order records and financial system data, as well as unstructured data such as customer service work orders and public opinion logs.

[0048] In one possible embodiment, the original marketing data is processed using a distributed message queue to obtain processed marketing data, including the following steps: The raw marketing data is categorized by event type to obtain data types; Assign message topics to data types, and partition the raw marketing data within message topics according to event core identifiers to obtain the first partition dataset; The size of the first subset is dynamically adjusted based on real-time data flow to obtain processed marketing data.

[0049] Specifically, after acquiring raw marketing data in real time through a multimodal input interface, the executing entity first classifies the raw marketing data according to event type to obtain data types, including promotional activity order data, user clickstream data, and advertising exposure data. Then, the executing entity assigns message topics to the data types and partitions the raw marketing data within the message topics according to the event core identifier to obtain the first partition dataset. The first partition dataset contains the first subset dataset divided according to event type and core identifier. Finally, the executing entity dynamically adjusts the quantity of the first subset dataset according to the real-time data flow to obtain the processed marketing data.

[0050] In one possible implementation, for different types of raw marketing data, such as "promotional activity orders," "user clickstreams," and "ad impression data," the executing entity can assign independent Kafka Topics, i.e., message topics. Within each message topic, it can be partitioned according to the core identifier of the event. For example, for promotional activity orders, the executing entity can partition by "promotion_id" or "product_category_id," and for user clickstreams, it can partition by "user_id." In this embodiment, by routing similar or related data to the same partition, it can be ensured that all related data of the same marketing event is processed by members of the same consumer group, thereby avoiding cross-partition data recombination and out-of-order problems, simplifying the real-time processing logic of the downstream Flink (real-time stream processing engine), and improving processing efficiency. In addition, this partitioning method is also conducive to achieving atomic processing and state management of events.

[0051] In one possible embodiment, the raw marketing data is processed using a distributed message queue to obtain processed marketing data, and the process further includes the following steps: The original marketing data is partitioned according to marketing regions or user groups to obtain the second partitioned dataset; The number of the second subset is dynamically adjusted based on real-time data flow to obtain processed marketing data.

[0052] Here, after the executing entity obtains the raw marketing data in real time through the multimodal input interface, it first partitions the raw marketing data according to marketing regions or user groups to obtain the second partition dataset. The second partition dataset contains the second sub-datasets divided according to marketing regions or user groups. Then, the executing entity dynamically adjusts the number of the second sub-datasets according to the real-time data flow to obtain the processed marketing data.

[0053] In one possible implementation, for raw marketing data from different marketing regions or different user segments, the executing entity can design the Kafka partition key as "region_id" or "user_segment_id". For example, order data from the North China region enters the "region_north_china" partition, and click data from VIP users enters the "user_vip" partition. This strategy enables real-time processing engines such as Flink to perform independent and parallel analysis of data from specific regions or user groups, reducing the data volume and computational complexity of a single processing task and further improving real-time processing performance. For example, for the analysis of the effectiveness of marketing activities in a certain region, it is only necessary to consume the Kafka partition data corresponding to that region, without having to scan the entire dataset.

[0054] In one possible implementation, the Kafka producer can distribute messages evenly to different partitions based on the hash value of the partition key, ensuring a uniform distribution of data within the Kafka cluster. Furthermore, the number of partitions for a topic can be dynamically adjusted to adapt to varying marketing data throughput. For example, when the system throughput reaches 100,000 messages per second, the number of partitions can be increased to horizontally scale processing capacity based on the actual load. Through reasonable partition design and dynamic adjustment, Kafka can maximize the utilization of cluster resources, ensuring that data does not back up or experience delays in high-concurrency, high-throughput marketing event scenarios, providing a stable and efficient data source for subsequent real-time stream processing.

[0055] Step 202: Perform real-time windowed aggregation on the processed marketing data to generate real-time calculated metrics, and perform offline statistics on the historically stored batch marketing data to build a historical benchmark database.

[0056] Here, the executing entity acquires raw marketing data in real time through a multimodal input interface, processes the raw marketing data through a distributed message queue, and obtains processed marketing data. After processing, the processed marketing data can be aggregated in real time using a windowed approach to generate real-time calculated metrics. Furthermore, offline statistics can be performed on historically stored batch marketing data to build a historical benchmark database.

[0057] In one possible embodiment, the processed marketing data is aggregated in real-time using a windowed format to generate real-time calculated metrics, including the following steps: Set the event time window parameters and perform streaming aggregation on the processed marketing data based on the event time window parameters; The processed marketing data is grouped and calculated according to business dimension identifiers, and the output is a real-time calculated indicator containing timestamps and dimension identifiers.

[0058] Specifically, the executing entity acquires raw marketing data in real time through a multimodal input interface, processes the raw marketing data through a distributed message queue, and obtains processed marketing data. After obtaining the processed marketing data, the executing entity sets event time window parameters and performs streaming aggregation on the processed marketing data based on the event time window parameters.

[0059] In one possible implementation, Flink, as a real-time stream computing engine, uses its EventTime Window, which is crucial for handling out-of-order data and achieving accurate marketing metric calculations. Based on business needs, the window can be dynamically adjusted in the following dimensions: It can be designed to be statistically analyzed by region and user group. For marketing metrics that require regional statistics, such as "sales revenue in each city" or "regional user activity," the executing entity can define a sliding or rolling window based on eventtime in Flink, building upon the regional partitioning in Kafka. For example, the executing entity defines a sliding window that rolls every 1 minute with a window length of 5 minutes to aggregate and calculate order data under a specific "region_id." The key is to introduce a watermark mechanism to handle potential out-of-order events.

[0060] In one possible implementation, similarly, for user segmentation metrics such as "VIP user conversion rate" and "new user retention rate," the executing entity can perform a "keyBy" operation on the user event stream according to "user_segment_id" in Flink, and then apply event time windows for statistics. The business side can dynamically adjust the window parameters through the configuration interface, for example, changing from "hourly statistics" to "every 10 minutes statistics," or setting different window lengths and sliding intervals for different marketing activities to adapt to the differences in real-time requirements of different marketing strategies. Flink can flexibly define window types and lengths through APIs. Dynamic adjustments for regions and user segments can be made through external configuration services, such as ZooKeeper or configuration centers, to distribute window parameters, which are then dynamically loaded and updated by Flink jobs.

[0061] In one possible embodiment, offline statistics are performed on historically stored bulk marketing data to build a historical benchmark database, including the following steps: Extract time-period metrics from historical bulk marketing data; Calculate the mean and standard deviation of time period indicators, and build a historical benchmark database based on the mean and standard deviation.

[0062] Specifically, the executing entity performs real-time windowed aggregation of the processed marketing data and generates real-time calculated indicators. First, the executing entity extracts time-period indicators from historical batch marketing data. Then, the executing entity calculates the average and standard deviation of the time-period indicators and builds a historical benchmark database based on the average and standard deviation.

[0063] In one possible implementation, the execution entity periodically reads massive amounts of historical marketing data from the data lake using Spark SQL. Exemplarily, this historical marketing data can include CRM orders, financial data, customer service tickets, and public opinion logs. The execution entity performs batch data cleaning, ETL (Extract, Transform, Load), and aggregation operations using Spark SQL. Then, the execution entity stores the aggregated historical marketing metrics data, such as historical average sales, historical conversion rates, and historical standard deviations of each metric, into a low-latency, queryable historical benchmark repository, such as Hive, Parquet, or HBase, or directly loads it into an in-memory computing framework such as Alluxio for fast access. It should be noted that Spark SQL is a structured data processing module in the Apache Spark ecosystem. It supports querying, analyzing, and aggregating large-scale structured and semi-structured data using SQL statements or the DataFrame / Dataset API. It combines the structured query capabilities of relational databases with the efficiency of distributed computing, capable of processing petabytes of data, and is a key tool for offline batch data processing in the system.

[0064] Step 203: Based on real-time calculated indicators, dynamically adjust indicator weights through a time-series prediction model; identify business anomalies through a historical benchmark database and an unsupervised detection algorithm; generate decision suggestions based on business anomalies, and trigger cross-system collaboration instructions based on the decision suggestions.

[0065] Here, the executing entity performs offline statistics on historically stored batch marketing data, constructs a historical benchmark database, and then dynamically adjusts the indicator weights based on real-time calculated indicators through a time-series prediction model. In turn, it identifies business anomalies through the historical benchmark database and unsupervised detection algorithms, generates decision suggestions based on business anomalies, and triggers cross-system collaboration instructions based on the decision suggestions.

[0066] In one possible embodiment, the indicator weights are dynamically adjusted based on real-time calculated indicators using a time-series prediction model, including the following steps: Real-time calculated metrics are input into the time series prediction model to capture the dynamic dependencies between metrics, and the importance of each metric to the decision objective is evaluated through the attention mechanism in the time series prediction model. Weighting coefficients are generated based on importance, and real-time calculated indicators are weighted and fused based on these weighting coefficients to obtain the target indicator value after weight adjustment.

[0067] Specifically, the executing entity performs offline statistics on historically stored batch marketing data to build a historical benchmark database. First, the executing entity inputs real-time calculated indicators into a time-series prediction model to capture the dynamic dependencies between indicators. Then, through the attention mechanism in the time-series prediction model, it assesses the importance of each indicator to the decision-making objective. Next, the executing entity generates weight allocation coefficients based on the importance and performs weighted fusion of the real-time calculated indicators based on the weight allocation coefficients to obtain the target indicator value after weight adjustment.

[0068] In one possible embodiment, the executing entity inputs the real-time calculated indicators into an LSTM time-series prediction model. This model has been trained with historical data, meaning it can capture the dynamic dependencies between indicators, i.e., the correlation between different indicators over time. For example, after one indicator rises, another indicator shows its trend within a specific time period. Using an attention mechanism, the model assesses the importance of each real-time indicator to the decision-making objective of this promotional activity, such as improving overall efficiency. This is represented by the attention score of each indicator. The executing entity generates weight allocation coefficients for each indicator based on the importance scores output by the attention mechanism. The sum of these coefficients is 1. Indicators positively correlated with the decision-making objective are assigned positive weights, while inverse indicators that need to be controlled, such as excessive spending costs, are assigned corresponding inverse weights. The executing entity performs weighted fusion of the real-time calculated indicators based on the weight allocation coefficients to obtain the target indicator value after weight adjustment. This value can intuitively reflect the overall performance of the current marketing activity.

[0069] In one possible embodiment, identifying business anomalies using a historical benchmark database and an unsupervised detection algorithm includes the following steps: The deviation is obtained by comparing the real-time calculated index with the standard deviation of the historical benchmark database; When the deviation exceeds the preset threshold, the real-time calculated indicator will be marked as a candidate anomaly. Verify and score the real-time calculated metrics labeled as candidate anomalies, and output a business anomaly identifier containing the anomaly score.

[0070] Specifically, after the executing entity dynamically adjusts the indicator weights through the time series prediction model, it first compares the standard deviation of the real-time calculated indicator with that of the historical benchmark database to obtain the deviation. When the deviation exceeds the preset threshold, the real-time calculated indicator is labeled as a candidate anomaly. Then, the executing entity verifies and scores the real-time calculated indicators labeled as candidate anomalies and outputs a business anomaly identifier containing the anomaly score.

[0071] In one possible embodiment, the executing entity uses the Flink real-time stream computing engine in the data processing layer to process the collected multi-source marketing data in real time and generate real-time calculated metrics, including sales revenue, conversion rate, click-through rate, etc. At the same time, the executing entity calls a historical benchmark database built by Spark SQL, which stores the historical mean and standard deviation of various metrics, such as the historical mean of sales revenue of a certain type of product and the standard deviation of click-through rate in different time periods. The executing entity compares the real-time calculated metrics with the historical mean and standard deviation of the corresponding metrics in the historical benchmark database. By calculating the difference between the real-time calculated metrics and the historical mean and then dividing it by the historical standard deviation, the deviation of the real-time calculated metrics is obtained, which is used to measure the degree of deviation of the real-time metrics from the historical normal level.

[0072] The executing entity judges the deviation based on the system's preset threshold. The preset threshold can be the standard deviation of the corresponding indicator in the historical benchmark database that is greater than 3 times the deviation threshold. If the deviation of a certain real-time calculated indicator exceeds the preset threshold, it indicates that the indicator is significantly different from the historical normal fluctuation range. The executing entity then marks the label of the real-time calculated indicator as a candidate anomaly and preliminarily determines that it has the possibility of being abnormal.

[0073] The executing entity invokes the isolated forest anomaly detection algorithm in the analysis and computing layer to further verify real-time calculated indicators labeled as candidate anomalies. This algorithm isolates anomalies by constructing random decision trees, assesses the degree of anomaly of candidate anomaly indicators, and generates anomaly scores. The higher the score, the greater the probability of an anomaly, and the accuracy rate can reach 92% or higher. After verification, the executing entity integrates relevant information of the anomaly indicators, including indicator name, occurrence time, anomaly score, etc., outputs a business anomaly identifier containing the anomaly score, and pushes the identifier to the visualization and interaction layer. It can also trigger cross-system linkage, such as sending pending tasks or risk warnings to the technical support center and the bidding management system.

[0074] It's important to note that LSTM, as a deep learning model adept at handling time-series data, can capture the complex patterns and intrinsic relationships of marketing metrics changing over time. As the market fluctuates, the importance of different metrics changes. LSTM dynamically learns the contribution of each metric to the prediction result through internal gate control mechanisms (input gate, forget gate, output gate) and self-attention mechanisms, achieving dynamic adjustment of weights. Specifically, LSTM assigns an "attention score" to each input marketing metric through its attention mechanism. These scores serve as dynamic weights. When the market changes, the model optimizes the attention parameters through backpropagation and gradient descent, tilting the weights towards metrics more important for the current decision, providing a more accurate reference for metric priority in subsequent anomaly detection.

[0075] Isolation Forest constructs "isolated trees" by randomly selecting features and split points to isolate outliers (outliers have shorter paths due to their ease of isolation). It generates an anomaly score of 0-1 for each data point (the closer to 1, the more likely it is to be an anomaly). In practical applications, it does not rely solely on algorithm scores but also combines an initial threshold defined by business experts: when the anomaly score exceeds this threshold, the indicator is marked as a potential anomaly. To reduce false alarms, the system introduces a multi-dimensional confirmation mechanism: combining the future trend of the indicator predicted by the LSTM model with the changes in related indicators dynamically weighted by LSTM, it comprehensively judges whether it is a true anomaly. At the same time, the system dynamically adjusts the threshold through statistical methods based on historical anomaly data and business feedback to improve the adaptability of anomaly detection.

[0076] In one possible implementation, Isolation Forest is specifically optimized for the sparsity and high dimensionality of marketing data. Marketing data often contains a large number of low-relevance or noisy features. Before constructing the isolation tree, the system evaluates the importance of features using methods such as information gain and mutual information, and prioritizes splitting features with high information gain to reduce noise interference and improve algorithm efficiency and accuracy. To avoid overfitting, the system introduces dynamic pruning rules: when constructing the isolation tree, if the tree depth reaches a preset threshold or the number of node samples is too low, splitting stops. After the tree is fully constructed, unnecessary splits are removed by evaluating the impact of pruning on outlier scores. At the same time, the maximum tree depth is set with reference to the number of samples to ensure the stability and computational efficiency of the algorithm when processing sparse data.

[0077] In one possible implementation, to achieve efficient deployment in resource-constrained environments, such as edge computing nodes, LSTM adopts a lightweight design and leverages a distributed architecture to improve overall performance. The system trains a high-precision, large-scale LSTM "teacher model" in the cloud, and then uses its output probability distribution to train small "student models" on edge nodes. This reduces the number of parameters while maintaining performance. By converting floating-point weights to low-precision integers, memory usage and computational overhead are reduced, making the model easier to deploy on edge devices. The distributed nature of the entire system permeates the entire data processing process: data undergoes edge preprocessing (cleaning, feature extraction) near each data source to reduce transmission latency; high-performance models are centrally trained in the cloud using frameworks such as Spark; lightweight models are distributed to edge nodes for real-time inference (such as local LSTM prediction and preliminary anomaly detection using isolated forests); complex anomalies are sent back to the cloud for in-depth analysis from edge nodes; and the results are aggregated and distributed to business teams via Kafka, achieving a collaborative model of "real-time edge response + deep cloud decision-making," supporting the real-time performance and accuracy of marketing metric monitoring.

[0078] In one possible embodiment, generating decision suggestions based on business anomalies and triggering cross-system collaboration instructions based on these decision suggestions includes the following steps: The system matches predefined response rules with business anomaly identifiers and generates decision recommendations that include action plans based on real-time calculated metrics and response rules. The decision recommendations are parsed into cross-system collaboration instructions, and external business systems are called through application programming interfaces to execute the cross-system collaboration instructions. It can acquire and output the execution status of cross-system collaborative instructions in real time.

[0079] Specifically, after identifying business anomalies through historical benchmark databases and unsupervised detection algorithms, the executing entity first matches predefined response rules based on the business anomaly identifier, and generates decision suggestions containing execution actions based on real-time calculated indicators and response rules. Then, the executing entity parses the decision suggestions into cross-system collaboration instructions, and calls external business systems through application programming interfaces to execute the cross-system collaboration instructions. Finally, the executing entity obtains and outputs the execution status of the cross-system collaboration instructions in real time.

[0080] In one possible embodiment, after the executing entity identifies a business anomaly through a historical benchmark database and an unsupervised detection algorithm—for example, an abnormal drop in the conversion rate of a marketing campaign in a certain region—and generates a business anomaly identifier containing an anomaly score, the executing entity automatically matches the business anomaly identifier with a predefined response rule base in the system. This response rule base stores handling strategies for different types and levels of business anomalies. For example, when the conversion rate drops abnormally and the anomaly score is ≥0.9, it is necessary to combine real-time sales figures, advertising volume, and other indicators to determine whether it is a traffic quality issue or a campaign strategy issue. The executing entity calls real-time calculated indicators (such as the region's concurrent ad clicks, user visit duration, order volume, etc.) and, combined with the matched response rules, generates specific decision suggestions. For example, for the business anomaly of "abnormal drop in conversion rate," the executing entity generates a decision suggestion of "pausing the current low-conversion advertising in the region, simultaneously pushing a list of high-potential users to the CRM system, and initiating targeted coupon distribution," clearly including the specific actions to be performed.

[0081] The executing entity performs structured parsing of the generated decision suggestions, breaking them down into cross-system collaborative instructions that can be recognized by external business systems. For example, the executing entity parses "pause low-conversion advertising" into an instruction format executable by the advertising system (including parameters such as advertising plan ID, pause time range, and operation type). The executing entity parses "push a list of high-potential users and issue targeted coupons" into user tag filtering instructions from the CRM system and coupon generation and issuance instructions from the marketing tool. Subsequently, the executing entity calls the corresponding external business systems (such as advertising platforms, CRM systems, and marketing automation tools) through preset application programming interfaces, sending the parsed collaborative instructions to each system to trigger cross-system collaborative operations.

[0082] The executing entity monitors the execution progress of cross-system collaboration instructions in real time through the status feedback interfaces of various external business systems. For example, the advertising system returns "The specified advertising plan has been paused", the CRM system returns "The list of high-potential users has been generated and synchronized to the marketing tools", and the marketing tools return "Targeted coupons have been issued to target users, with a completion rate of 95%". The executing entity summarizes and integrates these real-time execution statuses and displays them through a dynamic panel in the visual interaction layer, including the execution results, completion time, and exception prompts of each instruction. This ensures that relevant business teams can keep abreast of the collaboration progress and form a closed-loop management system from exception identification to decision execution.

[0083] Step 204: Real-time rendering of indicator weights, business anomalies, and the status of cross-system collaboration instructions, and uploading user feedback to the analysis and computing layer.

[0084] Here, after the executing entity triggers cross-system collaboration instructions based on decision recommendations, it can render the status of indicator weights, business anomalies, and cross-system collaboration instructions in real time, and upload user feedback to the analysis and computing layer.

[0085] In one possible embodiment, the execution entity renders the indicator weights in real time at the visualization interaction layer and the analysis and calculation layer. The weight heatmap dynamically displays the real-time weight distribution of each marketing indicator (such as click-through rate, conversion rate, average order value, etc.). The color depth intuitively reflects the weight level, with darker colors indicating higher weights. When the LSTM model dynamically adjusts the weights due to market fluctuations, the heatmap color changes in real time. For example, during periods of market fluctuation such as promotions, the weight of sales forecast-related indicators increases, and the corresponding area becomes darker. At the same time, dynamic numerical labels are attached next to the indicators, displaying the current weight value and the trend of change, clearly presenting the adjustment results of the LSTM model.

[0086] Real-time rendering of business anomalies is mainly reflected in the real-time monitoring area of ​​the data processing interface, the anomaly detection module of the analysis and calculation interface, and the alarm center of the visualization interaction layer. In the real-time indicator charts, abnormal fluctuation areas are specially marked, such as by using a red dashed box. The anomaly score output by the isolated forest algorithm is displayed next to the corresponding indicator in the form of a progress bar. When the score reaches a high level, the progress bar will turn completely red and trigger an alarm icon. In addition, the system will distinguish the anomaly type by status labels. Candidate anomaly indicators are marked with yellow labels, and business anomalies confirmed by multiple dimensions are marked with red labels. They are displayed in the alarm information list in order of time, including information such as the name of the anomaly indicator, the time of occurrence, and the anomaly score.

[0087] The real-time rendering of cross-system collaboration instructions is concentrated in the to-do task list, cross-system linkage event center, and marketing data display interface of the visualization interaction layer. The system displays the execution flow of cross-system collaboration instructions in the form of a flowchart, and different nodes are distinguished by color to indicate the execution status: gray indicates not initiated, blue indicates processing, green indicates completed, and red indicates failure. Feedback information, such as instruction reception status and execution progress, will be displayed next to the name of the called external business system (such as advertising platform, CRM system, etc.). After the instruction is executed, the final status will be summarized and displayed in the "Collaboration Response Area" of the 3D dashboard, including execution time, changes in relevant indicators, and the resolution status of business anomalies.

[0088] In one possible embodiment, uploading user feedback to the analytics layer includes the following steps: Receive feedback from users regarding service anomaly indicators, and calibrate the preset threshold based on the feedback information to obtain the calibrated threshold; The calibrated threshold is uploaded to the analysis and calculation layer.

[0089] Specifically, after the executing entity renders the status of indicator weights, business anomalies, and cross-system collaboration instructions in real time, it first receives feedback information from users regarding business anomaly identifiers, and calibrates the preset thresholds based on the feedback information to obtain calibrated thresholds. Then, the executing entity uploads the calibrated thresholds to the analysis and calculation layer.

[0090] In one possible embodiment, after the executing entity renders the status of indicator weights, business anomalies, and cross-system collaboration instructions in real time, it provides a user feedback entry point through a visual interaction layer to receive user feedback information on business anomaly indicators. This feedback information includes user confirmation of the anomaly indicator (e.g., "Confirmed that the anomaly is a real business problem") or denial (e.g., "The anomaly is a false alarm and belongs to normal fluctuations"). The executing entity adjusts the preset threshold according to the user feedback: if the user repeatedly reports that a certain type of anomaly indicator is a false alarm, it means that the original threshold is too low, and the system will appropriately increase the anomaly judgment threshold for that type of indicator; if the user reports that a certain type of anomaly is not identified in time, it means that the original threshold is too high, and the system will lower the corresponding threshold. Through this calibration, the executing entity obtains a calibrated threshold that is more in line with the actual business scenario.

[0091] The executing entity uploads the calibrated threshold to the analysis and computing layer through the internal data interface. After receiving it, the analysis and computing layer applies it to the subsequent anomaly detection process: when the isolated forest algorithm calculates anomaly scores, the calibrated threshold is used as the new judgment standard to redefine the labeling rules of anomaly indicators. At the same time, when the LSTM model dynamically adjusts the indicator weights, it also refers to the changes in the business scenario corresponding to the calibrated threshold, optimizes the attention mechanism's assessment of the importance of indicators, and improves the overall accuracy and adaptability of anomaly detection.

[0092] As described above, the marketing quantitative indicator monitoring method provided in this disclosure utilizes a multimodal input interface and a distributed message queue to process raw marketing data. It generates real-time calculated indicators through real-time windowed aggregation, combines this with a historical benchmark database built from offline statistics, and dynamically adjusts indicator weights based on a time-series prediction model. It uses the historical benchmark database and unsupervised detection algorithms to identify business anomalies, and finally generates decision suggestions based on these anomalies, triggering cross-system collaboration instructions. Simultaneously, it renders relevant statuses in real-time and uploads user feedback. This process fully integrates a distributed computing framework and intelligent algorithms, achieving end-to-end intelligent and collaborative management of marketing data from collection, processing, and analysis to decision execution. It effectively improves the real-time performance, accuracy, and automation level of marketing monitoring, enhances cross-system collaboration efficiency, reduces reliance on manual labor and operational costs, provides efficient support for enterprise marketing decisions, and adapts to diverse marketing scenario needs.

[0093] Example 3 Based on the above embodiments, such as Figure 3 The diagram shows a data acquisition layer interface. This interface is used to configure and manage the system to collect raw data from various marketing data sources (such as advertising platforms, e-commerce platforms, social media, CRM systems, etc.) in real time or near real time. This data serves as the system's data entry point, ensuring the breadth and timeliness of the data. It should be noted that... Figure 3 This is merely an illustrative diagram of the interface of a data acquisition layer. In practical applications, the interface of the data acquisition layer can include more functions, which are not limited here.

[0094] In one possible embodiment, the interface supports selecting data source types such as API interfaces, database connections, log files, and streaming media data, and can configure connection parameters such as access credentials, network addresses, and data formats. The interface can flexibly set the collection frequency according to the real-time requirements of marketing metrics, and can select marketing-related fields from the original data source and rename them according to system specifications. The interface allows defining conversion rules from original data fields to internal system data types, configuring missing value filling or deletion strategies, and defining outlier detection and filtering rules based on statistical methods or business rules. The interface displays the connection status of each data source in real time to ensure normal data flow, and visually displays data collection speed and latency, records error information during the collection process, and can trigger alarms according to preset rules.

[0095] As can be seen from the above, the data acquisition layer supports rapid access to new marketing data sources and adjustment of acquisition strategies through flexible configuration options, which greatly reduces the difficulty and cost of system expansion; its built-in preprocessing capabilities can perform preliminary cleaning and transformation of data before it enters the core processing stage, ensuring the quality and efficiency of subsequent data processing; at the same time, the real-time feedback mechanism can promptly detect and resolve data acquisition problems through real-time status monitoring and alarms, ensuring the accuracy and timeliness of marketing indicators.

[0096] Example 4 Based on the above embodiments, such as Figure 4 The diagram shows a schematic of the data processing layer interface. This interface is the configuration hub of the system's core data pipeline, responsible for defining and managing how raw data acquired from the data acquisition layer is processed, transformed, aggregated, and stored in real time using distributed computing frameworks (such as Flink and Spark Streaming). The goal is to transform raw, heterogeneous marketing data into structured, analyzable quantitative indicator data. It should be noted that... Figure 4 This is merely an illustrative diagram of a data processing layer interface. In practical applications, the data processing layer interface can include more functions, which are not limited here.

[0097] In one possible implementation, the interface allows users to select the underlying distributed stream processing framework (such as Apache Flink or Spark Streaming). The interface defines data stream processing stages (such as data parsing, filtering, marketing event identification, and metric aggregation) through a graphical or configurable approach, and supports features similar to Spark. This interface uses SQL query language for simple ETL tasks and integrates a programming interface or script editor for users to write custom processing functions to handle complex logic. It allows defining atomic metric calculation formulas (such as "clicks" and "impressions"), supports building composite metrics based on atomic metrics, and configures time windows (such as scrolling windows and sliding windows) and dimension aggregations (such as by campaign, channel, region, and time) for real-time metrics. It also allows configuring the storage location of processed marketing metric data (such as a real-time data warehouse, relational database, or data lake) and defining the table structure of the metric data in the target storage (including field names, data types, and indexes). Furthermore, it configures retry strategies, error data isolation, and alarm mechanisms for data processing failures. Finally, it manages the task lifecycle, monitors the CPU, memory, and network I / O resource usage of tasks in the distributed cluster, and displays the end-to-end latency from data entry into the processing stage to the output result in real time.

[0098] As can be seen from the above, the data processing layer has integrated stream and batch processing capabilities, which can process real-time stream data and historical batch data simultaneously to ensure comprehensive and accurate marketing metrics. The data processing layer supports flexible metric customization, allowing users to customize metrics and calculation logic according to business needs without modifying the underlying code, thus improving the system's adaptability and flexibility. In the data processing layer, the configuration and management of complex stream processing tasks can be simplified through visual task orchestration, reducing the barrier to entry for non-technical personnel.

[0099] Example 5 Based on the above embodiments, such as Figure 5 The diagram shows an interface for the analysis and computation layer. This interface is the core of the system's intelligent analysis, responsible for in-depth analysis, trend prediction, anomaly detection, and attribution analysis of the processed marketing quantitative indicator data. It utilizes machine learning, statistical models, and other technologies to provide intelligent insights for marketing decisions. It should be noted that... Figure 5 This is merely an illustrative diagram of the interface of the analysis and computing layer. In practical applications, the interface of the analysis and computing layer can include more functions, which are not limited here.

[0100] In one possible embodiment, the interface offers a variety of pre-built analysis models to choose from, such as time series prediction models, anomaly detection models, and attribution models. It allows users to adjust key model parameters, select historical data for model training, and provides performance evaluation metrics (such as accuracy and recall). The trained model can be deployed to a real-time analysis engine and updated online. The interface supports setting alarm rules based on fixed thresholds and dynamically adjusting alarm thresholds based on the confidence interval or historical fluctuation range output by the prediction model. It utilizes machine learning algorithms to identify abnormal behaviors in marketing metrics that do not conform to preset patterns (such as sudden drops in traffic or signs of click fraud). Within this interface, users can select different marketing channel attribution models (such as linear attribution and time decay attribution) to evaluate the contribution of different marketing touchpoints to the final conversion and define the time window for attribution analysis (such as all touchpoints within 7 days before conversion). Users can also select specific future periods (days / weeks / months) for data prediction (such as sales revenue and user growth), compare the prediction results (including predicted values ​​and confidence intervals) with actual data, and decompose marketing metric data into trend, seasonal, periodic, and residual components to help understand data patterns.

[0101] As described above, the analytics and computing layer integrates multiple machine learning models to achieve AI-driven intelligent analysis, automatically detecting data anomalies, predicting future trends, and performing multi-channel attribution analysis, transforming traditional passive monitoring into proactive intelligent insight. The analytics and computing layer supports users in selecting and configuring analysis models according to business scenarios, providing model training and evaluation results, enhancing the credibility and interpretability of analysis conclusions, and providing marketers with immediate warnings and decision-making basis through real-time anomaly detection and prediction, assisting in rapid response to market changes.

[0102] Example 6 Based on the above embodiments, such as Figure 6 As shown, a schematic diagram of a visual interaction layer is provided. This interface is the main window for users to interact with the system, aiming to present complex marketing quantitative indicators and analysis results to users in intuitive and easy-to-understand charts and reports. It supports users to perform multi-dimensional drill-down, filtering, and custom views, and is a key support for data-driven marketing decisions. It should be noted that... Figure 6 This is merely an illustrative diagram of a visual interaction layer interface. In practical applications, the interface of the visual interaction layer can contain more functions, which are not limited here.

[0103] In one possible embodiment, the interface allows users to filter and combine data based on marketing metrics such as time, channel, product, region, and user group. It also supports drilling down from a high-level overview (e.g., total sales) to specific details (e.g., sales of a particular product on a particular channel), facilitating in-depth data mining. The interface provides various commonly used chart types, including bar charts, line charts, pie charts, scatter plots, and heatmaps. Users can select metrics, dimensions, and chart types to create personalized data views, supporting the creation of personalized dashboards to centrally display multiple related charts and key metrics, achieving a "one-screen overview." The dashboard and charts can be configured with an automatic refresh rate to ensure the data is up-to-date. For indicators with high real-time requirements, the latest values ​​are displayed in a dynamic format. This interface centrally displays abnormal alarm information detected by the system, including alarm level, time, indicator name, and anomaly description. Clicking on an alarm allows users to view details and jump to the relevant data view for problem tracing and analysis. Users can mark, comment on, or close alarms, forming a closed-loop processing mechanism. This interface supports setting up scheduled generation of marketing analysis reports and sending them to a specified email address, and allows exporting the data and charts of the current view to multiple formats such as CSV, Excel, PNG, and PDF.

[0104] As can be seen from the above, users can explore multi-dimensional data without complex SQL queries by dragging, clicking, and other methods on the visual interaction layer. This lowers the threshold for data analysis, brings an intuitive and easy-to-use interactive experience, meets the personalized data display needs of different marketing roles, improves work efficiency, and can seamlessly integrate data visualization with intelligent alerts, enabling users to discover problems in a timely manner and quickly locate them.

[0105] Example 7 Based on the above embodiments, such as Figure 7 The image shows a schematic diagram of a marketing data display interface. This interface is a highly customized data display platform designed specifically for marketers. Its core function is to transform complex quantitative indicators into insights with business guidance by integrating them into the business context. It includes modules such as a marketing dashboard, marketing campaign effectiveness evaluation, and user profile analysis. It should be noted that... Figure 7 This is merely an illustrative diagram of a marketing data display interface. In practical applications, the marketing data display interface can include more functions, which are not limited here.

[0106] In one possible embodiment, the interface centrally displays key marketing campaign KPIs (such as ROI, CPA, LTV, etc.). This interface compares current metrics with historical averages or target values ​​to quickly assess marketing performance. It displays marketing metrics performance by product line, customer group, region, and other market segments, assisting in the development of differentiated strategies. The interface supports selecting specific marketing campaigns and setting evaluation periods, comparing the performance of different channels in the same campaign, evaluating the contribution of each channel, and visually displaying the user's journey from reach to conversion. It identifies key conversion nodes and bottlenecks, compares and analyzes the test results of marketing campaigns, and evaluates the advantages and disadvantages of different strategies. The interface can segment users based on user behavior data (such as purchase frequency, browsing preferences, etc.), display the marketing performance of each group, analyze the behavior patterns and paths of high-value users, provide a basis for precision marketing, assess the lifetime value (LTV) of users at different stages, and assist in the formulation of customer retention strategies. The interface supports drilling down from macro indicators to specific issues, such as drilling down from a decline in overall sales to an abnormal click-through rate of a certain advertising campaign. Combining the conclusions of the analysis model and business rules, it automatically generates or recommends marketing action suggestions, such as suggesting adjustments to the placement strategy for advertising costs exceeding the budget, and suggesting targeted promotions for regions with high user churn rates.

[0107] As can be seen from the above, the design of the marketing data display interface closely aligns with the marketing business process, directly transforming data analysis results into business insights and action recommendations, and achieving customized display driven by business needs. The marketing data display interface provides multi-dimensional and multi-level marketing campaign evaluation tools to help fully understand campaign effectiveness. By integrating analysis insights and action recommendations, the marketing data display interface elevates the system's function from "identifying problems" to "solving problems," significantly improving the intelligence level and efficiency of marketing decision-making.

[0108] As described above, this disclosed technical solution constructs a full-link real-time processing architecture by deeply integrating distributed technologies such as Kafka, Flink, and Spark SQL, enabling millisecond-level collection, aggregation, and dynamic computation of multi-source heterogeneous marketing data. This solution innovatively employs an LSTM time-series model to dynamically adjust indicator weights to adapt to market fluctuations, combined with an improved isolated forest algorithm to achieve high-precision anomaly detection, overcoming the dual bottlenecks of poor real-time performance and high false alarm rates inherent in traditional solutions. Simultaneously, this solution relies on a stream-batch integrated computing model to collaboratively analyze real-time data and historical benchmark databases, generating interpretable business anomaly identifiers and root cause localization. Through a 3D visualization interactive platform and automated response mechanism, monitoring results are transformed into cross-departmental collaborative instructions and executable decision suggestions, ultimately forming a closed loop of "data perception - intelligent analysis - decision execution," significantly improving the timeliness, accuracy, and business linkage efficiency of marketing monitoring, becoming a core infrastructure for enterprise data-driven marketing.

[0109] Example 8 Based on the above embodiments, this disclosure also provides an electronic device, including: at least one processor; a memory for storing executable instructions of the at least one processor; wherein the at least one processor is configured to execute the instructions to implement the above-described method disclosed in this disclosure.

[0110] Figure 8 This is a schematic diagram of the structure of an electronic device provided as an exemplary embodiment of this disclosure. For example... Figure 8 As shown, the electronic device 800 includes at least one processor 801 and a memory 802 coupled to the processor 801. The processor 801 can perform the corresponding steps in the methods disclosed in the embodiments of this disclosure.

[0111] The processor 801 described above can also be called a central processing unit (CPU), which can be an integrated circuit chip with signal processing capabilities. Each step in the method disclosed in this embodiment can be implemented by the integrated logic circuitry in the processor 801 or by software instructions. The processor 801 can be a general-purpose processor, a digital signal processor (DSP), an ASIC, a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in this embodiment can be directly implemented by a hardware decoding processor, or implemented by a combination of hardware and software modules in the decoding processor. The software modules can be located in the memory 802, such as random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, or other mature storage media in the art. The processor 801 reads information from the memory 802 and, in conjunction with its hardware, completes the steps of the method described above.

[0112] Furthermore, various operations / processes according to this disclosure, implemented via software and / or firmware, can be transmitted from a storage medium or network to a computer system with a dedicated hardware architecture, such as... Figure 9 The computer system 900 shown is equipped with the programs that constitute the software. When various programs are installed, the computer system is able to perform various functions, including functions such as those described above. Figure 9 A block diagram of a computer system provided for an exemplary embodiment of this disclosure.

[0113] Computer system 900 is intended to represent various forms of digital electronic computer devices, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. Electronic devices can also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the present disclosure described and / or claimed herein.

[0114] like Figure 9As shown, the computer system 900 includes a computing unit 901, which can perform various appropriate actions and processes based on a computer program stored in a read-only memory (ROM) 902 or a computer program loaded into a random access memory (RAM) 903 from a storage unit 908. The RAM 903 may also store various programs and data required for the operation of the computer system 900. The computing unit 901, ROM 902, and RAM 903 are interconnected via a bus 904. An input / output (I / O) interface 905 is also connected to the bus 904.

[0115] Multiple components in the computer system 900 are connected to the I / O interface 905, including: an input unit 906, an output unit 907, a storage unit 908, and a communication unit 909. The input unit 906 can be any type of device capable of inputting information into the computer system 900. The input unit 906 can receive input numerical or character information and generate key signal inputs related to user settings and / or function control of the electronic device. The output unit 907 can be any type of device capable of presenting information and may include, but is not limited to, a monitor, speaker, video / audio output terminal, vibrator, and / or printer. The storage unit 908 may include, but is not limited to, a hard disk and an optical disk. The communication unit 909 allows the computer system 900 to exchange information / data with other devices via a network such as the Internet, and may include, but is not limited to, a modem, network card, infrared communication device, wireless communication transceiver, and / or chipset, such as Bluetooth™ device, WiFi device, WiMax device, cellular communication device, and / or the like.

[0116] The computing unit 901 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 901 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 901 performs the various methods and processes described above. For example, in some embodiments, the methods disclosed in this disclosure can be implemented as a computer software program tangibly contained in a machine-readable medium, such as storage unit 908. In some embodiments, part or all of the computer program can be loaded and / or installed on the electronic device 800 via ROM 902 and / or communication unit 909. In some embodiments, the computing unit 901 can be configured to perform the methods disclosed in this disclosure by any other suitable means (e.g., by means of firmware).

[0117] This disclosure also provides a computer-readable storage medium, wherein when the instructions in the computer-readable storage medium are executed by a processor of an electronic device, the electronic device is able to perform the methods disclosed in this disclosure.

[0118] The computer-readable storage medium in this disclosure can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. The aforementioned computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specifically, the aforementioned computer-readable storage medium may include electrical connections based on one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0119] The aforementioned computer-readable medium may be included in the aforementioned electronic device; or it may exist independently and not assembled into the electronic device.

[0120] Figure 10 A computer program product 1000 is provided as an exemplary embodiment of the present disclosure. The computer program product 1000 includes a computer program 1001, wherein the computer program 1001, when executed by a processor, implements the methods disclosed in the embodiments of the present disclosure.

[0121] In embodiments of this disclosure, computer program code for performing the operations of this disclosure can be written in one or more programming languages ​​or a combination thereof. These programming languages ​​include, but are not limited to, object-oriented programming languages ​​such as Java, Smalltalk, and C++, as well as conventional procedural programming languages ​​such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network (including a local area network (LAN) or a wide area network (WAN)), or it can be connected to an external computer.

[0122] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0123] The modules, components, or units described in the embodiments of this disclosure can be implemented in software or hardware. The names of the modules, components, or units do not necessarily constitute a limitation on the module, component, or unit itself.

[0124] The functions described above in this document can be performed at least in part by one or more hardware logic components. For example, without limitation, exemplary hardware logic components that can be used include: field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), system-on-a-chip (SoCs), complex programmable logic devices (CPLDs), and so on.

[0125] The above description is merely an embodiment of this disclosure and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of this disclosure is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the above-described concept. For example, technical solutions formed by substituting the above features with (but not limited to) technical features disclosed in this disclosure that have similar functions.

[0126] While specific embodiments of this disclosure have been described in detail by way of example, those skilled in the art should understand that the examples are for illustrative purposes only and not intended to limit the scope of this disclosure. Those skilled in the art should understand that modifications can be made to the above embodiments without departing from the scope and spirit of this disclosure. The scope of this disclosure is defined by the appended claims.

Claims

1. A system for monitoring marketing quantitative indicators, characterized by, The system comprises a data collection layer, a data processing layer, an analysis calculation layer, and a visual interaction layer. The data collection layer is configured to acquire original marketing data in real time through a multi-modal input interface, and process the original marketing data through a distributed message queue to obtain processed marketing data. The data processing layer is configured to perform real-time window aggregation on the processed marketing data to generate real-time calculation indicators, and perform offline statistics on batch marketing data stored historically to construct a historical benchmark database. The analysis calculation layer is configured to dynamically adjust indicator weights based on the real-time calculation indicators through a time series prediction model. Business anomalies are identified based on the historical benchmark database and an unsupervised detection algorithm. Decision suggestions are generated based on the business anomalies, and cross-system collaboration instructions are triggered based on the decision suggestions. The visual interaction layer is configured to render the states of the indicator weights, the business anomalies, and the cross-system collaboration instructions in real time. The method comprises:

2. A method of monitoring marketing quantitative indicators, characterized in that, acquiring original marketing data in real time through a multi-modal input interface, and processing the original marketing data through a distributed message queue to obtain processed marketing data; performing real-time window aggregation on the processed marketing data to generate real-time calculation indicators, and performing offline statistics on batch marketing data stored historically to construct a historical benchmark database; dynamically adjusting indicator weights based on the real-time calculation indicators through a time series prediction model, identifying business anomalies based on the historical benchmark database and an unsupervised detection algorithm, generating decision suggestions based on the business anomalies, and triggering cross-system collaboration instructions based on the decision suggestions; rendering the states of the indicator weights, the business anomalies, and the cross-system collaboration instructions in real time, and uploading user feedback to the analysis calculation layer. The processing of the original marketing data through the distributed message queue to obtain processed marketing data comprises:

3. The method of claim 2, wherein, classifying the original marketing data by event type to obtain data types, wherein the data types include promotion activity order data, user click stream data, and advertisement exposure data; allocating a message topic for the data types, and partitioning the original marketing data by event core identifier within the message topic to obtain a first partitioned data set, wherein the first partitioned data set includes a first sub-data set divided according to the event type and the core identifier; dynamically adjusting the number of first sub-data sets according to real-time data traffic to obtain processed marketing data. The processing of the original marketing data through the distributed message queue to obtain processed marketing data comprises:

4. The method of claim 2, wherein, partitioning the original marketing data by marketing region or user group to obtain a second partitioned data set, wherein the second partitioned data set includes a second sub-data set divided according to the marketing region or the user group; dynamically adjusting the number of second sub-data sets according to real-time data traffic to obtain processed marketing data. The real-time window aggregation on the processed marketing data to generate real-time calculation indicators comprises:

5. The method of claim 2, wherein, ​ Setting an event time window parameter, and performing stream aggregation on the processed marketing data based on the event time window parameter; Identifying a business dimension to group and calculate the processed marketing data, and outputting real-time calculation indicators containing timestamps and dimension identifications.

6. The method of claim 2, wherein, The offline statistics of the historical batch marketing data include: Extracting time period indicators from the historical batch marketing data; Calculating the average and standard deviation of the time period indicators, and constructing the historical benchmark database based on the average and the standard deviation.

7. The method of claim 2, wherein, The dynamic adjustment of the indicator weights based on the real-time calculation indicators through a time series prediction model includes: Inputting the real-time calculation indicators into a time series prediction model to capture the dynamic dependency between indicators, and evaluating the importance of the indicators to the decision target through the attention mechanism in the time series prediction model; Generating a weight distribution coefficient according to the importance, and weighting and fusing the real-time calculation indicators based on the weight distribution coefficient to obtain the target indicator value after weight adjustment.

8. The method of claim 2, wherein, The identification of business anomalies through the historical benchmark database and unsupervised detection algorithms includes: Comparing the real-time calculation indicators with the standard deviation of the historical benchmark database to obtain a deviation degree; When the deviation degree exceeds a preset threshold, labeling the real-time calculation indicators as candidate anomalies; Verifying and scoring the real-time calculation indicators with the candidate anomaly label, and outputting business anomaly identifications containing anomaly scores.

9. The method of claim 8, wherein, The generation of decision suggestions based on the business anomalies, and the triggering of cross-system collaboration instructions based on the decision suggestions include: Matching predefined response rules according to the business anomaly identifications, and generating decision suggestions containing execution actions according to the real-time calculation indicators and the response rules; Parsing the decision suggestions into cross-system collaboration instructions, and calling external business systems through application program interfaces to execute the cross-system collaboration instructions; Real-time acquisition and output of the execution status of the cross-system collaboration instructions.

10. The method of claim 8, wherein, The uploading of user feedback to the analysis computing layer includes: Receiving user feedback information on the business anomaly identifications, and calibrating the preset threshold based on the feedback information to obtain a calibrated threshold; Uploading the calibrated threshold to the analysis computing layer.

Images