Computer sensitive data intelligent identification method based on artificial intelligence

Abstract

The invention discloses a computer sensitive data intelligent identification method based on artificial intelligence, and relates to the technical field of sensitive data identification, and the identification method comprises the steps: collecting different source data, and based on a multi-modal data fusion technology, integrating the different source data; extracting features based on a self-supervised learning technology; based on a deep learning model, model training is carried out, and sensitive data detection is carried out; a dynamic model updating mechanism is introduced, and online adjustment and optimization are carried out on the model; and identifying and classifying new data by using the trained model, and implementing a multi-level classification system. Multi-modal data fusion is provided, the multi-modal data fusion technology is utilized, different types of data are integrated into a unified structure, complementarity of information is ensured, meanwhile, a dynamic model updating mechanism is introduced, online adjustment and optimization are conducted on the model according to real-time data feedback, the method can adapt to data changes and emerging modes, and the method is high in practicability and easy to popularize. And the data detection is more accurate.

Description

Technical Field

[0001] The present invention relates to the technical field of sensitive data identification, and in particular to a method for intelligently identifying computer sensitive data based on artificial intelligence. Background Art

[0002] In today's digital age, the generation and storage of data are growing exponentially. All walks of life are relying on data for decision-making, operations and services. However, with the increase in data use, the protection of sensitive data has become increasingly important, especially in cases involving personal privacy, business secrets and compliance requirements. As an important part of data security and privacy protection, sensitive data identification technology is receiving widespread attention. Sensitive data identification technology usually includes the following aspects: Scan databases, file systems and applications through automated tools to identify stored sensitive data;

[0003] Although sensitive data identification technology has made certain progress, it still faces many challenges, including data diversity, different formats, the need for identification technology to be flexible and adaptable, and the inability to quickly and accurately identify sensitive data when data traffic is large. Summary of the invention

[0004] In order to solve the above technical problems, a method for intelligent identification of computer sensitive data based on artificial intelligence is provided. This technical solution solves the above problems.

[0005] In order to achieve the above purpose, the technical solution adopted by the present invention is:

[0006] Intelligent identification method for computer sensitive data based on artificial intelligence, the identification method includes:

[0007] Collect data from different sources, integrate them based on multimodal data fusion technology, and pre-process the integrated data;

[0008] Based on self-supervised learning technology, it extracts keywords and context information from texts and high-level features from images;

[0009] Based on deep learning models, model training is performed, and sensitive data detection is performed based on models trained on large-scale data sets;

[0010] Introducing a dynamic model update mechanism to adjust and optimize the model online based on real-time feedback;

[0011] Use trained models to identify and classify new data, implement a multi-level classification system, divide data into different sensitivity levels, and set different processing strategies.

[0012] Preferably, the collecting of different source data, integrating the different source data based on multimodal data fusion technology, and preprocessing the integrated data specifically include:

[0013] Identify data sources, including structured data, unstructured data, and semi-structured data, and acquire data;

[0014] Convert data from different sources into a unified format to ensure data consistency and align data from different sources based on associated fields;

[0015] Standardize the images, including resizing, normalization, and enhancement;

[0016] Duplicate data entries were identified and removed, missing values ​​were deleted, and categorical variables were encoded.

[0017] Preferably, the extracting keywords and context information from the text and extracting high-level features from the image based on the self-supervised learning technology specifically includes:

[0018] Based on BERT as a self-supervised learning model, unlabeled text data is used for model training to obtain contextual information;

[0019] Output the embedding vector of the trained model and extract keywords based on the clustering analysis algorithm;

[0020] Input different parts of the text through the model to obtain contextual embeddings, and then select the contextual information of a specific paragraph or sentence;

[0021] After clustering is completed, keywords are extracted from each cluster and the extracted keywords are summarized;

[0022] Get image features and combine image features with text features. The feature combination formula is:

[0023] F=α·T+β·I

[0024] Where F is the fused feature vector, α is the weight coefficient of the text feature vector T, T is the text feature vector, β is the weight coefficient of the image feature vector I, and I is the image feature vector.

[0025] Preferably, the model training is performed based on the deep learning model, and the sensitive data detection is performed based on the model trained on a large-scale data set, specifically including:

[0026] Obtain training data related to sensitive data, including data samples labeled with sensitive information;

[0027] Label the collected data, including the sensitive data category and location;

[0028] Based on the fused feature vector, a detection model for sensitive data is established;

[0029] Use back propagation and optimizer to train the model using the training set;

[0030] Deploy the trained model to the production environment for real-time batch data processing;

[0031] Input the data to be tested, and the model outputs the location and type of sensitive information;

[0032] Post-process the model output results to filter out false positives and false alarms.

[0033] Preferably, the step of establishing a sensitive data detection model based on the fused feature vector specifically includes:

[0034] Based on the activation function, the fused feature vector is used as the model input, and the prediction result of the model is used as the output to establish a sensitive data detection model. The formula of the sensitive data detection model is:

[0035]

[0036] In the formula, h(F) is the prediction result of the model, F is the fused feature vector, σ(z) is the Sigmoid function, z is the result of the linear combination, and e is a mathematical constant;

[0037] Based on the prediction results of the model, the sensitive probability of the output data is calculated.

[0038] Preferably, the introduction of a dynamic model update mechanism to adjust and optimize the model online based on real-time feedback specifically includes:

[0039] Collect data generated in real time, obtain user feedback, compare system output results with target values;

[0040] Clean and preprocess the collected feedback data and calculate the model loss function;

[0041] Use the training set to train the model, and based on the model's loss function, adjust the weights to minimize the loss function;

[0042] Set the frequency of model updates, including each new data point, each time window, and a fixed time interval;

[0043] Monitor model performance indicators in real time and evaluate the performance of the updated model based on accuracy;

[0044] Use real-time data as a validation set to test the performance of the updated model;

[0045] Set thresholds to monitor whether model performance is lower than expected, such as when accuracy drops, issue warnings, and establish model version control;

[0046] Further model adjustments are made based on user feedback on model results, incorporating user feedback into the decision-making process;

[0047] Record detailed information for each model update, including updated parameters, feedback sources, and performance changes.

[0048] Preferably, the cleaning and preprocessing of the collected feedback data and the calculation of the model loss function specifically include:

[0049] Among them, the model loss function calculation formula is:

[0050]

[0051] In the formula, J is the value of the loss function, N is the total number of samples, and h is i is the true label of the i-th sample, is the predicted probability of the kth sample.

[0052] Preferably, the real-time monitoring model performance indicators, based on the accuracy evaluation of the updated model performance specifically include:

[0053] Collect the latest input data and the corresponding true labels from the system, use the currently updated model to predict the collected new data, and calculate the accuracy. The accuracy calculation formula is:

[0054]

[0055] Where A is the accuracy, TP is the number of samples that the model correctly predicts as positive, TN is the number of samples that the model correctly predicts as negative, FP is the number of samples that the model incorrectly predicts as positive, and FN is the number of samples that the model incorrectly predicts as negative.

[0056] Calculate and record accuracy regularly to capture the performance of the model at different points in time;

[0057] Generate a report containing the latest accuracy, past accuracy history, timestamp, and number of data samples;

[0058] Analyze the changing trend of accuracy to see whether the model's performance is stable and improved after receiving new data;

[0059] Incorporate user feedback on prediction results into the subsequent model update process;

[0060] Use the dashboard’s visualization tools to display accuracy and performance metrics in real time.

[0061] Preferably, the method of using the trained model to identify and classify new data, implementing a multi-level classification system, dividing the data into different sensitivity levels, and setting different processing strategies according to specific business needs specifically includes:

[0062] Load a pre-trained classification model into the environment to make predictions on new data;

[0063] Input new data into the model for identification and classification, and obtain the predicted label and corresponding sensitivity level of each data sample;

[0064] According to the prediction results output by the model, the data samples are divided into different sensitivity levels;

[0065] Develop corresponding processing strategies according to different sensitivity levels;

[0066] Automatically execute corresponding processing policies according to sensitivity levels, including data encryption, access control and record auditing;

[0067] For highly sensitive data, a manual review mechanism is implemented to ensure compliance with policy and regulatory requirements;

[0068] Real-time monitoring of the classification and processing process to ensure the normal operation of the system;

[0069] Keep detailed records of all processing processes, including data classification results, processing strategy implementation and audit records.

[0070] Preferably, formulating corresponding processing strategies according to different sensitivity levels specifically includes:

[0071] Low sensitivity level: public, with routine storage and access;

[0072] Medium sensitivity level: encrypted storage, restricted access rights, and regular audits;

[0073] High sensitivity level: Strict access control, requiring additional monitoring and review.

[0074] Compared with the prior art, the present invention has the following beneficial effects:

[0075] The present invention proposes multimodal data fusion, which uses multimodal data fusion technology to integrate different types of data into a unified structure, ensuring the complementarity of information. At the same time, a dynamic model update mechanism is introduced to adjust and optimize the model online based on real-time data feedback, which can adapt to data changes and emerging patterns, making data detection more accurate. BRIEF DESCRIPTION OF THE DRAWINGS

[0076] Figure 1 It is a step flow framework diagram of the present invention. DETAILED DESCRIPTION

[0077] The following description is used to disclose the present invention so that those skilled in the art can implement the present invention. The preferred embodiments described below are only examples, and those skilled in the art may think of other obvious variations.

[0078] Reference Figure 1 As shown, the intelligent identification method of computer sensitive data based on artificial intelligence includes:

[0079] Collect data from different sources, integrate them based on multimodal data fusion technology, and pre-process the integrated data;

[0080] Based on self-supervised learning technology, it extracts keywords and context information from texts and high-level features from images;

[0081] Based on deep learning models, model training is performed, and sensitive data detection is performed based on models trained on large-scale data sets;

[0082] Introducing a dynamic model update mechanism to adjust and optimize the model online based on real-time feedback;

[0083] Use trained models to identify and classify new data, implement a multi-level classification system, divide data into different sensitivity levels, and set different processing strategies.

[0084] Furthermore, the collecting of different source data, integrating the different source data based on multimodal data fusion technology, and preprocessing the integrated data specifically include:

[0085] Identify data sources, including structured data, unstructured data, and semi-structured data, and acquire data;

[0086] Convert data from different sources into a unified format to ensure data consistency and align data from different sources based on associated fields;

[0087] Standardize the images, including resizing, normalization, and enhancement;

[0088] Duplicate data entries were identified and removed, missing values ​​were deleted, and categorical variables were encoded.

[0089] The method of extracting keywords and context information from text and extracting high-level features from images based on self-supervised learning technology specifically includes:

[0090] Based on BERT as a self-supervised learning model, unlabeled text data is used for model training to obtain contextual information;

[0091] Output the embedding vector of the trained model and extract keywords based on the clustering analysis algorithm;

[0092] Input different parts of the text through the model to obtain contextual embeddings, and then select the contextual information of a specific paragraph or sentence;

[0093] After clustering is completed, keywords are extracted from each cluster and the extracted keywords are summarized;

[0094] Get image features and combine image features with text features. The feature combination formula is:

[0095] F=α·T+β·I

[0096] Where F is the fused feature vector, α is the weight coefficient of the text feature vector T, T is the text feature vector, β is the weight coefficient of the image feature vector I, and I is the image feature vector.

[0097] Furthermore, the model training based on the deep learning model and the sensitive data detection based on the model trained on a large-scale data set specifically include:

[0098] Obtain training data related to sensitive data, including data samples labeled with sensitive information;

[0099] Label the collected data, including the sensitive data category and location;

[0100] Based on the fused feature vector, a detection model for sensitive data is established;

[0101] Use back propagation and optimizer to train the model using the training set;

[0102] Deploy the trained model to the production environment for real-time batch data processing;

[0103] Input the data to be tested, and the model outputs the location and type of sensitive information;

[0104] Post-process the model output results to filter out false positives and false alarms.

[0105] Furthermore, the establishment of a sensitive data detection model based on the fused feature vector specifically includes:

[0106] Based on the activation function, the fused feature vector is used as the model input, and the prediction result of the model is used as the output to establish a sensitive data detection model. The formula of the sensitive data detection model is:

[0107]

[0108] In the formula, h(F) is the prediction result of the model, F is the fused feature vector, σ(z) is the Sigmoid function, z is the result of the linear combination, and e is a mathematical constant;

[0109] Based on the prediction results of the model, the sensitive probability of the output data is calculated.

[0110] Furthermore, the introduction of a dynamic model update mechanism to adjust and optimize the model online based on real-time feedback specifically includes:

[0111] Collect data generated in real time, obtain user feedback, compare system output results with target values;

[0112] Clean and preprocess the collected feedback data and calculate the model loss function;

[0113] Use the training set to train the model, and based on the model's loss function, adjust the weights to minimize the loss function;

[0114] Set the frequency of model updates, including each new data point, each time window, and a fixed time interval;

[0115] Monitor model performance indicators in real time and evaluate the performance of the updated model based on accuracy;

[0116] Use real-time data as a validation set to test the performance of the updated model;

[0117] Set thresholds to monitor whether model performance is lower than expected, such as when accuracy drops, issue warnings, and establish model version control;

[0118] Further model adjustments are made based on user feedback on model results, incorporating user feedback into the decision-making process;

[0119] Record detailed information for each model update, including updated parameters, feedback sources, and performance changes.

[0120] Furthermore, the collected feedback data is cleaned and preprocessed, and the calculation model loss function specifically includes:

[0121] Among them, the model loss function calculation formula is:

[0122]

[0123] In the formula, J is the value of the loss function, N is the total number of samples, and h is i is the true label of the i-th sample, is the predicted probability of the kth sample.

[0124] Furthermore, the real-time monitoring model performance indicators, based on the accuracy evaluation of the updated model performance specifically include:

[0125] Collect the latest input data and the corresponding true labels from the system, use the currently updated model to predict the collected new data, and calculate the accuracy. The accuracy calculation formula is:

[0126]

[0127] Where A is the accuracy, TP is the number of samples that the model correctly predicts as positive, TN is the number of samples that the model correctly predicts as negative, FP is the number of samples that the model incorrectly predicts as positive, and FN is the number of samples that the model incorrectly predicts as negative.

[0128] Calculate and record accuracy regularly to capture the performance of the model at different points in time;

[0129] Generate a report containing the latest accuracy, past accuracy history, timestamp, and number of data samples;

[0130] Analyze the changing trend of accuracy to see whether the model's performance is stable and improved after receiving new data;

[0131] Incorporate user feedback on prediction results into the subsequent model update process;

[0132] Use the dashboard’s visualization tools to display accuracy and performance metrics in real time.

[0133] Furthermore, the trained model is used to identify and classify new data, a multi-level classification system is implemented, data is divided into different sensitivity levels, and different processing strategies are set according to specific business needs, including:

[0134] Load a pre-trained classification model into the environment to make predictions on new data;

[0135] Input new data into the model for identification and classification, and obtain the predicted label and corresponding sensitivity level of each data sample;

[0136] According to the prediction results output by the model, the data samples are divided into different sensitivity levels;

[0137] Develop corresponding processing strategies according to different sensitivity levels;

[0138] Automatically execute corresponding processing policies according to sensitivity levels, including data encryption, access control and record auditing;

[0139] For highly sensitive data, a manual review mechanism is implemented to ensure compliance with policy and regulatory requirements;

[0140] Real-time monitoring of the classification and processing process to ensure the normal operation of the system;

[0141] Keep detailed records of all processing processes, including data classification results, processing strategy implementation and audit records.

[0142] Furthermore, the formulation of corresponding processing strategies according to different sensitivity levels specifically includes:

[0143] Low sensitivity level: public, with routine storage and access;

[0144] Medium sensitivity level: encrypted storage, restricted access rights, and regular audits;

[0145] High sensitivity level: Strict access control, requiring additional monitoring and review.

[0146] The use process of the present invention is:

[0147] Step 1: Identify data sources, including structured data, unstructured data, and semi-structured data, and obtain data;

[0148] Step 2: Convert data from different sources into a unified format to ensure data consistency and align data from different sources based on related fields;

[0149] Step 3: Standardize the image, including resizing, normalization, and enhancement;

[0150] Step 4: Identify and remove duplicate data entries, delete missing values, and encode categorical variables.

[0151] Step 5: Based on BERT as a self-supervised learning model, use unlabeled text data to train the model and obtain context information;

[0152] Step 6: Output the embedding vector of the trained model and extract keywords based on the clustering analysis algorithm;

[0153] Step 7: Input different parts of the text through the model to obtain contextual embeddings, and then select the contextual information of a specific paragraph or sentence;

[0154] Step 8: After clustering is completed, extract keywords from each cluster and summarize the extracted keywords;

[0155] Step 9: Obtain image features and combine image features with text features;

[0156] Step 10: Obtain training data related to sensitive data, including data samples marked with sensitive information;

[0157] Step 11: Label the collected data, including the sensitive data category and location;

[0158] Step 12: Based on the fused feature vector, establish a sensitive data detection model;

[0159] Step 13: Use back propagation and optimizer to train the model using the training set;

[0160] Step 14: Deploy the trained model to the production environment for real-time batch data processing;

[0161] Step 15: Input the data to be tested, and the model outputs the location and type of sensitive information;

[0162] Step 16: Post-process the model output results to filter out false positives and false alarms;

[0163] Step 17: Collect data generated in real time, obtain user feedback, compare system output results with target values;

[0164] Step 18: Clean and preprocess the collected feedback data and calculate the model loss function;

[0165] Step 19: Use the training set to train the model, and adjust the weights to minimize the loss function based on the model's loss function;

[0166] Step 20: Set the frequency of model updates, including each new data point, each time window and a fixed time interval;

[0167] Step 21: Monitor model performance indicators in real time and evaluate the performance of the updated model based on accuracy;

[0168] Step 22: Use real-time data as a validation set to test the performance of the updated model;

[0169] Step 23: Set thresholds to monitor whether the model performance is lower than expected. If the accuracy drops, issue an alert and establish model version control.

[0170] Step 24: Further adjust the model based on user feedback on the model results and incorporate user feedback into the decision-making process;

[0171] Step 25: Record detailed information for each model update, including updated parameters, feedback sources, and performance changes;

[0172] Step 26: Collect the latest input data and the corresponding true labels from the system, use the currently updated model to predict the collected new data, and calculate the accuracy;

[0173] Step 27: Calculate and record the accuracy regularly to capture the performance of the model at different time points;

[0174] Step 28: Generate a report, including the latest accuracy, past accuracy history data, timestamp, and number of data samples;

[0175] Step 29: Analyze the trend of accuracy to see whether the model's performance is stable and improved after receiving new data;

[0176] Step 30: Incorporate user feedback on prediction results into the subsequent model update process;

[0177] Step 31: Use the dashboard’s visualization tools to display accuracy and performance metrics in real time.

[0178] Step 32: Load the pre-trained classification model into the environment and make predictions on new data;

[0179] Step 33: Input new data into the model for identification and classification to obtain the predicted label and corresponding sensitivity level of each data sample;

[0180] Step 34: Classify data samples into different sensitivity levels according to the prediction results output by the model;

[0181] Step 35: Develop corresponding processing strategies according to different sensitivity levels;

[0182] Step 36: Automatically execute corresponding processing strategies according to the sensitivity level, including data encryption, access control and record auditing;

[0183] Step 37: For highly sensitive data, implement a manual review mechanism to ensure compliance with policy and regulatory requirements;

[0184] Step 38: Monitor the classification and processing process in real time to ensure the normal operation of the system;

[0185] Keep detailed records of all processing processes, including data classification results, processing strategy implementation and audit records.

[0186] In summary, the advantages of the present invention are:

[0187] From multiple data sources, ensure the diversity and richness of data, and use multimodal data fusion technology to integrate different types of data into a unified structure to ensure the complementarity of information;

[0188] Apply self-supervised learning technology to extract keywords and context information from text data, and extract high-level features such as edges, shapes, and textures from image data, thereby increasing the data extraction rate;

[0189] Based on the extracted features, a deep learning model is built for training, and large-scale data sets are used to optimize the model to ensure the efficiency of the model in sensitive data detection.

[0190] A dynamic model update mechanism is introduced to adjust and optimize the model online based on real-time data feedback to adapt to data changes and emerging patterns.

[0191] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The above embodiments and descriptions only describe the principles of the present invention. The present invention may be subject to various changes and improvements without departing from the spirit and scope of the present invention. These changes and improvements fall within the scope of the present invention. The scope of protection claimed by the present invention is defined by the attached claims and their equivalents.

Claims

1. A computer sensitive data intelligent identification method based on artificial intelligence, characterized in that: Identification methods include: Collect data from different sources, integrate them based on multimodal data fusion technology, and pre-process the integrated data; Based on self-supervised learning technology, it extracts keywords and context information from texts and high-level features from images; Based on deep learning models, model training is performed, and sensitive data detection is performed based on models trained on large-scale data sets; Introducing a dynamic model update mechanism to adjust and optimize the model online based on real-time feedback; Use trained models to identify and classify new data, implement a multi-level classification system, divide data into different sensitivity levels, and set different processing strategies.

2. The method for intelligently identifying computer sensitive data based on artificial intelligence according to claim 1, characterized in that: The collecting of different source data, integrating the different source data based on multimodal data fusion technology, and preprocessing the integrated data specifically include: Identify data sources, including structured data, unstructured data, and semi-structured data, and acquire data; Convert data from different sources into a unified format to ensure data consistency and align data from different sources based on associated fields; Standardize the images, including resizing, normalization, and enhancement; Duplicate data entries were identified and removed, missing values ​​were deleted, and categorical variables were encoded.

3. The method for intelligently identifying computer sensitive data based on artificial intelligence according to claim 2, characterized in that: The method of extracting keywords and context information from text and extracting high-level features from images based on self-supervised learning technology specifically includes: Based on BERT as a self-supervised learning model, unlabeled text data is used for model training to obtain contextual information; Output the embedding vector of the trained model and extract keywords based on the clustering analysis algorithm; Input different parts of the text through the model to obtain contextual embeddings, and then select the contextual information of a specific paragraph or sentence; After clustering is completed, keywords are extracted from each cluster and the extracted keywords are summarized; Get image features and combine image features with text features. The feature combination formula is: F=α·T+β·I Where F is the fused feature vector, α is the weight coefficient of the text feature vector T, T is the text feature vector, β is the weight coefficient of the image feature vector I, and I is the image feature vector.

4. The method for intelligently identifying computer sensitive data based on artificial intelligence according to claim 3 is characterized in that: The model training based on the deep learning model and the sensitive data detection based on the model trained on a large-scale data set specifically include: Obtain training data related to sensitive data, including data samples labeled with sensitive information; Label the collected data, including the sensitive data category and location; Based on the fused feature vector, a detection model for sensitive data is established; Use back propagation and optimizer to train the model using the training set; Deploy the trained model to the production environment for real-time batch data processing; Input the data to be tested, and the model outputs the location and type of sensitive information; Post-process the model output results to filter out false positives and false alarms.

5. The method for intelligently identifying computer sensitive data based on artificial intelligence according to claim 4 is characterized in that: The establishment of a sensitive data detection model based on the fused feature vector specifically includes: Based on the activation function, the fused feature vector is used as the model input, and the prediction result of the model is used as the output to establish a sensitive data detection model. The formula of the sensitive data detection model is: In the formula, h(F) is the prediction result of the model, F is the fused feature vector, σ(z) is the Sigmoid function, z is the result of the linear combination, and e is a mathematical constant; Based on the prediction results of the model, the sensitive probability of the output data is calculated.

6. The method for intelligently identifying computer sensitive data based on artificial intelligence according to claim 5, characterized in that: The introduction of a dynamic model update mechanism to adjust and optimize the model online based on real-time feedback specifically includes: Collect data generated in real time, obtain user feedback, compare system output results with target values; Clean and preprocess the collected feedback data and calculate the model loss function; Use the training set to train the model, and based on the model's loss function, adjust the weights to minimize the loss function; Set the frequency of model updates, including each new data point, each time window, and a fixed time interval; Monitor model performance indicators in real time and evaluate the performance of the updated model based on accuracy; Use real-time data as a validation set to test the performance of the updated model; Set thresholds to monitor whether model performance is lower than expected, such as when accuracy drops, issue warnings, and establish model version control; Further model adjustments are made based on user feedback on model results, incorporating user feedback into the decision-making process; Record detailed information for each model update, including updated parameters, feedback sources, and performance changes.

7. The method for intelligently identifying computer sensitive data based on artificial intelligence according to claim 6 is characterized in that: The collected feedback data is cleaned and preprocessed, and the model loss function is calculated. include: Among them, the model loss function calculation formula is: In the formula, J is the value of the loss function, N is the total number of samples, and h i is the true label of the i-th sample, is the predicted probability of the kth sample.

8. The method for intelligently identifying computer sensitive data based on artificial intelligence according to claim 7 is characterized in that: The real-time monitoring model performance indicators, based on the accuracy evaluation of the updated model performance, specifically include: Collect the latest input data and the corresponding true labels from the system, use the currently updated model to predict the collected new data, and calculate the accuracy. The accuracy calculation formula is: Where A is the accuracy, TP is the number of samples that the model correctly predicts as positive, TN is the number of samples that the model correctly predicts as negative, FP is the number of samples that the model incorrectly predicts as positive, and FN is the number of samples that the model incorrectly predicts as negative. Calculate and record accuracy regularly to capture the performance of the model at different points in time; Generate a report containing the latest accuracy, past accuracy history, timestamp, and number of data samples; Analyze the changing trend of accuracy to see whether the model's performance is stable and improved after receiving new data; Incorporate user feedback on prediction results into the subsequent model update process; Use the dashboard’s visualization tools to display accuracy and performance metrics in real time.

9. The method for intelligently identifying computer sensitive data based on artificial intelligence according to claim 8, characterized in that: The trained model is used to identify and classify new data, a multi-level classification system is implemented, data is divided into different sensitivity levels, and different processing strategies are set according to specific business needs, including: Load a pre-trained classification model into the environment to make predictions on new data; Input new data into the model for identification and classification, and obtain the predicted label and corresponding sensitivity level of each data sample; According to the prediction results output by the model, the data samples are divided into different sensitivity levels; Develop corresponding processing strategies according to different sensitivity levels; Automatically execute corresponding processing policies according to sensitivity levels, including data encryption, access control and record auditing; For highly sensitive data, a manual review mechanism is implemented to ensure compliance with policy and regulatory requirements; Real-time monitoring of the classification and processing process to ensure the normal operation of the system; Keep detailed records of all processing processes, including data classification results, processing strategy implementation and audit records.

10. The method for intelligently identifying computer sensitive data based on artificial intelligence according to claim 9, characterized in that: The corresponding processing strategies formulated according to different sensitivity levels specifically include: Low sensitivity level: public, with routine storage and access; Medium sensitivity level: encrypted storage, restricted access rights, and regular audits; High sensitivity level: Strict access control, requiring additional monitoring and review.

Images