Data mining methods, apparatus and storage media

Abstract

This application relates to a data mining method, apparatus, and storage medium. The data mining method includes: acquiring real-time data; classifying registered modular algorithm models via an algorithm repository management platform, assembling the modular algorithm models under each category into independent card units, and dynamically sending each independent card unit to an application platform for configuration; detecting a first operation command for the independent card unit via the application platform; responding to the detected first operation command, determining the target algorithm model in the modular algorithm model, and logically configuring multiple target algorithm models via the algorithm repository management platform to obtain a target workflow; and invoking the target workflow to perform data mining on the real-time data to generate data mining results. This application solves the problem of low accuracy in data mining under complex application scenarios.

Description

Technical Field

[0001] This application relates to the field of big data, and in particular to data mining methods, apparatus and storage media. Background Technology

[0002] Data mining technology generally refers to methods for extracting hidden information from large amounts of data using algorithms. It is widely used in many industries, such as finance, telecommunications, and transportation. However, when processing massive amounts of data, data mining platforms are often designed by professional developers. Non-professional users find it difficult to quickly master the complex algorithm selection and configuration processes, which hinders the widespread adoption of big data mining and leads to lower accuracy in complex application scenarios.

[0003] Currently, no effective solution has been proposed to address the issue of low accuracy in data mining under complex application scenarios in related technologies. Summary of the Invention

[0004] This application provides a data mining method, apparatus, and storage medium to at least address the problem of low accuracy in data mining under complex application scenarios in related technologies.

[0005] In a first aspect, embodiments of this application provide a data mining method, the method comprising:

[0006] Obtain real-time data;

[0007] The registered modular algorithm models are classified through the algorithm repository management platform. The modular algorithm models under each category are assembled into independent card units, and each independent card unit is dynamically sent to the application platform for configuration.

[0008] The application platform detects a first operation instruction for the independent card unit; in response to the detected first operation instruction, the target algorithm model in the modular algorithm model is determined, and multiple target algorithm models are logically configured via the algorithm warehouse management platform to obtain a target workflow.

[0009] The target workflow is invoked to perform data mining on the real-time data and generate data mining results.

[0010] In some embodiments, the step of responding to the detected first operation instruction, determining the target algorithm model in the modular algorithm model, and logically configuring multiple target algorithm models via the algorithm repository management platform to obtain a target workflow includes:

[0011] Parse the first operation instruction to determine the operated card unit in the independent card unit indicated by the first operation instruction;

[0012] The target algorithm model corresponding to the operated card unit is determined from each of the modular algorithm models, the target algorithm model is configured according to the reference model parameters in the operated card unit, and the target workflow is obtained through the algorithm warehouse management platform.

[0013] In some embodiments, configuring the target algorithm model according to the reference model parameters in the operated card unit and obtaining the target workflow via the algorithm repository management platform includes:

[0014] The algorithm repository management platform detects the parameter adjustment instructions received by the application platform.

[0015] The reference model parameters are adjusted based on the parameter adjustment instructions to obtain the target model parameters, and the target algorithm model is configured according to the target model parameters to obtain the target workflow.

[0016] In some embodiments, the real-time data is multimodal data obtained from a multimodal data source.

[0017] In some embodiments, the step of invoking the target workflow to perform data mining processing on the real-time data and generate data mining results includes:

[0018] The target workflow is invoked to perform data mining processing on the real-time data, generating a first data mining result.

[0019] In response to the first data mining result, business logic rules are constructed based on the second operation instruction detected by the application platform;

[0020] Based on the attribute feature information of the multimodal data, the logical judgment conditions corresponding to the business logic rules are determined, and dynamic logical configuration information is generated;

[0021] Based on the dynamic logic configuration information, a detection event is triggered; in response to the triggered detection event, a second data mining result is generated; the data mining result includes the first data mining result and the second data mining result.

[0022] In some embodiments, determining the logical judgment conditions corresponding to the business logic rules based on the attribute feature information of the multimodal data and generating dynamic logical configuration information includes:

[0023] Obtain preset alarm rules;

[0024] Detect the rule adjustment instruction from the application platform; in response to the received rule adjustment instruction, adjust the alarm rule to obtain the alarm triggering condition;

[0025] Based on the attribute feature information of the multimodal data, the logical judgment conditions are determined, and the dynamic logical configuration information is generated according to the logical judgment conditions and the alarm triggering conditions.

[0026] In some embodiments, after generating the second data mining result, the method further includes:

[0027] In response to the second data mining result, target labeled data is obtained via the application platform;

[0028] Based on the target labeled data, an optimized configuration scheme is generated, and the target algorithm model is adjusted based on the optimized configuration scheme to obtain a new algorithm model.

[0029] In some embodiments, acquiring real-time data includes:

[0030] When the data acquisition task is a multi-concurrent task, the algorithm warehouse management platform sends the task information of each concurrent task to the corresponding algorithm service node, and calls each algorithm service node to execute the concurrent task in parallel to obtain the real-time data.

[0031] Secondly, embodiments of this application provide a data mining apparatus, comprising:

[0032] The acquisition unit is used to acquire real-time data;

[0033] The independent card configuration unit is used to classify the registered modular algorithm models through the algorithm warehouse management platform, assemble the modular algorithm models under each category into independent card units, and dynamically send each independent card unit to the application platform for configuration.

[0034] The workflow generation unit is configured to detect a first operation instruction for the independent card unit via the application platform; in response to the detected first operation instruction, determine the target algorithm model in the modular algorithm model, and logically configure multiple target algorithm models to obtain a target workflow;

[0035] The data mining department is used to invoke the target workflow to perform data mining on the real-time data and generate data mining results.

[0036] Thirdly, embodiments of this application provide a storage medium storing a computer program thereon, which, when executed by a processor, implements the data mining method as described in the first aspect above.

[0037] Compared to related technologies, the data mining method, apparatus, and storage medium provided in this application embodiment acquire real-time data; classify registered modular algorithm models through an algorithm warehouse management platform, assemble the modular algorithm models under each category into independent card units, and dynamically send each independent card unit to an application platform for configuration; detect a first operation instruction for the independent card unit through the application platform; in response to the detected first operation instruction, determine the target algorithm model in the modular algorithm model, and logically configure multiple target algorithm models through the algorithm warehouse management platform to obtain a target workflow; call the target workflow to perform data mining on the real-time data and generate data mining results.

[0038] By standardizing and card-based encapsulating algorithms through the algorithm repository management platform, the reusability and maintainability of algorithm components are significantly improved. Simultaneously, the dynamic card push mechanism enables front-end applications to quickly adapt to new algorithms, reducing technical integration costs. Combined with the application platform's visual orchestration capabilities, users can autonomously build workflows through drag-and-drop operations on a low-code interface, transforming professional algorithm capabilities into business-understandable configuration items. This constructs an end-to-end adaptive analysis system that meets personalized business needs while reducing manual intervention through automated configuration. This improves development efficiency and enhances the real-time performance and accuracy of data processing and analysis, effectively solving the problem of low accuracy in data mining under complex application scenarios.

[0039] Details of one or more embodiments of this application are set forth in the following drawings and description to make other features, objects and advantages of this application more readily apparent. Attached Figure Description

[0040] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings:

[0041] Figure 1 This is a hardware structure block diagram of a terminal for a data mining method according to an embodiment of this application;

[0042] Figure 2 This is a flowchart of a data mining method according to an embodiment of this application;

[0043] Figure 3 This is a schematic diagram of a graphical card interface according to an embodiment of this application;

[0044] Figure 4 This is a schematic diagram of the structure of a data mining method according to an embodiment of this application;

[0045] Figure 5This is a structural block diagram of a data mining apparatus according to an embodiment of this application. Detailed Implementation

[0046] To make the objectives, technical solutions, and advantages of this application clearer, the application is described and illustrated below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the application. All other embodiments obtained by those skilled in the art based on the embodiments provided in this application without inventive effort are within the scope of protection of this application. Furthermore, it is understood that although the efforts made in such a development process may be complex and lengthy, for those skilled in the art related to the content disclosed in this application, modifications to design, manufacturing, or production based on the technical content disclosed in this application are merely conventional technical means and should not be construed as insufficient disclosure of the content of this application.

[0047] In this application, the reference to "embodiment" means that a specific feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places in the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment that is mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described in this application may be combined with other embodiments without conflict.

[0048] Unless otherwise defined, the technical or scientific terms used in this application shall have the ordinary meaning understood by one of ordinary skill in the art to which this application pertains. The terms “a,” “an,” “an,” “the,” and similar words used in this application do not indicate quantity limitation and may indicate singular or plural. The terms “comprising,” “including,” “having,” and any variations thereof used in this application are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or device that includes a series of steps or modules (units) is not limited to the listed steps or units, but may also include steps or units not listed, or may include other steps or units inherent to these processes, methods, products, or devices. The terms “connected,” “linked,” “coupled,” and similar words used in this application are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. “Multiple” used in this application means two or more. “And / or” describes the relationship between related objects, indicating that three relationships may exist; for example, “A and / or B” can represent: A alone, A and B simultaneously, and B alone. The terms “first,” “second,” “third,” etc., used in this application are merely to distinguish similar objects and do not represent a specific ordering of the objects.

[0049] The method embodiments provided in this example can be executed on a terminal, computer, or similar computing device. Taking running on a terminal as an example, Figure 1 This is a hardware structure block diagram of a terminal for a data mining method according to an embodiment of this application. For example... Figure 1 As shown, a terminal may include one or more ( Figure 1 Only one is shown in the diagram. A processor 102 (which may include, but is not limited to, a microprocessor MCU or a programmable logic device FPGA, etc.) and a memory 104 for storing data are also shown. Optionally, the terminal may further include a transmission device 106 for communication functions and an input / output device 108. Those skilled in the art will understand that... Figure 1 The structure shown is for illustrative purposes only and does not limit the structure of the terminal described above. For example, the terminal may also include components that are larger than... Figure 1 The more or fewer components shown, or having the same Figure 1 The different configurations shown.

[0050] The memory 104 can be used to store computer programs, such as application software programs and modules, like the computer program corresponding to the data mining method in this embodiment. The processor 102 executes various functional applications and data processing by running the computer program stored in the memory 104, thus implementing the aforementioned method. The memory 104 may include high-speed random access memory and non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory 104 may further include memory remotely located relative to the processor 102, and these remote memories can be connected to the terminal via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0051] The transmission device 106 is used to receive or send data via a network. Specific examples of the network described above may include a wireless network provided by the terminal's communication provider. In one example, the transmission device 106 includes a Network Interface Controller (NIC), which can connect to other network devices via a base station to communicate with the Internet. In another example, the transmission device 106 may be a Radio Frequency (RF) module used for wireless communication with the Internet.

[0052] As mentioned in the background, while existing data mining platforms can perform basic data storage and management functions when processing massive amounts of view data, they still have the following shortcomings when dealing with complex and diverse application scenarios: 1. Low flexibility: They lack modular design, making it difficult to quickly integrate and replace the latest object detection, anomaly detection, and other algorithm models, and thus struggling to adapt to the rapid iteration of artificial intelligence algorithms; 2. High user threshold: Existing platforms are mostly designed by professional developers, making it difficult for non-professional users to quickly master the complex algorithm selection and configuration process, affecting the popularization of big data mining; 3. Lack of in-depth visualization and analysis capabilities: Existing platforms only provide basic detection result displays, failing to intuitively and dynamically present data characteristics and analysis results, limiting users' ability to deeply understand and make decisions regarding the causes of anomalies, data distribution, and optimization directions.

[0053] Based on this, this embodiment provides a data mining method. Figure 2 This is a flowchart of a data mining method according to an embodiment of this application, such as... Figure 2 As shown, the process includes the following steps:

[0054] Step S210: Obtain real-time data.

[0055] In this step, multi-source data is collected in real time using sensors and cameras deployed in scenarios such as high-speed rail turnstiles. For example, real-time video data can be input to the data mining platform via network video streams, such as Real-Time Streaming Protocol (RTSP), from cameras near the turnstiles. The data acquisition and preprocessing module of this data mining platform employs efficient video decoding technology, utilizing hardware decoding (such as a graphics processing unit, GPU) to decode the video stream and decompose it into consecutive frames. To reduce processing redundancy, the module can also set up a keyframe extraction mechanism, extracting one frame every five frames as a keyframe. This reduces the amount of data while preserving the continuity of the frame sequence, providing support for subsequent analysis.

[0056] After decoding, keyframes enter the data preprocessing module for standardization, including format conversion, size scaling, and pixel normalization. For example, the original video frames are scaled from 1920×1080 to 640×480 to fit the requirements of the object detection model's input layer. Simultaneously, the module supports data augmentation strategies, such as random cropping, color dithering, and brightness adjustment, to enhance the model's robustness under different lighting conditions. Furthermore, the module utilizes a graphical interface, allowing users to configure preprocessing parameters, such as selecting frame intervals and adjusting image resolution, to meet the specific needs of different gate environments.

[0057] Step S220: The registered modular algorithm models are classified through the algorithm repository management platform. The modular algorithm models under each category are assembled into independent card units, and each independent card unit is dynamically sent to the application platform for configuration.

[0058] It should be further explained that this application uses an algorithm repository management platform to uniformly manage models, process orchestration, and computing resources. Specifically, the application platform or business front-end submits the algorithm model process configuration to the algorithm repository management platform; the algorithm repository management platform is responsible for scheduling computing resources and includes the algorithm repository platform address in the startup command; after each algorithm service starts, it registers with the platform and connects to the algorithm repository; the algorithm repository management platform allocates or creates specific video analysis tasks, with parameters including: task ID, video channel address, business algorithm ID, algorithm version number, model version number, etc. If the algorithm ID does not exist or the version number expires, the model list and download address will be automatically obtained and updated; the analysis task is created and started only after all dependencies are correctly loaded; after the task is successfully created, it starts running and reports results. During this process, algorithm service nodes continuously report events according to front-end requirements or detection rules; if the detection results meet preset threshold conditions, subsequent business logic or alarms are triggered.

[0059] More specifically, the algorithm repository management platform categorizes registered modular algorithm models according to their functional type, including object detection algorithms, anomaly detection algorithms, and re-identification algorithms. Within each category, related models are organized into independent card units. For example, the object detection category includes models such as Faster Region-based Convolutional Neural Network (Faster R-CNN), YOLOv3 (You Only Look Once version 3) to YOLOv5 (You Only Look Once version 5), and these are distributed to the application platform via a dynamic push mechanism. Each card is dynamically generated through a graphical card interface configured on the application platform, and includes interactive functions such as hovering to view details or clicking to select. The cards can be configured declaratively, such as using JSONSchema (a JSON-based specification) to define input feature dimensions, supporting plug-and-play and cross-node collaboration, and providing a standardized component library for low-code configuration.

[0060] The aforementioned algorithm repository management platform mainly includes: a model repository, which stores various pre-packaged algorithm models (model 1, 2, 3, 4...) and provides version management and metadata management capabilities; an algorithm orchestration platform, which combines or orchestrates models in a process-oriented manner to form reusable business algorithms (business algorithms 1, 2, 3...) to meet the needs of different scenarios; and a computing power management / scheduling platform, which dynamically allocates computing power resources such as GPU / CPU to each algorithm service node (node ​​1, 2, 3...), and can perform real-time scheduling and scaling up / down based on load conditions.

[0061] With the above architectural design, this invention effectively solves the shortcomings of traditional platforms in model management and computing power scheduling, enabling front-end services to quickly call and deploy models through drag-and-drop, selection, configuration, and other methods.

[0062] Step S230: The application platform detects the first operation instruction for the independent card unit; in response to the detected first operation instruction, the target algorithm model in the modular algorithm model is determined, and multiple target algorithm models are logically configured through the algorithm warehouse management platform to obtain the target workflow.

[0063] Specifically, the application platform can capture the user's first operation command (such as drag-and-drop combination, click, hover, etc.) on individual card units in real time through a graphical card interface. After parsing the operation intent, the platform uses the algorithm repository's orchestration engine to chain the selected target algorithm models according to business logic, ultimately forming an executable target workflow. The workflow supports complex structures such as sequential execution, branch routing (such as conditional judgments), and loop iteration, and inherits automation nodes. The graphical card interface dynamically updates based on the model selection results, displaying the currently selected model and its parameter settings. In multi-model combinations, it supports selecting multiple models simultaneously and setting parameters separately to meet the logical requirements of complex business processes.

[0064] In addition, the application platform can interact with the algorithm repository management platform through a graphical card interface. The application platform (business layer) can configure algorithms in a drag-and-drop manner in a low-code environment, such as selecting a pedestrian detection algorithm or anomaly detection algorithm; the algorithm repository management platform uniformly manages algorithm services, and multiple algorithm service nodes (ARS) can access the platform and execute different analysis tasks; the algorithm service nodes maintain communication with the platform during execution, and the platform can monitor the task status and load of each node in real time to ensure real-time processing capabilities under high concurrency.

[0065] Step S240: Invoke the target workflow to perform data mining on the real-time data and generate data mining results.

[0066] The generated target workflow is invoked, and the real-time data obtained in step S210 is input into the target workflow to perform distributed computing. The structured mining results (such as the probability value of trailing events) are generated through pipelined inference (such as target detection to spatiotemporal feature extraction to behavior classification). The results are written to the database in real time and alarms are triggered. At the same time, performance logs are recorded for closed-loop optimization.

[0067] Through steps S210 to S240, the algorithm repository management platform standardizes, categorizes, and encapsulates algorithms into cards, significantly improving the reusability and maintainability of algorithm components. Simultaneously, the dynamic card push mechanism enables front-end applications to quickly adapt to new algorithms, reducing technical integration costs. Combined with the application platform's visual orchestration capabilities, users can autonomously build workflows through drag-and-drop operations on a low-code interface, transforming professional algorithm capabilities into business-understandable configuration items. This constructs an end-to-end adaptive analysis system that meets personalized business needs while reducing manual intervention through automated configuration. It improves development efficiency while enhancing the real-time performance and accuracy of data processing and analysis, effectively solving the problem of low accuracy in data mining under complex application scenarios.

[0068] In some embodiments, the process of responding to a detected first operation instruction, determining a target algorithm model in a modular algorithm model, and logically configuring multiple target algorithm models via an algorithm repository management platform to obtain a target workflow may further include the following steps:

[0069] The first operation instruction is parsed to determine the operated card unit indicated by the first operation instruction in the independent card unit; the target algorithm model corresponding to the operated card unit is determined from each modular algorithm model, and the target algorithm model is configured according to the reference model parameters in the operated card unit, and the target workflow is obtained through the algorithm warehouse management platform.

[0070] In this embodiment, precise configuration and workflow orchestration of the algorithm model are achieved by parsing user interaction commands: When the application platform detects the first operation command (such as drag and drop, parameter adjustment, or connection relationship change) for an independent card unit, the system first parses the operation type and target object in the command to locate the specific card unit being operated on; then, through the mapping relationship between card metadata and algorithm repository registration information, the system matches the target algorithm model corresponding to the card from the modular algorithm model library. For example, the "face recognition card" dragged by the user corresponds to a pre-trained face recognition network (FaceNet); next, the target algorithm model is initialized and configured according to the reference model parameters embedded in the card unit (such as confidence threshold and detection area coordinates) to ensure that the model input and output interfaces are aligned with business requirements; finally, through the orchestration engine of the algorithm repository management platform, the configured target algorithm model is combined according to the logical relationship (such as sequential execution and conditional branch) implied in the first operation command to generate a target workflow that includes data flow, model call order, and exception handling mechanism. This workflow can be deployed and executed after version verification and performance estimation.

[0071] Specifically, please refer to Figure 3 This system is used to display a graphical card interface. On the right side of this interface, a menu bar showing all components, including atomic capabilities, business processing, and other components, is displayed. Users can assemble these components by dragging and dropping them to the center of the interface to create a target workflow. For example, when a user connects a "license plate recognition card" and a "trajectory analysis card" through the visual interface, the system automatically configures the parameters of the two models, such as the Region of Interest (ROI) for license plate detection and the time window for trajectory analysis. It then generates a series of workflows including image preprocessing, license plate localization, character recognition, and trajectory modeling, realizing a complete analysis chain from video stream input to tracking of violating vehicles.

[0072] Through the above embodiments, the operated card unit is accurately located by parsing the user's first operation command. Combined with the mapping relationship between card metadata and algorithm model, the target algorithm model can be quickly matched. This mechanism transforms traditional code-level development into visual configuration, significantly reducing the technical threshold and enabling business personnel to directly participate in algorithm process design. Secondly, dynamic configuration is performed based on the reference model parameters embedded in the card unit. This parameterized configuration method achieves accurate adaptation between algorithm capabilities and business needs while keeping the core logic of the model unchanged.

[0073] In some embodiments, the above-mentioned configuration of the target algorithm model based on the reference model parameters in the operated card unit and the acquisition of the target workflow via the algorithm repository management platform may further include the following steps:

[0074] The algorithm repository management platform detects parameter adjustment instructions received by the application platform; based on the parameter adjustment instructions, the reference model parameters are adjusted to obtain the target model parameters; and the target algorithm model is configured according to the target model parameters to obtain the target workflow.

[0075] Specifically, when a user initiates parameter adjustment operations (such as modifying detection sensitivity, adjusting classification thresholds, or specifying input data sources) on a configured algorithm card through the application platform, the platform first identifies the parameter type and adjustment range through its built-in instruction parsing module, generating a structured parameter adjustment instruction. After receiving the instruction, the algorithm repository management platform extracts the reference model parameters (such as pre-trained weight files and hyperparameter configuration tables) corresponding to the operated card from the centralized parameter library based on the mapping relationship between card metadata and model parameters, and ensures that the adjusted values ​​meet the model constraints (e.g., the confidence threshold must be in the 0-1 range) through the parameter validation engine. Subsequently, the platform adopts an incremental update strategy, modifying only the parameter items affected by the instruction and retaining other default parameters to generate target model parameters that match the specific business scenario. Next, the algorithm repository management platform injects the target parameters into the corresponding target algorithm model through the model service interface, triggering the model's hot loading mechanism to ensure that parameter changes take effect without restarting the service. Finally, the platform re-evaluates the dependencies of each node in the workflow based on the updated model parameters, optimizes the execution order through dynamic programming algorithms, and integrates exception handling logic (such as a rollback strategy when parameters are invalid), generating a target workflow that meets new requirements while maintaining system stability. For example, when a user adjusts the "Pedestrian Detection Confidence" in the video analytics card from 0.8 to 0.95, the system will automatically update the classification threshold parameters of the corresponding YOLOv8 model and rearrange the workflow including preprocessing, detection, and postprocessing to ensure high-precision detection while maintaining real-time processing performance.

[0076] Through the above embodiments, by detecting the parameter adjustment instructions of the application platform in real time, it is possible to quickly respond to users' refined needs for model performance, thereby improving workflow execution efficiency while ensuring system stability.

[0077] In some embodiments, the aforementioned real-time data is multimodal data obtained from multimodal data sources. Specifically, the aforementioned data mining platform supports dynamically acquiring data from multimodal data sources (such as time-series data and geographic location information), including target detection results in real-time video streams, positioning data in Geographic Information Systems (GIS), and time-series signals collected by sensors; after access, the data undergoes preprocessing, including format conversion, data normalization, and timestamp alignment, to ensure the consistency of the fusion of different modal data.

[0078] Based on this, the aforementioned invocation of the target workflow to perform data mining processing on real-time data and generate data mining results may further include the following steps:

[0079] The system invokes the target workflow to perform data mining on real-time data, generating a first data mining result. In response to the first data mining result, it constructs business logic rules based on a second operation instruction detected by the application platform. Based on the attribute feature information of the multimodal data, it determines the logical judgment conditions corresponding to the business logic rules and generates dynamic logic configuration information. Based on the dynamic logic configuration information, it triggers a detection event. In response to the triggered detection event, it generates a second data mining result. The data mining result includes the first data mining result and the second data mining result.

[0080] In this embodiment, the business logic support module in the aforementioned data mining platform uses a visual rule definition tool to realize dynamic logical judgment and rule configuration based on multimodal data; the method includes the following steps:

[0081] First, a pre-configured target workflow is invoked to perform initial mining on the real-time data stream, generating first data mining results containing basic analysis conclusions (such as device status indicators, user behavior tags, or target detection coordinates). Then, based on the first operation instructions issued by the user on the application platform (such as setting anomaly thresholds or defining association rules), the first results are transformed into executable business logic rules. Regarding the provision of the rule definition tool: the tool uses a graphical card interface, allowing users to construct business logic rules by dragging and dropping components or selecting preset templates; interface components include, but are not limited to, condition nodes, logical operators (such as "AND" and "OR"), and action nodes, supporting users in intuitively defining complex logical relationships.

[0082] Next, multimodal data access and parsing are performed. It supports dynamic data acquisition from multimodal data sources (such as time-series data and geographic location information), including target detection results from real-time video streams, location data from Geographic Information Systems (GIS), and time-series signals collected by sensors. After access, the data undergoes preprocessing, including format conversion, data normalization, and timestamp alignment, to ensure consistency in the fusion of different modalities.

[0083] When real-time data meets dynamically configured detection conditions, the system automatically triggers preset detection events (such as initiating a secondary verification algorithm or invoking an emergency handling process). It then generates second data mining results (such as root cause analysis or predictive maintenance recommendations) by re-executing the target workflow or invoking a specialized algorithm model. Finally, the two analysis results are merged and output, forming a complete closed loop from real-time monitoring and rule-driven approaches to dynamic response. This mechanism enables the system to adaptively adjust business logic based on initial analysis results and deepen data value through secondary mining, thereby shortening the decision-making cycle while improving the accuracy and adaptability of the analysis.

[0084] In some embodiments, the process of determining the logical judgment conditions corresponding to the business logic rules and generating dynamic logical configuration information based on the attribute feature information of multimodal data may further include the following steps:

[0085] Obtain preset alarm rules; detect rule adjustment instructions from the application platform; respond to the received rule adjustment instructions, adjust the alarm rules, and obtain alarm trigger conditions; determine logical judgment conditions based on the attribute feature information of multimodal data, and generate dynamic logical configuration information according to the logical judgment conditions and alarm trigger conditions.

[0086] By constructing a collaborative adjustment mechanism between alarm rules and dynamic logic, intelligent processing from preset rules to scenario-based adaptation is achieved. Specifically, pre-configured alarm rules (such as triggering an alert when the device temperature exceeds 80℃) are first obtained as the basic judgment framework; then, rule adjustment instructions from the application platform are detected in real time. Users can dynamically adjust alarm rules through rule definition tools, for example, modifying alarm triggering conditions based on the number and distribution of detected personnel in real-time monitoring scenarios; the set rules are parsed and executed in real time by a rule engine (such as Drools) to ensure the accuracy and timeliness of logical judgment.

[0087] Based on this, the system further refines the logical judgment conditions by combining the attribute characteristics of multimodal data (such as target detection results in video streams and time-series signals collected by sensors) (such as data temporal correlation, spatial distribution range, and type differences). For example, it defines the correlation between temperature data and equipment load and sets the effective range of personnel detection area. Finally, the adjusted alarm trigger conditions are integrated with the logical judgment conditions determined based on multimodal data to generate dynamic logical configuration information containing conditional expressions, threshold parameters, and associated actions. This enables the system to automatically match the optimal alarm logic according to real-time data characteristics, improving scenario adaptability while ensuring the rigor of the rules.

[0088] In some embodiments, after generating the second data mining result, the data mining method may further include the following steps:

[0089] In response to the second data mining results, target labeled data is obtained through the application platform; based on the target labeled data, an optimized configuration scheme is generated, and the target algorithm model is adjusted based on the optimized configuration scheme to obtain a new algorithm model.

[0090] Specifically, the platform will provide feedback to users on the results of rule execution (i.e., the second data mining results mentioned above) through visual charts or alerts, such as highlighting the abnormal areas corresponding to the triggering conditions on the interface; users can adjust rule parameters or add new conditions based on the execution results to optimize business logic.

[0091] More specifically, the above-mentioned method, which allows users to mark missed detections and false detections in the model's detection results through a graphical card interface and generates targeted optimization schemes based on the user-marked data, includes the following steps:

[0092] The model's detection results are displayed through a graphical card interface, including: object detection tasks, which show the location and confidence level of the detected objects in the form of bounding boxes; anomaly detection tasks, which show the detection results in the form of heatmaps or overlays of anomaly regions, marking areas that may contain anomalies and providing anomaly scores or probabilities; and pedestrian re-identification tasks, which provide a list of matching results and corresponding confidence scores. Users can view every detail of the detection results in detail through zoom, pan, and click interaction functions.

[0093] Secondly, the graphical card interface provides a marking tool that allows users to manually select undetected targets (missed detections) or incorrectly detected targets (false detections).

[0094] For missed targets, users can add markers by drawing bounding boxes or directly clicking on the abnormal area; for false positives, users can directly click on the corresponding detection result to mark it and select the marker type (e.g., "region error" or "abnormal miss").

[0095] The data mining platform stores user-labeled data in a dedicated label database, recording label location, category, and user notes. After statistical analysis, the labeled data generates an error distribution report, including the distribution of false negatives, false positives, and error types. Finally, it automatically generates targeted optimization solutions based on the labeled data, including but not limited to: hyperparameter adjustment suggestions, such as adjusting threshold settings or the scoring range of the model output in anomaly detection tasks; data augmentation strategy recommendations, generating data augmentation strategies based on the distribution characteristics and background conditions of anomalous regions in the labeled data, such as simulating more anomalous samples or enhancing the features of anomalous regions; and constraint definition, defining constraints according to the error types of the labels, such as specifying specific regions as key detection areas or increasing the detection weight for certain types of anomalies.

[0096] The aforementioned optimization schemes can be directly applied to the model retraining or parameter tuning process. The graphical card interface allows users to manually modify or regenerate the optimization schemes to ensure their applicability and flexibility. After the optimization schemes are applied, the data mining platform re-evaluates the model and generates a performance improvement report, comparing the changes in metrics before and after optimization. Users can re-label the optimization results, forming a feedback loop to further improve model performance.

[0097] In some embodiments, the above-described acquisition of real-time data may further include the following steps:

[0098] When the data acquisition task is a multi-concurrent task, the algorithm warehouse management platform distributes the task information of each concurrent task to the corresponding algorithm service node, and calls each algorithm service node to execute the concurrent tasks in parallel to obtain real-time data.

[0099] Under high concurrency conditions, the platform uses a task scheduling mechanism to allocate video streams from multiple cameras to different nodes for parallel processing, ensuring real-time performance. At this time, the algorithm repository management platform will distribute the corresponding analysis tasks (including video channel addresses, algorithm IDs, and other information) to the corresponding algorithm service nodes.

[0100] During peak hours at turnstiles, the simultaneous acquisition of video streams from multiple cameras can increase the system's processing load. This invention achieves high concurrency support through resource optimization algorithms. The task scheduling system dynamically allocates video stream processing tasks based on the CPU and GPU load status of each node, ensuring that the computational load is evenly distributed across all nodes. Furthermore, a pipelined processing architecture is adopted, seamlessly integrating video decoding, data preprocessing, model inference, and alarm logic in separate steps to minimize latency.

[0101] More specifically, the data mining platform supports real-time processing of input data from multiple signal sources under high concurrency conditions, and its method includes the following steps:

[0102] Access and management of multiple signal sources: The platform supports access from multiple signal sources, including local video files, network video streams (such as RTSP streams) and real-time camera input; it uses a unified data access interface to standardize data sources of different formats, frame rates and resolutions to ensure that the system is compatible with multiple signal types; it manages concurrent connections of signal sources through connection pooling technology, dynamically allocates network bandwidth and decoding resources to ensure the stability of data access;

[0103] Concurrent scheduling for real-time data processing: Under high concurrency conditions, a distributed task scheduling system is adopted to distribute the input data of the signal source to multiple processing nodes in blocks; an advanced queue management mechanism is used to group the input data for processing, prioritizing the processing of high-priority data streams, while avoiding blocking caused by resource contention;

[0104] Hardware acceleration utilization: The platform integrates hardware acceleration modules (such as GPUs or dedicated hardware decoders) to achieve fast execution of video decoding and data preprocessing; it supports batch decoding at the hardware level to optimize decoding of high frame rate data streams and reduce the processing time per frame.

[0105] Application of resource optimization algorithms: The platform adopts a dynamic resource allocation algorithm, which adjusts the allocation of computing resources in real time during task execution based on the current load status of CPU and GPU; it uses a load balancing strategy to distribute high computing demand tasks evenly across multiple nodes to avoid single-point performance bottlenecks; and it introduces a resource reclamation mechanism in low-priority tasks to redistribute underutilized resources and improve overall resource utilization.

[0106] Real-time performance guarantee mechanism: The platform reduces latency in data processing through real-time optimization algorithms, including reducing network latency during data transmission and queuing time in task scheduling; it adopts a pipelined processing architecture to execute data access, decoding, preprocessing and inference tasks step by step and seamlessly connect them, thereby shortening the processing cycle;

[0107] Performance monitoring and feedback: Equipped with a performance monitoring module, it monitors the utilization of key resources such as CPU, GPU, memory and network bandwidth in real time; and provides users with feedback on resource usage status through data visualization tools, making it easy to manually adjust resource allocation strategies when needed.

[0108] Through the above embodiments, multiple concurrent tasks are broken down and executed in parallel across multiple algorithm service nodes, making full use of computing resources and thus significantly shortening task processing time. Furthermore, by using the algorithm repository management platform as a unified scheduling center, the load on each node can be monitored in real time and tasks can be dynamically allocated, ensuring maximum resource utilization.

[0109] In some embodiments, the above-described acquisition of real-time data may further include the following steps:

[0110] Acquire continuous video frames; set frame interval parameters according to preset task requirements; extract corresponding key frames from continuous video frames based on frame interval parameters; perform integrity verification on key frames; and acquire real-time data based on key frames if the integrity verification passes.

[0111] Specifically, the data acquisition and preprocessing module in the aforementioned data mining platform uses a video keyframe extraction method based on fixed frame intervals to extract keyframes from the video stream or stored video files, thereby reducing computational redundancy and improving data processing efficiency. The method includes the following steps:

[0112] First, the video frame sequence is read. Continuous video frames are parsed from the input video stream using hardware or software decoding to form a frame sequence. Frame interval settings are then implemented, with a parameter N defined to specify how many frames a keyframe is extracted. Next, data redundancy control is implemented to reduce the total number of frames to be processed, effectively decreasing the data processing load in video analysis tasks, particularly suitable for optimizing real-time stream processing and large-scale video data storage. Then, frame information verification is performed, checking the integrity of the extracted keyframes to ensure no data loss or corruption. If necessary, pixel normalization and color space conversion are performed on the keyframes to adapt to subsequent processing requirements. Finally, frame buffer management stores the keyframes in an efficient cache for subsequent model inference or further data preprocessing operations, such as object detection and image segmentation.

[0113] The present application will be described in detail below with reference to specific embodiments. Taking the tailgating detection scenario at high-speed railway station turnstiles as an example, this application provides an efficient, real-time, and intelligent solution that utilizes multi-module collaborative operation to handle video stream data acquisition, behavior determination, and alarm generation. Please refer to [link to relevant documentation]. Figure 4 The data mining platform includes a data management module, a container deployment module for Docker (a containerization platform), a user interface module, a model integration module, a business logic module for spatiotemporal business logic constraints, and a platform standard function module. This standard function module supports platform management, product management, user management, low-code application development, resource allocation, and standard continuous integration / continuous deployment (CI / CD) functions. The following is the implementation process of each part of the data mining platform:

[0114] The data management module comprises three sub-modules: data acquisition, data storage, and data preprocessing. This module automates the management of stored data types, formats, and sources through metadata services, supporting access to diverse data sources, including IoT sensor data, social media platform data, e-commerce transaction data, medical imaging data, traffic monitoring video data, and government public sector data. The module employs data sharding and distributed storage strategies, dividing large-scale data into multiple shards and distributing them across multiple nodes to improve data access speed and system fault tolerance. It supports dynamic expansion based on a distributed storage architecture, enabling online expansion of storage capacity to accommodate data storage needs ranging from terabytes (TB) to petabytes (PB). The module provides data access interfaces based on multiple access protocols, such as Structured Query Language (SQL) and the Hadoop Distributed File System API (HDFS API), allowing users to efficiently retrieve and process stored data, and further optimize retrieval performance using indexing technology. Finally, the module reduces storage space requirements while ensuring data integrity through data compression and deduplication technologies, making it suitable for efficient management in large-scale data scenarios.

[0115] The model integration module integrates various algorithm models, including abnormal behavior detection models, object detection models, and re-identification models. It provides a modular design and standardized interfaces, supporting flexible configuration of model parameters and business logic. For example, the module loads the YOLOv5 object detection model, specifically designed to identify pedestrian targets within the turnstile area. Users select models through a graphical card interface and can intuitively view model descriptions, applicable scenarios, and default parameters on the cards. To adapt to tailgating detection scenarios, users adjust the confidence threshold to 0.5 and set the coordinate range of the turnstile area in the interface. The module loads models through a standardized interface, supporting hardware-accelerated inference to improve real-time detection performance. Inference results include the coordinates of pedestrian bounding boxes and confidence scores, providing spatial distribution data of pedestrians within the turnstile area.

[0116] The user interaction interface module provides a drag-and-drop UI component library, visualization results, and statistical data. For example, users can select the corresponding algorithm by clicking on the card of the desired model; the system provides default parameter settings for each model, such as the confidence threshold for object detection or the region size for segmentation algorithms; users can adjust model parameters according to actual application scenarios, either by clicking on parameter fields and entering new values ​​or by using sliders.

[0117] The business logic module provides flexible logic rule definition capabilities for tail-following detection. Through a rule engine, it supports the definition of complex business logic, enabling logical judgments and event triggering across multimodal data. Users can construct business logic using a visual rule definition tool, for example:

[0118] Rule 1: If two or more pedestrians are detected in the gate area at the same time, and the distance between their bounding boxes is less than 30 pixels, it is judged as tailgating behavior;

[0119] Rule 2: If the bounding boxes of a pedestrian target are detected to partially overlap and exist continuously for more than 3 frames, an alarm signal will be issued.

[0120] Users can drag and drop components in the interface to set logical nodes and dynamically adjust rule parameters (such as distance thresholds and frame counts). Rules are parsed into event triggering conditions and executed in real time through the Drools rule engine.

[0121] When tailgating is detected, the business logic support module triggers an alarm signal. The model evaluation module highlights the relevant image frames of the trigger condition and displays the specific details of the abnormal behavior through a visual interface, including the bounding box location, the number of pedestrians detected, and the confidence score. Simultaneously, the alarm information is sent in text format to the security personnel's mobile device, including the gate number, event time, and anomaly type. The platform also supports storing alarm records in the operation log for subsequent tracking and analysis.

[0122] In addition, the aforementioned data mining platform may also include a model evaluation module and a user permission management module.

[0123] The model evaluation module allows users to label missed and false positive results. For example, users can label incorrectly identified tailing behavior as a missed detection, or incorrectly identified behavior as a false positive. The labeled data is statistically analyzed to generate an error distribution report, including the missed detection rate, false positive rate, and main error types. Based on the labeled data, the platform automatically generates optimization schemes, such as: suggesting adjustments to the model's confidence threshold to reduce missed detections; generating data augmentation samples simulating tailing behavior to enrich the model's training set; and defining more refined detection region weights to improve detection accuracy. These optimization schemes can be directly used for model retraining or parameter fine-tuning, achieving continuous performance improvement.

[0124] To ensure system security, the user access control module assigns tiered permissions to users with different roles. For example, ordinary security personnel can only view alarm records and image data, while administrators can modify business logic rules and model parameters. Users must use two-factor authentication (such as SMS verification code and password) to log in, and a unique audit log is generated for each operation, including the operation time, module, and details. The platform also supports anomaly detection, such as triggering account lockout after multiple failed login attempts, further enhancing system security.

[0125] More specifically, controlling user access permissions through hierarchical management and dynamic authorization mechanisms includes the following steps:

[0126] User authentication: Verify user identity through multiple authentication methods, including but not limited to username and password authentication, two-factor authentication, and biometric authentication (such as fingerprint or facial recognition); when a user logs in, the system generates a unique session identifier (Session ID) and stores it in encryption to ensure session security;

[0127] Permission hierarchy and role definition: The system predefines various user roles, such as administrator, developer, regular user and visitor, each role corresponding to a different permission range; it also supports custom roles according to business needs, and users can assign specific permissions to new roles through the interface;

[0128] Dynamic permission management: Provides dynamic authorization functionality based on user groups or individuals, allowing administrators to adjust user access permissions according to real-time needs; supports fine-grained permission control, such as setting access permissions for specific datasets, functional modules, or task operations (such as model training, parameter setting, or result export);

[0129] Operation log recording and auditing: The system records all user operations in real time, including login time, accessed modules, modified parameters and executed operations; operation logs are stored in encrypted form and support retrieval based on time period, user or operation type, which facilitates security auditing and problem tracing;

[0130] Access control strategy: A combination of role-based access control (RBAC) and attribute-based access control (ABAC) is adopted to ensure that different users' access to resources meets security and business requirements; the system has a built-in permission conflict detection mechanism to prevent over-authorization caused by overlapping permissions of multiple roles.

[0131] The above embodiments provide a low-code data mining platform for massive view big data. Employing hardware acceleration and a distributed task scheduling mechanism, it achieves real-time data processing under high concurrency conditions, meeting the needs of multiple signal source inputs and suitable for intelligent traffic monitoring and rapid response scenarios. The model calling and integration module, algorithm repository management platform, etc., adopt modular and standardized interfaces, supporting rapid integration and dynamic replacement of various algorithm models, and can be flexibly configured in the front-end visual interface. Through a visual rule definition tool, users can construct complex logical rules to achieve multimodal data fusion analysis and dynamic event triggering. The rule engine supports real-time parsing and adjustment, improving the system's flexibility and intelligence. A graphical user interface and drag-and-drop operation are provided, enabling non-professional users to easily complete complex algorithm configuration, logic definition, and data mining tasks, significantly lowering the barrier to entry. It supports the annotation and analysis of false positives and false negatives, and combined with optimization schemes (such as parameter adjustment and data augmentation strategies), forms a feedback loop to continuously improve model performance and stability. Unified management of algorithm and model versions and dependencies, automated downloading, updating, and registration ensure consistency and correctness when the front-end calls the model, improving system maintainability and iteration efficiency. In addition, the technical solution can be applied to various scenarios such as intelligent traffic monitoring, medical image analysis, and industrial anomaly detection.

[0132] It should be noted that the steps shown in the above process or in the flowchart of the accompanying figures 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.

[0133] This embodiment also provides a data mining apparatus for implementing the above embodiments and preferred embodiments, and details already described will not be repeated. As used below, the terms "module," "unit," "subunit," etc., can refer to a combination of software and / or hardware that performs a predetermined function. Although the apparatus described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0134] Figure 5 This is a structural block diagram of a data mining apparatus according to an embodiment of this application, such as... Figure 5 As shown, the device includes:

[0135] The acquisition unit 51 is used to acquire real-time data; the independent card configuration unit 52 is used to classify the registered modular algorithm models through the algorithm warehouse management platform, assemble the modular algorithm models under each category into independent card units, and dynamically send each independent card unit to the application platform for configuration; the workflow generation unit 53 is used to detect the first operation instruction for the independent card unit through the application platform; in response to the detected first operation instruction, it determines the target algorithm model in the modular algorithm model, and logically configures multiple target algorithm models to obtain the target workflow; the mining unit 54 is used to call the target workflow to perform data mining on the real-time data and generate data mining results.

[0136] In some embodiments, the workflow generation unit 53 is further configured to parse the first operation instruction, determine the operated card unit indicated by the first operation instruction in the independent card unit; determine the target algorithm model corresponding to the operated card unit from each modular algorithm model, configure the target algorithm model according to the reference model parameters in the operated card unit, and obtain the target workflow through the algorithm warehouse management platform.

[0137] In some embodiments, the workflow generation unit 53 is further configured to detect parameter adjustment instructions received by the application platform via the algorithm repository management platform; adjust the reference model parameters based on the parameter adjustment instructions to obtain the target model parameters; and configure the target algorithm model according to the target model parameters to obtain the target workflow.

[0138] In some embodiments, the mining unit 54 is further configured to invoke the target workflow, perform data mining processing on real-time data, and generate a first data mining result; in response to the first data mining result, construct business logic rules based on the second operation instruction detected by the application platform; determine the logical judgment conditions corresponding to the business logic rules based on the attribute feature information of the multimodal data, and generate dynamic logic configuration information; trigger a detection event based on the dynamic logic configuration information; and generate a second data mining result in response to the triggered detection event; the data mining result includes the first data mining result and the second data mining result.

[0139] In some embodiments, the mining unit 54 is further configured to acquire preset alarm rules; detect rule adjustment instructions from the application platform; adjust the alarm rules in response to the received rule adjustment instructions to obtain alarm triggering conditions; determine logical judgment conditions based on the attribute feature information of multimodal data, and generate dynamic logical configuration information according to the logical judgment conditions and alarm triggering conditions.

[0140] In some embodiments, the data mining apparatus further includes an optimization unit; the optimization unit is used to obtain target labeled data via an application platform in response to the second data mining result; generate an optimized configuration scheme based on the target labeled data; and adjust the target algorithm model based on the optimized configuration scheme to obtain a new algorithm model.

[0141] It should be noted that the above modules can be functional modules or program modules, and can be implemented by software or hardware. For modules implemented by hardware, the above modules can reside in the same processor; or the above modules can be located in different processors in any combination. Specific examples in this embodiment can be found in the examples described in the above embodiments and optional implementations, and will not be repeated in this embodiment.

[0142] This embodiment also provides an electronic device, including a memory and a processor, wherein the memory stores a computer program and the processor is configured to run the computer program to perform the steps in any of the above method embodiments.

[0143] Optionally, the electronic device may further include a transmission device and an input / output device, wherein the transmission device is connected to the processor and the input / output device is connected to the processor.

[0144] Optionally, in this embodiment, the processor can be configured to perform the following steps via a computer program:

[0145] S1, obtain real-time data.

[0146] S2, through the algorithm repository management platform, classifies the registered modular algorithm models, assembles the modular algorithm models under each category into independent card units, and dynamically sends each independent card unit to the application platform for configuration.

[0147] S3, via the application platform, detects the first operation instruction for the independent card unit; in response to the detected first operation instruction, it determines the target algorithm model in the modular algorithm model, and via the algorithm warehouse management platform, it logically configures multiple target algorithm models to obtain the target workflow.

[0148] S4 invokes the target workflow to perform data mining on real-time data and generates data mining results.

[0149] It should be noted that the specific examples in this embodiment can refer to the examples described in the above embodiments and optional implementations, and will not be repeated here.

[0150] Furthermore, in conjunction with the data mining methods described in the above embodiments, this application embodiment can provide a storage medium for implementation. This storage medium stores a computer program; when executed by a processor, the computer program implements any of the data mining methods described in the above embodiments.

[0151] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties.

[0152] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments described above. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0153] Those skilled in the art should understand that the technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments have been described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0154] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this patent application should be determined by the appended claims.

Claims

1. A data mining method characterized by, The method comprises: acquiring real-time data; the real-time data is multi-modal data acquired from a multi-modal data source; via an algorithm warehouse management platform, classifying registered modular algorithm models, assembling the modular algorithm models under each category into independent card units, and dynamically sending each independent card unit to an application platform for configuration; each independent card unit is dynamically generated through a graphical card interface configured by the application platform; the application platform interacts with the algorithm warehouse management platform through the graphical card interface; the algorithm warehouse management platform uniformly manages algorithm services; via the application platform, detecting a first operation instruction for the independent card unit; in response to the detected first operation instruction, determining a target algorithm model in the modular algorithm model, and via the algorithm warehouse management platform, logically configuring a plurality of target algorithm models to obtain a target workflow; calling the target workflow to perform data mining on the real-time data to generate a data mining result, comprising: calling the target workflow to perform data mining processing on the real-time data to generate a first data mining result; in response to the first data mining result, constructing a business logic rule based on a second operation instruction detected by the application platform; based on attribute feature information of the multi-modal data, determining a logical judgment condition corresponding to the business logic rule, and generating dynamic logic configuration information, including: acquiring a preset alarm rule; detecting a rule adjustment instruction of the application platform; in response to the received rule adjustment instruction, adjusting the alarm rule to obtain an alarm trigger condition; based on the attribute feature information of the multi-modal data, determining the logical judgment condition, and according to the logical judgment condition and the alarm trigger condition, generating dynamic logic configuration information containing a condition expression, a threshold parameter and an associated action; based on the dynamic logic configuration information, triggering a detection event; in response to the triggered detection event, generating a second data mining result; the data mining result includes the first data mining result and the second data mining result.

2. The data mining method of claim 1, wherein, In response to the detected first operation instruction, determining a target algorithm model in the modular algorithm model, and via the algorithm warehouse management platform, logically configuring a plurality of target algorithm models to obtain a target workflow, comprising: parsing the first operation instruction to determine an operated card unit in the independent card unit indicated by the first operation instruction; from each of the modular algorithm models, determining a target algorithm model corresponding to the operated card unit, configuring the target algorithm model according to the reference model parameters in the operated card unit, and obtaining the target workflow via the algorithm warehouse management platform.

3. The data mining method of claim 2, wherein, According to the reference model parameters in the operated card unit, configuring the target algorithm model, and obtaining the target workflow via the algorithm warehouse management platform, comprising: via the algorithm warehouse management platform, detecting a parameter adjustment instruction received by the application platform; Adjust the reference model parameters based on the parameter adjustment instruction to obtain target model parameters, and configure the target algorithm model based on the target model parameters to obtain the target workflow.

4. The data mining method of claim 1, wherein, After the second data mining result is generated, the method further includes: In response to the second data mining result, target labeling data is acquired via the application platform; Based on the target labeling data, an optimization configuration scheme is generated, and the target algorithm model is adjusted based on the optimization configuration scheme to obtain a new algorithm model.

5. The data mining method of claim 1, wherein, The real-time data is acquired, including: In the case of multiple concurrent tasks in the data collection task, the task information of each concurrent task is distributed to the corresponding algorithm service node via the algorithm warehouse management platform, and each algorithm service node is called to execute the concurrent tasks in parallel to acquire the real-time data.

6. The data mining method according to any one of claims 1 to 5, characterized in that, The real-time data is acquired, including: Acquiring continuous video frames; According to the preset task requirement information, a frame interval parameter is set; based on the frame interval parameter, corresponding key frames are extracted from the continuous video frames; The key frames are subjected to integrity verification, and in the case that the integrity verification is passed, real-time data is acquired based on the key frames.

7. A data mining apparatus characterized by comprising: Including: An acquisition unit for acquiring real-time data; The real-time data is multi-modal data acquired from a multi-modal data source; An independent card configuration unit for classifying registered modular algorithm models via an algorithm warehouse management platform, assembling the modular algorithm models under each classification into independent card units, and dynamically sending each independent card unit to an application platform for configuration; each independent card unit is dynamically generated through a graphical card interface configured by the application platform; the application platform interacts with the algorithm warehouse management platform through the graphical card interface; the algorithm warehouse management platform uniformly manages algorithm services; A workflow generation unit for detecting a first operation instruction for the independent card unit via the application platform; In response to the detected first operation instruction, a target algorithm model in the modular algorithm model is determined, and multiple target algorithm models are logically configured to obtain a target workflow; A mining unit for calling the target workflow to perform data mining on the real-time data to generate a data mining result, including: Calling the target workflow to perform data mining processing on the real-time data to generate a first data mining result; In response to the first data mining result, a business logic rule is constructed based on a second operation instruction detected by the application platform; Based on the attribute feature information of the multi-modal data, a logic judgment condition corresponding to the business logic rule is determined, and dynamic logic configuration information is generated, including: obtaining a preset alarm rule; detecting a rule adjustment instruction of the application platform; in response to the received rule adjustment instruction, adjusting the alarm rule to obtain an alarm trigger condition; based on the attribute feature information of the multi-modal data, the logic judgment condition is determined, and according to the logic judgment condition and the alarm trigger condition, the dynamic logic configuration information containing a condition expression, a threshold parameter and an associated action is generated; Based on the dynamic logic configuration information, a detection event is triggered; in response to the triggered detection event, a second data mining result is generated; the data mining result includes the first data mining result and the second data mining result.

8. A storage medium, characterized by The storage medium has a computer program stored therein, wherein the computer program is configured to execute the data mining method of any one of claims 1 to 6 when running.

Images