Data desensitization method, device, equipment and storage medium based on artificial intelligence

Abstract

The data desensitization method and apparatus, equipment and storage medium based on artificial intelligence provided by the present application relate to the field of artificial intelligence technology. In the present application, first, the data to be desensitized is obtained; secondly, for each unit to be desensitized in the data to be desensitized, the semantic mining model included in the target sensitive information recognition network is used to mine the target desensitized semantic features corresponding to the unit to be desensitized based on the sensitive information knowledge graph; then, for each unit to be desensitized in the data to be desensitized, the sensitive information recognition model included in the target sensitive information recognition network is used to identify and output the target desensitized semantic features corresponding to the unit to be desensitized, and obtain the sensitive information recognition result corresponding to the unit to be desensitized; finally, each unit to be desensitized belonging to sensitive information is desensitized separately. Based on the above content, the problem of relatively low reliability of data desensitization in the prior art can be improved.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and more specifically, to an artificial intelligence-based data desensitization method and apparatus, equipment, and storage medium. Background Art

[0002] Data masking is a method of protecting sensitive information, especially in the process of data sharing, data storage and processing. It replaces or modifies sensitive data to ensure that the data is not leaked during use while maintaining the availability of the data. Among them, a key link in data masking is to accurately identify sensitive information (such as sensitive characters, sensitive words, etc.). In the prior art, a sensitive word set is generally set, and then the words in the data to be masked are compared with the sensitive word set to determine the sensitive words in the data to be masked, and corresponding masking processing (such as replacement or modification, etc.) is performed. However, due to the rich semantics of words, the reliability of this simple comparative identification of sensitive data is not high (i.e., it is prone to misidentification problems), which makes the reliability of the masking processing based on the identified sensitive data also low. Summary of the Invention

[0003] In view of this, the purpose of this application is to provide a data desensitization method and device, equipment and storage medium based on artificial intelligence to improve the problem of relatively low reliability of data desensitization in the prior art.

[0004] To achieve the above objectives, this application adopts the following technical solutions:

[0005] A data desensitization method based on artificial intelligence, comprising:

[0006] Obtaining data to be desensitized, wherein the data to be desensitized is text data;

[0007] For each unit to be desensitized in the data to be desensitized, using the semantic mining model included in the target sensitive information identification network, based on a pre-constructed sensitive information knowledge graph, mining the target desensitization semantic features corresponding to the unit to be desensitized, wherein each of the units to be desensitized has the same granularity, which includes a word, a phrase, a sentence, or a paragraph, the target sensitive information identification network is a pre-trained neural network, and the sensitive information knowledge graph is determined based on the relationships between each data unit in the sensitive information database;

[0008] For each unit to be desensitized in the data to be desensitized, using the sensitive information recognition model included in the target sensitive information recognition network, identify and output the target desensitization semantic features corresponding to the unit to be desensitized, and obtain a sensitive information recognition result corresponding to the unit to be desensitized, wherein the sensitive information recognition result is used to reflect whether the unit to be desensitized is sensitive information;

[0009] Desensitizing is performed on each unit of sensitive information to be desensitized.

[0010] In a preferred embodiment of the present application, in the above-mentioned artificial intelligence-based data desensitization method, for each unit to be desensitized in the data to be desensitized, the semantic mining model included in the target sensitive information recognition network is used to mine the target desensitization semantic features corresponding to the unit to be desensitized based on the pre-built sensitive information knowledge graph, including:

[0011] For each unit to be desensitized in the data to be desensitized, determine each information entity having the same name as the unit to be desensitized in a pre-constructed sensitive information knowledge graph, and obtain at least one matching information entity corresponding to the unit to be desensitized, wherein each of the information entities corresponds to a data unit in the sensitive information database;

[0012] For each of the matching information entities, using the first semantic mining unit in the semantic mining model included in the target sensitive information identification network, mining the information entity semantic features corresponding to the matching information entity in the sensitive information knowledge graph;

[0013] For each unit to be desensitized in the data to be desensitized, using the second semantic mining unit in the semantic mining model, mining candidate desensitization semantic features corresponding to the unit to be desensitized in the data to be desensitized;

[0014] For each unit to be desensitized in the data to be desensitized, the association semantic mining unit in the semantic mining model is used to perform association mining on the candidate desensitized semantic features corresponding to the unit to be desensitized based on the information entity semantic features corresponding to each matching information entity corresponding to the unit to be desensitized, so as to obtain the target desensitized semantic features corresponding to the unit to be desensitized.

[0015] In a preferred embodiment of the present application, in the above-mentioned artificial intelligence-based data desensitization method, for each of the matching information entities, the step of mining the semantic features of the information entity corresponding to the matching information entity in the sensitive information knowledge graph using the first semantic mining unit in the semantic mining model included in the target sensitive information identification network includes:

[0016] For each of the matching information entities, using the first semantic mining unit in the semantic mining model included in the target sensitive information recognition network, semantic mining is performed on the entity attribute data corresponding to the matching information entity, and the candidate entity semantic features corresponding to the matching information entity are output;

[0017] Extracting a local sensitive information graph centered on the matching information entity from the sensitive information knowledge graph, wherein the graph scale of the local sensitive information graph matches a pre-configured reference graph scale, and the graph scale is correlated with the maximum path length in the shortest paths between the matching information entity and each other information entity;

[0018] The candidate entity semantic features corresponding to each other information entity in the sensitive information local graph are integrated into the candidate entity semantic features corresponding to the matching information entity to obtain the information entity semantic features corresponding to the matching information entity, wherein the candidate entity semantic features corresponding to each other information entity are obtained by semantic mining using the first semantic mining unit.

[0019] In a preferred embodiment of the present application, in the above-mentioned artificial intelligence-based data desensitization method, the step of fusing the candidate entity semantic features corresponding to each other information entity in the sensitive information local graph into the candidate entity semantic features corresponding to the matching information entity to obtain the information entity semantic features corresponding to the matching information entity includes:

[0020] For each information entity in the sensitive information local graph, based on the path length of the shortest path between the information entity and the matching information entity, determine the graph relationship hierarchy corresponding to the information entity, wherein the matching information entity corresponds to the lowest graph relationship hierarchy, and the other information entities with the largest path length correspond to the highest graph relationship hierarchy;

[0021] The candidate entity semantic features corresponding to each information entity are transferred along the direction from the highest to the lowest level of the graph relationship hierarchy, so that the candidate entity semantic features corresponding to each other information entity in the sensitive information local graph are integrated into the candidate entity semantic features corresponding to the matching information entity to obtain the information entity semantic features corresponding to the matching information entity.

[0022] In a preferred embodiment of the present application, in the above-mentioned artificial intelligence-based data desensitization method, the step of transferring the candidate entity semantic features corresponding to each information entity along the direction from the highest to the lowest level of the graph relationship hierarchy, so that the candidate entity semantic features corresponding to each other information entity in the sensitive information local graph are integrated into the candidate entity semantic features corresponding to the matching information entity to obtain the information entity semantic features corresponding to the matching information entity, includes:

[0023] For each information entity at the highest level of the graph relationship, performing focused feature mining on the semantic features of candidate entities corresponding to the information entity to obtain the focused features corresponding to the information entity;

[0024] For each information entity in each graph relationship layer other than the highest graph relationship layer, determine whether there are multiple other information entities connected to the information entity in each other information entity at a higher layer, and when there are not multiple other information entities connected to the information entity, perform focused feature mining on the semantic features of the candidate entity corresponding to the information entity based on the focused features corresponding to one of the other connected information entities to obtain the focused features corresponding to the information entity, and, when there are multiple other information entities connected to the information entity, determine whether there is a connection relationship between the other connected information entities, and perform focused feature mining on the semantic features of the candidate entity corresponding to the information entity based on the focused features corresponding to each of the other information entities that do not have a connection relationship to obtain the focused local features corresponding to the information entity, and, after superimposing the focused features corresponding to the other information entities that have a connection relationship, perform focused feature mining on the semantic features of the candidate entity corresponding to the information entity based on the features after the superimposed operation to obtain the focused local features corresponding to the information entity, and then perform superimposed operation on each of the focused local features corresponding to the information entity to obtain the focused features corresponding to the information entity;

[0025] After obtaining the focus feature corresponding to the matching information entity, the focus feature corresponding to the matching information entity and the candidate entity semantic feature corresponding to the matching information entity are superimposed to obtain the information entity semantic feature corresponding to the matching information entity.

[0026] In a preferred embodiment of the present invention, in the above-mentioned artificial intelligence-based data desensitization method, for each unit to be desensitized in the data to be desensitized, using the second semantic mining unit in the semantic mining model, the step of mining the candidate desensitization semantic features corresponding to the unit to be desensitized in the data to be desensitized includes:

[0027] For each unit to be desensitized in the data to be desensitized, embedding the unit to be desensitized using the embedding subunit in the second semantic mining unit in the semantic mining model, and outputting a desensitized embedding feature corresponding to the unit to be desensitized;

[0028] Along the direction from the first to-be-desensitized unit to the last to-be-desensitized unit in the to-be-desensitized data, the desensitized embedded feature corresponding to each to-be-desensitized unit is transferred and updated to obtain a first desensitized transfer feature corresponding to each to-be-desensitized unit;

[0029] Along the direction from the last unit to be desensitized in the data to be desensitized to the first unit to be desensitized, the first desensitization transfer feature corresponding to each unit to be desensitized is transferred and updated to obtain the second desensitization transfer feature corresponding to each unit to be desensitized;

[0030] The second desensitized transfer feature corresponding to each unit to be desensitized is used as the candidate desensitized semantic feature corresponding to each unit to be desensitized.

[0031] In a preferred embodiment of the present application, in the above-mentioned artificial intelligence-based data desensitization method, for each desensitized unit in the desensitized data, using the associated semantic mining unit in the semantic mining model, based on the information entity semantic features corresponding to each matching information entity corresponding to the desensitized unit, the candidate desensitized semantic features corresponding to the desensitized unit are associated mined to obtain the target desensitized semantic features corresponding to the desensitized unit, including the following steps:

[0032] For each unit to be desensitized in the data to be desensitized, using the associated semantic mining unit in the semantic mining model, based on the information entity semantic features corresponding to each matching information entity corresponding to the unit to be desensitized, the candidate desensitized semantic features corresponding to the unit to be desensitized are subjected to focused feature mining to obtain each intermediate desensitized semantic feature corresponding to the unit to be desensitized;

[0033] Superimposing the candidate desensitized semantic features corresponding to each to-be-desensitized unit in the to-be-desensitized data to obtain a superimposed desensitized semantic feature corresponding to the to-be-desensitized data;

[0034] For each unit to be desensitized in the data to be desensitized, the feature similarity between the information entity semantic feature corresponding to each matching information entity corresponding to the unit to be desensitized and the superimposed desensitized semantic feature corresponding to the data to be desensitized is determined respectively, and based on the feature similarity corresponding to each matching information entity, each intermediate desensitized semantic feature is weightedly superimposed to obtain the target desensitized semantic feature corresponding to the unit to be desensitized.

[0035] This application also provides an artificial intelligence-based data desensitization device, including:

[0036] A data acquisition module is used to acquire the data to be desensitized, wherein the data to be desensitized is text data;

[0037] A semantic feature mining module is used to mine target desensitization semantic features corresponding to each unit to be desensitized in the data to be desensitized, using a semantic mining model included in a target sensitive information recognition network and based on a pre-constructed sensitive information knowledge graph. Each unit to be desensitized has the same granularity, which includes a word, phrase, sentence, or paragraph. The target sensitive information recognition network is a pre-trained neural network, and the sensitive information knowledge graph is determined based on the relationships between each data unit in the sensitive information database.

[0038] A sensitive information identification module is configured to identify and output, for each unit to be desensitized in the data to be desensitized, a target desensitizing semantic feature corresponding to the unit to be desensitized using the sensitive information identification model included in the target sensitive information identification network, and obtain a sensitive information identification result corresponding to the unit to be desensitized, wherein the sensitive information identification result is used to reflect whether the unit to be desensitized is sensitive information;

[0039] The desensitization processing module is used to perform desensitization processing on each unit to be desensitized that belongs to sensitive information.

[0040] Based on the above, the present application further provides an electronic device, including:

[0041] memory for storing computer programs;

[0042] A processor connected to the memory is used to execute the computer program stored in the memory to implement the above-mentioned artificial intelligence-based data desensitization method.

[0043] Based on the above, the present application also provides a computer-readable storage medium, which stores a computer program. When the computer program is run, it executes the various steps of the above-mentioned artificial intelligence-based data desensitization method.

[0044] The data desensitization method and apparatus, equipment and storage medium based on artificial intelligence provided by the present application first obtain the data to be desensitized; secondly, for each unit to be desensitized in the data to be desensitized, the semantic mining model included in the target sensitive information recognition network is used to mine the target desensitized semantic features corresponding to the unit to be desensitized based on the sensitive information knowledge graph; then, for each unit to be desensitized in the data to be desensitized, the sensitive information recognition model included in the target sensitive information recognition network is used to identify and output the target desensitized semantic features corresponding to the unit to be desensitized, and obtain the sensitive information recognition result corresponding to the unit to be desensitized; finally, each unit to be desensitized belonging to sensitive information is desensitized separately. Based on the above content, since the powerful analysis and recognition ability of the target sensitive information recognition network can be utilized, the sensitive information recognition result obtained can be more reliable. Moreover, since in the process of semantic mining, it is not limited to the semantic information of the data to be desensitized itself, but also combines the relevant information in the sensitive information knowledge graph, the semantic information of the mined target desensitized semantic features is richer and more reliable, and the reliability of the obtained sensitive information recognition result can be further improved, thereby improving the relatively low reliability of data desensitization in the prior art. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] In order to make the above-mentioned objects, features and advantages of the present application more obvious and easy to understand, preferred embodiments are given below and described in detail with reference to the accompanying drawings.

[0046] Figure 1 This is a structural block diagram of the electronic device provided in an embodiment of the present application.

[0047] Figure 2 A flowchart of the artificial intelligence-based data desensitization method provided in an embodiment of the present application.

[0048] Figure 3 A schematic diagram of the graph relationship hierarchy provided in an embodiment of the present application.

[0049] Figure 4 A block diagram of an artificial intelligence-based data desensitization device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0050] To make the objectives, technical solutions, and advantages of the embodiments of the present application more clear, the technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the accompanying drawings of 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. Generally, the components of the embodiments of the present application described and shown in the drawings herein can be arranged and designed in various different configurations.

[0051] Therefore, the following detailed description of the embodiments of the present application provided in the accompanying drawings is not intended to limit the scope of the present application for protection, but merely represents selected embodiments of the present application. All other embodiments obtained by persons of ordinary skill in the art based on the embodiments in the present application without creative work are within the scope of protection of the present application.

[0052] like Figure 1 As shown, an embodiment of the present application provides an electronic device. The electronic device may include a memory, a processor, and an artificial intelligence-based data desensitization device.

[0053] In detail, the memory and the processor are electrically connected directly or indirectly to realize the transmission or interaction of data. For example, the memory and the processor can be electrically connected through one or more communication buses or signal lines. The data desensitization device based on artificial intelligence includes at least one software function module stored in the memory in the form of software or firmware. The processor is used to execute the executable computer program stored in the memory, for example, the software function module and computer program included in the data desensitization device based on artificial intelligence, so as to realize the data desensitization method based on artificial intelligence provided in the embodiment of the present application.

[0054] Optionally, the memory may be, but is not limited to, random access memory (RAM), read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), etc.

[0055] Furthermore, the processor may be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), a system on chip (SoC), etc.; it may also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.

[0056] I understand. Figure 1The structure shown is only for illustration, and the electronic device may also include Figure 1 More or fewer components than shown, or with Figure 1 The different configurations shown may, for example, further include a communication unit for exchanging information with other devices.

[0057] Combine Figure 2 , the embodiment of the present application further provides an artificial intelligence-based data desensitization method applicable to the above-mentioned electronic device. Among them, the method steps defined in the process related to the artificial intelligence-based data desensitization method can be implemented by the electronic device.

[0058] The following will Figure 2 The specific process shown is explained in detail.

[0059] Step S110: Obtain the data to be desensitized.

[0060] In an embodiment of the present application, the electronic device may obtain data to be desensitized, wherein the data to be desensitized is text data, such as "Zhang San's age is 10 years old."

[0061] Step S120: For each unit to be desensitized in the data to be desensitized, the semantic mining model included in the target sensitive information recognition network is used to mine the target desensitization semantic features corresponding to the unit to be desensitized based on the pre-built sensitive information knowledge graph.

[0062] In an embodiment of the present application, the electronic device can, for each unit to be desensitized in the data to be desensitized, use the semantic mining model included in the target sensitive information identification network, based on the pre-built sensitive information knowledge graph, to mine the target desensitization semantic features corresponding to the unit to be desensitized. Wherein, the granularity of each unit to be desensitized is the same, and the granularity includes words, phrases, sentences or paragraphs. The target sensitive information identification network belongs to a pre-trained neural network (for example, trained based on sample data and corresponding labels, and the labels are used to indicate whether each unit in the sample data belongs to sensitive information). The sensitive information knowledge graph is determined based on the relationship between each data unit and each data unit in the sensitive information database. For example, each data unit corresponds to an information entity in the sensitive information knowledge graph, and the information entities with a correlation relationship between the corresponding data units are connected together, and the information entities without a correlation relationship between the corresponding data units are not connected together. The correlation between the data units can be determined based on the co-occurrence frequency of the data units in the corresponding expected database. If the co-occurrence frequency is greater than a preset number of times, it can be considered to have a correlation. Otherwise, it is not considered to have a correlation.

[0063] Step S130: For each unit to be desensitized in the data to be desensitized, the sensitive information recognition model included in the target sensitive information recognition network is used to identify and output the target desensitizing semantic features corresponding to the unit to be desensitized, and obtain the sensitive information recognition result corresponding to the unit to be desensitized.

[0064] In an embodiment of the present application, the electronic device can use the sensitive information recognition model included in the target sensitive information recognition network to identify and output the target desensitized semantic features corresponding to each unit to be desensitized in the data to be desensitized, and obtain the sensitive information recognition result corresponding to the unit to be desensitized. The sensitive information recognition result is used to reflect whether the unit to be desensitized belongs to sensitive information. Exemplarily, the sensitive information recognition model may include a fully connected unit and a classification unit, wherein, after the fully connected unit processes the target desensitized semantic features to obtain the corresponding fully connected semantic features, the classification function (such as softmax function) included in the classification unit can process the fully connected semantic features to obtain the sensitive information recognition result.

[0065] Step S140 , performing desensitization processing on each unit to be desensitized that belongs to sensitive information.

[0066] In an embodiment of the present application, the electronic device can perform desensitization processing on each unit to be desensitized that belongs to sensitive information, that is, make corresponding modifications or replacements, such as replacing the "10" in "Zhang San's age is 10 years old" with "XX", that is, "Zhang San's age is XX years old".

[0067] Based on the above, by leveraging the powerful analytical and recognition capabilities of the target sensitive information identification network, the resulting sensitive information identification results can be made more reliable. Furthermore, since the semantic mining process is not limited to the semantic information of the data to be desensitized, but also incorporates relevant information from the sensitive information knowledge graph, the semantic information of the target desensitized semantic features mined is richer and more reliable, further improving the reliability of the sensitive information identification results, thereby addressing the relatively low reliability of data desensitization in existing technologies.

[0068] What needs to be explained about step S120 is that the specific method of mining the target desensitized semantic features corresponding to the unit to be desensitized is not limited and can be selected according to actual needs.

[0069] For example, in an alternative embodiment, in order to make the reliability of the mined target desensitized semantic features higher, the above-mentioned step S120 may further include step S121, step S122, step S123 and step S124, and the specific content of each step is described as follows.

[0070] Step S121: For each unit to be desensitized in the data to be desensitized, determine each information entity with the same name as the unit to be desensitized in the pre-constructed sensitive information knowledge graph, and obtain at least one matching information entity corresponding to the unit to be desensitized.

[0071] In an embodiment of the present application, for each unit to be desensitized in the data to be desensitized, each information entity with the same name as the unit to be desensitized can be determined in a pre-constructed sensitive information knowledge graph to obtain at least one matching information entity corresponding to the unit to be desensitized (the same name may have different meanings, and therefore, may have different information entities. For example, for "high-tech zone", there are corresponding high-tech zones in multiple prefecture-level cities, that is, there are multiple attribute information, and therefore, there are multiple corresponding information entities, which are then characterized separately). Each of the information entities corresponds to a data unit in the sensitive information database.

[0072] Step S122: For each of the matching information entities, the first semantic mining unit in the semantic mining model included in the target sensitive information recognition network is used to mine the information entity semantic features corresponding to the matching information entity in the sensitive information knowledge graph.

[0073] In an embodiment of the present application, after determining the matching information entity, the first semantic mining unit in the semantic mining model included in the target sensitive information identification network can be used to mine the information entity semantic features corresponding to the matching information entity in the sensitive information knowledge graph for each matching information entity, that is, to mine the semantics of the matching information entity itself and the related semantics in the sensitive information knowledge graph, so as to improve the richness of the semantic information.

[0074] Step S123: for each unit to be desensitized in the data to be desensitized, using the second semantic mining unit in the semantic mining model, mining candidate desensitizing semantic features corresponding to the unit to be desensitized in the data to be desensitized.

[0075] In an embodiment of the present application, for each unit to be desensitized in the data to be desensitized, the second semantic mining unit in the semantic mining model can be used to mine candidate desensitized semantic features corresponding to the unit to be desensitized in the data to be desensitized, that is, to mine the semantics of the unit to be desensitized itself and the related semantics in the data to be desensitized. The first semantic mining unit is different from the second semantic mining unit, and the specific processing process is described later.

[0076] Step S124: For each unit to be desensitized in the data to be desensitized, the association semantic mining unit in the semantic mining model is used to perform association mining on the candidate desensitized semantic features corresponding to the unit to be desensitized based on the information entity semantic features corresponding to each matching information entity corresponding to the unit to be desensitized, so as to obtain the target desensitized semantic features corresponding to the unit to be desensitized.

[0077] In an embodiment of the present application, for each unit to be desensitized in the data to be desensitized, the association semantic mining unit in the semantic mining model is used to perform association mining on the candidate desensitized semantic features corresponding to the unit to be desensitized based on the information entity semantic features corresponding to each matching information entity corresponding to the unit to be desensitized, thereby obtaining the target desensitized semantic features corresponding to the unit to be desensitized. In other words, the information entity semantic features corresponding to the matching information entity can be integrated into the candidate desensitized semantic features corresponding to the unit to be desensitized, so that the semantic representation ability of the obtained target desensitized semantic features is better and the real semantic information can be expressed more accurately.

[0078] It is understood that in step S122 above, the specific method for mining the semantic features of the information entity corresponding to the matching information entity in the sensitive information knowledge graph is not limited and can be selected based on actual needs. For example, in an alternative embodiment, step S122 above may further include steps S122a, S122b, and S122c, the specific contents of each step being described below.

[0079] Step S122a: For each of the matching information entities, use the first semantic mining unit in the semantic mining model included in the target sensitive information recognition network to perform semantic mining on the entity attribute data corresponding to the matching information entity, and output the candidate entity semantic features corresponding to the matching information entity.

[0080] In an embodiment of the present application, for each matching information entity, a first semantic mining unit in a semantic mining model included in a target sensitive information recognition network can be used to perform semantic mining on the entity attribute data corresponding to the matching information entity, thereby outputting candidate entity semantic features corresponding to the matching information entity. For example, word embedding processing can be performed on the entity attribute data corresponding to the matching information entity to obtain corresponding candidate entity semantic features.

[0081] Step S122b: extracting a sensitive information local graph centered on the matching information entity from the sensitive information knowledge graph.

[0082] In an embodiment of the present application, a local sensitive information graph centered on the matching information entity can be extracted from the sensitive information knowledge graph. The graph scale of the local sensitive information graph matches the pre-configured reference graph scale, and the graph scale is correlated with the maximum path length of the shortest path between the matching information entity and other information entities, for example, when the maximum path length is no greater than a preset length.

[0083] Step S122c: The candidate entity semantic features corresponding to each other information entity in the sensitive information local graph are integrated into the candidate entity semantic features corresponding to the matching information entity to obtain the information entity semantic features corresponding to the matching information entity.

[0084] In an embodiment of the present application, after obtaining the candidate entity semantic features, the candidate entity semantic features corresponding to each other information entity in the sensitive information local graph can be fused with the candidate entity semantic features corresponding to the matching information entity to obtain the information entity semantic features corresponding to the matching information entity. The candidate entity semantic features corresponding to each other information entity are obtained by semantic mining using the first semantic mining unit, i.e., by performing corresponding word embedding processing.

[0085] It is understood that in step S122c above, the specific method for fusing the candidate entity semantic features corresponding to each other information entity in the sensitive information local graph with the candidate entity semantic features corresponding to the matching information entity is not limited and can be selected based on actual needs. For example, in an alternative embodiment, to ensure the reliability of the fused information entity semantic features, step S122c above may further include steps c1 and c2, the specific contents of each step being described below.

[0086] Step c1: for each information entity in the sensitive information local graph, based on the path length of the shortest path between the information entity and the matching information entity, determine the graph relationship level corresponding to the information entity.

[0087] In the embodiment of the present application, for each information entity in the sensitive information local graph, the graph relationship level corresponding to the information entity can be determined based on the path length of the shortest path between the information entity and the matching information entity. Among them, the graph relationship level corresponding to the matching information entity is the lowest (such as 0), and the graph relationship level corresponding to the other information entity with the largest path length is the highest (such as 3). The specific representation of the graph relationship level is as follows: Figure 3 shown.

[0088] Step c2, transferring the candidate entity semantic features corresponding to each information entity along the direction from the highest to the lowest level of the graph relationship hierarchy, so that the candidate entity semantic features corresponding to each other information entity in the sensitive information local graph are integrated into the candidate entity semantic features corresponding to the matching information entity to obtain the information entity semantic features corresponding to the matching information entity.

[0089] In the embodiment of the present application, after determining the corresponding graph relationship hierarchy, the candidate entity semantic features corresponding to each information entity can be transferred along the direction from the highest to the lowest level of the graph relationship hierarchy (such as the direction from information entity 8 to the matching information entity), so that the candidate entity semantic features corresponding to each other information entity in the sensitive information local graph are fused into the candidate entity semantic features corresponding to the matching information entity to obtain the information entity semantic features corresponding to the matching information entity, that is, the fusion Figure 3 The candidate entity semantic features corresponding to each information entity shown.

[0090] It is understood that in the above step c2, the specific method of transmitting the candidate entity semantic features corresponding to each information entity is not limited and can be selected according to actual needs. For example, in an alternative embodiment, in order to ensure the reliability of the semantic feature transmission, the above step c2 may further include the following content:

[0091] First, for each information entity at the highest level of the graph relationship, the semantic features of the candidate entities corresponding to the information entity can be mined for focused features to obtain the focused features corresponding to the information entity. For example, Figure 3 For information entity 8 in the example, the candidate entity semantic feature corresponding to information entity 8 may be multiplied by the transposed result of the candidate entity semantic feature. Then, based on the multiplication result, the candidate entity semantic feature may be weighted to obtain the corresponding focus feature.

[0092] Secondly, for each information entity in each graph relationship layer other than the highest graph relationship layer, determine whether there are multiple other information entities connected to the information entity in each other information entity at a higher layer, and when there are not multiple other information entities connected to the information entity, perform focused feature mining on the candidate entity semantic features corresponding to the information entity based on the focused features corresponding to one of the other connected information entities (for example, the focused feature can be multiplied by the transposed result of the semantic features of the candidate entity, and then the semantic features of the candidate entity can be weighted based on the multiplication result) to obtain the focused features corresponding to the information entity, and, when there are multiple other information entities connected to the information entity, determine whether there is a connection relationship between the other connected information entities (such as information entities 5 and information entity 6 have a connection relationship), and based on the focus features corresponding to each other information entity that does not have a connection relationship, the candidate entity semantic features corresponding to the information entity are respectively mined to obtain the focused local features corresponding to the information entity, and, after superimposing the focus features corresponding to other information entities with a connection relationship, the candidate entity semantic features corresponding to the information entity are mined based on the features after the superimposition operation to obtain the focused local features corresponding to the information entity, and then superimposing each focused local feature corresponding to the information entity (such as mean superposition, etc.) to obtain the focused features corresponding to the information entity; It should be noted that, in each other information entity at a higher level, there may be no other information entities connected to the information entity, such as Figure 3 Information entity 1, information entity 4, information entity 5 and information entity 6 in the data, therefore, focus feature mining can also be performed on information entity 8 to obtain the corresponding focus features;

[0093] Finally, after obtaining the focus feature corresponding to the matching information entity, the focus feature corresponding to the matching information entity and the candidate entity semantic feature corresponding to the matching information entity are superimposed to obtain the information entity semantic feature corresponding to the matching information entity.

[0094] It is understandable that, in the above step S123, the specific method of mining the candidate desensitized semantic features corresponding to the unit to be desensitized in the data to be desensitized is not limited and can be selected according to actual needs. For example, in an alternative embodiment, in order to ensure that the mined candidate desensitized semantic features have better semantic representation capabilities, the above step S123 may include:

[0095] First, for each unit to be desensitized in the data to be desensitized, embedding processing (such as word embedding processing) can be performed on the unit to be desensitized using the embedding subunit in the second semantic mining unit in the semantic mining model, and a desensitized embedding feature corresponding to the unit to be desensitized can be output;

[0096] Secondly, the desensitized embedded features corresponding to each unit to be desensitized can be transferred and updated along the direction from the first unit to be desensitized to the last unit to be desensitized in the data to be desensitized, so as to obtain the first desensitized transfer feature corresponding to each unit to be desensitized; illustratively, in the process of transfer and update, the desensitized embedded features corresponding to the first unit to be desensitized are used as the corresponding first desensitized transfer feature, and for other units to be desensitized, the desensitized embedded features corresponding to the other units to be desensitized can be focused feature mining based on the first desensitized transfer feature corresponding to the previous unit to be desensitized, so as to obtain the corresponding first desensitized transfer feature;

[0097] Then, along the direction from the last unit to be desensitized in the data to be desensitized to the first unit to be desensitized, the first desensitization transfer feature corresponding to each unit to be desensitized is transferred and updated to obtain the second desensitization transfer feature corresponding to each unit to be desensitized, as described above;

[0098] Finally, the second desensitized transfer feature corresponding to each unit to be desensitized can be used as a candidate desensitized semantic feature corresponding to each unit to be desensitized.

[0099] It is understandable that, in the above step S124, the specific method of performing association mining on the candidate desensitized semantic features corresponding to the to-be-desensitized unit based on the information entity semantic features corresponding to each matching information entity corresponding to the to-be-desensitized unit is not limited and can be selected according to actual needs. For example, in an alternative embodiment, in order to make the obtained target desensitized semantic features have better semantic representation ability, the above step S124 can further include:

[0100] First, for each unit to be desensitized in the data to be desensitized, the associated semantic mining unit in the semantic mining model can be used to perform focused feature mining on the candidate desensitized semantic features corresponding to the unit to be desensitized based on the information entity semantic features corresponding to each matching information entity corresponding to the unit to be desensitized, so as to obtain each intermediate desensitized semantic feature corresponding to the unit to be desensitized (one-to-one correspondence with the matching information entity);

[0101] Secondly, the candidate desensitized semantic features corresponding to each to-be-desensitized unit in the to-be-desensitized data may be superimposed to obtain the superimposed desensitized semantic features corresponding to the to-be-desensitized data;

[0102] Then, for each desensitized unit in the data to be desensitized, the feature similarity (such as cosine similarity, etc.) between the information entity semantic features corresponding to each matching information entity corresponding to the unit to be desensitized and the superimposed desensitized semantic features corresponding to the data to be desensitized is determined respectively, and, based on the feature similarity corresponding to each matching information entity, each intermediate desensitized semantic feature is weightedly superimposed to obtain the target desensitized semantic feature corresponding to the unit to be desensitized.

[0103] Combine Figure 4 The present application also provides an artificial intelligence-based data desensitization device applicable to the above-mentioned electronic device. The artificial intelligence-based data desensitization device may include a data acquisition module, a semantic feature mining module, a sensitive information identification module, and a desensitization processing module.

[0104] In detail, the data acquisition module can be used to obtain the data to be desensitized, wherein the data to be desensitized is text data. In the embodiment of the present application, the data acquisition module can be used to execute Figure 2 As shown in step S110, for the relevant content of the data acquisition module, reference may be made to the above description of step S110.

[0105] In detail, the semantic feature mining module can be used to mine the target desensitized semantic features corresponding to each unit to be desensitized in the data to be desensitized, using the semantic mining model included in the target sensitive information recognition network, based on the pre-built sensitive information knowledge graph, wherein each unit to be desensitized has the same granularity, which includes words, phrases, sentences or paragraphs, the target sensitive information recognition network is a pre-trained neural network, and the sensitive information knowledge graph is determined based on the relationship between each data unit and each data unit in the sensitive information database. In an embodiment of the present application, the semantic feature mining module can be used to perform Figure 2 As shown in step S120, for the relevant content of the semantic feature mining module, reference can be made to the above description of step S120.

[0106] In detail, the sensitive information identification module can be used to identify and output the target desensitized semantic features corresponding to each unit to be desensitized in the data to be desensitized, using the sensitive information identification model included in the target sensitive information identification network, to obtain the sensitive information identification result corresponding to the unit to be desensitized, wherein the sensitive information identification result is used to reflect whether the unit to be desensitized belongs to sensitive information. In the embodiment of the present application, the sensitive information identification module can be used to perform Figure 2 As shown in step S130, for the relevant content of the sensitive information identification module, please refer to the description of step S130 above.

[0107] In detail, the desensitization processing module can be used to perform desensitization processing on each unit to be desensitized that belongs to sensitive information. Figure 2 As shown in step S140, for the relevant content of the desensitization processing module, please refer to the previous description of step S140.

[0108] In an embodiment of the present application, corresponding to the above-mentioned artificial intelligence-based data desensitization method applied to the electronic device, a computer-readable storage medium is also provided, in which a computer program is stored. When the computer program is run, the various steps of the artificial intelligence-based data desensitization method are executed.

[0109] Among them, the steps executed when the aforementioned computer program is running will not be described here one by one. Please refer to the previous explanation of the data desensitization method based on artificial intelligence.

[0110] In summary, the data desensitization method and apparatus, equipment and storage medium based on artificial intelligence provided by the present application, first, obtain the data to be desensitized; secondly, for each unit to be desensitized in the data to be desensitized, the semantic mining model included in the target sensitive information recognition network is used to mine the target desensitized semantic features corresponding to the unit to be desensitized based on the sensitive information knowledge graph; then, for each unit to be desensitized in the data to be desensitized, the sensitive information recognition model included in the target sensitive information recognition network is used to identify and output the target desensitized semantic features corresponding to the unit to be desensitized, and obtain the sensitive information recognition result corresponding to the unit to be desensitized; finally, each unit to be desensitized belonging to sensitive information is desensitized separately. Based on the above content, since the powerful analysis and recognition capabilities of the target sensitive information recognition network can be utilized, the obtained sensitive information recognition results can be more reliable. Moreover, since the semantic mining process is not limited to the semantic information of the data to be desensitized itself, but also combines the relevant information in the sensitive information knowledge graph, the semantic information of the target desensitized semantic features to be mined is richer and more reliable, and the reliability of the sensitive information identification results obtained can be further improved, thereby improving the reliability of data desensitization, and further improving the problem of relatively low reliability of data desensitization in the existing technology.

[0111] In the several embodiments provided in the embodiments of the present application, it should be understood that the disclosed devices and methods can also be implemented in other ways. The device and method embodiments described above are merely schematic. For example, the flowcharts and block diagrams in the accompanying drawings show the possible architectures, functions and operations of the devices, methods and computer program products according to the multiple embodiments of the present application. In this regard, each box in the flowchart or block diagram can represent a module, a program segment or a part of the code, and the module, program segment or a part of the code contains one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in an order different from that marked in the accompanying drawings. For example, two consecutive boxes can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flowchart, and the combination of boxes in the block diagram and / or flowchart, can be implemented with a dedicated hardware-based system that performs the specified function or action, or can be implemented with a combination of dedicated hardware and computer instructions.

[0112] In addition, the functional modules in each embodiment of the present application can be integrated together to form an independent part, or each module can exist independently, or two or more modules can be integrated to form an independent part.

[0113] If the functions are implemented in the form of software function modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or partly contributed to the prior art or part of the technical solution can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, electronic device, or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk. It should be noted that, in this article, the terms "include", "comprise" or any other variant 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 a process, method, article or device. Without further constraints, an element defined by the phrase "comprises a..." does not preclude the existence of additional identical elements in the process, method, article or apparatus that includes the element.

[0114] The above description is merely a preferred embodiment of the present application and is not intended to limit the present application. Various modifications and variations are possible for those skilled in the art. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present application shall be included within the scope of protection of the present application.

Claims

1. A data desensitization method based on artificial intelligence, characterized in that: include: Obtaining data to be desensitized, wherein the data to be desensitized is text data; For each unit to be desensitized in the data to be desensitized, in the pre-constructed sensitive information knowledge graph, each information entity with the same name as the unit to be desensitized is determined, and at least one matching information entity corresponding to the unit to be desensitized is obtained, wherein each of the information entities corresponds to a data unit in the sensitive information database; for each matching information entity, the first semantic mining unit in the semantic mining model included in the target sensitive information recognition network is used to mine the information entity semantic features corresponding to the matching information entity in the sensitive information knowledge graph; for each unit to be desensitized in the data to be desensitized, the second semantic mining unit in the semantic mining model is used to mine the semantic features of the information entity corresponding to the matching information entity in the data to be desensitized According to the data, candidate desensitized semantic features corresponding to the unit to be desensitized are mined; for each unit to be desensitized in the data to be desensitized, the association semantic mining unit in the semantic mining model is used to perform association mining on the candidate desensitized semantic features corresponding to the unit to be desensitized based on the information entity semantic features corresponding to each matching information entity corresponding to the unit to be desensitized, so as to obtain the target desensitized semantic features corresponding to the unit to be desensitized, wherein the granularity of each unit to be desensitized is the same, and the granularity includes words, phrases, sentences or paragraphs, the target sensitive information recognition network is a pre-trained neural network, and the sensitive information knowledge graph is determined based on the relationship between each data unit and each data unit in the sensitive information database; For each unit to be desensitized in the data to be desensitized, using the sensitive information recognition model included in the target sensitive information recognition network, identify and output the target desensitization semantic features corresponding to the unit to be desensitized, and obtain a sensitive information recognition result corresponding to the unit to be desensitized, wherein the sensitive information recognition result is used to reflect whether the unit to be desensitized is sensitive information; Desensitizing is performed on each unit of sensitive information to be desensitized.

2. The data desensitization method based on artificial intelligence according to claim 1 is characterized in that: The step of mining, for each of the matching information entities, the semantic features of the information entity corresponding to the matching information entity in the sensitive information knowledge graph using the first semantic mining unit in the semantic mining model included in the target sensitive information identification network includes: For each of the matching information entities, using the first semantic mining unit in the semantic mining model included in the target sensitive information recognition network, semantic mining is performed on the entity attribute data corresponding to the matching information entity, and the candidate entity semantic features corresponding to the matching information entity are output; Extracting a local sensitive information graph centered on the matching information entity from the sensitive information knowledge graph, wherein the graph scale of the local sensitive information graph matches a pre-configured reference graph scale, and the graph scale is correlated with the maximum path length in the shortest paths between the matching information entity and each other information entity; The candidate entity semantic features corresponding to each other information entity in the sensitive information local graph are integrated into the candidate entity semantic features corresponding to the matching information entity to obtain the information entity semantic features corresponding to the matching information entity, wherein the candidate entity semantic features corresponding to each other information entity are obtained by semantic mining using the first semantic mining unit.

3. The data desensitization method based on artificial intelligence according to claim 2, characterized in that: The step of fusing the candidate entity semantic features corresponding to each other information entity in the sensitive information local graph with the candidate entity semantic features corresponding to the matching information entity to obtain the information entity semantic features corresponding to the matching information entity includes: For each information entity in the sensitive information local graph, based on the path length of the shortest path between the information entity and the matching information entity, determine the graph relationship hierarchy corresponding to the information entity, wherein the matching information entity corresponds to the lowest graph relationship hierarchy, and the other information entities with the largest path length correspond to the highest graph relationship hierarchy; The candidate entity semantic features corresponding to each information entity are transferred along the direction from the highest to the lowest level of the graph relationship hierarchy, so that the candidate entity semantic features corresponding to each other information entity in the sensitive information local graph are integrated into the candidate entity semantic features corresponding to the matching information entity to obtain the information entity semantic features corresponding to the matching information entity.

4. The data desensitization method based on artificial intelligence according to claim 3 is characterized in that: The step of transferring the candidate entity semantic features corresponding to each information entity along the direction from the highest to the lowest level of the graph relationship hierarchy, so that the candidate entity semantic features corresponding to each other information entity in the sensitive information local graph are integrated into the candidate entity semantic features corresponding to the matching information entity, to obtain the information entity semantic features corresponding to the matching information entity, includes: For each information entity at the highest level of the graph relationship, performing focused feature mining on the semantic features of candidate entities corresponding to the information entity to obtain the focused features corresponding to the information entity; For each information entity in each graph relationship layer other than the highest graph relationship layer, determine whether there are multiple other information entities connected to the information entity in each other information entity at a higher layer, and when there are not multiple other information entities connected to the information entity, perform focused feature mining on the semantic features of the candidate entity corresponding to the information entity based on the focused features corresponding to one of the other connected information entities to obtain the focused features corresponding to the information entity, and, when there are multiple other information entities connected to the information entity, determine whether there is a connection relationship between the other connected information entities, and perform focused feature mining on the semantic features of the candidate entity corresponding to the information entity based on the focused features corresponding to each of the other information entities that do not have a connection relationship to obtain the focused local features corresponding to the information entity, and, after superimposing the focused features corresponding to the other information entities that have a connection relationship, perform focused feature mining on the semantic features of the candidate entity corresponding to the information entity based on the features after the superimposed operation to obtain the focused local features corresponding to the information entity, and then perform superimposed operation on each of the focused local features corresponding to the information entity to obtain the focused features corresponding to the information entity; After obtaining the focus feature corresponding to the matching information entity, the focus feature corresponding to the matching information entity and the candidate entity semantic feature corresponding to the matching information entity are superimposed to obtain the information entity semantic feature corresponding to the matching information entity.

5. The data desensitization method based on artificial intelligence according to claim 1, characterized in that: The step of mining, for each unit to be desensitized in the data to be desensitized, a candidate desensitized semantic feature corresponding to the unit to be desensitized in the data to be desensitized using the second semantic mining unit in the semantic mining model includes: For each unit to be desensitized in the data to be desensitized, embedding the unit to be desensitized using the embedding subunit in the second semantic mining unit in the semantic mining model, and outputting a desensitized embedding feature corresponding to the unit to be desensitized; Along the direction from the first to-be-desensitized unit to the last to-be-desensitized unit in the to-be-desensitized data, the desensitized embedded feature corresponding to each to-be-desensitized unit is transferred and updated to obtain a first desensitized transfer feature corresponding to each to-be-desensitized unit; Along the direction from the last unit to be desensitized in the data to be desensitized to the first unit to be desensitized, the first desensitization transfer feature corresponding to each unit to be desensitized is transferred and updated to obtain the second desensitization transfer feature corresponding to each unit to be desensitized; The second desensitized transfer feature corresponding to each unit to be desensitized is used as the candidate desensitized semantic feature corresponding to each unit to be desensitized.

6. The data desensitization method based on artificial intelligence according to claim 1, characterized in that: The step of performing association mining on candidate desensitized semantic features corresponding to each to-be-desensitized unit in the to-be-desensitized data using the association semantic mining unit in the semantic mining model based on the information entity semantic features corresponding to each matching information entity corresponding to the to-be-desensitized unit to obtain the target desensitized semantic features corresponding to the to-be-desensitized unit includes: For each unit to be desensitized in the data to be desensitized, using the associated semantic mining unit in the semantic mining model, based on the information entity semantic features corresponding to each matching information entity corresponding to the unit to be desensitized, the candidate desensitized semantic features corresponding to the unit to be desensitized are subjected to focused feature mining to obtain each intermediate desensitized semantic feature corresponding to the unit to be desensitized; Superimposing the candidate desensitized semantic features corresponding to each to-be-desensitized unit in the to-be-desensitized data to obtain a superimposed desensitized semantic feature corresponding to the to-be-desensitized data; For each unit to be desensitized in the data to be desensitized, the feature similarity between the information entity semantic feature corresponding to each matching information entity corresponding to the unit to be desensitized and the superimposed desensitized semantic feature corresponding to the data to be desensitized is determined respectively, and based on the feature similarity corresponding to each matching information entity, each intermediate desensitized semantic feature is weightedly superimposed to obtain the target desensitized semantic feature corresponding to the unit to be desensitized.

7. A data desensitization device based on artificial intelligence, characterized in that: include: A data acquisition module is used to acquire the data to be desensitized, wherein the data to be desensitized is text data; The semantic feature mining module is used to determine, for each unit to be desensitized in the data to be desensitized, each information entity with the same name as the unit to be desensitized in the pre-built sensitive information knowledge graph, and obtain at least one matching information entity corresponding to the unit to be desensitized, wherein each of the information entities corresponds to a data unit in the sensitive information database; for each matching information entity, the first semantic mining unit in the semantic mining model included in the target sensitive information recognition network is used to mine the information entity semantic features corresponding to the matching information entity in the sensitive information knowledge graph; for each unit to be desensitized in the data to be desensitized, the second semantic mining unit in the semantic mining model is used to mine the semantic features of the information entity corresponding to the matching information entity in the sensitive information knowledge graph; From the data to be desensitized, candidate desensitized semantic features corresponding to the unit to be desensitized are mined; for each unit to be desensitized in the data to be desensitized, the association semantic mining unit in the semantic mining model is used to perform association mining on the candidate desensitized semantic features corresponding to the unit to be desensitized based on the information entity semantic features corresponding to each matching information entity corresponding to the unit to be desensitized, so as to obtain the target desensitized semantic features corresponding to the unit to be desensitized, wherein each unit to be desensitized has the same granularity, which includes words, phrases, sentences or paragraphs, the target sensitive information recognition network is a pre-trained neural network, and the sensitive information knowledge graph is determined based on the relationship between each data unit and each data unit in the sensitive information database; A sensitive information identification module is configured to identify and output, for each unit to be desensitized in the data to be desensitized, a target desensitizing semantic feature corresponding to the unit to be desensitized using the sensitive information identification model included in the target sensitive information identification network, and obtain a sensitive information identification result corresponding to the unit to be desensitized, wherein the sensitive information identification result is used to reflect whether the unit to be desensitized is sensitive information; The desensitization processing module is used to perform desensitization processing on each unit to be desensitized that belongs to sensitive information.

8. An electronic device, characterized in that: include: Memory for storing computer programs; A processor connected to the memory, used to execute the computer program stored in the memory to implement the artificial intelligence-based data desensitization method described in any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, which, when running, executes the artificial intelligence-based data desensitization method according to any one of claims 1 to 6.

Images