Data desensitization method based on large model and related device

Abstract

The invention provides a data desensitization method based on a large model and a related device, which can perform sensitivity evaluation on a data acquisition request and obtain a first sensitivity score. For the original response data corresponding to the data acquisition request, the sensitivity evaluation of the structure context can be performed according to the nested information of the response field name in the storage data structure, and the second sensitivity score is obtained. And identifying potential sensitive content in the response value by adopting a semantic sensitivity evaluation model, and obtaining a third sensitivity score. After multi-dimensional sensitivity evaluation is carried out, the sensitivity score of the data acquisition request, the sensitivity score of the structure context and the sensitivity score of the response value can be integrated, whether the response value relates to sensitive information or not can be comprehensively judged, and data desensitization is carried out on the response value related to the sensitive information. Therefore, the sensitive information in the response data can be comprehensively identified and data desensitization can be performed, so that the data leakage risk can be reduced, and the data security can be improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data processing, and particularly relates to a data desensitization method based on a large model and a related device. BACKGROUND

[0002] Under the background of the continuous improvement of the current digitalization and informatization level, more and more electronic data are collected, transmitted, stored and used. These data may involve sensitive information such as account password, token, residence address, contact phone number, etc. In order to ensure information security, the server storing sensitive information must perform data desensitization processing on the response data when responding to the data acquisition request issued by the client device. Through data deformation, data hiding and other ways, the sensitive information in the response data can be effectively concealed, so as to prevent the leakage of sensitive information.

[0003] However, the inventors have found that the prior art is difficult to comprehensively identify sensitive information in the response data, which is prone to data leakage and has the problem of low data security. SUMMARY

[0004] The purpose of the present application is to at least solve one of the above technical defects, in particular the technical defect of low data security in the prior art.

[0005] In a first aspect, an embodiment of the present application provides a data desensitization method based on a large model, comprising:

[0006] In response to a data acquisition request sent by a request device, determining original response data; the original response data includes at least one response field name and a response value corresponding to each response field name;

[0007] Using a sensitive field library to evaluate the sensitivity of the data acquisition request, and obtaining a first sensitivity score;

[0008] Determining the nesting information of a target response field name in a storage data structure, and evaluating the sensitivity according to the nesting information to obtain a second sensitivity score corresponding to the target response field name; wherein the target response field name is any one of the at least one response field name;

[0009] Using a pre-trained semantic sensitivity evaluation model to evaluate the sensitivity of a target response value, and obtaining a third sensitivity score corresponding to the target response field name; wherein the target response value is a response value corresponding to the target response field name;

[0010] According to the first sensitivity score, the second sensitivity score corresponding to the target response field name and the third sensitivity score corresponding to the target response field name, it is judged whether the target response value is sensitive information;

[0011] If the target response value is sensitive information, the target response value in the original response data is subjected to data desensitization processing, and desensitized response data is obtained;

[0012] The desensitized response data is returned to the request device.

[0013] In some embodiments, the data acquisition request includes a request path and a query field name, and the first sensitivity score includes a path sensitivity score and a query field sensitivity score;

[0014] The sensitivity of the data acquisition request is evaluated using the sensitive field library, and a first sensitivity score is obtained, which includes:

[0015] According to the matching result of the preset sensitive path mode and the request path, the path sensitivity score is determined;

[0016] According to the matching result of the sensitive field library and the query field name, the query field sensitivity score is determined.

[0017] In some embodiments, the sensitive field library includes a complete sensitive field and a partial sensitive field;

[0018] According to the matching result of the sensitive field library and the query field name, the query field sensitivity score is determined, which includes:

[0019] If the query field name matches the complete sensitive field, the query field sensitivity score is determined as the maximum value;

[0020] If the query field name matches the partial sensitive field, a proportional matching score and a position matching score are determined respectively, and the query field sensitivity score is calculated based on the proportional matching score and the position matching score;

[0021] Wherein, the proportional matching score is positively correlated with the sensitive proportion, and the sensitive proportion is the proportion of the length of the partial sensitive field to the length of the query field name; the position matching score is associated with the position of the partial sensitive field in the query field name.

[0022] In some embodiments, the nested information includes a nested path and a nested depth;

[0023] According to the sensitivity evaluation of the nested information, a second sensitivity score corresponding to the target response field name is obtained, which includes:

[0024] The sensitive field library is used to perform field matching on the nested path, and a path matching result is obtained;

[0025] According to the path matching result and the nesting depth, a second sensitivity score corresponding to the target response field name is determined; wherein the second sensitivity score is positively correlated with the nesting depth.

[0026] In some embodiments, the determining, according to the first sensitivity score, the second sensitivity score corresponding to the target response field name, and the third sensitivity score corresponding to the target response field name, whether the target response value is sensitive information, comprises:

[0027] The first sensitivity score, the second sensitivity score corresponding to the target response field name, and the third sensitivity score corresponding to the target response field name are weighted and summed to obtain a target sensitivity score;

[0028] If the target sensitivity score is greater than a first preset score, it is determined that the target response value is sensitive information, and the original sensitive level is high;

[0029] If the target sensitivity score is less than or equal to the first preset score and greater than a second preset score, it is determined that the target response value is sensitive information, and the original sensitive level is medium;

[0030] If the target sensitivity score is less than or equal to the second preset score and greater than a third preset score, it is determined that the target response value is sensitive information, and the original sensitive level is low;

[0031] If the target sensitivity score is less than or equal to the third preset score, it is determined that the target response value is non-sensitive information.

[0032] In some embodiments, the data desensitization processing of the target response value in the original response data comprises:

[0033] According to the original sensitive level corresponding to the target response value, a target sensitive level is determined;

[0034] If the target sensitive level is high, a preset standardized placeholder is used to replace the target response value;

[0035] If the target sensitive level is medium, the target response value is partially masked, and a description identifier is added according to the data type of the target response value;

[0036] If the target sensitive level is low, the target response value is generalized.

[0037] In some embodiments, the determining, according to the original sensitive level corresponding to the target response value, a target sensitive level, comprises:

[0038] The source interface of the data acquisition request is obtained;

[0039] If the source interface is a high-risk interface, the original sensitivity level corresponding to the target response value is level-up adjusted to obtain the target sensitivity level;

[0040] If the source interface is a low-risk interface, the original sensitivity level corresponding to the target response value is level-down adjusted to obtain the target sensitivity level.

[0041] In a second aspect, an embodiment of the present application provides a data desensitization device based on a large model, comprising:

[0042] An original response data determination module is configured to determine original response data in response to a data acquisition request sent by a request device; the original response data comprises at least one response field name and a response value corresponding to each response field name;

[0043] A first evaluation module is configured to evaluate the sensitivity of the data acquisition request by using a sensitive field library, and obtain a first sensitivity score;

[0044] A second evaluation module is configured to determine the nesting information of a target response field name in a storage data structure, and evaluate the sensitivity according to the nesting information to obtain a second sensitivity score corresponding to the target response field name; wherein the target response field name is any one of the at least one response field name;

[0045] A third evaluation module is configured to evaluate the sensitivity of a target response value by using a pre-trained semantic sensitivity evaluation model, and obtain a third sensitivity score corresponding to the target response field name; wherein the target response value is a response value corresponding to the target response field name;

[0046] A sensitive judgment module is configured to judge whether the target response value is sensitive information according to the first sensitivity score, the second sensitivity score corresponding to the target response field name, and the third sensitivity score corresponding to the target response field name;

[0047] A data desensitization module is configured to perform data desensitization processing on the target response value in the original response data if the target response value is sensitive information, and obtain desensitized response data;

[0048] A response data return module is configured to return the desensitized response data to the request device.

[0049] In a third aspect, an embodiment of the present application provides a storage medium, the storage medium stores computer readable instructions, and the computer readable instructions are executed by one or more processors to make the one or more processors execute the steps of the data desensitization method based on a large model according to any one of the above embodiments.

[0050] In a fourth aspect, an embodiment of the present application provides a server, comprising: one or more processors, and a memory;

[0051] The memory stores computer readable instructions, which are executed by the one or more processors to perform the steps of the data desensitization method based on a large model according to any of the embodiments.

[0052] In the data desensitization method based on a large model and the related device provided by some embodiments of the present application, the sensitivity of the data acquisition request can be evaluated, and a first sensitivity score can be obtained. For the original response data corresponding to the data acquisition request, the sensitivity of the structural context can be evaluated according to the nesting information of the response field name in the storage data structure, and a second sensitivity score can be obtained. In addition, the semantic sensitivity evaluation model can be used to identify the potential sensitive content in the response value, and a third sensitivity score can be obtained. After multi-dimensional sensitivity evaluation, the sensitivity scores of the data acquisition request, the sensitivity scores of the structural context, and the sensitivity scores of the response value can be comprehensively judged to determine whether the response value involves sensitive information, and the response value involving sensitive information can be desensitized. In this way, the sensitive information in the response data can be comprehensively identified and desensitized, so as to reduce the risk of data leakage, and thus the data security can be improved. BRIEF DESCRIPTION OF DRAWINGS

[0053] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed in the embodiments or the prior art description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and those skilled in the art can obtain other drawings according to these drawings without creative labor.

[0054] Figure 1 For some embodiments, the application environment diagram of the data desensitization method based on a large model;

[0055] Figure 2 For some embodiments, the flowchart of the data desensitization method based on a large model;

[0056] Figure 3 For some embodiments, the structural diagram of the data desensitization device based on a large model;

[0057] Figure 4 For some embodiments, the internal structure diagram of the server. DETAILED DESCRIPTION

[0058] With reference to the drawings and embodiments of the present application, the technical solutions in the embodiments of the present application will be described clearly and completely. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative work fall within the scope of the present application.

[0059] It should be noted that the collection, storage, use, processing, transmission, provision and disclosure of data information involved in the present application comply with relevant laws and regulations and do not violate public order and good customs.

[0060] In some embodiments, the data desensitization method based on a large model provided by the present application can be applied to Figure 1 The application environment is shown. The collection device 102 refers to a device with data collection function and capable of transmitting the collected data to the server 104. It can be understood that the collection device 102 can be, but is not limited to, various personal computers, notebook computers, laptop computers, tablet computers, smart phones, Internet of Things devices and portable wearable devices. The Internet of Things device can be a smart speaker, a smart TV, a smart air conditioner, a smart vehicle device, etc., and the portable wearable device can be a smart watch, a smart bracelet, a head-mounted device, etc.

[0061] The server 104 refers to a device capable of receiving and storing collected data and having a data desensitization function. In the present application, the server 104 can be a single server or a server cluster. The request device 106 refers to a device requesting to obtain collected data from the server 104, which can be, but is not limited to, various personal computers, notebook computers, laptop computers, tablet computers, smart phones, Internet of Things devices and portable wearable devices. The Internet of Things device can be a smart speaker, a smart TV, a smart air conditioner, a smart vehicle device, etc., and the portable wearable device can be a smart watch, a smart bracelet, a head-mounted device, etc.

[0062] It can be understood that the data involved in the present application can be any type of data, which is not limited in the present application. In some examples, for the convenience of description, the browser data is taken as an example for description. It should be understood that in addition to the browser data, the data involved in the present application can also be other types of data or include more types of data.

[0063] Specifically, the collection device 102 can collect original browser data and transmit the original browser data to the server 104. The server 104 can receive and store the original browser data. In some examples, the server 104 can provide any one or any combination of data processing functions such as data truncation, data format standardization and conversion, data temporary storage, data analysis, etc.

[0064] When the requesting device 106 needs to acquire the browser data, the requesting device 106 can send a data acquisition request to the server 104 to request to acquire the target browser data. Upon receiving the data acquisition request, the server 104 can determine the original response data corresponding to the data acquisition request. For example, when the requesting device 106 requests to acquire the target browser data, the server 104 can take the target browser data as the original response data.

[0065] In view of the fact that the original response data can involve any one or any combination of sensitive information such as tokens, keys, identity card numbers, bank card numbers, contact numbers, residential addresses, account passwords, etc., in order to protect information security, the server 104 needs to perform data desensitization processing on the original response data to mask the sensitive information in the response data through data deformation, data hiding, etc., and obtain desensitized response data. The server 104 returns the desensitized response data to the requesting device 106 instead of the original response data, so as to ensure that the requesting device 106 cannot acquire the sensitive information from the received response data.

[0066] For example, in the example shown in Figure 1 In the example shown, the original browser data collected by the collection device 102 can be account password data. If the requesting device 106 requests to acquire the browser data including this account password data, the server 104 can perform data desensitization on the account password in the original response data to mask part or all of the account password information, and return the desensitized data to the requesting device 106. In this way, the requesting device 106 cannot learn the real account and the real password from the response data, so as to ensure information security and privacy.

[0067] In some embodiments, the present application provides a large model-based data desensitization method. The following embodiments are described with reference to the application of the method to Figure 1 The server is taken as an example for illustration. As shown in Figure 2 The method provided by the present application can include the following steps:

[0068] S202: In response to the data acquisition request sent by the requesting device, determining the original response data; wherein the original response data includes at least one response field name and a response value corresponding to each response field name.

[0069] It can be understood that the original response data refers to the data requested to be acquired by the requesting device, which can include sensitive information, and therefore it is necessary to identify the sensitive information in the original response data and perform data desensitization processing on the sensitive information.

[0070] The original response data can include one or more data nodes, and each data node can include a response field name and a response value. Among them, the data node can be a basic constituent unit of the original response data, and the original response data can be divided according to a predetermined basic data type. The basic data type can include a string, a numerical value, a Boolean value, a null value, an array, an object, a structure, a set, etc. The response field name of the data node can include the attribute name of the data node. The response value of the data node refers to the specific data content corresponding to the data node in the original response data, and can be used to reflect the actual storage information of the data node.

[0071] For example, when the original response data is account="Abc123" and password="456edfg*?", the original response data can include two response field names, namely account and password. The response value corresponding to account is Abc123, and the response value corresponding to password is 456edfg*?.

[0072] S204: The sensitivity of the data acquisition request is evaluated by using the sensitive field library, and a first sensitivity score is obtained.

[0073] Among them, the sensitive field library includes one or more predetermined sensitive fields, and the sensitive field can be a field name used to represent sensitive information, such as password, secret, token, key, etc.

[0074] In this step, the data acquisition request can carry screening information of the original response data, and in some cases, the screening information can reflect the information meaning of the requested data. For example, the data acquisition request can carry screening field names, request paths, and other screening information. When the screening field name is password or token, it indicates that the original response data requested has a high probability of involving sensitive information such as passwords or tokens.

[0075] Therefore, the present application can evaluate the sensitivity of the data acquisition request by using the sensitive field library, and obtain a first sensitivity score. The first sensitivity score is used to reflect the probability of the original response data involving sensitive information from the dimension of the data acquisition request.

[0076] It can be understood that the sensitive field in the sensitive field library can be determined in any manner, for example, can be pre-edited and input by an engineer. In some examples, the server can track the relevance of each field in the data stream using a pattern learning mechanism, and record the co-occurrence frequency of each pair of fields through an association matrix. If the co-occurrence frequency of the field pair is greater than a preset frequency threshold, and one of the fields involved in the field pair includes a sensitive field, it indicates that the other field involved in the field pair co-occurs with the known sensitive field at a high frequency. In this case, the server can increase the sensitive association score of the other field in the field pair. When the sensitive association score of a non-sensitive field is greater than or equal to a preset score threshold (for example, 0.75), the server can add the non-sensitive field to the sensitive field library and match the field as a sensitive field in subsequent processes.

[0077] In other examples, the server can also learn to identify application-specific naming habits. Specifically, the server can analyze the field name patterns of different applications and identify specific prefixes, suffixes, naming formats that can identify sensitive information, thereby determining field names that can involve sensitive information and adding them to the sensitive field library. For example, the server can identify whether an application uses field names such as secure prefix, key suffix, api_xyz_token, etc. If so, the field name is added to the sensitive field library for application of these fields for sensitive information identification in subsequent data desensitization processes.

[0078] S206: Determine the nesting information of the target response field name in the storage data structure, and perform sensitivity evaluation according to the nesting information to obtain a second sensitivity score corresponding to the target response field name; wherein the target response field name is any one of the at least one response field name.

[0079] Specifically, the server can store data using a corresponding storage data structure, such as a JSON data structure or an XML (Extensible Markup Language) data structure. In the process of sensitivity evaluation, the server can determine the nesting information of each response field name in the storage data structure. The nesting information can describe the reference of the response field name to other data objects in the storage data structure, so as to be used to determine the structural context information of the data node in the storage data structure, and further determine the information meaning of the corresponding response value.

[0080] Therefore, for each response field name in the original response data, the server can perform sensitivity evaluation according to the nesting information corresponding to the response field name, and obtain a second sensitivity score corresponding to the corresponding field name. The second sensitivity score is used to reflect the probability that the response value involves sensitive information from the structural context dimension.

[0081] S208: The semantic sensitivity evaluation model obtained by pre-training is used to evaluate the sensitivity of the target response value, and a third sensitivity score corresponding to the target response field name is obtained.

[0082] The semantic sensitivity evaluation model can be a large model for evaluating the content sensitivity of the response value, which can be obtained by pre-training. It can be understood that the model type, training method, loss calculation method, etc. of the semantic sensitivity evaluation model can be determined according to actual conditions, and the present application does not make specific limitations.

[0083] In this step, in order to avoid the influence of personalized and differentiated field naming methods on the accuracy of sensitive information identification, such as the password field name pass_w, the server can also perform value pattern analysis on each response value in the original response data to identify potential sensitive content.

[0084] Since the information content of sensitive information has certain rules, such as random token, key with high information entropy, bank card number, credit card number following Luhn algorithm 16-19 digits, authorization token prefixed with Bearer-, and encoded string generally satisfying specific encoding data rules, the present application can input each response value into the semantic sensitivity evaluation model obtained by pre-training, and obtain a third sensitivity score corresponding to each response value. It can be understood that the third sensitivity score is used to reflect the probability of the response value involving sensitive information from the semantic dimension of the response value.

[0085] S210: According to the first sensitivity score, the second sensitivity score corresponding to the target response field name, and the third sensitivity score corresponding to the target response field name, it is judged whether the target response value is sensitive information.

[0086] In this step, for each response value in the original response data, the server can integrate the sensitive analysis results of multiple dimensions to combine and evaluate whether the response value is sensitive information.

[0087] For example, the original response data includes a first response field name city and a second response field name interest, the response value corresponding to the first response field name is Beijing, and the response value corresponding to the second response field name is sports. In this case, the server can judge whether Beijing is sensitive information according to the first sensitivity score, the second sensitivity score corresponding to city, and the third sensitivity score corresponding to Beijing. Similarly, the server can judge whether sports is sensitive information according to the first sensitivity score, the second sensitivity score corresponding to interest, and the third sensitivity score corresponding to sports.

[0088] S212: If the target response value is sensitive information, performing data desensitization processing on the target response value in the original response data, and obtaining desensitized response data.

[0089] In this step, for each response value determined as sensitive information, the server can perform data desensitization processing on the response value. After performing data desensitization on each sensitive information, desensitized response data can be obtained. For example, in the above example, if Beijing and sports are both determined as sensitive information, the server can perform data desensitization processing on Beijing and sports respectively, and obtain desensitized response data. For another example, in the above example, if Beijing is determined as sensitive information and sports is determined as non-sensitive information, the server can perform data desensitization processing on Beijing, and sports is not changed, thereby obtaining desensitized response data.

[0090] It should be noted that the server does not need to modify the original value in the data storage structure, but modifies the response data returned to the requesting device, so that the server still retains the original true data of each sensitive information after performing data desensitization processing.

[0091] S214: Return the desensitized response data to the requesting device.

[0092] In this step, the server can return the desensitized response data to the requesting device instead of the original response data, so as to protect information security and data privacy.

[0093] In this application, the sensitivity scores of the data acquisition request, the structural context, and the response value can be comprehensively considered to comprehensively judge whether the response value involves sensitive information, and to perform data desensitization on the response value involving sensitive information. In this way, the application can comprehensively identify sensitive information in response data and perform data desensitization, thereby reducing the risk of data leakage and improving data security.

[0094] In some embodiments, the data acquisition request includes a request path and a query field name, both of which can be carried by the data acquisition request, and the query field name can be a screening field name described in other embodiments of the application. In some examples, the request path can be a URL (Uniform Resource Locator, Uniform Resource Locator) path. For ease of description, the following embodiments are described by taking the URL path as an example.

[0095] The first sensitivity score includes a path sensitivity score and a query field sensitivity score. The path sensitivity score is used to reflect the probability that the response value involves sensitive information from the dimension of the request path, and the query field sensitivity score is used to reflect the probability that the response value involves sensitive information from the dimension of the query field name.

[0096] The sensitivity of the data acquisition request is evaluated by using the sensitive field library, and a first sensitivity score is obtained, including:

[0097] Step A1: According to the matching result of the preset sensitive path mode and the request path, the path sensitivity score is determined.

[0098] Step A3: According to the matching result of the sensitive field library and the query field name, the query field sensitivity score is determined.

[0099] In this embodiment, in order to more accurately evaluate the sensitivity of the response value according to the data acquisition request, the request path and the query field name of the data acquisition request can be evaluated for sensitivity respectively, so as to realize multi-dimensional sensitivity analysis.

[0100] Specifically, the server can parse the data acquisition request to extract the URL path and the query field name respectively. The server can match the preset sensitive path mode with the URL path according to the matching result to determine the path sensitivity score. For example, the server can use a regular expression engine to realize the matching of the sensitive path mode and the request path.

[0101] In some examples, the server can split the URL path into multiple components, for example, the URL path / api / v1 / users / auth / token is split into ["api", "v1", "users", "auth", "token"]. After splitting, the server can load the predefined sensitive path mode, which can include common security-related path keywords such as " / auth / ", " / login / ", " / secure / ", " / token / ", etc. The server can use a regular expression engine to realize the matching of the sensitive path mode and the request path, and determine the sensitivity score of each security-related path, thereby obtaining the path sensitivity score. Further, when the path sensitivity score exceeds a preset threshold (for example, 0.7), the server can mark the context of the entire data acquisition request as sensitive, and increase the sensitivity evaluation standard for all data therein.

[0102] In addition, the server can match the sensitive field library with the query field name, and determine the query field sensitivity score based on the matching structure. For example, the server can compare and match the query field name with each sensitive field in the sensitive field library to determine whether the query field name is a sensitive field or whether the query field name includes a sensitive field. The server can determine the query field sensitivity score according to the number of matched sensitive fields and the matching condition.

[0103] In some embodiments, the sensitive field library includes complete sensitive fields and partial sensitive fields. The complete sensitive fields can be fields that can directly determine sensitive information, such as password, token, secret, etc. The partial sensitive fields can be fields that can possibly involve sensitive information, such as key, auth, access, etc.

[0104] According to the matching result of the sensitive field library and the query field name, the query field sensitivity score is determined, including:

[0105] Step A31: If the query field name matches the complete sensitive field, the query field sensitivity score is determined as the maximum value.

[0106] Step A33: If the query field name matches the partial sensitive field, the proportional matching score and the position matching score are determined respectively, and the query field sensitivity score is calculated based on the proportional matching score and the position matching score.

[0107] The proportional matching score is positively correlated with the sensitive proportion, and the sensitive proportion is the proportion of the length of the partial sensitive field to the length of the query field name. The position matching score is associated with the position of the partial sensitive field in the query field name.

[0108] In this embodiment, if the query field name matches the complete sensitive field, the original response data involving sensitive information can be directly determined, so the query field sensitivity score can be determined as the maximum value.

[0109] Considering that the partial sensitive field has more deformations, and the field including the partial sensitive field does not necessarily involve sensitive information, for example, the response values corresponding to the two field names key_name and keyboard_api can be non-sensitive information, therefore, if the query field name matches the partial sensitive field, the server can comprehensively determine the query field sensitivity score according to the length proportion of the partial sensitive field to the query field name and the position of the partial sensitive field in the query field name.

[0110] Specifically, in the case that the query field name matches the partially sensitive field, the server can determine a proportion matching score and a position matching score respectively, and calculate the query field sensitivity score by using, for example, weighted summation. The proportion matching score is positively correlated with the sensitive proportion, that is, the greater the proportion of the partially sensitive field in the length of the query field name, the higher the proportion matching score. The position matching score is associated with the position of the partially sensitive field. In some examples, if the partially sensitive field is a prefix of the query field name, the position matching score can be a first score; if the partially sensitive field is a suffix of the query field name, the position matching score can be a second score; if the partially sensitive field is located in the middle position of the query field name, the position matching score can be a third score. The first score is greater than the second score, and the second score is greater than the third score.

[0111] In this embodiment, different ways can be used to determine the query field sensitivity score for the completely sensitive field and the partially sensitive field. For the case that the query field name matches the partially sensitive field, the server can determine the query field sensitivity score based on the sensitive proportion and the position. In this way, the probability of sensitive data involved in the data acquisition request can be more accurately evaluated, so as to further reduce the risk of data leakage and improve data security.

[0112] In some embodiments, the nesting information includes a nesting path and a nesting depth. The nesting depth of the target response field name can be the maximum number of levels from the top layer (for example, the root node) of the stored data structure to the position of the target data node in the stored data structure. The target data node is the data node corresponding to the target response field name. The nesting path of the target response field name is the complete access path from the top layer of the stored data structure to the target data node. Based on the nesting path, the server can accurately locate and reference the target data node in the stored data structure.

[0113] According to the nesting information, the second sensitivity score corresponding to the target response field name is obtained by sensitivity evaluation, including:

[0114] Step B1: performing field matching on the nesting path by using the sensitive field library, and obtaining a path matching result;

[0115] Step B3: determining the second sensitivity score corresponding to the target response field name according to the path matching result and the nesting depth; wherein the second sensitivity score is positively correlated with the nesting depth.

[0116] In this embodiment, the server can track the complete access path of each response field name in the storage data structure and obtain a nested path. Based on this, the server can match the nested path with the sensitive fields in the sensitive field library and obtain a path matching result. It will be understood that the path matching structure can be used to reflect the matching status of the nested path with the sensitive field library, such as the number of sensitive fields in the nested path.

[0117] For example, if the nested path of the target response field name is user.credentials.apiKey, this indicates that the target response field is nested within the credentials object, which is in turn nested within the user object. The server can use the sensitive field library to perform field matching on user, credentials, and apiKey, respectively, and obtain path matching results. For example, if credentials is determined to be a sensitive field, the server can increase the second sensitivity score of the target response field name.

[0118] Considering the impact of nesting depth on sensitivity, deeply nested fields are more likely to contain sensitive information than top-level fields. Based on this, the server can comprehensively determine the second sensitivity score for the target response field name based on the path matching results and the nesting depth. It is understood that if the path matching results are the same, the second sensitivity score may increase as the nesting depth increases.

[0119] In some examples, the server may also evaluate the sensitivity of sibling fields of the target response field name based on the nesting depth. When multiple sibling fields are determined to be sensitive, the server may increase the second sensitivity score of the target response field name to increase its sensitivity possibility.

[0120] In this embodiment, by determining the corresponding second sensitivity score based on the nesting path and nesting depth of the target response field name in the storage data structure, the probability that the target response value involves sensitive information can be more accurately evaluated based on the structural context, thereby further reducing the risk of data leakage and improving data security.

[0121] In some embodiments, determining whether the target response value is sensitive information based on the first sensitivity score, the second sensitivity score corresponding to the target response field name, and the third sensitivity score corresponding to the target response field name includes:

[0122] Step C1: performing weighted summation on the first sensitivity score, the second sensitivity score corresponding to the target response field name, and the third sensitivity score corresponding to the target response field name to obtain the target sensitivity score;

[0123] Step C3: if the target sensitivity score is greater than a first preset score, the target response value is determined to be sensitive information, and the original sensitive level is high;

[0124] Step C5: if the target sensitivity score is less than or equal to the first preset score and greater than a second preset score, the target response value is determined to be sensitive information, and the original sensitive level is medium;

[0125] Step C7: if the target sensitivity score is less than or equal to the second preset score and greater than a third preset score, the target response value is determined to be sensitive information, and the original sensitive level is low;

[0126] Step C9: if the target sensitivity score is less than or equal to the third preset score, the target response value is determined to be non-sensitive information.

[0127] The first preset score is greater than the second preset score, and the second preset score is greater than the third preset score. It can be understood that the specific values of the first preset score, the second preset score and the third preset score can be determined according to actual conditions, and the present application does not make specific limitations thereon. Exemplarily, the first preset score can be 0.8, the second preset score can be 0.5, and the third preset score can be 0.3.

[0128] In the present embodiment, for each response field name in the original response data, the server can perform weighted summation calculation according to the first sensitivity score, the second sensitivity score corresponding to the response field name, and the third sensitivity score corresponding to the response value, and obtain a target sensitivity score. It can be understood that the weighting coefficients of the sensitivity scores can be determined according to actual conditions, and the weighting coefficients of any two sensitivity scores can be the same or different, and the present application does not make specific limitations thereon.

[0129] After obtaining the target sensitivity score, the present application can compare the target sensitivity score with a preset score to determine the original sensitive level of the target response value. When the target sensitivity score is greater than the first preset score, the target response value can be determined to be high sensitive data. When the target sensitivity score is less than or equal to the first preset score and greater than the second preset score, the target response value can be determined to be medium sensitive data. When the target sensitivity score is less than or equal to the second preset score and greater than the third preset score, the target response value can be determined to be low sensitive data. When the target sensitivity score is less than or equal to the third preset score, the target response value can be determined to be non-sensitive information. In this way, the data desensitization efficiency can be improved on the basis of ensuring the accuracy of sensitive information identification.

[0130] In some embodiments, after determining the original sensitive level of each response value in the original response data, the server can generate a sensitivity evaluation report according to the original sensitive level, and record the basis for judgment, the confidence and the factors affecting the final decision in the report.

[0131] In some embodiments, the data desensitization processing on the target response value in the original response data comprises:

[0132] Step D1: determining the target sensitive level according to the original sensitive level corresponding to the target response value;

[0133] Step D3: if the target sensitive level is high, using a preset standardized placeholder to replace the target response value;

[0134] Step D5: if the target sensitive level is medium, performing partial mask processing on the target response value, and adding a description identifier according to the data type of the target response value;

[0135] Step D7: if the target sensitive level is low, performing generalization processing on the target response value.

[0136] In this embodiment, in the process of performing data desensitization processing, the server can determine the target sensitive level of the target response value according to the original sensitive level corresponding to the target response value, and perform data desensitization according to the target sensitive level by using different processing methods, so as to balance the information security and data use requirements.

[0137] If the target sensitive level corresponding to the target response value is high, the server can perform the most stringent protection measures to minimize the risk of leakage of the target response value. Specifically, for the target response value of high sensitive level, the server can use a standard placeholder such as [REDACTED] to replace the original value, so as to completely remove the original value and not to retain any part, such as length, format, character distribution characteristics, etc. are not retained in the desensitized response data. Moreover, the present application avoids leaking information through replacement mode by using consistent replacement text for replacement.

[0138] In some examples, the server can also completely remove the data of high sensitive level when processing derived data such as logs, caches, error reports, etc. For example, the server only records the number of high sensitive fields that have been processed in the desensitized report, but does not contain the field name itself, preventing information leakage through metadata. For another example, when processing user login data containing password fields, the server completely replaces the value of the "password" field "Secure123!" with "[REDACTED]", leaving no trace of the original password.

[0139] If the target sensitive level corresponding to the target response value is medium, the server can perform mask processing on part of the information, so that the processed desensitized data can retain part of the information of the original response value, and a description identifier can be added according to the data type of the target response value, so as to maintain the understandability of the desensitized data.

[0140] For example, for an email address john.doe@example.com, the server can replace the original value with j@ple.com[EMAIL], which can help understand the data type while protecting the specific content. Further, the server can also preserve the format features of the data, such as the separator pattern of dates, the grouping format of numbers, etc., so that the processed data maintains a recognizable structure.

[0141] In some examples, the server can apply different masking patterns for different types of sensitive data. For example, for a string type of original response value, the server can preserve the first two characters and the last two characters of the original response value, and replace the middle characters with asterisks, the number of which is the same as the number of replaced characters, so as to maintain the original length feature. For example, for an original response value of an identified credit card number, the server can preserve the first 4 digits and the last 4 digits, and replace the middle characters. For example, the server can mask the domain name of an email address, and preserve the country code and the last 4 digits of a phone number.

[0142] If the target response value corresponds to a target sensitivity level of low, the server can perform generalization processing on it, balancing information utility and privacy protection by using light protection measures. That is, the server can preserve the basic structure and most of the content of the data, and only apply limited anonymization processing to specific parts. Among them, generalization processing can be to convert specific numerical values or text into more abstract concepts or ranges, to protect the privacy and security of the data.

[0143] In some examples, for a name, the server can replace the full name with the initial letter, for example, John Smith can be replaced with J.S. For specific location information, the server can convert it to a wider geographical area to generalize the information. For example, the server can convert the address 123 Main St, Boston, MA to Boston Area. For long text content, the server can preserve most of the text content, and only edit part of it when a specific paragraph or sentence contains sensitive information. For information involving numerical values such as age and amount, the server can perform fuzzification processing on it, for example, the exact age value can be replaced with an age range, and the exact amount can be replaced with an approximate value.

[0144] Further, in some embodiments, for non-sensitive information, the server can implement a minimal intervention observation policy. Specifically, the server keeps the original data content intact without any modification or masking on the values themselves. The server adds a flag in the data metadata indicating that this field can be sensitive in certain contexts, facilitating subsequent processing stages to add protection as needed. The server maintains an access log recording who accessed these potentially sensitive fields, when, and in what context, supporting security auditing and anomaly detection. The server implements environment-aware processing, potentially dynamically boosting the processing level in high-risk environments (e.g., public API responses) while maintaining minimal intervention in controlled environments (e.g., internal system-to-system communication). For example, a user display name johndoe is typically kept in its original form, but the server adds a metadata flag so that the system knows it is potentially personally identifiable information.

[0145] In some embodiments, the target sensitivity level is determined according to an original sensitivity level corresponding to the target response value, including:

[0146] Step D11: obtaining a source interface of the data acquisition request;

[0147] Step D13: if the source interface is a high-risk interface, performing level up according to the original sensitivity level corresponding to the target response value, and obtaining the target sensitivity level;

[0148] Step D15: if the source interface is a low-risk interface, performing level down according to the original sensitivity level corresponding to the target response value, and obtaining the target sensitivity level.

[0149] In this embodiment, the server can use an adaptive processing mechanism to dynamically adjust the sensitivity level and processing strategy according to the environment context. In this way, the source and destination of the request can be analyzed to allow the sensitivity level of certain fields to be reduced when transmitting between trusted internal systems, while the sensitivity level is increased when transmitting across the trust boundary, so that the information security and data usage requirements can be further balanced.

[0150] Specifically, the server can determine the source interface of the data acquisition request, which can be an API (Application Programming Interface) called by the request device when sending the data acquisition request. If the source interface is a high-risk interface, for example, the source interface is a public-facing interface, the server can automatically increase the sensitivity level, and use the data desensitization method corresponding to the medium sensitivity level to process data information with a low original sensitivity level, and use the data desensitization method corresponding to the high sensitivity level to process data information with a medium original sensitivity level.

[0151] If the source interface is a low-risk interface such as a partner API or an internal system API, the server can lower the sensitivity level and process data information with an original sensitivity level of high using a data desensitization method corresponding to a medium sensitivity level, and process data information with an original sensitivity level of medium using a data desensitization method corresponding to a low sensitivity level.

[0152] In some embodiments, considering that individual records can have different sensitivity requirements compared to a large amount of aggregated data, for example: a single user's purchase record can be a low sensitivity level, but a plurality of user's aggregated purchase data can require a higher level of protection, therefore the server can also check the aggregation state of the data, and determine the target sensitivity level according to the aggregation state and the original sensitivity level.

[0153] In some embodiments, the server maintains a central sensitivity configuration library, allowing dynamic updating of processing rules according to organizational policies and regulatory requirements. The server supports role-based sensitivity processing, determining the appropriate processing level according to the identity and permissions of the requester. The server monitors the effect of sensitive data processing in real time, and when it finds that the filtering is excessive (hindering legitimate functions) or insufficient (there is a potential risk of data leakage), it automatically adjusts the processing strategy. The server regularly reports newly discovered sensitive data patterns and the effectiveness of processing strategies to the security team to support continuous improvement of data protection mechanisms.

[0154] The following describes a large model-based data desensitization device provided by an embodiment of the present application. The large model-based data desensitization device described below can be mutually referred to with the large model-based data desensitization method described above.

[0155] In some embodiments, as shown in Figure 3 The present application provides a large model-based data desensitization device 300, which includes:

[0156] An original response data determination module 302 is configured to determine original response data in response to a data acquisition request sent by a requesting device; the original response data includes at least one response field name and a response value corresponding to each response field name;

[0157] A first evaluation module 304 is configured to evaluate the sensitivity of the data acquisition request using a sensitive field library, and obtain a first sensitivity score;

[0158] A second evaluation module 306 is configured to determine the nesting information of a target response field name in a storage data structure, and evaluate the sensitivity according to the nesting information to obtain a second sensitivity score corresponding to the target response field name; wherein the target response field name is any one of the at least one response field name;

[0159] The third evaluation module 308 is configured to perform sensitivity evaluation on the target response value by using the pre-trained semantic sensitivity evaluation model, and obtain a third sensitivity score corresponding to the target response field name; the target response value is a response value corresponding to the target response field name;

[0160] The sensitivity judgment module 310 is configured to judge whether the target response value is sensitive information according to the first sensitivity score, the second sensitivity score corresponding to the target response field name, and the third sensitivity score corresponding to the target response field name.

[0161] The data desensitization module 312 is configured to perform data desensitization processing on the target response value in the original response data if the target response value is sensitive information, and obtain desensitized response data.

[0162] The response data returning module 314 is configured to return the desensitized response data to the request device.

[0163] In some embodiments, the data acquisition request includes a request path and a query field name, and the first sensitivity score includes a path sensitivity score and a query field sensitivity score. The first evaluation module 304 includes:

[0164] The first path matching unit is configured to determine the path sensitivity score according to a matching result of a preset sensitive path mode and the request path.

[0165] The first field scoring unit is configured to determine the query field sensitivity score according to a matching result of the sensitive field library and the query field name.

[0166] In some embodiments, the sensitive field library includes a complete sensitive field and a partial sensitive field. The first field scoring unit includes:

[0167] The first matching unit is configured to determine the query field sensitivity score as a maximum value if the query field name matches the complete sensitive field.

[0168] The second matching unit is configured to determine a proportional matching score and a position matching score respectively if the query field name matches the partial sensitive field, and calculate the query field sensitivity score based on the proportional matching score and the position matching score.

[0169] The proportional matching score is positively correlated with a sensitive proportion, and the sensitive proportion is a proportion of a length of the partial sensitive field to a length of the query field name; and the position matching score is associated with a position of the partial sensitive field in the query field name.

[0170] In some embodiments, the nesting information comprises a nesting path and a nesting depth. The second evaluation module 306 of the present application comprises:

[0171] a second path matching unit, configured to perform field matching on the nesting path by using the sensitive field library, and obtain a path matching result;

[0172] a structure sensitivity scoring unit, configured to determine a second sensitivity score corresponding to the target response field name according to the path matching result and the nesting depth, wherein the second sensitivity score is positively correlated with the nesting depth.

[0173] In some embodiments, the second sensitivity judgment module 310 of the present application comprises:

[0174] a weighted sum unit, configured to perform weighted sum on the first sensitivity score, the second sensitivity score corresponding to the target response field name, and the third sensitivity score corresponding to the target response field name, and obtain a target sensitivity score;

[0175] a first judgment unit, configured to determine that the target response value is sensitive information and the original sensitive level is high if the target sensitivity score is greater than a first preset score;

[0176] a second judgment unit, configured to determine that the target response value is sensitive information and the original sensitive level is medium if the target sensitivity score is less than or equal to the first preset score and greater than a second preset score;

[0177] a third judgment unit, configured to determine that the target response value is sensitive information and the original sensitive level is low if the target sensitivity score is less than or equal to the second preset score and greater than a third preset score;

[0178] a fourth judgment unit, configured to determine that the target response value is non-sensitive information if the target sensitivity score is less than or equal to the third preset score.

[0179] In some embodiments, the data desensitization module 312 of the present application comprises:

[0180] a target level determination unit, configured to determine a target sensitive level according to the original sensitive level corresponding to the target response value;

[0181] a first desensitization unit, configured to replace the target response value with a preset standardized placeholder if the target sensitive level is high;

[0182] a second desensitization unit, configured to perform partial mask processing on the target response value and add a description identifier according to the data type of the target response value if the target sensitive level is medium;

[0183] a third desensitization unit, configured to perform generalization processing on the target response value if the target sensitivity level is low.

[0184] In some embodiments, the target level determination unit of the present application comprises:

[0185] a source interface acquisition unit, configured to acquire a source interface of the data acquisition request;

[0186] an up-regulation unit, configured to perform level up-regulation according to an original sensitivity level corresponding to the target response value, and obtain the target sensitivity level if the source interface is a high-risk interface;

[0187] a down-regulation unit, configured to perform level down-regulation according to an original sensitivity level corresponding to the target response value, and obtain the target sensitivity level if the source interface is a low-risk interface.

[0188] In one embodiment, the present application further provides a storage medium having computer readable instructions stored therein, and the computer readable instructions are executed by one or more processors to cause the one or more processors to perform the steps of the data desensitization method based on a large model in any embodiment.

[0189] In one embodiment, the present application further provides a server having computer readable instructions stored therein, and the computer readable instructions are executed by one or more processors to cause the one or more processors to perform the steps of the data desensitization method based on a large model in any embodiment.

[0190] Schematically, Figure 4 An internal structure schematic diagram of a server provided by an embodiment of the present application is shown in FIG. 9. Referring to FIG. 9, Figure 4 The server 900 comprises a processing component 902, which further comprises one or more processors, and a memory resource represented by a memory 901, for storing instructions executable by the processing component 902, such as an application program. The application program stored in the memory 901 can comprise one or more than one module each corresponding to a set of instructions. In addition, the processing component 902 is configured to execute the instructions to perform the steps of the data desensitization method based on a large model in any embodiment described above.

[0191] The server 900 can further include a power supply component 903 configured to perform power management for the server 900, a wired or wireless network interface 904 configured to connect the server 900 to a network, and an input output (I / O) interface 905. The server 900 can operate based on an operating system stored in the memory 901, such as Windows Server TM, Mac OS X TM, Unix TM, Linux TM, Free BSD TM, or the like.

[0192] Those skilled in the art can understand that the internal structure of the server shown in the present application is only a block diagram of part of the structure related to the scheme of the present application, and does not constitute a limitation on the server to which the scheme of the present application is applied. The specific server can include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0193] Finally, it should be noted that, in this document, the terms such as first and second 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 the entities or operations. Moreover, the terms "include", "contain" or any other variants thereof are intended to cover non-exclusive inclusion, so that the process, method, article or device including a series of elements not only includes those elements, but also includes other elements not explicitly listed or inherent to such process, method, article or device. Without more limitations, the element defined by the statement "including a" does not exclude the presence of additional identical elements in the process, method, article or device including the element. In this document, "a", "one", "said", "the" and "it" can also include plural forms, unless the context clearly indicates otherwise. A plurality means at least two, such as 2, 3, 5 or 8, etc. "And / or" includes any and all combinations of the related listed items.

[0194] The various embodiments in the specification are described in a progressive manner, each embodiment focusing on the differences from other embodiments, and the various embodiments can be combined as needed, and the same and similar parts refer to each other.

[0195] The above description of the disclosed embodiments enables a person skilled in the art to implement or use the present application. Various modifications to the embodiments will be apparent to those skilled in the art, and the general principles defined herein can 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 data desensitization method based on a large model, characterized in that: include: In response to the data acquisition request sent by the requesting device, determining original response data; The original response data includes at least one response field name and a response value corresponding to each response field name; Performing a sensitivity assessment on the data acquisition request using a sensitive field library and obtaining a first sensitivity score; Determining nesting information of a target response field name in a storage data structure, and performing a sensitivity assessment based on the nesting information to obtain a second sensitivity score corresponding to the target response field name; wherein the target response field name is any one of the at least one response field name; Using a pre-trained semantic sensitivity assessment model, a sensitivity assessment is performed on the target response value, and a third sensitivity score corresponding to the target response field name is obtained; wherein the target response value is a response value corresponding to the target response field name; Determining whether the target response value is sensitive information according to the first sensitivity score, the second sensitivity score corresponding to the target response field name, and the third sensitivity score corresponding to the target response field name; If the target response value is sensitive information, performing data desensitization processing on the target response value in the original response data to obtain desensitized response data; Return the desensitized response data to the requesting device.

2. The method according to claim 1, characterized in that The data acquisition request includes a request path and a query field name, and the first sensitivity score includes a path sensitivity score and a query field sensitivity score; The sensitive field library is used to perform a sensitivity assessment on the data acquisition request and obtain a first sensitivity score, including: Determining the path sensitivity score according to a matching result between a preset sensitive path pattern and the requested path; The sensitivity score of the query field is determined based on the matching result between the sensitive field library and the query field name.

3. The method according to claim 2, characterized in that The sensitive field library includes fully sensitive fields and partially sensitive fields; Determining the sensitivity score of the query field according to the matching result between the sensitive field library and the query field name includes: If the query field name matches the fully sensitive field, the query field sensitivity score is determined to be the maximum value; If the query field name matches the partial sensitive fields, determining a proportion match score and a position match score respectively, and calculating the query field sensitivity score based on the proportion match score and the position match score; Among them, the proportion matching score is positively correlated with the sensitivity ratio, and the sensitivity ratio is the ratio of the length of the partially sensitive field to the length of the query field name; the position matching score is associated with the position of the partially sensitive field in the query field name.

4. The method according to claim 1, wherein The nesting information includes a nesting path and a nesting depth; The performing a sensitivity assessment based on the nested information to obtain a second sensitivity score corresponding to the target response field name includes: Using the sensitive field library to perform field matching on the nested path and obtain a path matching result; Determine a second sensitivity score corresponding to the target response field name based on the path matching result and the nesting depth; wherein the second sensitivity score is positively correlated with the nesting depth.

5. The method according to any one of claims 1 to 4, characterized in that The determining, based on the first sensitivity score, the second sensitivity score corresponding to the target response field name, and the third sensitivity score corresponding to the target response field name, whether the target response value is sensitive information includes: Performing a weighted summation on the first sensitivity score, the second sensitivity score corresponding to the target response field name, and the third sensitivity score corresponding to the target response field name to obtain a target sensitivity score; If the target sensitivity score is greater than a first preset score, determining that the target response value is sensitive information and the original sensitivity level is high; If the target sensitivity score is less than or equal to the first preset score and greater than the second preset score, then the target response value is determined to be sensitive information and the original sensitivity level is medium; If the target sensitivity score is less than or equal to the second preset score and greater than a third preset score, determining that the target response value is sensitive information and the original sensitivity level is low; If the target sensitivity score is less than or equal to the third preset score, the target response value is determined to be non-sensitive information.

6. The method according to claim 5, characterized in that The performing data desensitization processing on the target response value in the original response data includes: Determining a target sensitivity level according to an original sensitivity level corresponding to the target response value; If the target sensitivity level is high, a preset standardized placeholder is used to replace the target response value; If the target sensitivity level is medium, partially mask the target response value and add a description identifier according to the data type of the target response value; If the target sensitivity level is low, generalization processing is performed on the target response value.

7. The method according to claim 6, characterized in that The determining the target sensitivity level according to the original sensitivity level corresponding to the target response value includes: Obtaining the source interface of the data acquisition request; If the source interface is a high-risk interface, the level is increased according to the original sensitivity level corresponding to the target response value, and the target sensitivity level is obtained; If the source interface is a low-risk interface, the level is downgraded according to the original sensitivity level corresponding to the target response value to obtain the target sensitivity level.

8. A data desensitization device based on a large model, characterized in that: include: an original response data determining module, configured to determine original response data in response to a data acquisition request sent by a requesting device; The original response data includes at least one response field name and a response value corresponding to each response field name; A first evaluation module, configured to perform a sensitivity evaluation on the data acquisition request using a sensitive field library and obtain a first sensitivity score; a second evaluation module, configured to determine nesting information of a target response field name in a storage data structure, and perform a sensitivity evaluation based on the nesting information to obtain a second sensitivity score corresponding to the target response field name; wherein the target response field name is any one of the at least one response field name; a third evaluation module, configured to perform a sensitivity evaluation on a target response value using a pre-trained semantic sensitivity evaluation model, and obtain a third sensitivity score corresponding to the target response field name; wherein the target response value is a response value corresponding to the target response field name; a sensitivity determination module, configured to determine whether the target response value is sensitive information based on the first sensitivity score, the second sensitivity score corresponding to the target response field name, and the third sensitivity score corresponding to the target response field name; a data desensitization module, configured to perform data desensitization processing on the target response value in the original response data if the target response value is sensitive information, and obtain desensitized response data; The response data returning module is used to return the desensitized response data to the requesting device.

9. A storage medium, characterized in that: The storage medium stores computer-readable instructions, which, when executed by one or more processors, enable the one or more processors to perform the steps of the large model-based data desensitization method as described in any one of claims 1 to 7.

10. A server, characterized in that: include: one or more processors, and memory; The memory stores computer-readable instructions, which, when executed by the one or more processors, perform the steps of the large model-based data desensitization method as described in any one of claims 1 to 7.

Images