A security attribute calibration method, system, device and computer storage medium

Abstract

The present application discloses a security attribute calibration method, system, device and computer storage medium, which obtains labeled source domain data and unlabeled target domain data; determines the target domain data label of the target domain data based on the source domain data through a pre-trained neural network model; determines the candidate security attribute set of the target domain data based on the WordNet dictionary and the target domain data label; screens out the target security attributes of the target domain data at each vocabulary depth from the candidate security attribute set; and determines the security attribute calibration result of the target domain data from the target security attributes at each vocabulary depth according to the access control granularity. In the present application, the target domain data label of the target domain data is determined based on the labeled source domain data through a pre-trained neural network model, and the security attribute calibration result that meets the access control granularity can be obtained based on the WordNet dictionary and the access control granularity, thereby realizing fine-grained control of security attributes.

Description

Technical Field

[0001] The present application relates to the field of access control technology, and more specifically, to a security attribute calibration method, system, device and computer storage medium. Background Art

[0002] Access control technology is a key technology to achieve safe and controllable sharing and circulation of big data. The calibration of security attributes of access objects is the basis of ABAC (attribute based access control) mechanism. The sources of big data resources are diverse and massive. If the traditional manual method is used to calibrate the security attributes of big data resources, it will be time-consuming and laborious, and it is easy to miss or miscalibrate, which will bring security risks. To avoid this problem, keyword mining technology can be used to extract text candidate keywords from all text information, and three categories of features can be extracted from the candidate keywords: general features, structural features, such as whether they appear in the title, and language features, such as part of speech and word class. These three categories of features total 40 types; then these 40 types of features are converted into 100-dimensional vectors and presented in the form of grayscale images; then the classification network based on CNN (Convolutional Neural Networks) is used to determine which candidate security attributes are the security attributes of the text.

[0003] However, when there are many types of data features, directly using keywords as security attributes will lead to the expansion of the security attribute space due to the richness of the language, which will seriously affect the accurate formulation of access control policies and the efficiency of judgment.

[0004] In summary, how to ensure that security attributes are controllable is an urgent problem to be solved by those skilled in the art. Summary of the invention

[0005] The purpose of this application is to provide a security attribute calibration method, which can solve the technical problem of how to ensure that security attributes are controllable to a certain extent. This application also provides a security attribute calibration system, device and computer-readable storage medium.

[0006] In order to achieve the above objectives, this application provides the following technical solutions:

[0007] A security attribute calibration method, comprising:

[0008] Obtain labeled source domain data and unlabeled target domain data;

[0009] Determining a target domain data label of the target domain data based on the source domain data by using a pre-trained neural network model;

[0010] Determine a candidate security attribute set of the target domain data based on the WordNet dictionary and the target domain data label;

[0011] Filtering out target security attributes of the target domain data at various vocabulary depths from the candidate security attribute set;

[0012] According to the access control granularity, the security attribute calibration result of the target domain data is determined in the target security attribute of each vocabulary depth.

[0013] Preferably, the step of determining a candidate security attribute set of the target domain data based on the WordNet dictionary and the target domain data label includes:

[0014] Determining a target synonym set of the target domain data label based on the WordNet dictionary;

[0015] Taking each element in the target synonym set as a starting point in turn, based on the hypernymy relationship between synonym sets in the WordNet dictionary, searching for hypernymy sets layer by layer upwards from the corresponding vocabulary depth until the entire WordNet network is traversed to obtain a target hypernymy set of the target domain data label;

[0016] Determine a first synonym set name corresponding to the target synonym set;

[0017] Determine a second synonym set name corresponding to the target hypernymous synonym set;

[0018] The union of the first synonym set name and the second synonym set name is used as the candidate security attribute set of the target domain data.

[0019] Preferably, the step of screening out target security attributes of the target domain data at various vocabulary depths from the candidate security attribute set includes:

[0020] Determining each of the vocabulary depths of the target domain data labels;

[0021] Determine the candidate security attributes with the same vocabulary depth in the candidate security attribute set as the security attributes to be selected corresponding to the vocabulary depth;

[0022] For each of the candidate security attributes corresponding to the vocabulary depth, the similarity between the target domain data label and each of the candidate security attributes is calculated, and the candidate security attribute corresponding to the maximum similarity value is used as the target security attribute of the target domain data at the vocabulary depth.

[0023] Preferably, the calculating the similarity between the target domain data label and each of the selected security attributes includes:

[0024] Based on the cosine distance calculation formula, the similarity between the target domain data label and each of the selected security attributes is calculated;

[0025] The cosine distance calculation formula includes:

[0026]

[0027] Among them, dist(vec1,vec2) represents the similarity; vec1 represents the word vector value of the target domain data label; vec2 represents the word vector value of the selected security attribute; and n represents the word vector length.

[0028] Preferably, after screening out the target security attributes of the target domain data at each vocabulary depth from the candidate security attribute set, and before determining the security attribute calibration result of the target domain data from the target security attributes at each vocabulary depth according to the access control granularity, the method further includes:

[0029] The target security attributes corresponding to the target domain data are sorted in order of the vocabulary depth from small to large to obtain the upper and lower node relationship;

[0030] Adding the target domain data label to the end of the upper and lower node relationship to obtain the security attribute branch corresponding to the target domain data;

[0031] Taking the target domain data labels in all the security attribute branches as leaf nodes;

[0032] Based on the upper and lower node relationships, starting from the maximum vocabulary depth among all the security attribute branches, traverse layer by layer in the direction where the vocabulary depth becomes smaller, and compare in turn whether there are branches with the same security attributes at each vocabulary depth. If there are branches with the same security attributes, merge the branches, and use the target security attributes corresponding to the merged branches as corresponding child nodes. If there are no branches with the same security attributes, use the target security attributes corresponding to each branch as corresponding child nodes, until the target security attribute corresponding to the minimum vocabulary depth is used as the root node to obtain the target security attribute tree.

[0033] Preferably, the neural network model includes a target CAN network, the target CAN network extracts features based on a residual network, and a CBAM layer is arranged between the connections of the residual network Bottleneck, and the CBAM layer includes a channel attention block and a spatial attention block.

[0034] Preferably, the training process of the neural network model includes:

[0035] Get the first type of data with labels and the second type of data without labels;

[0036] Training the neural network model based on the first category of data and the second category of data;

[0037] Extracting first-category data features, first-category data labels, second-category data features, and second-category data labels generated by the neural network model during training;

[0038] Based on the loss function, the first type of data features, the first type of data labels, the second type of data features, and the second type of data labels, updating the network parameters of the neural network model until the training of the neural network model is completed;

[0039] The loss function includes:

[0040]

[0041] in,

[0042]

[0043]

[0044]

[0045]

[0046]

[0047]

[0048]

[0049] Wherein, w represents the network parameters of the neural network model; X s represents the first type of data features; X T Represents the second type of data features; y s represents the first type of data label, represents the i-th label value in the first category of data labels, represents the jth label value in the first category of data labels; represents the i-th feature value in the first type of data features; represents the j-th feature value in the first type of data features; represents the i-th feature value in the second type of data features; represents the j-th feature value in the second type of data features; represents the second type of data label, represents the i-th label value in the second type of data label, represents the jth label value in the second type of data labels; represents the i-th sample The predicted label of s Indicates the number of the first type of data; n t represents the number of the second category of data; β represents the weight of the penalty item; L represents the total number of fully connected layers, l represents the lth layer in the fully connected layer; M represents the number of entity labels of the first category of data and the second category of data; δ represents the bandwidth of the Gaussian kernel.

[0050] A safety attribute calibration system, comprising:

[0051] A first acquisition module is used to acquire labeled source domain data and unlabeled target domain data;

[0052] A first determination module, configured to determine a target domain data label of the target domain data based on the source domain data by using a pre-trained neural network model;

[0053] A second determination module, configured to determine a candidate security attribute set of the target domain data based on a WordNet dictionary and the target domain data label;

[0054] A first screening module, configured to screen out target security attributes of the target domain data at various vocabulary depths from the candidate security attribute set;

[0055] The third determination module is used to determine the security attribute calibration result of the target domain data in the target security attribute of each vocabulary depth according to the access control granularity.

[0056] A safety attribute calibration device, comprising:

[0057] Memory for storing computer programs;

[0058] A processor is used to implement the steps of any of the security attribute calibration methods described above when executing the computer program.

[0059] A computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of any of the security attribute calibration methods described above are implemented.

[0060] The present application provides a method for calibrating security attributes, which obtains labeled source domain data and unlabeled target domain data; determines the target domain data label of the target domain data based on the source domain data through a pre-trained neural network model; determines the candidate security attribute set of the target domain data based on the WordNet dictionary and the target domain data label; selects the target security attributes of the target domain data at each vocabulary depth from the candidate security attribute set; and determines the security attribute calibration result of the target domain data from the target security attributes at each vocabulary depth according to the access control granularity. In the present application, the target domain data label of the target domain data can be determined based on the labeled source domain data through a pre-trained neural network model to achieve label prediction, and the candidate security attribute set of the target domain data can be determined based on the WordNet dictionary and the target domain data label, and the target security attributes of the target domain data at each vocabulary depth can be selected. Finally, according to the access control granularity, the security attribute calibration result of the target domain data is determined from the target security attributes at each vocabulary depth, so that the obtained security attribute calibration result meets the access control granularity and realizes fine-grained control of security attributes. The present application provides a security attribute calibration system, device and computer-readable storage medium that also solves the corresponding technical problems. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying any creative work.

[0062] Figure 1 A first flow chart of a security attribute calibration method provided in an embodiment of the present application;

[0063] Figure 2 A second flow chart of a security attribute calibration method provided in an embodiment of the present application;

[0064] Figure 3 A third flow chart of a security attribute calibration method provided in an embodiment of the present application;

[0065] Figure 4 It is the connection relationship diagram between Bottleneck and CBAM layers;

[0066] Figure 5 A schematic diagram of the structure of a security attribute calibration system provided in an embodiment of the present application;

[0067] Figure 6 A schematic diagram of the structure of a security attribute calibration device provided in an embodiment of the present application;

[0068] Figure 7 Another structural schematic diagram of a security attribute calibration device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0069] 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 ordinary technicians in this field without creative work are within the scope of protection of this application.

[0070] See also Figure 1 , Figure 1 A first flow chart of a security attribute calibration method provided in an embodiment of the present application.

[0071] A security attribute calibration method provided in an embodiment of the present application may include the following steps:

[0072] Step S101: Acquire labeled source domain data and unlabeled target domain data.

[0073] In practical applications, labeled source domain data and unlabeled target domain data can be obtained first. It should be noted that the types of source domain data and target domain data can be determined according to actual needs. For example, both source domain data and target domain data can be image data, etc. In addition, the labels of source domain data and target domain data can also be determined according to actual needs. For example, the labels can be the categories of image data, etc. This application does not make specific limitations here.

[0074] Step S102: Determine the target domain data label of the target domain data based on the source domain data through the pre-trained neural network model.

[0075] In practical applications, in order to facilitate the prediction of target domain data labels of target domain data, a pre-trained neural network model can be used to determine the target domain data labels of target domain data based on labeled source domain data with the help of the autonomous learning function of the neural network model.

[0076] It should be noted that the type, structure and training process of the neural network model can be determined according to actual needs, and this application does not make any specific limitations here.

[0077] Step S103: Determine a candidate security attribute set of the target domain data based on the WordNet dictionary and the target domain data label.

[0078] In practical applications, after the target domain data label of the target domain data is determined based on the source domain data through a pre-trained neural network model, the candidate security attribute set of the target domain data can be determined based on the WordNet dictionary and the target domain data label. It should be noted that the WordNet dictionary in this application is an English vocabulary database, and its basic unit is a synonym set, which is a group of words with the same meaning. Each synonym set has a synonym set name, representing a different concept, and there are superordinate and subordinate relationships between synonym sets. The process of determining the candidate security attribute set of the target domain data based on the WordNet dictionary and the target domain data label can be flexibly determined according to actual needs, and this application does not make specific limitations here; in addition, the candidate security attribute set obtained in this application is used to determine the target security attribute of the target domain data therefrom.

[0079] Step S104: Filter out target security attributes of target domain data at various vocabulary depths from the candidate security attribute set.

[0080] Step S105: According to the access control granularity, the security attribute calibration result of the target domain data is determined in the target security attributes of each vocabulary depth.

[0081] In practical applications, after determining the candidate security attribute set of the target domain data based on the WordNet dictionary and the target domain data label, the target security attributes of the target domain data at each vocabulary depth can be screened out from the candidate security attribute set, and the security attribute calibration results of the target domain data can be determined from the target security attributes at each vocabulary depth according to the access control granularity, so as to realize flexible calibration of security attributes with adjustable granularity. Finally, access control can be performed on the target domain data based on the final security attribute calibration results.

[0082] In specific application scenarios, in the process of determining the security attribute calibration results of the target domain data in the target security attributes of each vocabulary depth according to the access control granularity, the security attribute calibration results can be determined according to the principle that the finer the access control granularity, the target security attributes with a larger vocabulary depth will be determined as the security attribute calibration results of the target domain data. This application does not make specific limitations here.

[0083] The present application provides a method for calibrating security attributes, which obtains labeled source domain data and unlabeled target domain data; determines the target domain data label of the target domain data based on the source domain data through a pre-trained neural network model; determines the candidate security attribute set of the target domain data based on the WordNet dictionary and the target domain data label; screens out the target security attributes of the target domain data at each vocabulary depth in the candidate security attribute set; and determines the security attribute calibration result of the target domain data in the target security attributes at each vocabulary depth according to the access control granularity. In the present application, the target domain data label of the target domain data can be determined based on the labeled source domain data through a pre-trained neural network model to achieve label prediction, and the candidate security attribute set of the target domain data can be determined based on the WordNet dictionary and the target domain data label, and the target security attributes of the target domain data at each vocabulary depth can be screened out. Finally, according to the access control granularity, the security attribute calibration result of the target domain data is determined in the target security attributes at each vocabulary depth, so that the obtained security attribute calibration result meets the access control granularity and realizes fine-grained control of security attributes.

[0084] See also Figure 2 , Figure 2 A second flow chart of a security attribute calibration method provided in an embodiment of the present application.

[0085] A security attribute calibration method provided in an embodiment of the present application may include the following steps:

[0086] Step S201: Acquire labeled source domain data and unlabeled target domain data.

[0087] Step S202: Determine the target domain data label of the target domain data based on the source domain data through the pre-trained neural network model.

[0088] Step S203: Determine a target synonym set of the target domain data label based on the WordNet dictionary.

[0089] In practical applications, in the process of determining the candidate security attribute set of the target domain data based on the WordNet dictionary and the target domain data label, the target synonym set of the target domain data label can be determined based on the WordNet dictionary. Specifically, the target synonym set of the target domain data label can be obtained through the Synsets() method of the WordNet dictionary.

[0090] It should be noted that the definition of the synonym set in the WordNet dictionary is: entity tag l i The set of synonym sets in WordNet (1≤i≤M) is called the synonym set of the entity tag, denoted by S i and Among them, S i,j Indicates l i The jth synset in WordNet, n represents l i The number of synonym sets in WordNet. i,j The vocabulary depth is represented by D ij , 1≤j≤n.

[0091] Step S204: Take each element in the target synonym set as a starting point in turn, and based on the hypernym relationship between synonym sets in the WordNet dictionary, search for hypernym synonym sets layer by layer from the corresponding vocabulary depth until the entire WordNet network is traversed to obtain the target hypernym synonym set of the target domain data label.

[0092] In practical applications, after determining the target synonym set of the target domain data label based on the WordNet dictionary, each element in the target synonym set can be used as a starting point in turn, and based on the hypernym relationship between synonym sets in the WordNet dictionary, the hypernym set is searched upward layer by layer starting from the corresponding lexical depth until the entire WordNet network is traversed to obtain the target hypernym set of the target domain data label.

[0093] It should be noted that the definition of the hypernym set in the WordNet dictionary is: entity tag l i The set of synonym sets that form the hypernymy relationship of (1≤i≤M) is called the hypernymy set of the entity tag, denoted as H i and Among them, H i,j Indicates S i,j The set consisting of the hypernymous sets of (1≤j≤n).

[0094] Step S205: Determine the first synonym set name corresponding to the target synonym set.

[0095] Step S206: Determine the name of the second synonym set corresponding to the target hypernymous synonym set.

[0096] In practical applications, each element in the target synonym set is taken as the starting point in turn, and based on the hypernym relationship between synonym sets in the WordNet dictionary, the hypernym synonym set is searched upward layer by layer starting from the corresponding vocabulary depth until the entire WordNet network is traversed and the target hypernym synonym set of the target domain data label is obtained. Then, the name of the first synonym set corresponding to the target synonym set can be determined, and the name of the second synonym set corresponding to the target hypernym synonym set can be determined.

[0097] It should be noted that in a specific application scenario, the name() method in the WordNet dictionary can be used to determine the first synonym set name corresponding to the target synonym set and determine the second synonym set name corresponding to the target hypernymous synonym set.

[0098] Step S207: taking the union of the first synonym set name and the second synonym set name as a candidate security attribute set for the target domain data.

[0099] In practical applications, after determining the first synonym set name corresponding to the target synonym set and determining the second synonym set name corresponding to the target superordinate synonym set, the union of the first synonym set name and the second synonym set name can be used as a candidate security attribute set for the target domain data.

[0100] It should be noted that the candidate security attribute set in the WordNet dictionary is defined as: entity tag l i (1≤i≤M) at depth The set of candidate security attributes on is called the candidate security attribute set of the entity tag, denoted as A i and Indicates S i,j The name of the synonym set of Indicates H i,j The name of the synonym set.

[0101] Step S208: Determine the vocabulary depths of target domain data labels.

[0102] Step S209: Determine the candidate security attributes with the same vocabulary depth in the candidate security attribute set as the security attributes to be selected corresponding to the vocabulary depth.

[0103] Step S210: For each candidate security attribute corresponding to each vocabulary depth, calculate the similarity between the target domain data label and each candidate security attribute, and use the candidate security attribute corresponding to the maximum similarity value as the target security attribute of the target domain data at the vocabulary depth.

[0104] In practical applications, in the process of screening out the target security attributes of the target domain data at each lexical depth from the candidate security attribute set, the lexical depths of the target domain data label can be determined; the candidate security attributes of the same lexical depth in the candidate security attribute set are determined as the candidate security attributes corresponding to the lexical depth, so as to facilitate the subsequent determination of the security attributes of the target domain data at the lexical depth from the candidate security attributes corresponding to the lexical depth; finally, for the candidate security attributes corresponding to each lexical depth, the similarity between the target domain data label and each candidate security attribute is calculated, and the candidate security attribute corresponding to the maximum similarity value is used as the target security attribute of the target domain data at the lexical depth. In this way, at each lexical depth, there is only one security attribute of the target domain data label, which can simplify the scale of the target security attributes.

[0105] In a specific application scenario, in the process of calculating the similarity between the target domain data label and each candidate security attribute, the similarity between the target domain data label and each candidate security attribute can be calculated based on the cosine distance calculation formula;

[0106] The cosine distance calculation formula includes:

[0107]

[0108] Among them, dist(vec1,vec2) represents the similarity; vec1 represents the word vector value of the target domain data label, vec 1i represents the i-th value in the word vector value of the target domain data label; vec2 represents the word vector value of the selected security attribute, vec 2i Represents the i-th value in the word vector value of the selected security attribute; n represents the length of the word vector.

[0109] Step S211: Determine the security attribute calibration result of the target domain data in the target security attributes of each vocabulary depth according to the access control granularity.

[0110] See also Figure 3 , Figure 3 A third flow chart of a security attribute calibration method provided in an embodiment of the present application.

[0111] A security attribute calibration method provided in an embodiment of the present application may include the following steps:

[0112] Step S301: Acquire labeled source domain data and unlabeled target domain data.

[0113] Step S302: Determine the target domain data label of the target domain data based on the source domain data through the pre-trained neural network model.

[0114] Step S303: Determine a candidate security attribute set of the target domain data based on the WordNet dictionary and the target domain data label.

[0115] Step S304: Filter out target security attributes of target domain data at various vocabulary depths from the candidate security attribute set.

[0116] Step S305: sorting the target security attributes corresponding to the target domain data in order of vocabulary depth from small to large to obtain the superior-subordinate node relationship.

[0117] In practical applications, in order to facilitate the application of security attributes, after screening out the target security attributes of the target domain data at each vocabulary depth in the candidate security attribute set, and before screening out the target security attributes of the target domain data at each vocabulary depth in the candidate security attribute set according to the access control granularity, the security attributes corresponding to the target domain data can also be used to generate a security attribute tree, so as to use the security attribute tree to view the relationship between security attributes, etc. In this process, the target security attributes corresponding to the target domain data can be sorted in order from small to large according to the vocabulary depth to obtain the relationship between superior and subordinate nodes, that is, the security attributes in the security attribute tree are arranged in nodes based on the vocabulary depth layer.

[0118] Step S306: Add the target domain data label to the end of the upper and lower node relationship to obtain the security attribute branch corresponding to the target domain data.

[0119] In practical applications, after sorting the target security attributes corresponding to the target domain data in ascending order of vocabulary depth and obtaining the superior-subordinate node relationship, the target domain data label can be added to the end of the superior-subordinate node relationship to obtain the security attribute branch corresponding to the target domain data, that is, the element attr at position j in the security attribute branch. j Attr is the element at position j+1 j+1 The parent node of the target domain, and the security attribute branch reflects the security attribute information corresponding to the target domain data.

[0120] Step S307: Take the target domain data labels in all security attribute branches as leaf nodes.

[0121] In practical applications, after adding the target domain data label to the end of the superior-subordinate node relationship and obtaining the security attribute branch corresponding to the target domain data, the target domain data labels in all security attribute branches can be used as leaf nodes, that is, each target domain data label is stored at the bottom of the security attribute tree.

[0122] Step S308: Based on the relationship between superior and subordinate nodes, starting from the maximum vocabulary depth in all security attribute branches, traverse layer by layer in the direction of decreasing vocabulary depth, and compare whether there are branches with the same security attributes at each vocabulary depth. If there are branches with the same security attributes, merge the branches and use the target security attributes corresponding to the merged branches as the corresponding child nodes. If there are no branches with the same security attributes, use the target security attributes corresponding to each branch as the corresponding child nodes, until the target security attribute corresponding to the minimum vocabulary depth is used as the root node to obtain the target security attribute tree.

[0123] In practical applications, after taking the target domain data labels in all security attribute branches as leaf nodes, we can start from the maximum vocabulary depth in all security attribute branches based on the relationship between superior and subordinate nodes, and traverse layer by layer in the direction of decreasing vocabulary depth, and compare whether there are branches with the same security attributes in each vocabulary depth. If there are branches with the same security attributes, the branches are merged, and the target security attributes corresponding to the merged branches are used as corresponding child nodes, that is, the security attributes of the two branches in the previous layer point to the same parent node in this layer. If there are no branches with the same security attributes, the target security attributes corresponding to each branch are used as corresponding child nodes, until the target security attribute corresponding to the minimum vocabulary depth is used as the root node to obtain the target security attribute tree. That is, the final target security attribute tree is a security attribute tree with the target domain data label of the target domain data as the leaf node, the security attribute at the vocabulary depth of 0 as the root node, and the security attribute at the vocabulary depth of non-0 as the branch node. It should be noted that the vocabulary depth of 0 here refers to the vocabulary depth of the 0th layer in the WordNet dictionary, and the corresponding security attribute has the coarsest granularity; and in the security attribute tree, the lower you go, the higher the number of layers of vocabulary depth, and accordingly, the finer the granularity of the security attribute corresponding to the vocabulary depth.

[0124] Step S309: According to the access control granularity, the security attribute calibration result of the target domain data is determined in the target security attribute tree.

[0125] In actual applications, after generating the target security attribute tree, in the process of determining the security attribute calibration results of the target domain data in the target security attributes of each vocabulary depth according to the access control granularity, the security attribute calibration results of the target domain data can be determined in the target security attribute tree according to the access control granularity. For example, when the access control granularity is finer, the target security attributes at the later end of the target security attribute tree will be used as the security attribute calibration results of the target domain data.

[0126] In a security attribute calibration method provided in an embodiment of the present application, in order to enhance the generalization ability of the entire security attribute calibration method, the neural network model in the present application can be a target CAN (Contrastive Adaptation Network) network for unsupervised domain adaptation; and in order to extract high-dimensional features of data samples and prevent the problem of gradient disappearance caused by excessive accumulation of convolutional layers during feature extraction, the target CAN network extracts features based on a residual network, and the residual network can be a residual network such as ResNet50, ResNet101, ResNet152, etc. Of course, the target CAN network can also be based on a non-residual network to extract features, etc.; and in order to improve the extraction accuracy of data labels, a CBAM (Convolutional Block Attention Module) layer is provided between the connections of the residual network Bottleneck, and the CBAM layer includes a channel attention block (Channel Attention Module, CAM) and a spatial attention block (Spatial Attention Module, SAM), and the connection relationship between the Bottleneck and the CBAM layer can be as follows Figure 4 shown.

[0127] In specific application scenarios, the training process of the neural network model can include the following steps:

[0128] Get the first type of data with labels and the second type of data without labels;

[0129] Training the neural network model based on the first type of data and the second type of data;

[0130] Extracting first-class data features, first-class data labels, second-class data features, and second-class data labels generated by the neural network model during the training process;

[0131] Based on the loss function, the first type of data features, the first type of data labels, the second type of data features, and the second type of data labels, the network parameters of the neural network model are updated until the training of the neural network model is completed;

[0132] The loss functions include:

[0133]

[0134] in,

[0135]

[0136]

[0137]

[0138]

[0139]

[0140]

[0141]

[0142] Where w represents the network parameters of the neural network model; X S Represents the first type of data features; X T Represents the second type of data features; y s represents the first type of data label, represents the i-th label value in the first category of data labels, Represents the jth label value in the first category of data labels; Represents the i-th eigenvalue in the first category of data features; Represents the jth eigenvalue in the first category of data features; Represents the i-th eigenvalue in the second type of data features; Represents the jth eigenvalue in the second type of data features; represents the second type of data label, represents the i-th label value in the second category of data labels, Represents the jth label value in the second category of data labels; represents the i-th sample The predicted label of s Indicates the number of the first type of data; n t Represents the number of second-category data; β represents the penalty weight; L represents the total number of fully connected layers, l represents the lth layer in the fully connected layer; M represents the number of entity labels for the first-category data and the second-category data. It indicates the CDD (Contrastive Domain Discrepancy) difference calculated after the extracted features pass through the lth layer FC (Full Connection), and when c1=c2, It can represent the intra-class difference. When c1≠c2, Can represent differences between classes; It can be the second type of data label extracted based on the spherical K-means clustering algorithm; k l (x u ,x' u ) represents the kernel function that maps data to RKHS (reproducing kernel Hilbert space); δ represents the bandwidth of the Gaussian kernel.

[0143] See also Figure 5 , Figure 5 A schematic diagram of the structure of a security attribute calibration system provided in an embodiment of the present application.

[0144] A security attribute calibration system provided in an embodiment of the present application may include:

[0145] A first acquisition module 101 is used to acquire labeled source domain data and unlabeled target domain data;

[0146] A first determination module 102 is used to determine a target domain data label of the target domain data based on the source domain data by using a pre-trained neural network model;

[0147] A second determination module 103 is used to determine a candidate security attribute set of the target domain data based on the WordNet dictionary and the target domain data label;

[0148] The first screening module 104 is used to screen out target security attributes of target domain data at various vocabulary depths from the candidate security attribute set;

[0149] The third determination module 105 is used to determine the security attribute calibration result of the target domain data in the target security attributes at each vocabulary depth according to the access control granularity.

[0150] For the relevant description of each module in a security attribute calibration system provided in an embodiment of the present application, please refer to the above embodiment and will not be repeated here.

[0151] The present application also provides a security attribute calibration device and a computer-readable storage medium, both of which have the corresponding effects of the security attribute calibration method provided in the embodiment of the present application. Figure 6 , Figure 6 A schematic diagram of the structure of a security attribute calibration device provided in an embodiment of the present application.

[0152] A fan speed processing device provided in an embodiment of the present application includes a memory 201 and a processor 202. The memory 201 stores a computer program. When the processor 202 executes the computer program, the steps of the security attribute calibration method described in any of the above embodiments are implemented.

[0153] See also Figure 7, another security attribute calibration device provided in the embodiment of the present application may also include: an input port 203 connected to the processor 202, used to transmit commands input from the outside to the processor 202; a display unit 204 connected to the processor 202, used to display the processing results of the processor 202 to the outside; a communication module 205 connected to the processor 202, used to realize the communication between the security attribute calibration device and the outside. The display unit 204 can be a display panel, a laser scanning display, etc.; the communication mode adopted by the communication module 205 includes but is not limited to mobile high-definition link technology (HML), universal serial bus (USB), high-definition multimedia interface (HDMI), wireless connection: wireless fidelity technology (WiFi), Bluetooth communication technology, low-power Bluetooth communication technology, and communication technology based on IEEE802.11s.

[0154] An embodiment of the present application provides a computer-readable storage medium, in which a computer program is stored. When the computer program is executed by a processor, the steps of the security attribute calibration method described in any of the above embodiments are implemented.

[0155] The computer-readable storage medium involved in the present application includes random access memory (RAM), internal memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disks, removable disks, CD-ROMs, or any other form of storage medium known in the technical field.

[0156] For the description of the relevant parts of the security attribute calibration system, device and computer-readable storage medium provided in the embodiments of the present application, please refer to the detailed description of the corresponding parts in the security attribute calibration provided in the embodiments of the present application, which will not be repeated here. In addition, the parts of the above technical solutions provided in the embodiments of the present application that are consistent with the corresponding technical solutions in the prior art are not described in detail to avoid excessive elaboration.

[0157] It should also be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the presence of other identical elements in the process, method, article or device including the elements.

[0158] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A security attribute calibration method, characterized in that: include: Obtain labeled source domain data and unlabeled target domain data; Determining a target domain data label of the target domain data based on the source domain data by using a pre-trained neural network model; Determine a candidate security attribute set of the target domain data based on the WordNet dictionary and the target domain data label; Filtering out target security attributes of the target domain data at various vocabulary depths from the candidate security attribute set; Determining the security attribute calibration result of the target domain data in the target security attribute of each vocabulary depth according to the access control granularity; The step of determining a candidate security attribute set of the target domain data based on the WordNet dictionary and the target domain data label includes: Determining a target synonym set of the target domain data label based on the WordNet dictionary; Taking each element in the target synonym set as a starting point in turn, based on the hypernymy relationship between synonym sets in the WordNet dictionary, searching for hypernymy sets layer by layer upwards from the corresponding vocabulary depth until the entire WordNet network is traversed to obtain a target hypernymy set of the target domain data label; Determine a first synonym set name corresponding to the target synonym set; Determine a second synonym set name corresponding to the target hypernymous synonym set; taking the union of the first synonym set name and the second synonym set name as the candidate security attribute set of the target domain data; The step of selecting target security attributes of the target domain data at various vocabulary depths from the candidate security attribute set includes: Determining each of the vocabulary depths of the target domain data labels; Determine the candidate security attributes with the same vocabulary depth in the candidate security attribute set as the security attributes to be selected corresponding to the vocabulary depth; For each of the candidate security attributes corresponding to the vocabulary depth, the similarity between the target domain data label and each of the candidate security attributes is calculated, and the candidate security attribute corresponding to the maximum similarity value is used as the target security attribute of the target domain data at the vocabulary depth.

2. The method according to claim 1, characterized in that The calculating the similarity between the target domain data label and each of the selected security attributes includes: Based on the cosine distance calculation formula, the similarity between the target domain data label and each of the selected security attributes is calculated; The cosine distance calculation formula includes: Among them, dist(vec1,vec2) represents the similarity; vec1 represents the word vector value of the target domain data label; vec2 represents the word vector value of the selected security attribute; and n represents the word vector length.

3. The method according to claim 2, characterized in that After the target security attributes of the target domain data at each vocabulary depth are screened out from the candidate security attribute set, and before the security attribute calibration result of the target domain data is determined from the target security attributes at each vocabulary depth according to the access control granularity, the method further includes: The target security attributes corresponding to the target domain data are sorted in order of the vocabulary depth from small to large to obtain the upper and lower node relationship; Adding the target domain data label to the end of the upper and lower node relationship to obtain the security attribute branch corresponding to the target domain data; Taking the target domain data labels in all the security attribute branches as leaf nodes; Based on the upper and lower node relationships, starting from the maximum vocabulary depth among all the security attribute branches, traverse layer by layer in the direction where the vocabulary depth becomes smaller, and compare in turn whether there are branches with the same security attributes at each vocabulary depth. If there are branches with the same security attributes, merge the branches, and use the target security attributes corresponding to the merged branches as corresponding child nodes. If there are no branches with the same security attributes, use the target security attributes corresponding to each branch as corresponding child nodes, until the target security attribute corresponding to the minimum vocabulary depth is used as the root node to obtain the target security attribute tree.

4. The method according to any one of claims 1 to 3, characterized in that: The neural network model includes a target CAN network, which extracts features based on a residual network, and a CBAM layer is arranged between the connections of the residual network Bottleneck, and the CBAM layer includes a channel attention block and a spatial attention block.

5. The method according to claim 4, characterized in that The training process of the neural network model includes: Get the first type of data with labels and the second type of data without labels; Training the neural network model based on the first category of data and the second category of data; Extracting first-category data features, first-category data labels, second-category data features, and second-category data labels generated by the neural network model during training; Based on the loss function, the first type of data features, the first type of data labels, the second type of data features, and the second type of data labels, updating the network parameters of the neural network model until the training of the neural network model is completed; The loss function includes: in, Wherein, w represents the network parameters of the neural network model; X s represents the first type of data features; X T Represents the second type of data features; y s represents the first type of data label, represents the i-th label value in the first category of data labels, represents the jth label value in the first category of data labels; represents the i-th feature value in the first type of data features; represents the j-th feature value in the first type of data features; represents the i-th feature value in the second type of data features; represents the j-th feature value in the second type of data features; represents the second type of data label, represents the i-th label value in the second type of data label, represents the jth label value in the second type of data labels; represents the i-th sample X si The predicted label of s Indicates the number of the first type of data; n t represents the number of the second category of data; β represents the weight of the penalty item; L represents the total number of fully connected layers, l represents the lth layer in the fully connected layer; M represents the number of entity labels of the first category of data and the second category of data; δ represents the bandwidth of the Gaussian kernel.

6. A security attribute calibration system, characterized in that: include: A first acquisition module is used to acquire labeled source domain data and unlabeled target domain data; A first determination module, configured to determine a target domain data label of the target domain data based on the source domain data by using a pre-trained neural network model; A second determination module, configured to determine a candidate security attribute set of the target domain data based on a WordNet dictionary and the target domain data label; A first screening module, configured to screen out target security attributes of the target domain data at various vocabulary depths from the candidate security attribute set; A third determination module is used to determine the security attribute calibration result of the target domain data in the target security attribute of each vocabulary depth according to the access control granularity; The second determination module is used to: determine the target synonym set of the target domain data label based on the WordNet dictionary; take each element in the target synonym set as a starting point in turn, and based on the hypernym relationship between synonym sets in the WordNet dictionary, search for hypernym sets layer by layer from the corresponding vocabulary depth until the entire WordNet network is traversed to obtain the target hypernym set of the target domain data label; determine the first synonym set name corresponding to the target synonym set; determine the second synonym set name corresponding to the target hypernym set; and use the union of the first synonym set name and the second synonym set name as the candidate security attribute set of the target domain data; Among them, the first screening module is used to: determine the various vocabulary depths of the target domain data labels; determine the candidate security attributes with the same vocabulary depth in the candidate security attribute set as the candidate security attributes corresponding to the vocabulary depth; for the candidate security attributes corresponding to each vocabulary depth, calculate the similarity between the target domain data label and each candidate security attribute, and use the candidate security attribute corresponding to the maximum similarity value as the target security attribute of the target domain data at the vocabulary depth.

7. A security attribute calibration device, characterized in that: include: Memory for storing computer programs; A processor, used to implement the steps of the security attribute calibration method as described in any one of claims 1 to 5 when executing the computer program.

8. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the security attribute calibration method according to any one of claims 1 to 5 are implemented.

Images