Sensitive data processing method and device, electronic equipment, storage medium and product

Abstract

The invention discloses a sensitive data processing method and device, electronic equipment, a storage medium and a product. The sensitive data processing method comprises the following steps: acquiring data to be detected, wherein the data to be detected comprises digital data and / or text data; for digital type data, digital type sensitive data are determined according to the conversion probability matrix of the sensitive type attributes, and elements in the ith row and the jth column in the conversion probability matrix represent the probability of transferring from the sensitive type attribute i to the sensitive type attribute j; for the text type data, text type sensitive data is determined based on a preset model, the preset model is constructed based on a semantic representation model and a dictionary tree, and the semantic representation model is used for determining word vectors of segmented words in the text type data. According to the technical scheme, the sensitive data is recognized by adopting the transition probability matrix for the digital data, the sensitive data is recognized by adopting the model constructed by the semantic representation model and the dictionary tree for the text data, and the sensitive data is accurately and efficiently recognized for different types of data.

Description

Technical Field

[0001] The embodiments of the present application relate to the field of data processing technology, and in particular, to a sensitive data processing method, device, electronic device, storage medium and product. Background Art

[0002] Power sensitive information protection refers to the process of protecting sensitive data in the power system to prevent unauthorized access or leakage. As the digitalization and intelligence of the power system continue to increase, the sensitive information contained in it is becoming more and more important, such as the power supply network topology, power load data, system operation status, etc. The leakage or tampering of such information may lead to serious consequences, such as power system paralysis, information security threats, etc., so it is particularly important to protect this information.

[0003] At present, the protection of sensitive information in the power system faces many challenges. First, the complexity and large scale of the power system make it difficult for traditional security measures to meet the needs. For example, cryptographic-based encryption algorithms have excessive computing and storage overhead when protecting large-scale data, and the same algorithm also has differences in the recognition effect of different types of sensitive data. Secondly, the real-time nature of the power system requires that the protection of sensitive information must have high efficiency and response speed to deal with emergencies or malicious attacks. The existing sensitive data processing process cannot guarantee the accurate and efficient identification of different types of sensitive data. Summary of the invention

[0004] The present application provides a sensitive data processing method, device, electronic device, storage medium and product to accurately and efficiently identify sensitive data.

[0005] In a first aspect, an embodiment of the present application provides a sensitive data processing method, including:

[0006] Acquire data to be detected, wherein the data to be detected includes digital data and / or text data;

[0007] For the digital data, the digital sensitive data is determined according to a conversion probability matrix of the sensitive attribute, wherein the element in the i-th row and the j-th column of the conversion probability matrix represents the probability of transferring from the sensitive attribute i to the sensitive attribute j;

[0008] For the text data, the text sensitive data is determined based on a preset model, wherein the preset model is constructed based on a semantic representation model and a dictionary tree, the semantic representation model is used to determine the word vectors of the word segments in the text data, and the dictionary tree is used to store sensitive words.

[0009] In a second aspect, an embodiment of the present application further provides a sensitive data processing device, including:

[0010] An acquisition module, used for acquiring data to be detected, wherein the data to be detected includes digital data and / or text data;

[0011] A first processing module is used to determine the digital sensitive data according to the conversion probability matrix of the sensitive attribute for the digital data, wherein the element in the i-th row and the j-th column in the conversion probability matrix represents the probability of transferring from the sensitive attribute i to the sensitive attribute j;

[0012] The second processing module is used to determine text-type sensitive data for the text-type data based on a preset model, wherein the preset model is constructed based on a semantic representation model and a dictionary tree, the semantic representation model is used to determine the word vectors of the word segments in the text-type data, and the dictionary tree is used to store sensitive words.

[0013] In a third aspect, an embodiment of the present application provides an electronic device, including:

[0014] one or more processors;

[0015] A storage device for storing one or more programs;

[0016] When the one or more programs are executed by the one or more processors, the one or more processors implement the sensitive data processing method as described in the first aspect.

[0017] In a fourth aspect, an embodiment of the present application further provides a computer-readable storage medium on which a computer program is stored, and when the program is executed by a processor, the sensitive data processing method as described in the first aspect is implemented.

[0018] In a fifth aspect, an embodiment of the present application further provides a computer program product, comprising a computer program and / or instructions, which, when executed by a processor, implements a sensitive data processing method as described in any of the above embodiments.

[0019] The embodiments of the present application provide a sensitive data processing method, device, electronic device, storage medium and product, the sensitive data processing method comprising: obtaining data to be detected, wherein the data to be detected includes digital data and / or text data; for the digital data, determining the digital sensitive data according to the conversion probability matrix of the sensitive attribute, wherein the element in the i-th row and the j-th column of the conversion probability matrix represents the probability of transferring from sensitive attribute i to sensitive attribute j; for the text data, determining the text sensitive data based on a preset model, wherein the preset model is constructed based on a semantic representation model and a dictionary tree, the semantic representation model is used to determine the word vector of the word segmentation in the text data, and the dictionary tree is used to store sensitive words. The above technical solution uses a conversion probability matrix to identify sensitive data for digital data, and uses a semantic representation model and a model constructed by a dictionary tree to identify sensitive data for text data, thereby realizing accurate and efficient identification of sensitive data for different types of data. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] The above and other features, advantages and aspects of the embodiments of the present disclosure will become more apparent with reference to the following detailed description in conjunction with the accompanying drawings. Throughout the accompanying drawings, the same or similar reference numerals represent the same or similar elements. It should be understood that the drawings are schematic and the originals and elements are not necessarily drawn to scale.

[0021] Figure 1 A flowchart of a sensitive data processing method provided in an embodiment of the present application;

[0022] Figure 2 A flowchart of another sensitive data processing method provided in an embodiment of the present application;

[0023] Figure 3 A schematic diagram of data desensitization provided in an embodiment of the present application;

[0024] Figure 4 A schematic diagram of a sensitive data sharing process provided in an embodiment of the present application;

[0025] Figure 5 A schematic diagram of a blockchain identity review data retrieval process provided in an embodiment of the present application;

[0026] Figure 6 A schematic diagram of a sensitive data processing process provided in an embodiment of the present application;

[0027] Figure 7 A schematic diagram of the structure of a sensitive data processing device provided in an embodiment of the present application;

[0028] Figure 8 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0029] The present application will be further described in detail below in conjunction with the accompanying drawings and embodiments. It is to be understood that the specific embodiments described herein are only used to explain the present application, rather than to limit the present application. It should also be noted that, for ease of description, only the parts related to the present application, rather than all structures, are shown in the accompanying drawings.

[0030] It should be mentioned before discussing the exemplary embodiments in more detail that some exemplary embodiments are described as processes or methods depicted as flow charts. Although the flow charts describe the steps as sequential processes, many of the steps therein can be implemented in parallel, concurrently or simultaneously. In addition, the order of the steps can be rearranged. The process can be terminated when its operation is completed, but can also have additional steps not included in the accompanying drawings. The process can correspond to a method, function, procedure, subroutine, subprogram, etc.

[0031] It should be noted that the concepts of "first", "second", etc. mentioned in the embodiments of the present application are only used to distinguish different devices, modules, units or other objects, and are not used to limit the order or interdependence of the functions performed by these devices, modules, units or other objects.

[0032] In addition, the embodiments in the present application and the features in the embodiments may be combined with each other if there is no conflict.

[0033] The acquisition, storage, use, and processing of data in the technical solution of this application comply with the relevant provisions of national laws and regulations.

[0034] It should be noted that in the embodiments of the present application, certain software, components, models and other existing solutions in the industry may be mentioned, and they should be regarded as exemplary. Their purpose is only to illustrate the feasibility of implementing the technical solution of the present application, but it does not mean that the applicant has or will necessarily use the relevant content of the solution.

[0035] The purpose of the present invention is to solve the problems of secure transmission and permission sharing in the process of sharing sensitive information in practical application industries. In order to improve the existing security issues in the access of sensitive information and the imperfect tracing of heterogeneous data classification and post-data sharing usage logs, a detection technology for customized identification and early warning of sensitive data is proposed. Through innovative models and methods, this technology realizes the automated classification, detection, desensitization of sensitive data, and blockchain smart contract control access and log tracking, aiming to ensure the security and compliance of data.

[0036] Figure 1A flowchart of a sensitive data processing method provided in an embodiment of the present application, which can be applied to the case of identifying sensitive data. Specifically, the sensitive data processing method can be executed by a sensitive data processing device, which can be implemented by software and / or hardware and integrated in an electronic device. Electronic devices include but are not limited to computers, smart phones, servers, and other devices with data processing functions.

[0037] like Figure 1 As shown, the method specifically comprises the following steps:

[0038] S110, obtaining data to be detected, wherein the data to be detected includes digital data and / or text data.

[0039] Specifically, the data to be detected refers to data that may contain sensitive data, specifically power data. The data to be detected may include digital data or text data. Digital data mainly refers to data that represents quantity or can perform numerical operations, such as integers, decimals and / or fractions; text data mainly refers to data of any character type used to store text and numbers, such as "Hello World", "1234", etc.

[0040] S120. For the digital data, determine the digital sensitive data according to a conversion probability matrix of sensitive attributes, wherein the element in the i-th row and j-th column of the conversion probability matrix represents the probability of transferring from sensitive attribute i to sensitive attribute j.

[0041] Specifically, for digital data, the transition probability matrix, that is, the transition probability between different sensitive class attributes, can be used to identify digital sensitive data.

[0042] S130. For the text data, determine the text sensitive data based on a preset model, wherein the preset model is constructed based on a semantic representation model and a dictionary tree, the semantic representation model is used to determine the word vectors of the word segments in the text data, and the dictionary tree is used to store sensitive words.

[0043] Specifically, for text data, that is, the part involving text data in sensitive fields, text-type sensitive data can be identified based on a preset model (A-ELMo model). This preset model mainly introduces the attention mechanism (Attention Mechanism) based on the semantic representation model (Embeddings from Language Model, ELMo), and combines it with the sensitivity trigger threshold to achieve accurate and efficient identification of sensitive data.

[0044] Figure 2A flowchart of a sensitive data processing method provided in an embodiment of the present application. Figure 2 As shown, the method specifically comprises the following steps:

[0045] S210, obtaining data to be detected. The data to be detected includes digital data and / or text data.

[0046] S220: Determine the support count corresponding to each attribute in the digital data, wherein the support count is the number of times the data of the corresponding attribute appears.

[0047] Specifically, the data to be tested is preprocessed to establish a data set containing various attribute information. For digital data, the support count of each attribute is calculated. The initial data set is set to A, and its expression is {A 1 ,A 2 ,...,A k}, there are k data in total, the record sample data set is T, T is the attribute data set X. Let y i Represents the value of the data at attribute i. The value of each attribute in the data set T is represented by {y 1 ,y 2 ,...,y n}, count the support of all (n) attribute values.

[0048] The support count refers to the number of times the data of an attribute appears in a data set. By calculating the support count, we can determine which attribute data is more common in the data set, and thus identify which attribute values ​​may be sensitive information. For example, if the support count of some attribute values ​​is particularly high or low, it may mean that these values ​​have special significance in the data set, and thus may be defined as sensitive information. On this basis, the support count can be used to calculate the transition probability (i.e., the transition probability matrix) between different sensitive attributes.

[0049] S230: Determine a probability matrix of the digital data for each sensitive attribute according to the support count and the conversion probability matrix corresponding to each attribute.

[0050] S240. Determine the probability distribution of the digital sensitive data according to the probability matrix, and determine the digital sensitive data according to the probability distribution.

[0051] In the data sensitivity classification and determination algorithm, in order to ensure the subsequent effective desensitization, it is first necessary.

[0052] Specifically, a conversion probability matrix for sensitive attributes is established: Assume that the data set A contains n different types of attributes, where the sensitive attribute is represented by a and its value is a g (1≤g≤n).A It is represented as an n×m matrix as follows: Among them, the element λ in the i-th row and j-th column ij Represents the sensitive attribute a i Transfer to sensitive attribute a j The word probability matrix reflects the transfer between sensitive class attribute information in the data set.

[0053] According to different attributes, the data can be split into two information intervals, namely (a 1 ,a g ) and (a g+1 ,a n ), where if there are two different values ​​for an attribute, a summary calculation can be performed. Assume that the sensitive class attribute matrix of data set A is: Among them, rk (1≤k≤n) represents the state distribution probability of the attribute value ak in the initial information data. Based on this, the probability matrix of each type of data in the data set A is calculated in turn.

[0054] S250, segment the data to be detected into multiple sentences, segment each sentence into words, and obtain an input character sequence corresponding to the sentence.

[0055] Divide all the long texts of the data to be detected into short sentences (input character sequences). For example, the jieba function can be used to segment each sentence, and each segmentation unit is an input character sequence.

[0056] S260. For each sentence, determine the word vector of the input character sequence based on the semantic representation model.

[0057] Specifically, word embedding is the result of mapping words or phrases in natural language (i.e. the input character sequence mentioned above) to a high-dimensional real vector space. Word embedding not only captures the semantic information of the character sequence, but also enables computers to understand and process natural language. By representing the character sequence as a vector, various vector space operations can be performed. The generation of word embeddings depends on machine learning models, which can be unsupervised (such as Word2Vec or GloVe) or supervised (such as word embeddings learned in specific task training).

[0058] Optionally, a bidirectional long short-term memory network (LSTM) is set in the semantic representation model; a word vector of the input character sequence is determined based on the semantic representation model, including:

[0059] S2610, outputting a context-related first word vector generated by the forward language model and a second word vector generated by the backward language model at each position of each LSTM;

[0060] S2620, outputting a probability value according to the first word vector and the second word vector, wherein the probability value is used to predict a corresponding input character;

[0061] S2630. Superimpose all layers of the input character according to the attention weight to obtain a word vector corresponding to the input character.

[0062] In this embodiment, ELMo is used to replace traditional models such as Word2Vec or GloVe, and dynamic word vectors are used to represent polysemous words, thereby improving the processing effect of polysemous words. First, a bidirectional LSTM is set in ELMo, and the characteristics of the forward language model and the backward language model are combined to generate word embedding x using the static word vector model Word2Vec. k ; Then, at each LSTM layer, at each position k, a context-dependent first word vector (also called an intermediate word vector) generated by the forward language model is output. And the second word vector generated by the backward language model L represents the number of layers of the language model; finally, the intermediate word vector and The Softmax layer outputs the probability value, which is used to predict the input character t k+1 .

[0063] The forward language model is given a text sequence of length N (t 1 ,t 2 ,...,t N ) to perform probability modeling and calculate the current token t k The probability of is calculated as follows:

[0064] The backward language model is for the subsequent sequence (tk +1 ,tk +2 ,...,t N ) to perform probability modeling and calculate the current token t k Probability:

[0065] The forward and backward language models are modeled by maximizing the joint likelihood function, which is formulated as follows:

[0066]

[0067] Among them, Θ x is the parameter of the token representation layer, Θ sis the parameter of the Softmax layer, Θ LSTM are the parameters of the LSTM layer.

[0068] For each token t k , the bidirectional language model is composed of 2L+1 word vectors composed of forward and backward intermediate vectors and input tokens into L+1 representations:

[0069] where h k,0 =x k and

[0070] In addition, an attention mechanism is introduced on top of the ELMo language model. In order to meet the needs of downstream classification tasks, all layers in R are stacked into a single vector and V is used k Represents this dynamic word vector. Considering that the information learned at different levels may have different importance, the attention mechanism is introduced to assign different weights to words. The weights of the task-specific bidirectional language model layer are calculated as follows: Among them, s j is the maximum normalized weight, indicating how much attention should be placed on each layer. By introducing a global scaling factor γ, the task model is allowed to scale the entire V vector to further adjust the performance of the model.

[0071] The attention mechanism adjusts the weight of each word in the ELMo vector according to its sensitivity, so that the more sensitive the word, the higher its weight, so as to better capture sensitive information. The fully connected neural network layer and Softmax function are used to convert the ELMo vector into a classification result, thereby completing the detection and classification of sensitive information.

[0072] S270. Match the word vector of the input character sequence with the dictionary tree character by character to determine the sensitive word segmentation in the sentence.

[0073] Specifically, a dictionary tree matching method is introduced on top of the ELMo language model, and the word segmentation results (input character sequence) are input into the dictionary tree. In the dictionary tree, all possible sensitive words are pre-stored. The word segmentation results and the sensitive words in the dictionary tree are compared one by one, that is, in the dictionary tree, starting from the root node, the input character sequence is matched downward character by character. If a matching sensitive word is found, the input character sequence is marked. On this basis, the number of words that need to be analyzed using A-ELMo is effectively reduced, thereby improving detection efficiency.

[0074] In one embodiment, the method further comprises:

[0075] S140, for any digital or text sensitive data, query the real result of the sensitive data on the sensitive database;

[0076] S150. Based on the deep convolutional generative adversarial network, Laplace noise is added to the real result to replace the sensitive data and obtain the desensitized data.

[0077] In this embodiment, in order to desensitize the identified sensitive data and prevent data information leakage, a deep convolutional generative adversarial network (DCGAN) data desensitization model based on the characteristics of power system data is established. By combining the interval division of key information, sensitive information and normal information are reconstructed in the corresponding interval, and the probability after reconstruction is calculated to assist the subsequent reconstruction conversion. The data is desensitized.

[0078] Specifically, in order to better share data, differential privacy methods can be used to desensitize the data. The local differential privacy Laplace mechanism is used to protect data privacy. By adding Laplace noise to the calculation results, user behavior pattern information is further protected. The data desensitization process is mainly as follows:

[0079] (1) Calculate the true result of the query function f on the sensitive database D, denoted as f(D).

[0080] (2) To protect privacy, Laplace noise is added to the true result f(D). The size of the Laplace noise is determined by the parameter b and the sensitivity S of the query function f. f Decide.

[0081] (3) The query result finally published is f(D)+Lap(b / S f ), where Lap(b / S f ) represents the Laplace distribution Lap(0,b / S f ) to generate a random number.

[0082] The probability density function of the Laplace distribution is: Where b>0 is the scale parameter and x is a random variable.

[0083] Figure 3 A schematic diagram of data desensitization provided for an embodiment of the present application. In this embodiment, based on the deep convolutional generative adversarial network, differential privacy is combined to perform data desensitization protection. Random noise is added to the gradient passed back to the generator from the discriminator. By adding noise, the contribution of the generator to the real data can be reduced when updating parameters, thereby improving the privacy protection of the generated data.

[0084] like Figure 3 As shown in the figure, the data desensitization process is as follows:

[0085] First, initialize the required parameters and set the discriminator learning rate to α d , the generator learning rate is set to α g , parameter clipping constant b p , batch size m, total number of training data points at each discriminator iteration M, number of discriminator iterations at each generator iteration n d , generator iteration number n g , the noise scale σ n , on this basis:

[0086] (1) Initialize the system’s discriminator parameters, k 0 , the generator parameter is β 0

[0087] (2) For each h 1 ∈n g 、h 2 ∈n d , each batch of prior samples and real data points exist separately For each sample i, we have:

[0088]

[0089] (3)

[0090] (4)

[0091] (5)

[0092] (6) For another batch of prior samples

[0093] (7)

[0094] (8) Using the chain rule, the desensitization range is accurately narrowed, and the gradient returned to the generator is the part transmitted by the discriminator and the Jacobian matrix J of the generator itself. β It consists of two parts, and the specific formula is as follows:

[0095]

[0096] (9)

[0097] Among them, Generative Adversarial Network (GAN) is a powerful machine learning framework that generates samples that approximate the real data distribution by training two models at the same time. These two models are the generative model G and the discriminative model D. The task of the generative model G is to convert the input random noise vector into an output that approximates the real data distribution. The goal of G is to learn the mapping function of the real data distribution so that new samples similar to the real data can be generated. The input random noise vector is usually drawn from a known probability distribution, which is denoted by p. z (z); The task of the discriminant model D is to judge and output whether the sample comes from real data, and then output a probability, indicating the probability that the sample comes from real data. The training process of GANs is a game of mutual competition, in which G and D compete with each other. G tries to generate samples that are as real as possible to deceive D, while D tries to identify real samples and generated samples as accurately as possible. This process can be achieved by minimizing the following objective function V(G,D):

[0098]

[0099] On the basis of the adversarial network, Gaussian noise is used to desensitize the data. For the data generated by the generator, the corresponding data set in the parameter β calculation process is desensitized. The desensitization calculation process is as follows:

[0100] M (σ,c) (g)=clip(g,C)+N(0,σ 2 ,C 2 I) (12)

[0101]

[0102] Among them, g β represents the gradient of the generator.

[0103] On the basis of the above embodiments, considering that the openness and interconnectedness of the power system make it vulnerable to external attacks or internal threats, the present application also adopts multi-level and multi-faceted security measures to deal with different types of risks.

[0104] In one embodiment, the method further comprises:

[0105] S160, storing the desensitized data in the Inter Planetary File System (IPFS);

[0106] S170: upon receiving a data sharing request, verifying the validity of a decryption key for the sensitive data to be shared, and if the sensitive data to be shared is successfully decrypted, sending the sensitive data to be shared to a sharing object;

[0107] S180. The sharing information of the desensitized data and the corresponding digital signature information are packaged and uploaded to the chain after consensus through the smart contract service.

[0108] Specifically, to address the problem of data sharing trace tracing, a blockchain-based sensitive data security monitoring method was proposed. The method is implemented based on multiple components, including alliance blockchain, cryptographic operation unit, private IPFS cluster, system services, alliance members, application systems and users.

[0109] IPFS is a content-addressed peer-to-peer distributed file system that uses IPFS private clusters to provide data storage services and off-chain storage for desensitized data. IPFS uses distributed hash tables (DHT) and Merkle DAG (directed acyclic graph) technology to achieve efficient and reliable distributed file storage.

[0110] The consortium blockchain is composed of physical nodes and basic platform software, providing blockchain basic services and smart contract services to the outside world. These services include node authentication, data consensus, data on-chain, and query functions. Node authentication is usually implemented using Public Key Infrastructure (PKI), and data consensus can use consensus algorithms such as Byzantine Fault Tolerance (BFT) or Proof of Stake (PoS). Data on-chain and query are implemented through smart contracts and the blockchain's Application Programming Interface (API).

[0111] The cryptographic operation unit works in collaboration with blockchain nodes to provide cryptographic services such as signing and signature verification for the nodes, and provides cryptographic services such as symmetric encryption and decryption, asymmetric encryption and decryption, and hash operations for the application system. It can be implemented in hardware or software according to the application scenario and computing speed requirements.

[0112] Figure 4 A schematic diagram of a sensitive data sharing process provided in an embodiment of the present application. Figure 4 As shown, during the sharing process, users can view all sensitive data information that has been added through on-chain data, and select data to share or request according to their needs.

[0113] In the case of actively sharing data, the data owner needs to identify the sensitive data to be shared and provide the encrypted sensitive data object address ciphertext Sendata add Decryption key key user ; Perform asymmetric decryption operations through cryptographic services to verify the key user The validity of add Once decryption is successful, the data owner can share the sensitive data with any legitimate user; Decrypt Sendata add Upon success, sensitive information is obtained and stored in the blockchain after consensus.

[0114] Optionally, in the process of sharing sensitive data, different people are allowed to share data with different sensitivity thresholds. Sensitive information with low thresholds is open to ordinary managers and users, and sensitive data with high thresholds is open to higher-level managers. The ciphertext of the sensitive data object address held by managers at different levels is Sendata. add Decryption key key user The threshold range of sensitive data that can be obtained is also different.

[0115] The smart contract service will package the above information with the corresponding digital signature information and upload it to the chain after consensus. The data owner will know the user's application based on the information on the chain, and then authorize the sensitive data to the applicant based on the identity registration.

[0116] The system service provides interface calls for the application system in the form of RESTful API to realize functions such as secure storage, verification, sharing, and flow traceability of sensitive data. Legally registered users can access corresponding functions and services through the application system. These components together build a safe and reliable sensitive data sharing platform, which aims to protect the security and privacy of sensitive data while realizing the effective sharing and utilization of data.

[0117] Figure 5 A schematic diagram of a blockchain identity verification data retrieval process provided in an embodiment of the present application. Figure 5 As shown in the figure, after the normal data is desensitized and submitted to the user, a real-time tracking system is established, and the shared data is continuously tracked using the real-time monitoring system, so that the flow of sensitive data can be traced, and real-time monitoring and early warning of highly sensitive data outbound transmission can be achieved, thereby improving data security. A data access log recording mechanism is set up to achieve detailed tracing of sensitive data access history, provide full support for security audits, and help organizations meet compliance requirements.

[0118] In one embodiment, the method further comprises:

[0119] S190: Match the number of fields of the digital sensitive data and the text sensitive data with a preset threshold, and determine the approval level of the data to be detected according to the matched preset threshold.

[0120] In this embodiment, an effective data sensitivity assessment model is established, which classifies the sensitivity level according to the characteristics of the data and sets corresponding weights and thresholds for different types of sensitive fields. The number of sensitive fields is automatically detected by the algorithm and associated with the set threshold. When the number of matched data exceeds a specific threshold, the approval process of the corresponding level will be triggered, thereby realizing the hierarchical management of sensitive data to more effectively control data security.

[0121] Figure 6 A schematic diagram of a sensitive data processing process provided in an embodiment of the present application. Figure 6 As shown, the sensitivity level of data is classified according to the characteristics of the data, and an effective model for evaluating data sensitivity is established. Different weights are set for different types of sensitive fields, and different thresholds are set according to the weights. After the number of sensitive fields (i.e. sensitive data) is detected by the algorithm, it is automatically associated with the corresponding threshold. When the number of matched sensitive data is higher than the threshold, the corresponding approver level corresponding to the threshold is triggered for approval. When it is lower than the threshold, the ordinary approver is triggered for approval to complete the hierarchical classification. On this basis, by establishing an effective data sensitivity assessment model, setting weights for different types of sensitive fields and determining thresholds, the automatic classification and hierarchical approval of sensitive data are realized, thereby effectively controlling data security.

[0122] An exemplary sensitivity level of sensitive data is shown in Table 1.

[0123] Table 1 Information table of sensitivity levels

[0124]

[0125] The sensitive data processing method of the embodiment of the present application designs a sensitive information detection method based on the A-ELMo model for text data in sensitive fields. By introducing the attention mechanism and combining the sensitivity trigger threshold, accurate identification of different sensitive information is achieved. In addition, a data desensitization model based on the characteristics of power system data is proposed. Through interval division and probability calculation, effective desensitization of identified sensitive data is achieved to prevent information leakage. In addition, a real-time tracking system and a data access log recording mechanism are established. The shared data is continuously tracked by the real-time monitoring system, which realizes real-time monitoring and early warning of sensitive data, improves data security, and provides support for security audits. Finally, according to the above steps, a power data desensitization security protection system is constructed. Through systematic data retrieval, detection, and tracking processes, the process control process of data is improved, and data security and compliance are further improved. In summary, the comprehensive and effective detection technology for customized identification and early warning of sensitive data provided by the embodiment of the present application can be widely used in the field of data security, and the scientificity and security of data management are improved. On this basis, a complete power data desensitization security protection system can be constructed, and the process control process of data can be improved by designing a system system for retrieval, detection, and tracking of the entire process of data from the database.

[0126] Figure 7 A schematic diagram of the structure of a sensitive data processing device provided in an embodiment of the present application. The acquisition module in this embodiment is used to acquire data to be detected, wherein the data to be detected includes digital data and / or text data;

[0127] The first processing module 310 is used to determine the digital sensitive data according to the conversion probability matrix of the sensitive attribute for the digital data, wherein the element in the i-th row and the j-th column in the conversion probability matrix represents the probability of transferring from the sensitive attribute i to the sensitive attribute j;

[0128] The second processing module 320 is used to determine text-type sensitive data for the text-type data based on a preset model, wherein the preset model is constructed based on a semantic representation model and a dictionary tree, the semantic representation model is used to determine the word vectors of the word segments in the text-type data, and the dictionary tree is used to store sensitive words.

[0129] In one embodiment, the first processing module 310 includes:

[0130] A counting unit, used to determine the support count corresponding to each attribute in the digital data, wherein the support count is the number of times the data of the corresponding attribute appears;

[0131] A matrix determination unit, used to determine the probability matrix of the digital data for each sensitive class attribute according to the support count and the conversion probability matrix corresponding to each attribute;

[0132] A data determination unit is used to determine the probability distribution of digital sensitive data according to the probability matrix, and determine the digital sensitive data according to the probability distribution.

[0133] In one embodiment, the second processing module 320 includes:

[0134] A sentence segmentation unit, used for segmenting the data to be detected into multiple sentences, and performing word segmentation on each sentence to obtain an input character sequence corresponding to the sentence;

[0135] A word vector determination unit, configured to determine, for each of the sentences, a word vector of the input character sequence based on a semantic representation model;

[0136] A matching unit is used to match the word vector of the input character sequence with the dictionary tree character by character to determine the sensitive word segmentation in the sentence.

[0137] In one embodiment, a bidirectional LSTM is set in the semantic representation model;

[0138] Word vector determination unit, used for:

[0139] Output a context-dependent first word vector generated by the forward language model and a second word vector generated by the backward language model at each position of each LSTM;

[0140] Output a probability value according to the first word vector and the second word vector, wherein the probability value is used to predict a corresponding input character;

[0141] All layers of the input character are superimposed according to the attention weight to obtain the word vector corresponding to the input character.

[0142] In one embodiment, the device further comprises:

[0143] A query module, used to query the actual result of any digital or text sensitive data on the sensitive database;

[0144] The desensitization module is used to add Laplace noise to the true result based on a deep convolutional generative adversarial network to replace the sensitive data and obtain desensitized data.

[0145] In one embodiment, the device further comprises:

[0146] A storage module, used for storing the desensitized data in the InterPlanetary File System IPFS;

[0147] The sharing module is used to verify the validity of the decryption key for the sensitive data to be shared when receiving a data sharing request, and if the sensitive data to be shared is decrypted successfully, send the sensitive data to be shared to the sharing object;

[0148] The on-chain module is used to package the sharing information of the desensitized data and the corresponding digital signature information through the smart contract service, and upload them to the chain after consensus.

[0149] In one embodiment, the device further comprises:

[0150] The approval module is used to match the number of fields of the digital sensitive data and the text sensitive data with a preset threshold, and determine the approval level of the data to be detected according to the matched preset threshold.

[0151] The sensitive data processing device provided in the embodiments of the present application can be used to execute the sensitive data processing method provided in any of the above embodiments, and has corresponding functions and beneficial effects.

[0152] Figure 8 A schematic diagram of an electronic device 10 that can be used to implement an embodiment of the present application is shown. The electronic device 10 is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device 10 can also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, user equipment, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present application described and / or required herein.

[0153] like Figure 8 As shown, the electronic device 10 includes at least one processor 11, and a memory connected to the at least one processor 11, such as a read-only memory (ROM) 12, a random access memory (RAM) 13, etc., wherein the memory stores a computer program that can be executed by at least one processor, and the processor 11 can perform various appropriate actions and processes according to the computer program stored in the read-only memory (ROM) 12 or the computer program loaded from the storage unit 18 to the random access memory (RAM) 13. In the RAM 13, various programs and data required for the operation of the electronic device 10 can also be stored. The processor 11, the ROM 12, and the RAM 13 are connected to each other through a bus 14. An input / output (I / O) interface 15 is also connected to the bus 14.

[0154] Multiple components in the electronic device 10 are connected to the I / O interface 15, including: an input unit 16, such as a keyboard, a mouse, etc.; an output unit 17, such as various types of displays, speakers, etc.; a storage unit 18, such as a disk, an optical disk, etc.; and a communication unit 19, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 19 allows the electronic device 10 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks, wireless networks.

[0155] The processor 11 may be a variety of general and / or special processing components with processing and computing capabilities. Some examples of the processor 11 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any appropriate processor, controller, microcontroller, etc. The processor 11 performs the various methods and processes described above.

[0156] In some embodiments, the method of the above embodiment may be implemented as a computer program, which is tangibly contained in a computer-readable storage medium, such as a storage unit 18. In some embodiments, part or all of the computer program may be loaded and / or installed on the electronic device 10 via the ROM 12 and / or the communication unit 19. When the computer program is loaded into the RAM 13 and executed by the processor 11, one or more steps of the method described above may be performed. Alternatively, in other embodiments, the processor 11 may be configured to perform any of the above embodiment methods in any other appropriate manner (e.g., by means of firmware).

[0157] Various implementations of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), load programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various implementations can include: being implemented in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0158] The computer programs for implementing the methods of the present application may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, so that when the computer programs are executed by the processor, the functions / operations specified in the flow charts and / or block diagrams are implemented. The computer programs may be executed entirely on the machine, partially on the machine, partially on the machine and partially on a remote machine as a stand-alone software package, or entirely on a remote machine or server.

[0159] In the context of the present application, a computer readable storage medium may be a tangible medium that may contain or store a computer program for use by or in conjunction with an instruction execution system, device or equipment. A computer readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices or equipment, or any suitable combination of the foregoing. Alternatively, a computer readable storage medium may be a machine readable signal medium. A more specific example of a machine readable storage medium may include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0160] To provide interaction with a user, the systems and techniques described herein may be implemented on an electronic device 10 having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user can provide input to the electronic device 10. Other types of devices may also be used to provide interaction with the user; for example, the feedback provided to the user may be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user may be received in any form (including acoustic input, voice input, or tactile input).

[0161] The systems and techniques described herein may be implemented in a computing system that includes backend components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes frontend components (e.g., a user computer with a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such backend components, middleware components, or frontend components. The components of the system may be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: a local area network (LAN), a wide area network (WAN), a blockchain network, and the Internet.

[0162] A computing system may include a client and a server. The client and the server are generally remote from each other and usually interact through a communication network. The client and server relationship is generated by computer programs running on the corresponding computers and having a client-server relationship with each other. The server may be a cloud server, also known as a cloud computing server or cloud host, which is a host product in the cloud computing service system to solve the defects of difficult management and weak business scalability in traditional physical hosts and VPS services.

[0163] An embodiment of the present application also provides a computer program product, including a computer program and / or instructions, which, when executed by a processor, implements the sensitive data processing method described in any of the above embodiments.

[0164] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps recorded in this application can be executed in parallel, sequentially or in different orders, as long as the expected results of the technical solution of this application can be achieved, and this document is not limited here.

[0165] The above specific implementations do not constitute a limitation on the protection scope of this application. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions and improvements made within the spirit and principles of this application should be included in the protection scope of this application.

Claims

1. A method for processing sensitive data, It is characterized in that include: Acquire data to be detected, wherein the data to be detected includes digital data and / or text data; For the digital data, the digital sensitive data is determined according to a conversion probability matrix of the sensitive attribute, wherein the element in the i-th row and the j-th column of the conversion probability matrix represents the probability of transferring from the sensitive attribute i to the sensitive attribute j; For the text data, the text sensitive data is determined based on a preset model, wherein the preset model is constructed based on a semantic representation model and a dictionary tree, the semantic representation model is used to determine the word vectors of the word segments in the text data, and the dictionary tree is used to store sensitive words.

2. The method according to claim 1, It is characterized in that For the digital data, determining the digital sensitive data according to the conversion probability matrix of the sensitive class attribute includes: Determine a support count corresponding to each attribute in the digital data, wherein the support count is the number of times the data of the corresponding attribute appears; Determine the probability matrix of the digital data for each sensitive attribute according to the support count and conversion probability matrix corresponding to each attribute; The probability distribution of the digital sensitive data is determined according to the probability matrix, and the digital sensitive data is determined according to the probability distribution.

3. The method according to claim 1, It is characterized in that Determine text-based sensitive data based on preset models, including: Segment the data to be detected to obtain a plurality of sentences, segment each sentence to obtain an input character sequence corresponding to the sentence; For each of the sentences, determining a word vector of the input character sequence based on a semantic representation model; The word vector of the input character sequence is matched with the dictionary tree character by character to determine the sensitive word segmentation in the sentence.

4. The method according to claim 3, It is characterized in that A bidirectional LSTM is set in the semantic representation model; The determining of the word vector of the input character sequence based on the semantic representation model includes: Output a context-dependent first word vector generated by the forward language model and a second word vector generated by the backward language model at each position of each LSTM; Output a probability value according to the first word vector and the second word vector, wherein the probability value is used to predict a corresponding input character; All layers of the input character are superimposed according to the attention weight to obtain the word vector corresponding to the input character.

5. The method according to claim 1, It is characterized in that Also includes: For any digital or text sensitive data, query the actual result of the sensitive data in the sensitive database; Based on a deep convolutional generative adversarial network, Laplace noise is added to the true result to replace the sensitive data to obtain desensitized data.

6. The method according to claim 5, It is characterized in that Also includes: Storing the desensitized data in the InterPlanetary File System (IPFS); When receiving a data sharing request, verify the validity of the decryption key for the sensitive data to be shared, and if the sensitive data to be shared is successfully decrypted, send the sensitive data to be shared to the sharing object; Through the smart contract service, the sharing information of the desensitized data and the corresponding digital signature information are packaged and uploaded to the chain after consensus.

7. The method according to claim 1, It is characterized in that Also includes: The number of fields of the digital sensitive data and the text sensitive data is matched with a preset threshold, and the approval level of the data to be detected is determined according to the matched preset threshold.

8. A sensitive data processing device, It is characterized in that include: An acquisition module, used for acquiring data to be detected, wherein the data to be detected includes digital data and / or text data; A first processing module is used to determine the digital sensitive data according to the conversion probability matrix of the sensitive attribute for the digital data, wherein the element in the i-th row and the j-th column in the conversion probability matrix represents the probability of transferring from the sensitive attribute i to the sensitive attribute j; The second processing module is used to determine text-type sensitive data for the text-type data based on a preset model, wherein the preset model is constructed based on a semantic representation model and a dictionary tree, the semantic representation model is used to determine the word vectors of the word segments in the text-type data, and the dictionary tree is used to store sensitive words.

9. An electronic device, It is characterized in that include: at least one processor; as well as a memory communicatively connected to the at least one processor; wherein, The memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute the sensitive data processing method as described in any one of claims 1-7.

10. A computer-readable storage medium having a computer program stored thereon, It is characterized in that When the program is executed by a processor, the sensitive data processing method as described in any one of claims 1-7 is implemented.

11. A computer program product, It is characterized in that It comprises a computer program / instruction, which, when executed by a processor, implements the sensitive data processing method as described in any one of claims 1-7.

Images