Data processing method and device based on data desensitization, terminal and medium

Abstract

The invention discloses a data processing method and device based on data desensitization, a terminal and a medium, and the method comprises the steps: calling at least two original data files from a data file database, and obtaining confidentiality grade labels corresponding to the original data files; performing desensitization processing on the original data file based on the confidentiality level label to obtain a processed original data file; obtaining a confidentiality evaluation result corresponding to the processed original data file according to the processed original data file; determining a desensitization deviation value corresponding to the original data file based on the confidentiality level label and the confidentiality evaluation result; and if a desensitization deviation value which is not smaller than a first preset proportion in the desensitization deviation values is smaller than a preset desensitization deviation threshold value, determining the processed original data file as a target data file. The invention aims to ensure that the data desensitization processing effect of the same batch of original data files reaches an expected level and improve the confidentiality management level of the data files.

Description

Technical Field

[0001] The present application relates to the technical field of data desensitization, and in particular to a data processing method, device, terminal and medium based on data desensitization. Background Art

[0002] Data Masking is a technology that protects privacy by modifying sensitive data. It ensures that the data does not reveal personal or other sensitive information while maintaining its original format and statistical characteristics.

[0003] However, for data files with different confidentiality levels in a batch or a database, the existing data desensitization methods have the following shortcomings: the existing data confidentiality management strategy cannot perform differentiated processing for data files with different confidentiality levels. This may result in high-confidentiality data files not being adequately protected, while low-confidentiality data files may be over-desensitized, affecting the availability of data; the lack of quantitative evaluation indicators makes the evaluation results highly subjective and difficult to compare and verify objectively; the existing desensitization methods usually do not establish an effective feedback mechanism, and cannot promptly discover and correct problems in the desensitization process, resulting in unstable desensitization effects, and thus unable to improve the overall level of data confidentiality management. Summary of the invention

[0004] The main purpose of this application is to provide a data processing method, device, terminal and medium based on data desensitization, aiming to ensure that the data desensitization processing effect of the same batch of original data files reaches the expected level, so that files with different confidentiality levels are adapted to different data confidentiality effects, and by quantifying the errors of desensitization processing, continuously enhance the data desensitization effect, thereby improving the overall confidentiality management level of data files.

[0005] To achieve the above-mentioned purpose, the present application provides a data processing method based on data desensitization, which is applied to a data middle station, wherein the data middle station includes a data file database and a server, and the data file database and the server are both in the same network;

[0006] The method comprises:

[0007] Retrieving at least two original data files from a data file database, and obtaining confidentiality level labels corresponding to the original data files, wherein the confidentiality level label is used to represent a confidentiality level for the original data file selected from a preset confidentiality level set, and the preset confidentiality level set includes at least two different confidentiality levels;

[0008] Based on the confidentiality level label, desensitizing the original data file to obtain a processed original data file;

[0009] By presetting a confidentiality evaluation model, according to the processed original data file, a confidentiality evaluation result corresponding to the processed original data file is obtained;

[0010] Based on the confidentiality level label and confidentiality evaluation results, determine the desensitization deviation value corresponding to the original data file;

[0011] If no less than a first preset proportion of the desensitization deviation values ​​is less than a preset desensitization deviation threshold, the processed original data file is determined to be a target data file.

[0012] Specifically, the server is equipped with a preset desensitization processing model set, the preset desensitization processing model set includes at least two different preset desensitization processing models, and the preset desensitization processing models correspond to the confidentiality levels one by one;

[0013] The desensitizing the original data file based on the confidentiality level label to obtain the processed original data file includes:

[0014] Based on the confidentiality level label, selecting a preset desensitization processing model corresponding to the confidentiality level label from the preset desensitization processing model set;

[0015] The processed original data file is obtained according to the original data file through the preset desensitization processing model corresponding to the confidentiality level label.

[0016] Specifically, the preset desensitization processing model includes a first input layer, a first embedding layer, a first convolutional layer, a pooling layer, a first fully connected layer and a first output layer;

[0017] The method of obtaining the processed original data file according to the original data file by using the preset desensitization processing model corresponding to the confidentiality level label includes:

[0018] Obtaining a first sensitive information tensor according to the sensitive information in the original data file through the first input layer;

[0019] Obtaining a first floating-point number tensor according to the first sensitive information tensor through the first embedding layer;

[0020] Obtaining a second floating-point tensor according to the first floating-point tensor through the first convolutional layer;

[0021] Obtaining a third floating-point tensor according to the second floating-point tensor through the pooling layer;

[0022] Obtaining a fourth floating-point tensor according to the third floating-point tensor through the first fully connected layer;

[0023] Obtaining a first output tensor according to the fourth floating-point number tensor through the first output layer, wherein the first output tensor is used to represent the sensitive information after desensitization processing;

[0024] Based on the original data file and the first output tensor, the processed original data file is obtained.

[0025] Specifically, the preset desensitization processing model includes a second input layer, a second embedding layer, a second convolutional layer, a recurrent neural network layer, an attention mechanism layer, a second fully connected layer and a second output layer;

[0026] The method of obtaining the processed original data file according to the original data file by using the preset desensitization processing model corresponding to the confidentiality level label includes:

[0027] Obtaining a second sensitive information tensor according to the sensitive information in the original data file through the second input layer;

[0028] Obtaining a fifth floating-point number tensor according to the second sensitive information tensor through the second embedding layer;

[0029] Obtaining a sixth floating-point tensor according to the fifth floating-point tensor through the second convolutional layer;

[0030] Obtaining a seventh floating-point tensor according to the sixth floating-point tensor through the recurrent neural network layer;

[0031] Obtaining an eighth floating-point tensor according to the seventh floating-point tensor through the attention mechanism layer;

[0032] Obtaining a ninth floating-point tensor according to the eighth floating-point tensor through the second fully connected layer;

[0033] Obtaining a second output tensor according to the ninth floating-point number tensor through the second output layer, wherein the second output tensor is used to represent the sensitive information after desensitization processing;

[0034] Based on the original data file and the second output tensor, the processed original data file is obtained.

[0035] Specifically, the preset confidentiality evaluation model includes a third input layer, a third embedding layer, a self-attention mechanism layer, a multi-layer perceptron layer and a third output layer;

[0036] The method of obtaining a confidentiality evaluation result corresponding to the processed original data file by using a preset confidentiality evaluation model according to the processed original data file includes:

[0037] Obtaining an input tensor according to the processed original data file through the third input layer;

[0038] Obtaining a tenth floating-point number tensor according to the input tensor through the third embedding layer;

[0039] Obtaining an eleventh floating-point number tensor according to the tenth floating-point number tensor through the self-attention mechanism layer;

[0040] Obtaining a twelfth floating-point tensor according to the eleventh floating-point tensor through the multi-layer perceptron layer;

[0041] A confidentiality evaluation score is obtained according to the twelfth floating-point number tensor through the third output layer, and the confidentiality evaluation score is determined to be the confidentiality evaluation result.

[0042] Specifically, determining the desensitization deviation value corresponding to the original data file based on the confidentiality level label and the confidentiality evaluation result includes:

[0043] The confidentiality level corresponding to the confidentiality level label is converted into the confidentiality level value corresponding to the confidentiality level label by a preset linear mapping function;

[0044] The absolute value of the difference between the confidentiality level value and the confidentiality evaluation score is calculated to obtain the desensitization deviation value.

[0045] Specifically, after determining the desensitization deviation value corresponding to the original data file based on the confidentiality level label and the confidentiality evaluation result, the method further includes:

[0046] If a proportion of the desensitization deviation values ​​that is not less than a second preset proportion is greater than a preset desensitization deviation threshold, the preset desensitization processing model is updated based on the desensitization deviation value to obtain an updated preset desensitization processing model, wherein the updated preset desensitization model is used to perform desensitization processing on the original data file.

[0047] To achieve the above-mentioned purpose, the present application also provides a data processing device based on data desensitization, which is applied to a data middle station, wherein the data middle station includes a data file database and a server, and the data file database and the server are both in the same network;

[0048] The device comprises:

[0049] A first unit is used to retrieve at least two original data files from a data file database and obtain confidentiality level labels corresponding to the original data files, wherein the confidentiality level label is used to represent a confidentiality level for the original data file selected from a preset confidentiality level set, and the preset confidentiality level set includes at least two different confidentiality levels;

[0050] The second unit is used to perform desensitization processing on the original data file based on the confidentiality level label to obtain a processed original data file;

[0051] The third unit is used to obtain a confidentiality evaluation result corresponding to the processed original data file according to the processed original data file by using a preset confidentiality evaluation model;

[0052] The fourth unit is used to determine the desensitization deviation value corresponding to the original data file based on the confidentiality level label and the confidentiality evaluation result;

[0053] The fifth unit is used to determine that the processed original data file is a target data file if no less than a first preset proportion of the desensitization deviation values ​​is less than a preset desensitization deviation threshold.

[0054] To achieve the above-mentioned purpose, the present application also provides a terminal, including a memory storing multiple instructions; the processor loads instructions from the memory to execute the steps in any one of the methods provided in the present application.

[0055] To achieve the above objectives, the present application also provides a medium, wherein the medium stores a plurality of instructions, wherein the instructions are suitable for a processor to load to execute the steps in any one of the methods provided in the present application.

[0056] The present application provides a data processing method, device, terminal and medium based on data desensitization. First, at least two original data files can be retrieved from a data file database, and confidentiality level labels corresponding to the original data files can be obtained, wherein the confidentiality level label is used to characterize a confidentiality level for the original data file selected from a preset confidentiality level set, and the preset confidentiality level set includes at least two different confidentiality levels; then, based on the confidentiality level label, the original data file is desensitized to obtain a processed original data file; then, through a preset confidentiality evaluation model, according to the processed original data file, a confidentiality evaluation result corresponding to the processed original data file is obtained; then, based on the confidentiality level label and the confidentiality evaluation result, a desensitization deviation value corresponding to the original data file is determined; finally, if a desensitization deviation value having a desensitization deviation value not less than a first preset proportion is less than a preset desensitization deviation threshold, the processed original data file is determined as a target data file to ensure that the data desensitization processing effect of the same batch of original data files reaches the expected level.

[0057] Compared with the existing data desensitization methods, the present application can ensure that the data desensitization processing effect of the same batch of original data files reaches the expected level, so that files with different confidentiality levels are adapted to different data confidentiality effects, and by quantifying the errors of the desensitization processing, the data desensitization effect is continuously enhanced, thereby improving the overall confidentiality management level of data files. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1 A schematic diagram of a process flow provided for an embodiment of the present application;

[0059] Figure 2 A schematic diagram of the structure of a data processing device based on data desensitization provided in an embodiment of the present application;

[0060] Figure 3 A schematic diagram of the structure of a terminal provided in an embodiment of the present application. DETAILED DESCRIPTION

[0061] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of this application.

[0062] Due to the different confidentiality levels of data files in a batch or a database, the existing data desensitization methods have the following shortcomings: the existing data confidentiality management strategy cannot perform differentiated processing for data files of different confidentiality levels. This may result in high-confidentiality data files not being adequately protected, while low-confidentiality data files may be over-desensitized, affecting the availability of data; the lack of quantitative evaluation indicators makes the evaluation results highly subjective and difficult to compare and verify objectively; the existing desensitization methods usually do not establish an effective feedback mechanism, and cannot promptly discover and correct problems in the desensitization process, resulting in unstable desensitization effects, and thus unable to improve the overall level of data confidentiality management.

[0063] Therefore, the embodiments of the present application provide a data processing method, device, terminal and medium based on data desensitization to solve practical technical problems.

[0064] In some embodiments, the device may be specifically integrated into an electronic device, which may be a terminal, a server or other device.

[0065] In some embodiments, the server may also be implemented in the form of a terminal.

[0066] Among them, the server can be an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, CDN (Content Delivery Network), as well as big data and artificial intelligence platforms.

[0067] The terminal may be a smart phone, a tablet computer, a laptop computer, a desktop computer, a smart speaker, a smart watch, etc., but is not limited thereto. The terminal and the server may be directly or indirectly connected via wired or wireless communication, and this application does not limit this.

[0068] It should be noted that the serial numbers of the following embodiments are not intended to limit the preferred order of the embodiments.

[0069] An embodiment of the present application provides a data processing method based on data desensitization, which can ensure that the data desensitization processing effect of the same batch of original data files reaches the expected level, so that files with different confidentiality levels are adapted to different data confidentiality effects, and by quantifying the errors of the desensitization processing, the data desensitization effect is continuously enhanced, thereby improving the overall confidentiality management level of the data files.

[0070] A data processing method based on data desensitization is applied to a data middle platform, wherein the data middle platform includes a data file database and a server, and the data file database and the server are both in the same network.

[0071] Data Middleware Platform is a type of enterprise-level data management and application architecture model that plays a core role in the modern data-driven business environment. The goal of the data middleware platform is to unify and optimize the management and use of data resources to support more efficient and flexible data analysis and business decision-making. Specifically, the data middleware platform has the following features:

[0072] Data integration: The data center integrates and standardizes data from different sources and formats to form a unified data view to eliminate information silos and ensure data consistency and accuracy;

[0073] Data governance: provides a complete set of data management processes, including data quality control, data security, data lifecycle management, etc., to ensure data compliance and availability;

[0074] Data analysis and services: The data center platform supports real-time analysis, predictive analysis and intelligent decision-making by providing pre-processed data, models and algorithms, helping enterprises to respond quickly to market changes;

[0075] Data sharing and reuse: It promotes cross-departmental and cross-system data sharing, reduces duplication of construction, increases data reuse rate, and reduces the cost of data acquisition and processing;

[0076] Flexibility and scalability: The data center adopts a microservice architecture, which can be easily expanded and adjusted to adapt to the rapid development and changes of the business;

[0077] Technology stack: Data middle platform usually involves components such as big data processing technology, data warehouse, data lake, data API, machine learning platform, etc.

[0078] In some embodiments, the data middle platform includes a data file database and a server, and the data file database and the server are both in the same network. The data file database can be used to store different original data files corresponding to different confidentiality level labels, and the server is used to manage and process the original data files and the original data files after desensitization processing.

[0079] like Figure 1 As shown, the specific process of the method can be as follows:

[0080] S110. Retrieve at least two original data files from a data file database, and obtain confidentiality level labels corresponding to the original data files, wherein the confidentiality level label is used to represent a confidentiality level for the original data file selected from a preset confidentiality level set, and the preset confidentiality level set includes at least two different confidentiality levels.

[0081] In some embodiments, the batch ID of each raw data file can be recorded in the data file database, and a unique batch ID (Batch ID) can be assigned to each batch of raw data files. This ID can be a string or a number, which is used to uniquely identify a batch of raw data files; the batch ID can also be specified when retrieving data files to ensure that the retrieved raw data files belong to the same batch.

[0082] In some embodiments, a preset confidentiality level set may be predefined, that is, a preset confidentiality level set, such as {'Low', 'High'}, may be determined to represent the confidentiality level of the original data file. At the same time, the corresponding confidentiality level label may be stored for each original data file in the data file database, which may usually be implemented through a database table or other data structure.

[0083] Specifically, the corresponding confidentiality level label for each original data file can be determined in the following manner:

[0084] Manual annotation:

[0085] Data security experts or business personnel manually mark confidentiality levels based on the content and sensitivity of data files. This method is suitable for situations where the amount of data is small and high accuracy is required; manually mark confidentiality levels for data files based on existing business rules and standards. For example, financial data files may be marked as "High" by default, while user basic information files may be marked as "Low";

[0086] Automatic annotation:

[0087] Natural Language Processing (NLP): Use NLP technology to automatically analyze the content of data files, identify sensitive information therein, and automatically label the confidentiality level according to preset rules; Keyword Matching: Automatically label the confidentiality level by matching the preset sensitive keyword list. For example, files containing keywords such as "bank account" and "password" can be labeled as "High"; Entity Recognition: Use Named Entity Recognition (NER) technology to identify sensitive entities in files (such as names, addresses, phone numbers, etc.), and automatically label the confidentiality level based on the recognition results; Machine Learning Model: Train machine learning models to automatically predict the confidentiality level of new data files based on the labeling results of historical data files; Supervised Learning: Use labeled data files to train classification models (such as logistic regression, decision trees, random forests, etc.), and then use the model to predict the confidentiality level of new data files; Unsupervised Learning: Use clustering algorithms (such as K-means, DBSCAN, etc.) to cluster data files, and manually label or further analyze based on the clustering results;

[0088] Metadata annotation:

[0089] File metadata: Use file metadata (such as file type, creation date, modification date, etc.) to assist in determining confidentiality levels; for example, financial report files may be marked as "High" by default; File path: Determine confidentiality levels based on the path where files are stored; for example, files stored in a specific security directory may be marked as "High" by default;

[0090] Rules Engine:

[0091] Rule definition: Define a set of rules to automatically mark the confidentiality level of files based on their content and attributes. Rules can include multiple conditions such as keyword matching, file path, file type, etc. Example rules: If a file contains the keyword "bank account", mark it as "High"; if the file is stored in the "financial data" directory, mark it as "High"; if the file was created within the last 30 days, mark it as "High".

[0092] S120: Based on the confidentiality level label, desensitize the original data file to obtain a processed original data file.

[0093] In some embodiments, the server is equipped with a set of preset desensitization processing models, and the set of preset desensitization processing models includes at least two different preset desensitization processing models, and the preset desensitization processing models correspond one-to-one to the confidentiality levels.

[0094] Specifically, the desensitizing process is performed on the original data file based on the confidentiality level label to obtain the processed original data file, including the steps S121 to S122 as shown below:

[0095] S121. Based on the confidentiality level label, a preset desensitization processing model corresponding to the confidentiality level label is selected from the preset desensitization processing model set.

[0096] S122. Obtain the processed original data file according to the original data file through a preset desensitization processing model corresponding to the confidentiality level label.

[0097] In some embodiments, the preset desensitization processing model includes a first input layer, a first embedding layer, a first convolutional layer, a pooling layer, a first fully connected layer and a first output layer.

[0098] In some embodiments, the preset desensitization processing model including a first input layer, a first embedding layer, a first convolutional layer, a pooling layer, a first fully connected layer and a first output layer can be a neural network model suitable for original data files with a low confidentiality level. The neural network model suitable for original data files with a low confidentiality level is used in scenarios with lower data confidentiality requirements, such as internal testing and development environments. The model structure is relatively simple and the processing speed is fast, but the desensitization intensity is slightly low.

[0099] Specifically, the preset desensitization processing model corresponding to the confidentiality level label is used to obtain the processed original data file according to the original data file, including the steps A1 to A7 as shown below:

[0100] A1. Obtain a first sensitive information tensor according to the sensitive information in the original data file through the first input layer.

[0101] In some embodiments, through the first input layer, the sensitive information in the original data file can be converted into a two-dimensional tensor with a shape of (batch_size, sequence_length), where batch_size is the size of the batch of sensitive information, and sequence_length is the length of the information sequence in the sensitive information. The two-dimensional tensor is the first sensitive information tensor.

[0102] A2. Obtain a first floating-point tensor according to the first sensitive information tensor through the first embedding layer.

[0103] In some embodiments, each word in the text data of the sensitive information is converted into a vector representation of a fixed length through the first embedding layer, that is, a discrete vocabulary index is converted into a representation in a continuous vector space, which facilitates subsequent feature extraction.

[0104] Specifically, the first floating point tensor is a floating point tensor with a shape of (batch_size, sequence_length, embedding_dim), where embedding_dim is the dimension of the embedding vector.

[0105] A3. Obtain a second floating-point tensor according to the first floating-point tensor through the first convolutional layer.

[0106] In some embodiments, the first convolution layer may use multiple convolution kernels of different sizes to extract local features. Each convolution kernel slides on the input data to generate a feature map; multiple convolution kernels can capture local features of different scales to form a floating point tensor.

[0107] Specifically, through the first convolution layer, according to the first floating-point tensor, a floating-point tensor with a shape of (batch_size, sequence_length-kernel_size+1, num_filters) is obtained, that is, the second floating-point tensor, where kernel_size is the size of the convolution kernel and num_filters is the number of convolution kernels.

[0108] A4. Obtain a third floating-point tensor according to the second floating-point tensor through the pooling layer.

[0109] In some embodiments, the pooling layer reduces the feature dimension through maximum pooling or average pooling, retains the most important information, and can reduce the complexity of the model and reduce the risk of overfitting.

[0110] Specifically, through the pooling layer, according to the second floating-point tensor, a floating-point tensor with a shape of (batch_size, (sequence_length-kernel_size+1) / / pool_size, num_filters) is obtained, that is, the third floating-point tensor, where pool_size is the size of the pooling window.

[0111] A5. Obtain a fourth floating-point tensor according to the third floating-point tensor through the first fully connected layer.

[0112] In some embodiments, the first fully connected layer performs feature fusion and classification on the feature vector after convolution and pooling, and can capture global features to facilitate the generation of the final desensitization result.

[0113] Specifically, through the first fully connected layer, a floating point tensor with a shape of (batch_size, output_dim) is obtained according to the third floating point tensor, that is, the fourth floating point tensor, where output_dim is the output dimension of the first fully connected layer.

[0114] A6. Obtain a first output tensor according to the fourth floating-point tensor through the first output layer, wherein the first output tensor is used to represent the sensitive information after desensitization processing.

[0115] In some embodiments, the fourth floating-point tensor is input into the first output layer, and an integer or string tensor with a shape of (batch_size, sequence_length) is output, i.e., the first output tensor, which is used to represent the sensitive information after desensitization processing, for example, it can replace the asterisk part in the middle part of the phone number.

[0116] A7. Obtain the processed original data file based on the original data file and the first output tensor.

[0117] In some embodiments, the portion of the original data file containing sensitive information may be replaced with the desensitized sensitive information to obtain a processed original data file.

[0118] In some embodiments, the preset desensitization processing model includes a second input layer, a second embedding layer, a second convolutional layer, a recurrent neural network layer, an attention mechanism layer, a second fully connected layer and a second output layer.

[0119] In some embodiments, the preset desensitization processing model including a second input layer, a second embedding layer, a second convolutional layer, a recurrent neural network layer, an attention mechanism layer, a second fully connected layer and a second output layer can be a neural network model suitable for original data files with a high confidentiality level. The neural network model suitable for original data files with a high confidentiality level is used in scenarios with high data confidentiality requirements, such as external data sharing and compliance audits. The model has a complex structure and a relatively slow processing speed, but has high desensitization intensity and stronger security.

[0120] Specifically, the preset desensitization processing model corresponding to the confidentiality level label is used to obtain the processed original data file according to the original data file, including the steps B1 to B8 as shown below:

[0121] B1. Obtain a second sensitive information tensor through the second input layer according to the sensitive information in the original data file.

[0122] In some embodiments, through the second input layer, the sensitive information in the original data file can be converted into a two-dimensional tensor with a shape of (batch_size, sequence_length), where batch_size is the size of the batch of sensitive information, and sequence_length is the length of the information sequence in the sensitive information. The two-dimensional tensor is the second sensitive information tensor.

[0123] B2. Obtain a fifth floating-point tensor according to the second sensitive information tensor through the second embedding layer.

[0124] In some embodiments, each word in the text data of the sensitive information is converted into a vector representation of a fixed length through the second embedding layer, that is, a discrete vocabulary index is converted into a representation in a continuous vector space, which facilitates subsequent feature extraction.

[0125] Specifically, the fifth floating-point tensor is a floating-point tensor with a shape of (batch_size, sequence_length, embedding_dim), where embedding_dim is the dimension of the embedding vector.

[0126] B3. Obtain a sixth floating-point tensor according to the fifth floating-point tensor through the second convolutional layer.

[0127] In some embodiments, the second convolution layer may use multiple convolution kernels of different sizes to extract local features. Each convolution kernel slides on the input data to generate a feature map; multiple convolution kernels can capture local features of different scales to form a floating point tensor.

[0128] Specifically, through the second convolutional layer, according to the fifth floating-point tensor, a floating-point tensor with a shape of (batch_size, sequence_length-kernel_size+1, num_filters) is obtained, that is, the sixth floating-point tensor, wherein kernel_size is the size of the convolution kernel, and num_filters is the number of convolution kernels.

[0129] B4. Obtain a seventh floating-point tensor according to the sixth floating-point tensor through the recurrent neural network layer.

[0130] In some embodiments, the recurrent neural network layer uses a bidirectional LSTM layer to capture long-distance dependencies. The bidirectional LSTM can capture sequence information from both the forward and backward directions to improve the expressiveness of the model.

[0131] Specifically, through the recurrent neural network layer, according to the sixth floating-point tensor, a floating-point tensor with a shape of (batch_size, sequence_length-kernel_size+1, hidden_dim*2) is obtained, that is, the seventh floating-point tensor, where hidden_dim is the hidden state dimension of the LSTM unit, and *2 represents a bidirectional LSTM.

[0132] B5. Through the attention mechanism layer, an eighth floating-point tensor is obtained according to the seventh floating-point tensor.

[0133] In some embodiments, the attention mechanism layer focuses on key information by calculating the importance weight of each time step to improve the accuracy of the model and thereby improve the effect of the desensitization process.

[0134] Specifically, through the attention mechanism layer, a floating-point tensor with a shape of (batch_size, hidden_dim*2) is obtained according to the seventh floating-point tensor, that is, the eighth floating-point tensor.

[0135] B6. Obtain a ninth floating-point tensor according to the eighth floating-point tensor through the second fully connected layer.

[0136] In some embodiments, the second fully connected layer performs feature fusion and classification on the feature vector after the attention mechanism, and can capture global features to facilitate the generation of the final desensitization result.

[0137] Specifically, through the second fully connected layer, a floating point tensor with a shape of (batch_size, output_dim) is obtained according to the eighth floating point tensor, that is, the ninth floating point tensor, where output_dim is the output dimension of the fully connected layer.

[0138] B7. Obtain a second output tensor through the second output layer according to the ninth floating-point tensor, wherein the second output tensor is used to represent the sensitive information after desensitization processing.

[0139] In some embodiments, the ninth floating-point tensor is input into the second output layer, and an integer or string tensor with a shape of (batch_size, sequence_length) is output, i.e., the second output tensor, which is used to represent the sensitive information after desensitization processing, for example, it can be a random number part that replaces the middle part of the phone number.

[0140] B8. Obtain the processed original data file based on the original data file and the second output tensor.

[0141] In some embodiments, the portion of the original data file where the sensitive information is located can be replaced with the desensitized sensitive information to obtain the processed original data file.

[0142] S130 , obtaining a confidentiality evaluation result corresponding to the processed original data file according to the processed original data file by using a preset confidentiality evaluation model.

[0143] In some embodiments, the preset confidentiality evaluation model includes a third input layer, a third embedding layer, a self-attention mechanism layer, a multi-layer perceptron layer and a third output layer.

[0144] Specifically, the method of obtaining the confidentiality evaluation result corresponding to the processed original data file by presetting the confidentiality evaluation model according to the processed original data file includes the following steps S131 to S135:

[0145] S131. Obtain an input tensor through the third input layer according to the processed original data file.

[0146] In some embodiments, the input tensor may be a two-dimensional tensor of shape (batch_size, sequence_length), where batch_size is the batch size of the processed sensitive information, and sequence_length is the length of each sequence of the processed sensitive information.

[0147] S132. Obtain a tenth floating-point tensor according to the input tensor through the third embedding layer.

[0148] In some embodiments, through the third embedding layer, a floating point tensor with a shape of (batch_size, sequence_length, embedding_dim) is obtained according to the input tensor, that is, the tenth floating point tensor, where embedding_dim is the dimension of the embedding vector.

[0149] S133. Obtain an eleventh floating-point tensor according to the tenth floating-point tensor through the self-attention mechanism layer.

[0150] In some embodiments, the self-attention mechanism layer can capture long-distance dependencies in the sequence and focus on key information by calculating the importance weight of each time step to improve the accuracy of the model.

[0151] Specifically, through the third embedding layer, according to the tenth floating-point tensor, a floating-point tensor with a shape of (batch_size, sequence_length, attention_dim) is obtained, that is, the eleventh floating-point tensor, where attention_dim is the output dimension of the attention mechanism.

[0152] S134. Obtain a twelfth floating-point tensor according to the eleventh floating-point tensor through the multi-layer perceptron layer.

[0153] In some embodiments, feature fusion and classification are performed through multi-layer perceptron layers to capture global features to facilitate the generation of final confidentiality evaluation results.

[0154] Specifically, through the multi-layer perceptron layer, according to the eleventh floating-point tensor, a floating-point tensor with a shape of (batch_size, hidden_dim) is obtained, that is, the twelfth floating-point tensor, wherein hidden_dim is the hidden layer dimension of the multi-layer perceptron layer.

[0155] S135. Obtain a confidentiality evaluation score according to the twelfth floating-point number tensor through the third output layer, and determine the confidentiality evaluation score as the confidentiality evaluation result.

[0156] In some embodiments, through the third output layer, a floating point tensor with a shape of (batch_size, 1) is obtained according to the twelfth floating point tensor to represent the confidentiality evaluation score of each data file.

[0157] S140. Determine the desensitization deviation value corresponding to the original data file based on the confidentiality level label and the confidentiality evaluation result.

[0158] In some embodiments, the step of determining the desensitization deviation value corresponding to the original data file based on the confidentiality level label and the confidentiality evaluation result includes the following steps S141 to S142:

[0159] S141. Convert the confidentiality level corresponding to the confidentiality level label into the confidentiality level value corresponding to the confidentiality level label through a preset linear mapping function.

[0160] S142. Calculate the absolute value of the difference between the confidentiality level value and the confidentiality evaluation score to obtain the desensitization deviation value.

[0161] In some embodiments, the confidentiality level label corresponding to the original data file may include "High" and "Low", and the confidentiality evaluation score may be a value in the range of [0,1]. "High" means a high confidentiality level, and its interval range is [0.8,1.0]; "Low" means a low confidentiality level, and its interval range is [0.6,0.8].

[0162] Specifically, the preset linear mapping function includes an interval range, a coefficient and a preset normalized value, and the coefficient can be determined by the size of the interval range.

[0163] If a raw data file has a confidentiality level label of "High", the coefficient is set to 0.2, and the pre-set normalized value is 0.75, the corresponding confidentiality level value can be expressed as:

[0164] Confidentiality level value = 0.8 + 0.2 × normalized value

[0165] If a raw data file has a confidentiality level label of "Low", the coefficient is set to 0.2, and the pre-set normalized value is 0.75, then the corresponding confidentiality level value can be expressed as:

[0166] Confidentiality level value = 0.6 + 0.2 × normalized value

[0167] After obtaining the confidentiality level value, the desensitization deviation value is expressed by the following expression:

[0168] Desensitization deviation value = | confidentiality level value - actual confidentiality score |

[0169] Then, the desensitization deviation value is used to evaluate whether the effect of the desensitization processing has achieved the expected effect, that is, to evaluate whether the adaptive desensitization operation is performed according to the existing confidentiality level.

[0170] S150: If no less than a first preset proportion of the desensitization deviation values ​​is less than a preset desensitization deviation threshold, the processed original data file is determined to be a target data file.

[0171] In some embodiments, five original data files are retrieved from the data file database for desensitization processing, and corresponding desensitization deviation values ​​are obtained through the above embodiment content, as shown in the following content:

[0172] The first raw data file file1.txt after processing has a desensitization deviation value of 0.05;

[0173] The desensitization deviation value of the second raw data file file2.txt after processing is 0.02;

[0174] The desensitization deviation value of the processed third raw data file file3.txt is 0.02;

[0175] The desensitization deviation value of the fourth raw data file file4.txt after processing is 0.00;

[0176] The desensitization deviation value of the fifth raw data file file5.txt after processing is 0.05.

[0177] Now, the first preset ratio is set to 80%, and the preset desensitization deviation threshold is set to 0.06. Then, the desensitization deviation values ​​of the original data files after the above processing are all lower than the preset desensitization deviation threshold, and the proportion of the original data files after the processing lower than the preset desensitization deviation threshold to the retrieved original data files is 100%, which is more than 80%. Therefore, it is determined that the original data files after the above processing are the target data files that have successfully completed the desensitization processing.

[0178] In some embodiments, after determining the desensitization deviation value corresponding to the original data file based on the confidentiality level label and the confidentiality evaluation result, the method further includes the following specific implementation process:

[0179] If a proportion of the desensitization deviation values ​​that is not less than a second preset proportion is greater than a preset desensitization deviation threshold, the preset desensitization processing model is updated based on the desensitization deviation value to obtain an updated preset desensitization processing model, wherein the updated preset desensitization model is used to perform desensitization processing on the original data file.

[0180] In some embodiments, five original data files are retrieved from the data file database for desensitization processing, and corresponding desensitization deviation values ​​are obtained through the above embodiment content, as shown in the following content:

[0181] The first raw data file file1.txt after processing has a desensitization deviation value of 0.05;

[0182] The desensitization deviation value of the second raw data file file2.txt after processing is 0.02;

[0183] The desensitization deviation value of the processed third raw data file file3.txt is 0.02;

[0184] The desensitization deviation value of the fourth raw data file file4.txt after processing is 0.00;

[0185] The desensitization deviation value of the fifth raw data file file5.txt after processing is 0.05.

[0186] Now set the second preset ratio to 20%, and set the preset desensitization deviation threshold to 0.04, then two of the desensitization deviation values ​​of the original data files processed above are higher than the preset desensitization deviation threshold, and the proportion of processed original data files higher than the preset desensitization deviation threshold to the retrieved original data files is 40%, which is more than 20%. Therefore, next, the preset desensitization processing model can be updated based on the desensitization deviation value to obtain an updated preset desensitization processing model, wherein the updated preset desensitization model is used to desensitize the original data files.

[0187] Specifically, if the desensitization deviation values ​​of file1.txt and file5.txt are high, it is because some sensitive information is not fully desensitized. The following steps can be taken to update the preset desensitization processing model:

[0188] First, analyze the cause of the deviation. File1.txt and file5.txt contain some sensitive keywords, such as "bank account" and "ID number", but these keywords are not fully desensitized. Then, adjust the desensitization strategy and add detection and replacement rules for sensitive keywords to ensure that these keywords are completely replaced during the desensitization process. You can also introduce more powerful desensitization algorithms, such as text desensitization algorithms based on deep learning, to improve the desensitization effect. Use new data sets (including file1.txt and file5.txt) to retrain the preset desensitization processing model, optimize the model's hyperparameters, and improve the model's generalization ability and stability. Finally, use the updated preset desensitization model to re-desensitize the original data file and recalculate the desensitization deviation value until it is ensured that the updated preset desensitization model can effectively reduce the desensitization deviation value to achieve the expected desensitization processing level, and ultimately improve the overall data confidentiality management level.

[0189] As can be seen from the above, the embodiment of the present application can first retrieve at least two original data files from the data file database, and obtain the confidentiality level labels corresponding to the original data files, wherein the confidentiality level label is used to characterize a confidentiality level for the original data file selected from a preset confidentiality level set, and the preset confidentiality level set includes at least two different confidentiality levels; then, based on the confidentiality level label, the original data file is desensitized to obtain a processed original data file; then, through a preset confidentiality evaluation model, according to the processed original data file, a confidentiality evaluation result corresponding to the processed original data file is obtained; then, based on the confidentiality level label and the confidentiality evaluation result, a desensitization deviation value corresponding to the original data file is determined; finally, if a desensitization deviation value of not less than a first preset proportion among the desensitization deviation values ​​is less than a preset desensitization deviation threshold, the processed original data file is determined to be the target data file, so as to ensure that the data desensitization processing effect of the same batch of original data files reaches the expected level.

[0190] In summary, compared with the existing data desensitization processing or data confidentiality management methods, the embodiments of the present application can ensure that the data desensitization processing effect of the same batch of original data files reaches the expected level, so that files of different confidentiality levels are adapted to different data confidentiality effects, and by quantifying the errors of the desensitization processing, the data desensitization effect is continuously enhanced, thereby improving the overall confidentiality management level of data files.

[0191] In order to better implement the above method, the embodiment of the present application also provides a data processing device based on data desensitization, which can be integrated in an electronic device, which can be a terminal, a server, etc. Among them, the terminal can be a mobile phone, a tablet computer, a smart Bluetooth device, a laptop, a personal computer, etc.; the server can be a single server or a server cluster composed of multiple servers.

[0192] For example, in this embodiment, the method of the embodiment of the present application will be described in detail by taking the specific integration of a data processing device based on data desensitization in a terminal as an example.

[0193] For example, Figure 2 As shown, the data processing device based on data desensitization may include a first unit 201, a second unit 202, a third unit 203, a fourth unit 204 and a fifth unit 205, which are applied to a data middle station, wherein the data middle station includes a data file database and a server, and the data file database and the server are both in the same network;

[0194] The device comprises:

[0195] The first unit 201 is used to retrieve at least two original data files from a data file database and obtain confidentiality level labels corresponding to the original data files, wherein the confidentiality level label is used to represent a confidentiality level for the original data file selected from a preset confidentiality level set, and the preset confidentiality level set includes at least two different confidentiality levels;

[0196] The second unit 202 is used to perform desensitization processing on the original data file based on the confidentiality level label to obtain a processed original data file;

[0197] The third unit 203 is used to obtain a confidentiality evaluation result corresponding to the processed original data file according to the processed original data file by using a preset confidentiality evaluation model;

[0198] The fourth unit 204 is used to determine the desensitization deviation value corresponding to the original data file based on the confidentiality level label and the confidentiality evaluation result;

[0199] The fifth unit 205 is used to determine that the processed original data file is a target data file if no less than a first preset proportion of the desensitization deviation values ​​is less than a preset desensitization deviation threshold.

[0200] In specific implementation, the above units can be implemented as independent entities, or can be arbitrarily combined to be implemented as the same or several entities. The specific implementation of the above units can refer to the previous method embodiments, which will not be repeated here.

[0201] From the above, it can be seen that the embodiment of the present application can ensure that the data desensitization processing effect of the same batch of original data files reaches the expected level, so that files with different confidentiality levels are adapted to different data confidentiality effects, and by quantifying the errors of the desensitization processing, the data desensitization effect is continuously enhanced, thereby improving the overall confidentiality management level of the data files.

[0202] The embodiment of the present application also provides an electronic device, which can be a terminal, a server, etc. The terminal can be a mobile phone, a tablet computer, a smart Bluetooth device, a laptop, a personal computer, etc. The server can be a single server or a server cluster composed of multiple servers, etc.

[0203] In some embodiments, the data processing device based on data desensitization can also be integrated into multiple electronic devices. For example, the data processing device based on data desensitization can be integrated into multiple servers, and the data processing method based on data desensitization of the present application can be implemented by multiple servers.

[0204] In this embodiment, the electronic device of this embodiment is a terminal as an example for detailed description, for example, Figure 3 As shown, it shows a schematic diagram of the structure of the terminal involved in the embodiment of the present application, specifically:

[0205] The terminal may include one or more processors 301 of processing cores, one or more storage media 302, a power supply 303, an input module 304, and a communication module 305. Those skilled in the art will appreciate that Figure 3 The terminal structure shown in the figure does not constitute a limitation on the terminal, and may include more or less components than shown in the figure, or combine certain components, or arrange the components differently.

[0206] in:

[0207] The processor 301 is the control center of the terminal. It uses various interfaces and lines to connect various parts of the entire terminal. It executes various functions of the terminal and processes data by running or executing software programs and / or modules stored in the memory 302, and calling data stored in the memory 302, so as to monitor the terminal as a whole. In some embodiments, the processor 301 may include one or more processing cores; in some embodiments, the processor 301 may integrate an application processor and a modem processor, wherein the application processor mainly processes the operating system, user interface, and application programs, and the modem processor mainly processes wireless communications. It is understandable that the above-mentioned modem processor may not be integrated into the processor 301.

[0208] The memory 302 can be used to store software programs and modules. The processor 301 executes various functional applications and data processing by running the software programs and modules stored in the memory 302. The memory 302 may mainly include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application required for at least one function (such as a sound playback function, an image playback function, etc.), etc.; the data storage area may store data created according to the use of the terminal, etc. In addition, the memory 302 may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other volatile solid-state storage devices. Accordingly, the memory 302 may also include a memory controller to provide the processor 301 with access to the memory 302.

[0209] The terminal also includes a power supply 303 for supplying power to various components. In some embodiments, the power supply 303 can be logically connected to the processor 301 through a power management system, so as to manage charging, discharging, and power consumption through the power management system. The power supply 303 can also include any components such as one or more DC or AC power supplies, recharging systems, power failure detection circuits, power converters or inverters, and power status indicators.

[0210] The terminal may further include an input module 304, which may be used to receive input digital or character information and generate keyboard, mouse, joystick, optical or trackball signal inputs related to user settings and function controls.

[0211] The terminal may also include a communication module 305. In some embodiments, the communication module 305 may include a wireless module. The terminal may perform short-range wireless transmission through the wireless module of the communication module 305, thereby providing the user with wireless broadband Internet access. For example, the communication module 305 may be used to help the user send and receive emails, browse web pages, and access streaming media.

[0212] Although not shown, the terminal may further include a display unit, etc., which will not be described in detail herein. Specifically in this embodiment, the processor 301 in the terminal will load the executable files corresponding to the processes of one or more applications into the memory 302 according to the following instructions, and the processor 301 will run the applications stored in the memory 302, thereby realizing various functions, as follows:

[0213] Retrieving at least two original data files from a data file database, and obtaining confidentiality level labels corresponding to the original data files, wherein the confidentiality level label is used to represent a confidentiality level for the original data file selected from a preset confidentiality level set, and the preset confidentiality level set includes at least two different confidentiality levels;

[0214] Based on the confidentiality level label, desensitizing the original data file to obtain a processed original data file;

[0215] By presetting a confidentiality evaluation model, according to the processed original data file, a confidentiality evaluation result corresponding to the processed original data file is obtained;

[0216] Based on the confidentiality level label and confidentiality evaluation results, determine the desensitization deviation value corresponding to the original data file;

[0217] If no less than a first preset proportion of the desensitization deviation values ​​is less than a preset desensitization deviation threshold, the processed original data file is determined to be a target data file.

[0218] The specific implementation of the above operations can be found in the previous embodiments, which will not be described in detail here.

[0219] From the above, it can be seen that the embodiment of the present application can ensure that the data desensitization processing effect of the same batch of original data files reaches the expected level, so that files with different confidentiality levels are adapted to different data confidentiality effects, and by quantifying the errors of the desensitization processing, the data desensitization effect is continuously enhanced, thereby improving the overall confidentiality management level of the data files.

[0220] Those skilled in the art will appreciate that all or part of the steps in the various methods of the above embodiments may be completed by instructions, or by controlling related hardware through instructions. The instructions may be stored in a medium and loaded and executed by a processor.

[0221] To this end, an embodiment of the present application provides a medium in which a plurality of instructions are stored, and the instructions can be loaded by a processor to execute the steps in any one of the data processing methods based on data desensitization provided in the embodiment of the present application. For example, the instructions can execute the following steps:

[0222] Retrieving at least two original data files from a data file database, and obtaining confidentiality level labels corresponding to the original data files, wherein the confidentiality level label is used to represent a confidentiality level for the original data file selected from a preset confidentiality level set, and the preset confidentiality level set includes at least two different confidentiality levels;

[0223] Based on the confidentiality level label, desensitizing the original data file to obtain a processed original data file;

[0224] By presetting a confidentiality evaluation model, according to the processed original data file, a confidentiality evaluation result corresponding to the processed original data file is obtained;

[0225] Based on the confidentiality level label and confidentiality evaluation results, determine the desensitization deviation value corresponding to the original data file;

[0226] If no less than a first preset proportion of the desensitization deviation values ​​is less than a preset desensitization deviation threshold, the processed original data file is determined to be a target data file.

[0227] The medium may include: a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, etc.

[0228] According to one aspect of the present application, a computer program product or a computer program is provided, the computer program product or the computer program includes a computer instruction, the computer instruction is stored in a medium. A processor of a computer device reads the computer instruction from the medium, and the processor executes the computer instruction, so that the computer device executes the method provided in various optional implementations provided in the above embodiments.

[0229] Since the instructions stored in the medium can execute the steps of any data processing method based on data desensitization provided in the embodiments of the present application, the beneficial effects that can be achieved by any data processing method based on data desensitization provided in the embodiments of the present application can be achieved. Please refer to the previous embodiments for details and will not be repeated here.

[0230] The above is a detailed introduction to a data processing method, device, terminal and medium based on data desensitization provided in an embodiment of the present application. Specific examples are used in this article to illustrate the principles and implementation methods of the present application. The description of the above embodiments is only used to help understand the method of the present application and its core idea; at the same time, for technical personnel in this field, according to the idea of ​​the present application, there will be changes in the specific implementation method and application scope. In summary, the content of this specification should not be understood as a limitation on the present application.

Claims

1. A data processing method based on data desensitization, characterized in that: Applied to a data middle station, the data middle station includes a data file database and a server, and the data file database and the server are both in the same network; The method comprises: Retrieving at least two original data files from a data file database, and obtaining confidentiality level labels corresponding to the original data files, wherein the confidentiality level label is used to represent a confidentiality level for the original data file selected from a preset confidentiality level set, and the preset confidentiality level set includes at least two different confidentiality levels; Based on the confidentiality level label, desensitizing the original data file to obtain a processed original data file; By presetting a confidentiality evaluation model, according to the processed original data file, a confidentiality evaluation result corresponding to the processed original data file is obtained; Based on the confidentiality level label and confidentiality evaluation results, determine the desensitization deviation value corresponding to the original data file; If no less than a first preset proportion of the desensitization deviation values ​​is less than a preset desensitization deviation threshold, the processed original data file is determined to be a target data file.

2. The method according to claim 1, characterized in that The server is equipped with a preset desensitization processing model set, the preset desensitization processing model set includes at least two different preset desensitization processing models, and the preset desensitization processing models correspond to the confidentiality levels one by one; The desensitizing the original data file based on the confidentiality level label to obtain the processed original data file includes: Based on the confidentiality level label, selecting a preset desensitization processing model corresponding to the confidentiality level label from the preset desensitization processing model set; The processed original data file is obtained according to the original data file through the preset desensitization processing model corresponding to the confidentiality level label.

3. The method according to claim 2, characterized in that The preset desensitization processing model includes a first input layer, a first embedding layer, a first convolutional layer, a pooling layer, a first fully connected layer and a first output layer; The method of obtaining the processed original data file according to the original data file by using the preset desensitization processing model corresponding to the confidentiality level label includes: Obtaining a first sensitive information tensor according to the sensitive information in the original data file through the first input layer; Obtaining a first floating-point number tensor according to the first sensitive information tensor through the first embedding layer; Obtaining a second floating-point tensor according to the first floating-point tensor through the first convolutional layer; Obtaining a third floating-point tensor according to the second floating-point tensor through the pooling layer; Obtaining a fourth floating-point tensor according to the third floating-point tensor through the first fully connected layer; Obtaining a first output tensor according to the fourth floating-point number tensor through the first output layer, wherein the first output tensor is used to represent the sensitive information after desensitization processing; Based on the original data file and the first output tensor, the processed original data file is obtained.

4. The method according to claim 2, characterized in that The preset desensitization processing model includes a second input layer, a second embedding layer, a second convolutional layer, a recurrent neural network layer, an attention mechanism layer, a second fully connected layer and a second output layer; The method of obtaining the processed original data file according to the original data file by using the preset desensitization processing model corresponding to the confidentiality level label includes: Obtaining a second sensitive information tensor according to the sensitive information in the original data file through the second input layer; Obtaining a fifth floating-point number tensor according to the second sensitive information tensor through the second embedding layer; Obtaining a sixth floating-point tensor according to the fifth floating-point tensor through the second convolutional layer; Obtaining a seventh floating-point tensor according to the sixth floating-point tensor through the recurrent neural network layer; Obtaining an eighth floating-point tensor according to the seventh floating-point tensor through the attention mechanism layer; Obtaining a ninth floating-point tensor according to the eighth floating-point tensor through the second fully connected layer; Obtaining a second output tensor according to the ninth floating-point number tensor through the second output layer, wherein the second output tensor is used to represent the sensitive information after desensitization processing; Based on the original data file and the second output tensor, the processed original data file is obtained.

5. The method according to claim 1, characterized in that The preset confidentiality evaluation model includes a third input layer, a third embedding layer, a self-attention mechanism layer, a multi-layer perceptron layer and a third output layer; The method of obtaining a confidentiality evaluation result corresponding to the processed original data file by using a preset confidentiality evaluation model according to the processed original data file includes: Obtaining an input tensor according to the processed original data file through the third input layer; Obtaining a tenth floating-point number tensor according to the input tensor through the third embedding layer; Obtaining an eleventh floating-point number tensor according to the tenth floating-point number tensor through the self-attention mechanism layer; Obtaining a twelfth floating-point tensor according to the eleventh floating-point tensor through the multi-layer perceptron layer; A confidentiality evaluation score is obtained according to the twelfth floating-point number tensor through the third output layer, and the confidentiality evaluation score is determined to be the confidentiality evaluation result.

6. The method according to claim 5, characterized in that The method of determining the desensitization deviation value corresponding to the original data file based on the confidentiality level label and the confidentiality evaluation result includes: The confidentiality level corresponding to the confidentiality level label is converted into the confidentiality level value corresponding to the confidentiality level label by a preset linear mapping function; The absolute value of the difference between the confidentiality level value and the confidentiality evaluation score is calculated to obtain the desensitization deviation value.

7. The method according to claim 2, characterized in that After determining the desensitization deviation value corresponding to the original data file based on the confidentiality level label and the confidentiality evaluation result, the method further includes: If a proportion of the desensitization deviation values ​​that is not less than a second preset proportion is greater than a preset desensitization deviation threshold, the preset desensitization processing model is updated based on the desensitization deviation value to obtain an updated preset desensitization processing model, wherein the updated preset desensitization model is used to perform desensitization processing on the original data file.

8. A data processing device based on data desensitization, characterized in that: Applied to a data middle station, the data middle station includes a data file database and a server, and the data file database and the server are both in the same network; The device comprises: A first unit is used to retrieve at least two original data files from a data file database and obtain confidentiality level labels corresponding to the original data files, wherein the confidentiality level label is used to represent a confidentiality level for the original data file selected from a preset confidentiality level set, and the preset confidentiality level set includes at least two different confidentiality levels; The second unit is used to perform desensitization processing on the original data file based on the confidentiality level label to obtain a processed original data file; The third unit is used to obtain a confidentiality evaluation result corresponding to the processed original data file according to the processed original data file by using a preset confidentiality evaluation model; The fourth unit is used to determine the desensitization deviation value corresponding to the original data file based on the confidentiality level label and the confidentiality evaluation result; The fifth unit is used to determine that the processed original data file is a target data file if no less than a first preset proportion of the desensitization deviation values ​​is less than a preset desensitization deviation threshold.

9. A terminal, characterized in that: The method comprises a processor and a memory, wherein the memory stores a plurality of instructions; the processor loads instructions from the memory to execute the steps in the method according to any one of claims 1 to 7.

10. A medium, characterized in that The medium stores a plurality of instructions, and the instructions are suitable for being loaded by a processor to execute the steps in the method according to any one of claims 1 to 7.

Images