Method, system and equipment for realizing data security classification automation based on large model and medium

Abstract

The invention discloses a method, a system and equipment for realizing data security classification automation based on a large model, and a medium, mainly relates to the technical field of classification automation, and is used for solving the problems that an existing scheme depends on a rule engine or a simple machine learning model, the performance is poor when unstructured data is processed, and the classification efficiency is poor. The problems that data transmission is delayed and data processing speed is low due to the fact that semantics of data cannot be accurately captured and data needs to be transmitted to a central server to be processed in the prior art are solved. Comprising the steps of inputting to-be-processed data into a trained large language model, and obtaining an output prediction classification result and confidence; obtaining all prediction classification results and confidence in a preset time period, and verifying whether the prediction classification results meet a preset verification mechanism; when a preset verification mechanism is met, determining the predicted classification result as a final classification result; otherwise, the corresponding collected data are uploaded to the manual annotation terminal, and a returned final classification result is obtained.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of classification automation, in particular to a method and system for realizing data security classification automation based on a large model, equipment and a medium. BACKGROUND

[0002] In today's data-driven era, data security has become a core pillar of enterprise operations. With the explosive growth of data volume, traditional manual classification methods are not only inefficient, but also difficult to meet the processing needs of massive data. Manual classification methods usually rely on experienced experts to manually check data content, metadata and usage scenarios to classify and grade data. This method not only consumes time and effort, but is also susceptible to human factors, leading to inconsistency and inaccuracy of classification results. In addition, with the diversification of data types, such as the mixing of structured data, semi-structured data and unstructured data, manual classification methods are not up to the task of handling complex data, and are difficult to meet the real-time and accuracy requirements of enterprise data security management.

[0003] In order to overcome the limitations of manual classification methods, the industry has begun to explore automated solutions. However, existing automated methods often rely on rule engines or simple machine learning models, which perform poorly when dealing with unstructured data and cannot accurately capture the semantics of data. For example, rule engines rely on predefined rule sets and are difficult to adapt to dynamic changes in data types; simple machine learning models lack a deep understanding of complex data patterns, resulting in low accuracy of classification results. In addition, existing methods often require data to be transmitted to a central server for processing when dealing with large-scale data, resulting in delays in data transmission and slow data processing speed. SUMMARY

[0004] The application provides a method, system, device and medium for realizing data security classification automation based on a large model, to solve the problem that existing solutions rely on rule engines or simple machine learning models, which perform poorly when dealing with unstructured data and cannot accurately capture the semantics of data, and need to transmit data to a central server for processing, resulting in delays in data transmission and slow data processing speed.

[0005] In a first aspect, the application provides a method for realizing data security classification automation based on a large model, the method comprising: acquire specific types of initial data in a preset historical time period and preset classification keywords; determine corresponding edge computing nodes according to the specific types, and send the initial data and the preset classification keywords to the corresponding edge computing nodes according to the specific types; process the initial data into to-be-classified data in a preset unified format through the edge computing nodes; determine a classification result and a classification accuracy of the to-be-classified data by using the preset classification keywords; acquire the to-be-classified data, the classification result and the classification accuracy returned by all the edge computing nodes; upload initial data with a classification accuracy lower than a preset threshold to a preset manual labeling terminal, and update the classification result by using returned data of the manual labeling terminal; train a large language model by using the to-be-classified data and the classification result; in the training process, use a cross-entropy loss function as an optimization target, and constantly adjust model parameters through back propagation to obtain the trained large language model; acquire real-time collected data, determine corresponding edge computing nodes according to specific types of the collected data, and acquire to-be-processed data in a preset unified format returned by the edge computing nodes; input the to-be-processed data into the trained large language model to obtain output predicted classification results and confidence levels; acquire all the predicted classification results and the confidence levels in a preset time period, verify whether the predicted classification results meet a preset verification mechanism, determine the predicted classification results as final classification results when the predicted classification results meet the preset verification mechanism, or upload corresponding collected data to a manual labeling terminal to obtain returned final classification results when the predicted classification results do not meet the preset verification mechanism.

[0006] In an implementation manner of the present application, before the initial data and the preset classification keywords are sent to the corresponding edge computing nodes according to the specific types, the method further includes: configuring corresponding edge computing nodes for each specific type; processing the initial data into to-be-classified data in a preset unified format, and determining a classification result and a classification accuracy of the to-be-classified data by using the preset classification keywords, specifically including: segmenting the initial data into a plurality of initial fields; converting the initial fields into preset standard description data; splicing the preset standard description data to obtain to-be-classified data in a preset unified format; comparing the preset standard description data with the preset classification keywords to determine a classification result corresponding to a preset classification keyword with the maximum similarity as the classification result of the current to-be-classified data; determining the maximum similarity as the classification accuracy.

[0007] In an implementation manner of the present application, before the initial data is segmented into a plurality of initial fields, the method further includes: remove the repeated information in the initial data; obtain a preset normal range of the numerical values in the initial data, and remove numerical data not belonging to the preset normal range; When there are missing values in the initial data, use an artificial intelligence model to make a prediction and fill in the missing values; convert the format to a text format.

[0008] In an implementation form of the present application, after determining the corresponding edge computing node according to the specific type of the collected data, the method further comprises: extracting sample data of a preset proportion from all collected data; obtaining a first classification result and a classification accuracy of the sample data using a preset classification keyword; retaining the first classification result with a classification accuracy greater than a preset accuracy threshold; obtaining a predicted classification result corresponding to the first classification result in the large language model; obtaining a difference rate between the first classification result and the predicted classification result; when the difference rate is greater than a preset difference threshold, updating the large language model.

[0009] In an implementation form of the present application, the total predicted classification result and the confidence in a preset time period are obtained, and it is verified whether the predicted classification result meets the preset verification mechanism, specifically comprising: obtaining all predicted classification results in a preset time period, and obtaining the proportion of each predicted classification result; comparing the proportion of each predicted classification result with the proportion of each predicted classification result in the previous preset time period, and when the difference value is greater than a preset proportion difference threshold, it is determined that the preset verification mechanism is not met; when the difference value is less than or equal to the preset proportion difference threshold, obtaining the confidence of each predicted classification result; determining that the predicted classification result with a confidence greater than a preset confidence threshold meets the preset verification mechanism; determining that the predicted classification result with a confidence less than or equal to the preset confidence threshold does not meet the preset verification mechanism.

[0010] In a second aspect, the present application provides a system for realizing data security classification automation based on a large model, which comprises: A determination module is configured to obtain the specific type of initial data and a preset classification keyword in a preset historical time period, determine the corresponding edge computing node according to the specific type, send the initial data and the preset classification keyword to the corresponding edge computing node according to the specific type, process the initial data into a preset uniform format of to-be-classified data through the edge computing node, and determine the classification result and the classification accuracy of the to-be-classified data using the preset classification keyword; The training module is used to acquire the data to be classified, classification results, and classification accuracy returned by all edge computing nodes; upload the initial data with a classification accuracy lower than a preset threshold to a preset manual annotation terminal, and update the classification results using the manually returned data; train a large language model using the data to be classified and the classification results; during the training process, the cross-entropy loss function is used as the optimization objective, and the model parameters are continuously adjusted through backpropagation to obtain a well-trained large language model. The acquisition module is used to acquire real-time collected data, determine the corresponding edge computing node based on the specific type of collected data, acquire the data to be processed in a preset unified format returned by the edge computing node, input the data to be processed into the trained large language model, and obtain the output prediction classification result and confidence score; acquire all prediction classification results and confidence scores within a preset time period, and verify whether the prediction classification result meets the preset verification mechanism; when the preset verification mechanism is met, the prediction classification result is determined as the final classification result; otherwise, the corresponding collected data is uploaded to the manual annotation terminal to obtain the returned final classification result.

[0011] In one implementation of this application, the determining module includes a determining unit. Used to divide initial data into several initial fields; Convert the initial fields into preset standard descriptive data; Based on preset standard description data, splicing is used to obtain data to be classified in a preset unified format; The preset standard description data is compared with the preset classification keywords, and the classification result corresponding to the preset classification keyword with the highest similarity is determined as the classification result of the current data to be classified. The maximum similarity was determined as the classification accuracy.

[0012] In one implementation of this application, the obtaining module includes a judgment unit. Used to obtain all predicted classification results for a preset time period and obtain the proportion of each predicted classification result; The proportion of each predicted classification result is compared with the proportion of each predicted classification result in the previous preset time period. When the difference value is greater than the preset proportion difference threshold, it is determined that the preset verification mechanism is not met. When the difference value is less than or equal to the preset proportion difference threshold, obtain the confidence level of each predicted classification result; Determine if the predicted classification result with a confidence level greater than a preset confidence threshold satisfies the preset verification mechanism; The predefined verification mechanism is used to determine the classification results whose confidence level is less than or equal to a pre-set confidence threshold.

[0013] Thirdly, this application provides a device for automating data security classification based on a large model, the device comprising: processor; And a memory containing executable code, which, when executed, causes the processor to execute a method for automating data security classification based on a large model, as described above.

[0014] Fourthly, this application provides a non-volatile computer storage medium storing computer instructions thereon, which, when executed, implement a method for automating data security classification based on a large model, as described above.

[0015] As can be seen from the above technical solutions, this application has the following advantages: This application enhances the semantic understanding capability of unstructured data classification by introducing a large language model and edge computing nodes for collaborative processing. Traditional solutions rely on rule engines or simple machine learning models, which struggle to capture the deep semantics of data, resulting in low classification accuracy. This application leverages the powerful natural language processing capabilities of a large language model, combining it with pre-set classification keywords for initial classification, and then iteratively optimizes low-accuracy data through manual annotation terminals, forming a dynamic training loop. This design enables the model to continuously learn the semantic features of the data, effectively identifying implicit information in text, such as accurately distinguishing between patient privacy and public information in medical data, or identifying sensitive transaction records in financial data. Simultaneously, the localized processing of edge computing nodes avoids network transmission latency during semantic understanding, ensuring the real-time performance and accuracy of classification results.

[0016] This application employs a distributed edge computing architecture, reducing data transmission latency and improving processing efficiency. Traditional solutions require all data to be transmitted to a central server, leading to network congestion and response delays. This application intelligently allocates data to the nearest edge node based on data type, achieving localized data processing and unified format conversion, uploading only necessary results to the cloud. This design reduces network traffic, allowing workshop equipment data to be classified in real-time at edge nodes in smart manufacturing scenarios, without waiting for cloud responses. Furthermore, the collaborative training mechanism between edge nodes and large models adjusts model parameters through backpropagation, enabling the system to dynamically optimize based on local data characteristics, forming a collaborative intelligent processing system of "end-edge-cloud". This architecture ensures data security while achieving millisecond-level response times, meeting the stringent real-time requirements of applications such as the Industrial Internet and smart cities. Attached Figure Description

[0017] To more clearly illustrate the technical solution of the present invention, the accompanying drawings used in the description will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart of a method for automating data security classification based on a large model, provided in an embodiment of this application.

[0019] Figure 2 This is a schematic diagram of the internal structure of a system for automating data security classification based on a large model, provided in an embodiment of this application.

[0020] Figure 3 This is a schematic diagram of the internal structure of a device that automates data security classification based on a large model, as provided in an embodiment of this application. Detailed Implementation

[0021] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] Those skilled in the art should understand that the embodiments described below are merely preferred embodiments of this disclosure and do not imply that this disclosure can only be implemented through these preferred embodiments. These preferred embodiments are merely used to explain the technical principles of this disclosure and are not intended to limit the scope of protection of this disclosure. Based on the preferred embodiments provided by this disclosure, all other embodiments obtained by those skilled in the art without creative effort should still fall within the scope of protection of this disclosure.

[0023] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0024] The technical solutions proposed in the embodiments of this application will be described in detail below with reference to the accompanying drawings.

[0025] The embodiment provides a method for automating data security classification based on a large model, such as Figure 1 As shown in the embodiments of this application, the method mainly includes the following steps: Step 110: Obtain the specific type and preset classification keywords of the initial data within the preset historical time period; determine the corresponding edge computing node according to the specific type, and send the initial data and preset classification keywords to the corresponding edge computing node according to the specific type; process the initial data into data to be classified in a preset unified format through the edge computing node; use the preset classification keywords to determine the classification result and classification accuracy of the data to be classified.

[0026] It should be noted that each specific type corresponds to a preset category keyword; for example, furniture corresponds to tables and chairs.

[0027] In some embodiments, before sending the initial data and preset classification keywords to the corresponding edge computing node according to a specific type, the method further includes: Configure corresponding edge computing nodes for each specific type; The initial data is processed into a pre-defined, uniform format for classification; pre-defined classification keywords are used to determine the classification results and accuracy of the data, specifically including: The initial data is divided into several initial fields (using either dictionary-based or large-model-based word segmentation methods for word segmentation). Convert the initial fields into preset standard descriptive data; Based on preset standard description data, splicing is used to obtain data to be classified in a preset unified format; The preset standard description data is compared with the preset classification keywords, and the classification result corresponding to the preset classification keyword with the highest similarity is determined as the classification result of the current data to be classified. The maximum similarity was determined as the classification accuracy.

[0028] In addition, before splitting the initial data into several initial fields, the method also includes: Remove duplicate information from the initial data; obtain the preset normal range of values ​​in the initial data and remove values ​​that do not belong to the preset normal range; when there are missing values ​​in the initial data, use an artificial intelligence model to predict and fill in the missing values; convert the format to text format.

[0029] As an example, diverse data samples were collected from multiple fields such as finance, healthcare, and e-commerce, including structured data (such as table data in relational databases), semi-structured data (such as XML and JSON files), and unstructured data (such as text documents, images, audio, and video descriptions). Based on industry data security classification standards (such as the "Information Security Technology Data Security Classification Guidelines"), professional data security personnel set preset classification keywords.

[0030] In addition, after determining the corresponding edge computing nodes based on the specific type of collected data, the method also includes: Extract a preset proportion of sample data from all collected data; obtain the first classification result and classification accuracy of the sample data using preset classification keywords; retain the first classification result with a classification accuracy greater than a preset accuracy threshold; obtain the predicted classification result corresponding to the first classification result in the large language model; obtain the difference rate between the first classification result and the predicted classification result; when the difference rate is greater than a preset difference threshold, update the large language model.

[0031] Step 120: Obtain the data to be classified, classification results, and classification accuracy returned by all edge computing nodes; upload the initial data with a classification accuracy lower than the preset threshold to the preset manual annotation terminal, and update the classification results using the manually returned data; train the large language model using the data to be classified and the classification results; during the training process, use the cross-entropy loss function as the optimization objective, and continuously adjust the model parameters through backpropagation to obtain the trained large language model.

[0032] Understandably, this step achieves continuous optimization of data classification through the following mechanisms: First, the system aggregates the data to be classified, classification results, and accuracy rates returned by edge computing nodes. For initial data below a preset threshold, manual annotation intervention is implemented to ensure that low-quality data classification results are corrected. Second, the data to be classified and the manually corrected classification results are used as the training set. The cross-entropy loss function is used as the optimization objective, and the parameters of the large language model are dynamically adjusted through the backpropagation algorithm, enabling the model to learn more accurate semantic features and classification logic. The direct effects of this process include: 1) The manual annotation process effectively corrects potential classification biases in edge computing nodes, improving the reliability of data annotation; 2) The model training process based on the corrected data enables the large language model to continuously optimize its classification capabilities, forming a closed-loop iterative mechanism of "data-annotation-training"; 3) The combination of the cross-entropy loss function and backpropagation ensures the mathematical rigor of model parameter adjustments, making the classification results more consistent with actual semantic features.

[0033] As an example, a pre-trained large model with strong semantic understanding and feature extraction capabilities is selected as the base model.

[0034] All data to be classified, classification results, and classification accuracy are split into: training set, validation set, and test set.

[0035] Fine-tuning of the base model on the training set: First, set the model training parameters, including a learning rate of 1e-5 to 5e-5, a batch size of 8 to 32, and a training epoch of 10 to 30. During training, use the cross-entropy loss function as the optimization objective and continuously adjust the model parameters through backpropagation. Simultaneously, monitor the model's training effect in real time using the validation set. Stop training when the loss function value on the validation set no longer decreases or the model performance no longer improves. After training, evaluate the model's performance using the test set. If the model's classification accuracy and grading accuracy both reach the preset thresholds (e.g., classification accuracy ≥ 95%, grading accuracy ≥ 92%), then use this model as the dedicated model for data security classification. If the preset thresholds are not reached, adjust the training parameters or expand the training dataset and retrain fine-tune until the model performance meets the requirements.

[0036] Step 130: Acquire real-time collected data; determine the corresponding edge computing node based on the specific type of collected data; acquire the data to be processed in a preset unified format returned by the edge computing node; input the data to be processed into the trained large language model to obtain the output prediction classification result and confidence score; acquire all prediction classification results and confidence scores within a preset time period to verify whether the prediction classification result meets the preset verification mechanism; when the preset verification mechanism is met, determine the prediction classification result as the final classification result; otherwise, upload the corresponding collected data to the manual annotation terminal to obtain the returned final classification result.

[0037] Understandably, this step achieves efficient and reliable data classification processing through the following mechanism: First, the system intelligently allocates data to corresponding edge computing nodes based on the specific type of the real-time collected data. These edge nodes then convert the raw data into data to be processed in a pre-defined, uniform format, ensuring standardized data input. Second, the data to be processed is input into a pre-trained large language model, which directly outputs the predicted classification results and confidence scores. A verification mechanism is used, combining multiple rounds of prediction results within a pre-defined time period. Only when the results meet pre-defined consistency conditions is the final classification result confirmed; otherwise, a manual annotation process is triggered. The direct effects of this process include: 1) The type matching and format conversion functions of the edge computing nodes effectively reduce the complexity of data preprocessing and improve processing efficiency; 2) The prediction results and confidence scores output by the large language model provide quantifiable evidence for classification decisions; 3) The synergistic effect of the pre-defined verification mechanism and manual annotation ensures the reliability of the results while minimizing the need for manual intervention, forming a robust classification system that combines automation and manual correction.

[0038] In some embodiments, all predicted classification results and confidence levels within a preset time period are obtained, and the prediction classification results are verified to meet the preset verification mechanism, specifically including: Obtain all predicted classification results for a preset time period, and obtain the proportion of each predicted classification result; The proportion of each predicted classification result is compared with the proportion of each predicted classification result in the previous preset time period. When the difference value is greater than the preset proportion difference threshold, it is determined that the preset verification mechanism is not met. When the difference value is less than or equal to the preset proportion difference threshold, obtain the confidence level of each predicted classification result; Determine if the predicted classification result with a confidence level greater than a preset confidence threshold (e.g., 0.8) meets the preset verification mechanism; The predefined verification mechanism is used to determine the classification results whose confidence level is less than or equal to a pre-set confidence threshold.

[0039] As described above, this embodiment improves the semantic understanding capability of unstructured data classification by introducing a large language model and edge computing nodes for collaborative processing. Traditional solutions rely on rule engines or simple machine learning models, which struggle to capture the deep semantics of data, resulting in low classification accuracy. This application utilizes the powerful natural language processing capabilities of a large language model, combining it with preset classification keywords for initial classification, and then iteratively optimizes low-accuracy data through manual annotation terminals, forming a dynamic training loop. This design enables the model to continuously learn the semantic features of the data, effectively identifying implicit information in text, such as accurately distinguishing between patient privacy and public information in medical data, or identifying sensitive transaction records in financial data. Simultaneously, the localized processing of edge computing nodes avoids network transmission latency during semantic understanding, ensuring the real-time performance and accuracy of classification results.

[0040] This application employs a distributed edge computing architecture, reducing data transmission latency and improving processing efficiency. Traditional solutions require all data to be transmitted to a central server, leading to network congestion and response delays. This application intelligently allocates data to the nearest edge node based on data type, achieving localized data processing and unified format conversion, uploading only necessary results to the cloud. This design reduces network traffic, allowing workshop equipment data to be classified in real-time at edge nodes in smart manufacturing scenarios, without waiting for cloud responses. Furthermore, the collaborative training mechanism between edge nodes and large models adjusts model parameters through backpropagation, enabling the system to dynamically optimize based on local data characteristics, forming a collaborative intelligent processing system of "end-edge-cloud". This architecture ensures data security while achieving millisecond-level response times, meeting the stringent real-time requirements of applications such as the Industrial Internet and smart cities.

[0041] In addition, this application Figure 2 This application provides a system for automating data security classification based on a large model. For example... Figure 2 As shown in the embodiments of this application, the system mainly includes: The determination module 210 is used to obtain the specific type and preset classification keywords of the initial data within a preset historical time period; determine the corresponding edge computing node according to the specific type; send the initial data and preset classification keywords to the corresponding edge computing node according to the specific type; process the initial data into data to be classified in a preset unified format through the edge computing node; and determine the classification result and classification accuracy of the data to be classified using the preset classification keywords.

[0042] The determining module 210 includes a determining unit. Used to divide initial data into several initial fields; Convert the initial fields into preset standard descriptive data; Based on preset standard description data, splicing is used to obtain data to be classified in a preset unified format; The preset standard description data is compared with the preset classification keywords, and the classification result corresponding to the preset classification keyword with the highest similarity is determined as the classification result of the current data to be classified. The maximum similarity was determined as the classification accuracy.

[0043] The training module 220 is used to acquire the data to be classified, classification results, and classification accuracy returned by all edge computing nodes; upload the initial data with a classification accuracy lower than a preset threshold to a preset manual annotation terminal, and update the classification results using the manually returned data; train a large language model using the data to be classified and the classification results; during the training process, the cross-entropy loss function is used as the optimization objective, and the model parameters are continuously adjusted through backpropagation to obtain a well-trained large language model.

[0044] The acquisition module 230 is used to acquire real-time collected data, determine the corresponding edge computing node according to the specific type of collected data, acquire the data to be processed in a preset unified format returned by the edge computing node, input the data to be processed into the trained large language model, and obtain the output prediction classification result and confidence score; acquire all prediction classification results and confidence scores within a preset time period, and verify whether the prediction classification result meets the preset verification mechanism; when the preset verification mechanism is met, the prediction classification result is determined as the final classification result; otherwise, the corresponding collected data is uploaded to the manual annotation terminal to obtain the returned final classification result.

[0045] The module 230 includes a judgment unit. Used to obtain all predicted classification results for a preset time period and obtain the proportion of each predicted classification result; The proportion of each predicted classification result is compared with the proportion of each predicted classification result in the previous preset time period. When the difference value is greater than the preset proportion difference threshold, it is determined that the preset verification mechanism is not met. When the difference value is less than or equal to the preset proportion difference threshold, obtain the confidence level of each predicted classification result; Determine if the predicted classification result with a confidence level greater than a preset confidence threshold satisfies the preset verification mechanism; The predefined verification mechanism is used to determine the classification results whose confidence level is less than or equal to a pre-set confidence threshold.

[0046] The above are method embodiments of this application. Based on the same inventive concept, embodiments of this application also provide a device for automating data security classification based on a large model. Figure 3 As shown, the device includes: a processor; and a memory storing executable code thereon, which, when executed, causes the processor to perform a method for automating data security classification based on a large model, as described in the above embodiment.

[0047] Specifically, the server obtains the specific data type and preset classification keywords of the initial data within a preset historical time period; determines the corresponding edge computing node based on the specific type, and sends the initial data and preset classification keywords to the corresponding edge computing node according to the specific type; the edge computing node processes the initial data into a preset uniform format for classification; using the preset classification keywords, it determines the classification result and classification accuracy of the data to be classified; it obtains the data to be classified, classification results, and classification accuracy returned by all edge computing nodes; the initial data with a classification accuracy lower than a preset threshold is uploaded to a preset manual annotation terminal, and the classification results are updated using the manually returned data; and a large language model is trained using the data to be classified and the classification results. During training, the cross-entropy loss function is used as the optimization objective, and the model parameters are continuously adjusted through backpropagation to obtain a well-trained large language model. Real-time data is acquired, and corresponding edge computing nodes are determined according to the specific type of the acquired data. Data to be processed in a preset uniform format returned by the edge computing nodes is acquired. The data to be processed is input into the well-trained large language model to obtain the output predicted classification results and confidence scores. All predicted classification results and confidence scores within a preset time period are acquired to verify whether the predicted classification results meet the preset verification mechanism. When the preset verification mechanism is met, the predicted classification result is determined as the final classification result; otherwise, the corresponding acquired data is uploaded to the manual annotation terminal to obtain the returned final classification result.

[0048] In addition, embodiments of this application also provide a non-volatile computer storage medium storing executable instructions, which, when executed, implement the method described above for automating data security classification based on a large model.

[0049] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for implementing data security classification automation based on a large model, characterized in that, The method comprises: acquiring specific types of initial data in a preset historical time period and preset classification keywords; determining corresponding edge computing nodes according to the specific types, and sending the initial data and the preset classification keywords to the corresponding edge computing nodes according to the specific types; processing the initial data into preset uniform format to-be-classified data by the edge computing nodes; determining a classification result and a classification accuracy of the to-be-classified data by using the preset classification keywords; acquiring the to-be-classified data, the classification result and the classification accuracy returned by all edge computing nodes; uploading initial data with a classification accuracy lower than a preset threshold to a preset manual labeling terminal, and updating the classification result by using returned data of the manual labeling; training a large language model by using the to-be-classified data and the classification result; and obtaining a trained large language model by taking a cross-entropy loss function as an optimization target and continuously adjusting model parameters through back propagation during the training process. acquiring real-time collected data, determining corresponding edge computing nodes according to specific types of the collected data; acquiring preset uniform format to-be-processed data returned by the edge computing nodes; inputting the to-be-processed data into the trained large language model to obtain output predicted classification results and confidence; acquiring all predicted classification results and confidence within a preset time period, verifying whether the predicted classification results meet a preset verification mechanism; when the preset verification mechanism is met, determining the predicted classification results as final classification results; otherwise, uploading corresponding collected data to a manual labeling terminal to obtain returned final classification results.

2. The method for implementing data security classification automation based on a large model according to claim 1, characterized in that, Before sending the initial data and the preset classification keywords to the corresponding edge computing nodes according to the specific types, the method further comprises: configuring corresponding edge computing nodes for each specific type; processing the initial data into preset uniform format to-be-classified data; and determining a classification result and a classification accuracy of the to-be-classified data by using the preset classification keywords, specifically comprising: segmenting the initial data into a plurality of initial fields; converting the initial fields into preset standard description data; splicing the preset uniform format to-be-classified data based on the preset standard description data; determining a classification result corresponding to a preset classification keyword with the maximum similarity as the classification result of the current to-be-classified data by comparing the preset standard description data with the preset classification keywords; determining the maximum similarity as the classification accuracy.

3. The method for implementing data security classification automation based on a large model according to claim 2, characterized in that, Before segmenting the initial data into a plurality of initial fields, the method further comprises: removing duplicate information in the initial data; acquiring a preset normal range of numerical values in the initial data, and removing numerical data not belonging to the preset normal range; when there are missing values in the initial data, predicting and filling the missing values by using an artificial intelligence model; converting the format into a text format.

4. The method for realizing data security classification automation based on a large model according to claim 1, characterized in that: after determining the corresponding edge computing nodes according to the specific types of the collected data, the method further comprises: extracting sample data of a preset proportion from all collected data; acquiring a first classification result and a classification accuracy of the sample data by using the preset classification keywords; retaining the first classification result with a classification accuracy greater than a preset accuracy threshold. Obtaining a prediction classification result corresponding to the first classification result in the large language model; Obtaining a difference rate of the first classification result and the prediction classification result; When the difference rate is greater than a preset difference threshold, updating the large language model.

5. The method for implementing data security classification automation based on a large model according to claim 1, characterized in that, Obtaining all prediction classification results and confidence levels within a preset time period, verifying whether the prediction classification results meet a preset verification mechanism, specifically including: Obtaining all prediction classification results within a preset time period, obtaining a proportion of each prediction classification result; Comparing the proportion of each prediction classification result with the proportion of each prediction classification result in the last preset time period, when the difference value is greater than a preset proportion difference threshold, determining that the preset verification mechanism is not met; When the difference value is less than or equal to the preset proportion difference threshold, obtaining the confidence level of each prediction classification result; Determining that the prediction classification result with a confidence level greater than a preset confidence threshold meets the preset verification mechanism; Determining that the prediction classification result with a confidence level less than or equal to the preset confidence threshold does not meet the preset verification mechanism.

6. A system for implementing data security classification automation based on large models, characterized in that, The system comprises: A determination module is configured to obtain specific types of initial data and preset classification keywords within a preset historical time period, determine corresponding edge computing nodes according to the specific types, send the initial data and the preset classification keywords to the corresponding edge computing nodes according to the specific types, process the initial data into preset uniform format data to be classified by the edge computing nodes, and determine classification results and classification accuracies of the data to be classified by using the preset classification keywords; A training module is configured to obtain the data to be classified, the classification results and the classification accuracies returned by all edge computing nodes, upload initial data with a classification accuracy lower than a preset threshold to a preset manual labeling terminal, update the classification results by using the returned data from the manual labeling terminal, train a large language model by using the data to be classified and the classification results, and obtain a trained large language model by taking a cross-entropy loss function as an optimization target and continuously adjusting model parameters through back propagation during the training process; An obtaining module is configured to obtain real-time collected data, determine corresponding edge computing nodes according to specific types of the collected data, obtain preset uniform format data to be processed returned by the edge computing nodes, input the data to be processed into the trained large language model to obtain output prediction classification results and confidence levels, verify whether the prediction classification results meet a preset verification mechanism by obtaining all prediction classification results and confidence levels within a preset time period, determine the prediction classification results as final classification results when the preset verification mechanism is met, or upload corresponding collected data to a manual labeling terminal to obtain returned final classification results.

7. The system for implementing data security classification automation based on large models according to claim 6, characterized in that, The determination module comprises a determination unit, configured to split the initial data into a plurality of initial fields; convert the initial fields into preset standard description data; splice the preset uniform format data to be classified based on the preset standard description data; compare the preset standard description data and the preset classification keywords for similarity, and determine a classification result corresponding to a preset classification keyword with the maximum similarity as a classification result of the current data to be classified; determine the maximum similarity as the classification accuracy.

8. The system for implementing data security classification automation based on large models according to claim 6, wherein, The obtaining module comprises a judgment unit, configured to obtain all prediction classification results within a preset time period, and obtain a proportion of each prediction classification result. The proportion of each prediction classification result is compared with the proportion of each prediction classification result in the previous preset time period. When the difference value is greater than the preset proportion difference threshold, it is determined that the preset verification mechanism is not met; When the difference value is less than or equal to the preset proportion difference threshold, the confidence of each prediction classification result is obtained; It is determined that the prediction classification result with a confidence greater than the preset confidence threshold meets the preset verification mechanism; It is determined that the prediction classification result with a confidence less than or equal to the preset confidence threshold does not meet the preset verification mechanism.

9. An apparatus for implementing data security classification automation based on a large model, characterized in that, The device comprises: a processor; and a memory having stored thereon executable code that, when executed, causes the processor to perform a method for implementing data security classification automation based on a large model according to any one of claims 1-5.

10. A non-transitory computer storage medium, comprising, having stored thereon computer instructions that, when executed, implement a method for implementing data security classification automation based on a large model according to any one of claims 1-5.

Images