Content verification method and system based on LLM security protection

Abstract

The present invention discloses a content verification method and system based on LLM security protection. The method comprises: determining data to be reviewed from all data corresponding to the LLM security protection; performing a preliminary review operation on the data to be reviewed based on a predetermined review mechanism to obtain a preliminary review result; and performing a data verification operation on the preliminary review result based on a predetermined target analysis algorithm and data review strategy to obtain a data verification result corresponding to the data to be reviewed. It can be seen that the implementation of the present invention can intelligently verify the data of the LLM protection system, which is conducive to improving the accuracy and reliability of data verification, and further helps to improve the operational security of the LLM security protection system.

Description

Technical Field

[0001] The present invention relates to the field of data verification technology, and in particular to a content verification method and system based on LLM security protection. Background Art

[0002] Large Language Model (LLM) is an artificial intelligence model based on deep learning technology, which is designed to understand and generate human language. In recent years, the research and application of Large Language Model (LLM) technology has been developing rapidly.

[0003] As LLM technology continues to develop, LLM security products will also require continuous upgrades and improvements. Currently, most LLM data verification methods rely on pre-defined software testing methods to verify LLM data. However, this approach primarily scans the program under test, detecting the presence of security verification functions based on a large set of rules and static analysis strategies to generate data verification results. This leads to technical issues such as low data verification accuracy. Therefore, it is particularly important to provide a new content verification method based on LLM security protection to improve the accuracy and reliability of data verification. Summary of the Invention

[0004] The present invention provides a content verification method and system based on LLM security protection, which can intelligently verify the data of the LLM protection system, which is beneficial to improving the accuracy and reliability of data verification, and further beneficial to improving the operational security of the LLM security protection system.

[0005] In order to solve the above technical problems, the first aspect of the present invention discloses a content verification method based on LLM security protection, the method comprising:

[0006] Determine the data to be reviewed from all data corresponding to the LLM security protection, perform a preliminary review operation on the data to be reviewed based on a predetermined review mechanism, and obtain a preliminary review result;

[0007] According to a predetermined target analysis algorithm and data review strategy, a data verification operation is performed on the preliminary review result to obtain a data verification result corresponding to the data to be reviewed.

[0008] As an optional embodiment, in the first aspect of the present invention, the method further comprises:

[0009] According to the data verification result, determining whether the data verification result meets a predetermined protection verification condition;

[0010] When it is determined that the data verification result does not satisfy the predetermined protection verification condition, determining a target reason why the data verification result does not satisfy the protection verification condition based on the data verification result and the protection verification condition;

[0011] An optimization operation is performed on the data review strategy based on the target reason.

[0012] As an optional embodiment, in the first aspect of the present invention, before performing a data verification operation on the preliminary review result according to the predetermined target analysis algorithm and data review strategy and obtaining the data verification result corresponding to the data to be reviewed, the method further includes:

[0013] Obtaining comprehensive review information, wherein the comprehensive review information includes one or more of language review information, cultural review information, and background review information;

[0014] performing an information preprocessing operation on the comprehensive review information to obtain preprocessed review information;

[0015] Determining whether the pre-processed review information meets the preset information review conditions;

[0016] When it is determined that the pre-processing review information meets the preset information review condition, a data review strategy is generated based on the pre-processing review information and the information pre-processing operation.

[0017] As an optional embodiment, in the first aspect of the present invention, the method further comprises:

[0018] When it is determined that the pre-processing review information does not meet the preset information review condition, determining the operation parameters corresponding to the information pre-processing operation, and determining the difference parameters between the pre-processing review information and the information review condition;

[0019] Based on the operation parameters and the difference parameters, operation adjustment parameters for the information preprocessing operation are generated, and based on the operation adjustment parameters, the information preprocessing operation is updated, and the operation of performing the information preprocessing operation on the comprehensive review information is re-triggered to obtain the preprocessing review information and to trigger the execution of the operation of determining whether the preprocessing review information meets the preset information review conditions.

[0020] As an optional implementation manner, in the first aspect of the present invention, performing an optimization operation on the data review strategy according to the target cause includes:

[0021] Determining the review effectiveness of the data review strategy based on the target reason, and judging whether the review effectiveness is greater than or equal to a preset review effectiveness threshold;

[0022] When it is determined that the review effectiveness is greater than or equal to the preset review effectiveness threshold, at least one first optimization factor matching the key verification factor is determined in a predetermined optimization parameter database based on the target cause, and an optimization operation is performed on the data review strategy based on all of the first optimization factors;

[0023] When it is determined that the review effectiveness is less than the preset review effectiveness threshold, the review performance indicators in the data review strategy are analyzed based on the target cause to obtain the indicator analysis results, and the reasons to be optimized are generated based on the indicator analysis results and the target cause. Based on the reasons to be optimized, at least one second optimization factor matching the key verification factor is determined in a predetermined optimization parameter database, and the data review strategy is optimized based on all the second optimization factors.

[0024] As an optional implementation manner, in the first aspect of the present invention, performing a preliminary review operation on the data to be reviewed based on a predetermined review mechanism to obtain a preliminary review result includes:

[0025] Generate a target review model based on the predetermined review mechanism and all data corresponding to the LLM security protection;

[0026] Inputting the data to be reviewed into the target review model to obtain a model review result, wherein the model review result includes a review result after performing a review operation on the data to be reviewed by the review mechanism;

[0027] Based on the model review result, a preliminary review result corresponding to the data to be reviewed is generated.

[0028] As an optional implementation manner, in the first aspect of the present invention, before performing the optimization operation on the data review strategy according to the target cause, the method further includes:

[0029] Obtaining policy feedback information for the data review policy, and determining whether the policy feedback information meets a preset policy optimization condition;

[0030] When it is determined that the policy feedback information satisfies the preset policy optimization condition, generating feedback optimization parameters for the data review policy based on the policy feedback information;

[0031] The performing of an optimization operation on the data review strategy according to the target reason includes:

[0032] An optimization operation is performed on the data review strategy based on the target cause and the feedback optimization parameter.

[0033] A second aspect of the present invention discloses a content verification system based on LLM security protection, the system comprising:

[0034] A determination module is used to determine the data to be reviewed from all the data corresponding to the LLM security protection;

[0035] A review module, configured to perform a preliminary review operation on the data to be reviewed based on a predetermined review mechanism to obtain a preliminary review result;

[0036] The verification module is used to perform a data verification operation on the preliminary review result according to a predetermined target analysis algorithm and data review strategy to obtain a data verification result corresponding to the data to be reviewed.

[0037] As an optional embodiment, in the second aspect of the present invention, the system further includes:

[0038] A judgment module, configured to judge whether the data verification result satisfies a predetermined protection verification condition according to the data verification result;

[0039] The determining module is further configured to, when the judging module determines that the data verification result does not satisfy the predetermined protection verification condition, determine, based on the data verification result and the protection verification condition, a target reason why the data verification result does not satisfy the protection verification condition;

[0040] An optimization module is used to perform optimization operations on the data review strategy according to the target cause.

[0041] As an optional embodiment, in the second aspect of the present invention, the system further includes:

[0042] an acquisition module configured to acquire comprehensive review information before the verification module performs a data verification operation on the preliminary review result according to a predetermined target analysis algorithm and data review strategy to obtain a data verification result corresponding to the data to be reviewed, wherein the comprehensive review information includes one or more of language review information, cultural review information, and background review information;

[0043] a processing module, configured to perform information preprocessing on the comprehensive review information to obtain preprocessed review information;

[0044] The judgment module is further configured to judge whether the pre-processed review information satisfies a preset information review condition;

[0045] A generating module is configured to generate a data review strategy based on the preprocessing review information and the information preprocessing operation when the judging module judges that the preprocessing review information satisfies the preset information review condition.

[0046] As an optional embodiment, in the second aspect of the present invention, the determining module is further configured to, when the judging module determines that the pre-processing review information does not satisfy the preset information review condition, determine an operation parameter corresponding to the information pre-processing operation, and determine a difference parameter between the pre-processing review information and the information review condition;

[0047] The generation module is also used to generate operation adjustment parameters for the information preprocessing operation based on the operation parameters and the difference parameters, and update the information preprocessing operation based on the operation adjustment parameters, and re-trigger the processing module to perform the information preprocessing operation on the comprehensive review information to obtain the preprocessing review information and trigger the judgment module to perform the operation of judging whether the preprocessing review information meets the preset information review conditions.

[0048] As an optional implementation, in the second aspect of the present invention, the specific manner in which the optimization module performs the optimization operation on the data review strategy according to the target cause includes:

[0049] Determining the review effectiveness of the data review strategy based on the target reason, and judging whether the review effectiveness is greater than or equal to a preset review effectiveness threshold;

[0050] When it is determined that the review effectiveness is greater than or equal to the preset review effectiveness threshold, at least one first optimization factor matching the key verification factor is determined in a predetermined optimization parameter database based on the target cause, and an optimization operation is performed on the data review strategy based on all of the first optimization factors;

[0051] When it is determined that the review effectiveness is less than the preset review effectiveness threshold, the review performance indicators in the data review strategy are analyzed based on the target cause to obtain the indicator analysis results, and the reasons to be optimized are generated based on the indicator analysis results and the target cause. Based on the reasons to be optimized, at least one second optimization factor matching the key verification factor is determined in a predetermined optimization parameter database, and the data review strategy is optimized based on all the second optimization factors.

[0052] As an optional implementation, in the second aspect of the present invention, the review module performs a preliminary review operation on the data to be reviewed based on a predetermined review mechanism, and a specific manner of obtaining the preliminary review result includes:

[0053] Generate a target review model based on the predetermined review mechanism and all data corresponding to the LLM security protection;

[0054] Inputting the data to be reviewed into the target review model to obtain a model review result, wherein the model review result includes a review result after performing a review operation on the data to be reviewed by the review mechanism;

[0055] Based on the model review result, a preliminary review result corresponding to the data to be reviewed is generated.

[0056] As an optional implementation, in the second aspect of the present invention, the acquisition module is further configured to obtain policy feedback information for the data review policy before the optimization module performs the optimization operation on the data review policy according to the target reason;

[0057] The judgment module is further configured to judge whether the strategy feedback information satisfies a preset strategy optimization condition;

[0058] The generating module is further configured to generate feedback optimization parameters for the data review strategy based on the strategy feedback information when the judging module judges that the strategy feedback information satisfies the preset strategy optimization condition;

[0059] The specific manner in which the optimization module performs the optimization operation on the data review strategy according to the target reason includes:

[0060] An optimization operation is performed on the data review strategy based on the target cause and the feedback optimization parameter.

[0061] A third aspect of the present invention discloses another content verification system based on LLM security protection, the system comprising:

[0062] a memory storing executable program code;

[0063] a processor coupled to the memory;

[0064] The processor calls the executable program code stored in the memory to execute the content verification method based on LLM security protection disclosed in the first aspect of the present invention.

[0065] The fourth aspect of the present invention discloses a computer storage medium, which stores computer instructions. When the computer instructions are called, they are used to execute the content verification method based on LLM security protection disclosed in the first aspect of the present invention.

[0066] Compared with the prior art, the embodiments of the present invention have the following beneficial effects:

[0067] In this embodiment of the present invention, data to be reviewed is determined from all data corresponding to the LLM security protection. A preliminary review operation is performed on the data to be reviewed based on a predetermined review mechanism to obtain a preliminary review result. Based on a predetermined target analysis algorithm and data review strategy, a data verification operation is performed on the preliminary review result to obtain a data verification result corresponding to the data to be reviewed. This shows that the implementation of the present invention enables intelligent verification of data in the LLM protection system, which is beneficial for improving the accuracy and reliability of data verification and further enhancing the operational security of the LLM security protection system. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0069] Figure 1 This is a flowchart of a content verification method based on LLM security protection disclosed in an embodiment of the present invention;

[0070] Figure 2 This is a flow chart of another content verification method based on LLM security protection disclosed in an embodiment of the present invention;

[0071] Figure 3 This is a schematic diagram of the structure of a content verification system based on LLM security protection disclosed in an embodiment of the present invention;

[0072] Figure 4 This is a schematic structural diagram of another content verification system based on LLM security protection disclosed in an embodiment of the present invention;

[0073] Figure 5 This is a structural diagram of another content verification system based on LLM security protection disclosed in an embodiment of the present invention. DETAILED DESCRIPTION

[0074] In order to help those skilled in the art better understand the solutions of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below based on the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0075] The terms "first," "second," and so on, in the description and claims of the present invention and the accompanying drawings are used to distinguish between different items, not to describe a specific order. Furthermore, the terms "including," "having," and any variations thereof, are intended to cover non-exclusive inclusions. For example, a process, method, system, product, or end comprising a series of steps or elements is not limited to the listed steps or elements but may optionally include steps or elements not listed, or may optionally include other steps or elements inherent to the process, method, product, or end.

[0076] References herein to "embodiments" mean that a particular feature, structure, or characteristic described in connection with the embodiments may be included in at least one embodiment of the present invention. The appearance of this phrase in various places in the specification does not necessarily refer to the same embodiment, nor does it constitute an independent or alternative embodiment that is mutually exclusive of other embodiments. It is understood, both explicitly and implicitly, by those skilled in the art that the embodiments described herein may be based on other embodiments.

[0077] The present invention discloses a content verification method and system based on LLM security protection, which can intelligently verify the data of the LLM protection system, thereby improving the accuracy and reliability of data verification and further improving the operational security of the LLM security protection system. A detailed description is provided below.

[0078] Example 1

[0079] See also Figure 1 , Figure 1 This is a flow chart of a content verification method based on LLM security protection disclosed in an embodiment of the present invention. Figure 1 The content verification method based on LLM security protection described above can be applied to a content verification system based on LLM security protection, wherein the content verification system based on LLM security protection can be integrated in a cloud server or a local server, which is not limited in the embodiment of the present invention. Figure 1 As shown, the content verification method based on LLM security protection may include the following operations:

[0080] 101. Determine the data to be reviewed from all data corresponding to the LLM security protection, perform a preliminary review operation on the data to be reviewed based on a predetermined review mechanism, and obtain a preliminary review result.

[0081] In an embodiment of the present invention, optionally, all data corresponding to LLM security protection may include all data included in the LLM security protection system. Furthermore, all data corresponding to LLM security protection may include one or more of input data, training data, output data, personal privacy data, plug-in data, and code data.

[0082] In the embodiment of the present invention, optionally, the number of data to be reviewed may be one, may be multiple, or may be all data corresponding to the LLM security protection, which is not specifically limited in the embodiment of the present invention.

[0083] 102. According to the predetermined target analysis algorithm and data review strategy, a data verification operation is performed on the preliminary review result to obtain a data verification result corresponding to the data to be reviewed.

[0084] In an embodiment of the present invention, optionally, the predetermined target analysis algorithm may include one or more of a multi-dimensional data analysis algorithm and a machine learning algorithm.

[0085] In the embodiment of the present invention, optionally, the data verification result corresponding to the data to be reviewed may include a data review result indicating the compliance and rationality of the data to be reviewed.

[0086] It can be seen that implementation Figure 1 The described content verification method based on LLM security protection can identify data to be reviewed from all data corresponding to the LLM security protection, perform a preliminary review operation on the data to be reviewed based on a predetermined review mechanism to obtain a preliminary review result, and perform a data verification operation on the preliminary review result based on a target analysis algorithm and a data review strategy to obtain a data verification result corresponding to the data to be reviewed. This method can effectively identify and filter out potentially harmful content through the predetermined review mechanism and data review strategy, thereby enhancing the security of the LLM. Through the preliminary review and data verification operations on the data to be reviewed, the method helps improve the LLM's resistance to malicious input and enhance the robustness of the model. While enhancing security, it can also ensure high-quality responses to benign instructions and maintain the usability of the LLM. The review mechanism and data review strategy can be flexibly adjusted to meet different security protection requirements according to different application scenarios and security requirements. By ensuring that the LLM output complies with laws, regulations, and industry standards, the method helps promote LLM compliance. It can intelligently verify the data of the LLM protection system, which is conducive to improving the accuracy and reliability of data verification, and further helps to improve the operational security of the LLM security protection system.

[0087] Example 2

[0088] See also Figure 2 , Figure 2 This is a flow chart of a content verification method based on LLM security protection disclosed in an embodiment of the present invention. Figure 2The content verification method based on LLM security protection described above can be applied to a content verification system based on LLM security protection, wherein the content verification system based on LLM security protection can be integrated in a cloud server or a local server, which is not limited in the embodiment of the present invention. Figure 2 As shown, the content verification method based on LLM security protection may include the following operations:

[0089] 201. Determine the data to be reviewed from all data corresponding to the LLM security protection, perform a preliminary review operation on the data to be reviewed based on a predetermined review mechanism, and obtain a preliminary review result.

[0090] 202. According to the predetermined target analysis algorithm and data review strategy, a data verification operation is performed on the preliminary review result to obtain a data verification result corresponding to the data to be reviewed.

[0091] In the embodiment of the present invention, for the detailed description of step 201 - step 202, please refer to the other descriptions of step 101 - step 102 in the first embodiment, and the embodiment of the present invention will not be repeated.

[0092] 203. Based on the data verification result, determine whether the data verification result meets the pre-determined protection verification condition.

[0093] In the embodiment of the present invention, optionally, the determining whether the data verification result satisfies a predetermined protection verification condition may include:

[0094] Determine the data security level corresponding to the data verification result, and judge whether the data security level is greater than or equal to the security level threshold corresponding to the pre-determined protection verification condition;

[0095] When it is determined that the data security level is greater than or equal to the security level threshold corresponding to the predetermined protection verification condition, it is determined that the data verification result meets the predetermined protection verification condition; when it is determined that the data security level is less than the security level threshold corresponding to the predetermined protection verification condition, it is determined that the data verification result does not meet the predetermined protection verification condition.

[0096] 204. When it is determined that the data verification result does not meet the predetermined protection verification condition, a target reason why the data verification result does not meet the protection verification condition is determined based on the data verification result and the protection verification condition.

[0097] In the embodiment of the present invention, further optionally, when it is determined that the data verification result meets the predetermined protection verification condition, the process can be terminated.

[0098] In the embodiment of the present invention, optionally, the determining, based on the data verification result and the protection verification condition, of the target reason why the data verification result does not meet the protection verification condition may include:

[0099] Based on the data verification results and the protection verification conditions, the difference information between the data verification results and the protection verification conditions is determined, and the target reason why the data verification results do not meet the protection verification conditions is generated according to the difference information.

[0100] 205. Perform optimization operations on data review strategies based on target reasons.

[0101] It can be seen that implementation Figure 2 The described content verification method based on LLM security protection can determine whether the data verification result meets the predetermined protection verification condition. If not, the target reason for the data verification result not meeting the protection verification condition is determined based on the data verification result and the protection verification condition, and the data review strategy is optimized according to the target reason. By determining whether the data verification result meets the predetermined protection verification condition, it can accurately identify which data failed the security review, thereby improving the accuracy of security protection. After determining that the data verification result does not meet the protection verification condition, it can determine the target reason for the non-satisfaction based on the verification result and the protection condition, and optimize the data review strategy according to the target reason. Dynamic adjustment can be achieved to ensure that the review strategy can adapt to the ever-changing security threats and data environment, thereby improving overall protection capabilities. By optimizing the data review strategy, unnecessary reviews and processing can be reduced, the overall efficiency of the system can be improved, and the rational use of resources can be ensured. Effective security protection measures and continuous optimization processes can also enhance user trust in the system and improve user satisfaction. The data of the LLM protection system can be intelligently verified, which is conducive to improving the accuracy and reliability of data verification, and further conducive to improving the operational security of the LLM security protection system.

[0102] In an optional embodiment, according to a predetermined target analysis algorithm and data review strategy, a data verification operation is performed on the preliminary review result. Before obtaining the data verification result corresponding to the data to be reviewed, the method further includes:

[0103] Obtaining comprehensive review information, where the comprehensive review information includes one or more of language review information, cultural review information, and background review information;

[0104] Performing information preprocessing operations on the comprehensive review information to obtain preprocessed review information;

[0105] Determine whether the pre-processed review information meets the preset information review conditions;

[0106] When it is determined that the pre-processed review information meets the preset information review conditions, a data review strategy is generated based on the pre-processed review information and the information pre-processing operation.

[0107] In this optional embodiment, the comprehensive review information may be obtained in real time or periodically according to a preset time period, which is not specifically limited in the embodiment of the present invention.

[0108] In this optional embodiment, the language review information may optionally include language information corresponding to the review strategy, for example, Chinese, English, etc.; the cultural review information may include normative information of the cultural group corresponding to the review strategy; and the background review information may include normative information corresponding to the social background.

[0109] In this optional embodiment, the information preprocessing operation may optionally include one or more of a sensitive information detection operation, a sensitive information marking operation, a sensitive information deletion operation, and a sensitive information replacement operation; for example, the information preprocessing operation can automatically detect and mark biased, discriminatory, or other content that does not comply with ethical standards, thereby ensuring the ethical compliance of the output.

[0110] In this optional embodiment, optionally, the above-mentioned determination of whether the pre-processed review information meets the preset information review conditions may include:

[0111] Determine whether the pre-processed review information contains ethically sensitive information that corresponds to the preset information review conditions;

[0112] When it is determined that the pre-processing review information does not contain any ethical norm sensitive information corresponding to the preset information review conditions, it is determined that the pre-processing review information meets the preset information review conditions; when it is determined that the pre-processing review information contains ethical norm sensitive information corresponding to the preset information review conditions, it is determined that the pre-processing review information does not meet the preset information review conditions.

[0113] In this optional embodiment, optionally, the generation of a data review strategy based on the pre-processed review information and the information pre-processing operation may include:

[0114] According to the information preprocessing operation, a first preprocessing parameter is determined, and a data review strategy is generated based on the preprocessing review information and the first preprocessing parameter; wherein the data review strategy at least includes the preprocessing review information and the first preprocessing parameter.

[0115] It can be seen that the implementation of this optional embodiment can obtain comprehensive review information and perform information preprocessing operations on the comprehensive review information to obtain preprocessed review information, determine whether the preprocessed review information meets the preset information review conditions, and if so, generate a data review strategy based on the preprocessed review information and the information preprocessing operations. By obtaining comprehensive review information including language, culture, and background, it can cover all aspects of the review more comprehensively, improve the level of detail and accuracy of the review, and through the operation of preprocessing the review information, it can reduce the interference of redundant and irrelevant information, making the review process more efficient and speeding up the review process. Through the preset information review conditions, it can ensure that only information that meets the specific requirements is reviewed. Standardized pre-processed review information is used to generate data review strategies, thereby improving the accuracy of review results. Data review strategies generated based on pre-processed review information that meets preset conditions can better adapt to review requirements of different languages, cultures, and backgrounds, and improve the applicability and flexibility of strategies. By comprehensively considering various aspects of review information and performing pre-processing and conditional judgment, more accurate, efficient, and adaptable data review strategies can be generated, thereby improving the overall review effect. The data of the LLM protection system can be intelligently verified, which is conducive to improving the accuracy and reliability of data verification, and further conducive to improving the operational security of the LLM security protection system.

[0116] In another optional embodiment, the method further includes:

[0117] When it is determined that the pre-processed review information does not meet the preset information review conditions, determining the operation parameters corresponding to the information pre-processing operation, and determining the difference parameters between the pre-processed review information and the information review conditions;

[0118] Based on the operation parameters and the difference parameters, the operation adjustment parameters for the information preprocessing operation are generated, and the information preprocessing operation is updated based on the operation adjustment parameters, and the information preprocessing operation on the comprehensive review information is re-triggered to obtain the operation of preprocessing the review information and trigger the execution of the operation of judging whether the preprocessing review information meets the preset information review conditions.

[0119] In this optional embodiment, optionally, the difference parameter between the pre-processed review information and the information review condition may include a difference parameter between the pre-processed review information and the benchmark review information corresponding to the information review condition.

[0120] In this optional embodiment, optionally, generating the operation adjustment parameter for the information preprocessing operation based on the operation parameter and the difference parameter may include:

[0121] Based on the operation parameters and the difference parameters, an operation difference parameter between the operation parameters and the difference parameters is determined, and based on the operation difference parameter, an operation adjustment parameter for the information preprocessing operation that matches the operation difference parameter is determined in a predetermined adjustment parameter library.

[0122] It can be seen that the implementation of this optional embodiment can determine the operating parameters corresponding to the information preprocessing operation and determine the difference parameters between the preprocessing review information and the information review conditions when it is determined that the preprocessing review information does not meet the information review conditions, and generate the operation adjustment parameters for the information preprocessing operation based on the operating parameters and the difference parameters, update the information processing operation, and re-judge whether the preprocessing review information meets the preset information review conditions. By determining the operating parameters corresponding to the information preprocessing operation and the difference parameters between the preprocessing review information and the information review conditions, the preprocessing operation can be more accurately identified and adjusted, thereby improving the precision and accuracy of the review. The operation adjustment parameters generated based on the operating parameters and the difference parameters enhance the adaptability and flexibility of the system, and by re-triggering the information preprocessing operation, invalid or inefficient data processing can be reduced, resource utilization can be optimized, and review efficiency can be improved. It can also self-optimize the preprocessing operation based on the difference parameters to achieve continuous performance improvement and adapt to new review requirements. It can intelligently verify the data of the LLM protection system, which is conducive to improving the accuracy and reliability of data verification, and further conducive to improving the operational security of the LLM security protection system.

[0123] In yet another optional embodiment, performing optimization operations on the data review strategy according to the target reason includes:

[0124] Determine the review effectiveness of the data review strategy based on the target reason and judge whether the review effectiveness is greater than or equal to the preset review effectiveness threshold;

[0125] When it is determined that the review effectiveness is greater than or equal to a preset review effectiveness threshold, at least one first optimization factor matching the key verification factor is determined in a predetermined optimization parameter database based on the target reason, and an optimization operation is performed on the data review strategy based on all the first optimization factors;

[0126] When it is determined that the review effectiveness is less than the preset review effectiveness threshold, the review performance indicators in the data review strategy are analyzed based on the target cause to obtain the indicator analysis results, and the reasons to be optimized are generated based on the indicator analysis results and the target cause. Based on the reasons to be optimized, at least one second optimization factor that matches the key verification factor is determined in a predetermined optimization parameter database, and optimization operations are performed on the data review strategy based on all the second optimization factors.

[0127] In this optional embodiment, the review validity of the data review policy may optionally include a parameter for indicating the validity of reviewing the data to be reviewed through the data review policy.

[0128] In this optional embodiment, key verification factors are optionally included in the data review strategy as core evaluation indicators or key elements for measuring and verifying the review effectiveness, accuracy or reliability.

[0129] In this optional embodiment, optionally, determining at least one first optimization factor that matches the key verification factor in a predetermined optimization parameter database based on the target reason may include:

[0130] A first cause keyword of the target cause is extracted, and based on the first cause keyword, an optimization factor matching the first cause keyword is determined in a predetermined optimization parameter database, and all optimization factors matching the first cause keyword are determined as first optimization factors matching the key verification factor.

[0131] In this optional embodiment, optionally, analyzing the review performance indicators in the data review strategy based on the target cause to obtain the indicator analysis results, and generating the reasons to be optimized based on the indicator analysis results and the target cause may include:

[0132] Extract the second reason keyword of the target reason, and extract the target review performance index in the data review strategy, generate an indicator analysis model based on the target review performance index, input the second reason keyword into the indicator analysis model, obtain the indicator analysis result, and generate the reason to be optimized based on the indicator analysis result and the second reason keyword.

[0133] It can be seen that the implementation of this optional embodiment can determine the review effectiveness of the data review strategy according to the target cause and judge whether it is greater than or equal to the preset review effectiveness threshold. If so, the first optimization factor is determined in the optimization parameter database based on the target cause, and then the optimization operation is performed on the data review strategy. Otherwise, the review performance index in the data review strategy is analyzed based on the target cause to obtain the index analysis result and the cause to be optimized is generated based on the index analysis result and the target cause. Based on the cause to be optimized, the second optimization factor matching the key verification factor is determined in the optimization parameter database, and then the optimization operation is performed on the data review strategy. By judging the review effectiveness and selecting the optimization factor accordingly, the review strategy can be improved. effect, ensuring the accuracy and reliability of the review results, and through precise optimization operations, it can reduce the occurrence of false positives and missed reports, and improve the accuracy of the review. Through continuous optimization operations, the review strategy can be continuously improved to adapt to new review challenges, and it can also be based on the optimized review strategy to better comply with laws, regulations and industry standards, reduce compliance risks, and dynamic optimization based on target reasons not only improves the effectiveness and adaptability of the review strategy, but also improves the system's automation level and self-improvement capabilities. It can intelligently verify the data of the LLM protection system, which is conducive to improving the accuracy and reliability of data verification, and further helps to improve the operational security of the LLM security protection system.

[0134] In yet another optional embodiment, a preliminary review operation is performed on the data to be reviewed based on a predetermined review mechanism to obtain a preliminary review result, including:

[0135] Generate a target review model based on the pre-determined review mechanism and all data corresponding to LLM security protection;

[0136] Input the data to be reviewed into the target review model to obtain the model review result, wherein the model review result includes the review result after the review operation is performed on the data to be reviewed through the review mechanism;

[0137] Based on the model review results, generate preliminary review results corresponding to the data to be reviewed.

[0138] In this optional embodiment, optionally, generating a target review model based on the predetermined review mechanism and all data corresponding to the LLM security protection may include:

[0139] Based on the predetermined review mechanism and all data corresponding to the LLM security protection, a predetermined backup review model is input to obtain the model output result;

[0140] Determine whether the model output results meet the preset model operation conditions;

[0141] When it is determined that the model output result meets the preset model operation conditions, the backup review model is determined as the target review model;

[0142] When it is determined that the model output result does not meet the preset model operating conditions, the result difference value between the model output result and the preset model operating conditions is analyzed, and the backup review model is updated based on the result difference value, and the operation of determining whether the model output result meets the preset model operating conditions is re-triggered.

[0143] In this optional embodiment, optionally, the preliminary review results corresponding to the data to be reviewed include at least model review results.

[0144] It can be seen that the implementation of this optional embodiment can generate a target review model based on a predetermined review mechanism and all data corresponding to the LLM security protection, and input the data to be reviewed into the target review model to obtain a model review result, and generate a preliminary review result corresponding to the data to be reviewed based on the model review result. By generating a target review model and inputting the data to be reviewed, the model review result can be automatically obtained, which not only improves the efficiency of the review, but also improves the accuracy of the review by utilizing the capabilities of LLM, and combined with all data corresponding to the LLM security protection, it can better identify and prevent potential security threats and enhance overall security. By analyzing the model review results, the review mechanism and model can be continuously optimized, supporting the continuous improvement and updating of the review strategy, and can intelligently verify the data of the LLM protection system, which is conducive to improving the accuracy and reliability of data verification, and further conducive to improving the operational security of the LLM security protection system.

[0145] In yet another optional embodiment, before performing the optimization operation on the data review strategy according to the target reason, the method further includes:

[0146] Obtaining policy feedback information for the data review policy and determining whether the policy feedback information meets the preset policy optimization conditions;

[0147] When it is determined that the policy feedback information meets the preset policy optimization conditions, feedback optimization parameters for the data review policy are generated based on the policy feedback information;

[0148] Among them, according to the target reasons, the data review strategy is optimized, including:

[0149] Optimize the data review strategy based on the target reasons and feedback optimization parameters.

[0150] In this optional embodiment, the policy feedback information may optionally include usage feedback information for the data review policy.

[0151] In this optional embodiment, optionally, the above-mentioned determination of whether the policy feedback information satisfies a preset policy optimization condition may include:

[0152] Determine the feedback satisfaction level corresponding to the strategy feedback information, and judge whether the feedback satisfaction level is greater than or equal to a preset feedback satisfaction level threshold corresponding to the strategy optimization condition;

[0153] When it is determined that the feedback satisfaction is greater than or equal to the feedback satisfaction threshold corresponding to the preset strategy optimization condition, it is determined that the strategy feedback information meets the preset strategy optimization condition; when it is determined that the feedback satisfaction is less than the feedback satisfaction threshold corresponding to the preset strategy optimization condition, it is determined that the strategy feedback information does not meet the preset strategy optimization condition.

[0154] In this optional embodiment, further optionally, when it is determined that the policy feedback information does not meet the preset policy optimization conditions, the process may be terminated.

[0155] In this optional embodiment, optionally, feedback optimization parameters for the data review strategy are matched with the strategy feedback information.

[0156] In this optional embodiment, optionally, performing the optimization operation on the data review strategy based on the target cause and the feedback optimization parameter may include:

[0157] Extracting a first optimization keyword from the target reason and a second optimization keyword from the feedback optimization parameter, and generating a target optimization keyword based on the first optimization keyword and the second optimization keyword;

[0158] Based on the target optimization keywords, target optimization parameters are determined, and optimization operations are performed on the data review strategy based on the target optimization parameters.

[0159] It can be seen that the implementation of this optional embodiment can obtain policy feedback information for the data review strategy and determine whether the preset policy optimization conditions are met. If so, feedback optimization parameters for the data review strategy are generated based on the policy feedback information, and optimization operations are performed on the data review strategy according to the target cause and the feedback optimization parameters. By collecting policy feedback information and determining whether it meets the optimization conditions, it can be ensured that only effective feedback is used in the optimization process, thereby improving the effectiveness of the data review strategy. Based on real-time or periodic policy feedback information, the data review strategy can be dynamically adjusted to adapt to the ever-changing data review needs and environment. The optimized strategy can more effectively allocate review resources, reduce resource waste in low-risk areas, and concentrate resources on high-risk areas. The optimized review strategy can better comply with laws, regulations and industry standards, and reduce compliance risks. By combining policy feedback information and target causes, not only the effectiveness and adaptability of the review strategy are improved, but also the automation level and self-improvement ability of the system are improved. The data of the LLM protection system can be intelligently verified, which is conducive to improving the accuracy and reliability of data verification, and further conducive to improving the operational security of the LLM security protection system.

[0160] Example 3

[0161] See also Figure 3 , Figure 3 This is a schematic diagram of the structure of a content verification system based on LLM security protection disclosed in an embodiment of the present invention. Figure 3 As shown, the content verification system based on LLM security protection may include:

[0162] A determination module 301 is used to determine the data to be reviewed from all data corresponding to the LLM security protection;

[0163] The review module 302 is configured to perform a preliminary review operation on the data to be reviewed based on a predetermined review mechanism to obtain a preliminary review result;

[0164] The verification module 303 is used to perform a data verification operation on the preliminary review result according to a predetermined target analysis algorithm and data review strategy to obtain a data verification result corresponding to the data to be reviewed.

[0165] It can be seen that implementation Figure 3The described system can identify data to be reviewed from all data corresponding to the LLM security protection and perform a preliminary review operation on the data to be reviewed based on a predetermined review mechanism to obtain a preliminary review result. The system then performs a data verification operation on the preliminary review result based on a target analysis algorithm and a data review strategy to obtain a data verification result corresponding to the data to be reviewed. The system can effectively identify and filter out potentially harmful content through the predetermined review mechanism and data review strategy, thereby enhancing the security of the LLM. Through the preliminary review and data verification operations on the data to be reviewed, the method helps improve the LLM's resistance to malicious input and enhance the robustness of the model. While enhancing security, it can ensure high-quality responses to benign instructions and maintain the LLM's usability. The system can flexibly adjust the review mechanism and data review strategy to meet different security protection requirements based on different application scenarios and security requirements. Furthermore, by ensuring that the LLM's output complies with laws, regulations, and industry standards, the method helps promote LLM compliance. The system can intelligently verify the data of the LLM protection system, thereby improving the accuracy and reliability of data verification and further enhancing the operational security of the LLM security protection system.

[0166] In an optional embodiment, if Figure 4 As shown, the system also includes:

[0167] The judgment module 304 is used to judge whether the data verification result meets the predetermined protection verification condition according to the data verification result;

[0168] The determination module 301 is further configured to determine a target reason why the data verification result does not meet the predetermined protection verification condition based on the data verification result and the protection verification condition when the judgment module 304 determines that the data verification result does not meet the predetermined protection verification condition;

[0169] The optimization module 305 is used to perform optimization operations on the data review strategy according to the target reason.

[0170] It can be seen that implementation Figure 4The described system can determine whether the data verification result meets the predetermined protection verification conditions. If not, the system can determine the target reason why the data verification result does not meet the protection verification conditions based on the data verification result and the protection verification conditions, and perform optimization operations on the data review strategy based on the target reason. By determining whether the data verification result meets the predetermined protection verification conditions, it can accurately identify which data failed the security review, thereby improving the accuracy of security protection. After determining that the data verification result does not meet the protection verification conditions, it can determine the target reason for the non-satisfaction based on the verification result and the protection conditions, and optimize the data review strategy based on the target reason. Dynamic adjustment can be achieved to ensure that the review strategy can adapt to the ever-changing security threats and data environment, thereby improving overall protection capabilities. By optimizing the data review strategy, unnecessary reviews and processing can be reduced, the overall efficiency of the system can be improved, and the rational use of resources can be ensured. Effective security protection measures and continuous optimization processes can also enhance user trust in the system and improve user satisfaction. The system can intelligently verify the data of the LLM protection system, which is conducive to improving the accuracy and reliability of data verification, and further conducive to improving the operational security of the LLM security protection system.

[0171] In another optional embodiment, as Figure 4 As shown, the system also includes:

[0172] Acquisition module 306 is configured to obtain comprehensive review information, including one or more of language review information, cultural review information, and background review information, before verification module 303 performs data verification on the preliminary review results according to a predetermined target analysis algorithm and data review strategy to obtain the data verification results corresponding to the data to be reviewed.

[0173] Processing module 307, configured to perform information preprocessing on the comprehensive review information to obtain preprocessed review information;

[0174] The judgment module 304 is further used to judge whether the pre-processed review information meets the preset information review conditions;

[0175] The generating module 308 is configured to generate a data review strategy based on the pre-processing review information and the information pre-processing operation when the judging module 304 judges that the pre-processing review information meets the preset information review conditions.

[0176] It can be seen that implementation Figure 4The described system can obtain comprehensive review information and perform information preprocessing operations on the comprehensive review information to obtain preprocessed review information, determine whether the preprocessed review information meets preset information review conditions, and if so, generate a data review strategy based on the preprocessed review information and the information preprocessing operations. By obtaining comprehensive review information including language, culture, and background, it can more comprehensively cover all aspects of the review, improving the level of detail and accuracy of the review. The preprocessing of the review information can also reduce the interference of redundant and irrelevant information, making the review process more efficient and accelerating the review process. The preset information review conditions can ensure that only preprocessed review information that meets specific standards is used to generate the data review strategy, thereby improving the accuracy of the review results. The data review strategy generated based on the preprocessed review information that meets the preset conditions can better adapt to the review needs of different languages, cultures, and backgrounds, improving the applicability and flexibility of the strategy. By comprehensively considering multiple aspects of review information and performing preprocessing and conditional judgment, it can generate a more accurate, efficient, and adaptable data review strategy, thereby improving the overall review effect. The system can also intelligently verify the data of the LLM protection system, which is conducive to improving the accuracy and reliability of data verification and further improving the operational security of the LLM security protection system.

[0177] In another optional embodiment, Figure 4 As shown, the determination module 301 is further configured to determine the operation parameters corresponding to the information preprocessing operation and the difference parameters between the preprocessing review information and the information review condition when the judgment module 304 determines that the preprocessing review information does not meet the preset information review condition;

[0178] The generation module 308 is also used to generate operation adjustment parameters for the information preprocessing operation based on the operation parameters and the difference parameters, and update the information preprocessing operation based on the operation adjustment parameters, and re-trigger the processing module 307 to perform the information preprocessing operation on the comprehensive review information to obtain the preprocessed review information and trigger the execution judgment module 304 to perform the operation of judging whether the preprocessed review information meets the preset information review conditions.

[0179] It can be seen that implementation Figure 4The described system can determine the operating parameters corresponding to the information preprocessing operation and the difference parameters between the preprocessing review information and the information review conditions when it is determined that the preprocessing review information does not meet the information review conditions, and generate operation adjustment parameters for the information preprocessing operation based on the operating parameters and the difference parameters, update the information processing operation, and re-judge whether the preprocessing review information meets the preset information review conditions. By determining the operating parameters corresponding to the information preprocessing operation and the difference parameters between the preprocessing review information and the information review conditions, the preprocessing operation can be more accurately identified and adjusted, thereby improving the precision and accuracy of the review. The operation adjustment parameters generated based on the operating parameters and the difference parameters enhance the adaptability and flexibility of the system, and by re-triggering the information preprocessing operation, invalid or inefficient data processing can be reduced, resource utilization can be optimized, and review efficiency can be improved. It can also self-optimize the preprocessing operation based on the difference parameters to achieve continuous performance improvement and adapt to new review requirements. It can intelligently verify the data of the LLM protection system, which is conducive to improving the accuracy and reliability of data verification, and further conducive to improving the operational safety of the LLM security protection system.

[0180] In another optional embodiment, Figure 4 As shown, the optimization module 305 performs optimization operations on the data review strategy according to the target reason in the following specific manners:

[0181] Determine the review effectiveness of the data review strategy based on the target reason and judge whether the review effectiveness is greater than or equal to the preset review effectiveness threshold;

[0182] When it is determined that the review effectiveness is greater than or equal to a preset review effectiveness threshold, at least one first optimization factor matching the key verification factor is determined in a predetermined optimization parameter database based on the target reason, and an optimization operation is performed on the data review strategy based on all the first optimization factors;

[0183] When it is determined that the review effectiveness is less than the preset review effectiveness threshold, the review performance indicators in the data review strategy are analyzed based on the target cause to obtain the indicator analysis results, and the reasons to be optimized are generated based on the indicator analysis results and the target cause. Based on the reasons to be optimized, at least one second optimization factor that matches the key verification factor is determined in a predetermined optimization parameter database, and optimization operations are performed on the data review strategy based on all the second optimization factors.

[0184] It can be seen that implementation Figure 4The described system can determine the review effectiveness of the data review strategy based on the target cause and judge whether it is greater than or equal to the preset review effectiveness threshold. If so, a first optimization factor is determined in the optimization parameter database based on the target cause and then the optimization operation is performed on the data review strategy. Otherwise, the review performance index in the data review strategy is analyzed based on the target cause to obtain the index analysis result and the cause to be optimized is generated based on the index analysis result and the target cause. Based on the cause to be optimized, a second optimization factor matching the key verification factor is determined in the optimization parameter database and then the optimization operation is performed on the data review strategy. By judging the review effectiveness and selecting the optimization factor accordingly, the effect of the review strategy can be improved. , ensuring the accuracy and reliability of the review results, and through precise optimization operations, it can reduce the occurrence of false positives and missed reports, and improve the accuracy of the review. Through continuous optimization operations, the review strategy can be continuously improved to adapt to new review challenges, and it can also be based on the optimized review strategy to better comply with laws, regulations and industry standards, reduce compliance risks, and dynamic optimization based on target reasons. It not only improves the effectiveness and adaptability of the review strategy, but also improves the system's automation level and self-improvement capabilities. It can intelligently verify the data of the LLM protection system, which is conducive to improving the accuracy and reliability of data verification, and further helps to improve the operational security of the LLM security protection system.

[0185] In another optional embodiment, Figure 4 As shown, the review module 302 performs a preliminary review operation on the data to be reviewed based on a predetermined review mechanism. The specific method of obtaining the preliminary review result includes:

[0186] Generate a target review model based on the pre-determined review mechanism and all data corresponding to LLM security protection;

[0187] Input the data to be reviewed into the target review model to obtain the model review result, wherein the model review result includes the review result after the review operation is performed on the data to be reviewed through the review mechanism;

[0188] Based on the model review results, generate preliminary review results corresponding to the data to be reviewed.

[0189] It can be seen that implementation Figure 4The described system can generate a target review model based on a predetermined review mechanism and all data corresponding to LLM security protection, and input the data to be reviewed into the target review model to obtain a model review result, and generate a preliminary review result corresponding to the data to be reviewed based on the model review result. By generating a target review model and inputting the data to be reviewed, the model review result can be automatically obtained, which not only improves the efficiency of the review, but also improves the accuracy of the review by utilizing the capabilities of LLM, and combined with all data corresponding to LLM security protection, it can better identify and prevent potential security threats and enhance overall security. By analyzing the model review results, the review mechanism and model can be continuously optimized, supporting the continuous improvement and updating of the review strategy, and can intelligently verify the data of the LLM protection system, which is conducive to improving the accuracy and reliability of data verification, and further conducive to improving the operational security of the LLM security protection system.

[0190] In another optional embodiment, Figure 4 As shown, the acquisition module 306 is further used to obtain policy feedback information for the data review policy before the optimization module 305 performs the optimization operation on the data review policy according to the target reason;

[0191] The judgment module 304 is further used to judge whether the policy feedback information meets the preset policy optimization conditions;

[0192] The generating module 308 is further configured to generate feedback optimization parameters for the data review strategy based on the strategy feedback information when the judging module 304 judges that the strategy feedback information satisfies the preset strategy optimization conditions;

[0193] The specific manner in which the optimization module 305 performs optimization operations on the data review strategy according to the target reason includes:

[0194] Optimize the data review strategy based on the target reasons and feedback optimization parameters.

[0195] It can be seen that implementation Figure 4The described system can obtain policy feedback information for the data review strategy and determine whether it meets the preset policy optimization conditions. If so, it generates feedback optimization parameters for the data review strategy based on the policy feedback information, and performs optimization operations on the data review strategy according to the target cause and the feedback optimization parameters. By collecting policy feedback information and determining whether it meets the optimization conditions, it can ensure that only effective feedback is used in the optimization process, thereby improving the effectiveness of the data review strategy. Based on real-time or periodic policy feedback information, the data review strategy can be dynamically adjusted to adapt to the ever-changing data review needs and environment. The optimized strategy can more effectively allocate review resources, reduce resource waste in low-risk areas, and concentrate resources on high-risk areas. The optimized review strategy can better comply with laws, regulations and industry standards, and reduce compliance risks. By combining policy feedback information and target causes, not only the effectiveness and adaptability of the review strategy are improved, but also the automation level and self-improvement ability of the system are improved. It can intelligently verify the data of the LLM protection system, which is conducive to improving the accuracy and reliability of data verification, and further conducive to improving the operational security of the LLM security protection system.

[0196] Example 4

[0197] See also Figure 5 , Figure 5 This is a schematic diagram of the structure of another content verification system based on LLM security protection disclosed in an embodiment of the present invention. Figure 5 As shown, the content verification system based on LLM security protection may include:

[0198] A memory 401 storing executable program code;

[0199] a processor 402 coupled to the memory 401;

[0200] The processor 402 calls the executable program code stored in the memory 401 to execute the steps of the content verification method based on LLM security protection described in the first embodiment of the present invention or the second embodiment of the present invention.

[0201] Example 5

[0202] An embodiment of the present invention discloses a computer storage medium, which stores computer instructions. When the computer instructions are called, they are used to execute the steps of the content verification method based on LLM security protection described in Example 1 or Example 2 of the present invention.

[0203] Example 6

[0204] An embodiment of the present invention discloses a computer program product, which includes a non-transitory computer-readable storage medium storing a computer program, and the computer program is operable to enable a computer to execute the steps of the content verification method based on LLM security protection described in Example 1 or Example 2.

[0205] The system embodiments described above are merely illustrative. The modules described as separate components may or may not be physically separate, and the components shown as modules may or may not be physical modules, i.e., they may be located in one place or distributed across multiple network modules. Some or all of the modules may be selected based on actual needs to achieve the objectives of the present embodiment. Persons of ordinary skill in the art will be able to understand and implement the present embodiment without inventive effort.

[0206] Through the detailed description of the above embodiments, those skilled in the art will clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform, or of course, by means of hardware. Based on this understanding, the above technical solution, in essence, or the portion that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as a read-only memory (ROM), a random access memory (RAM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), a one-time programmable read-only memory (OTPROM), an electronically erasable programmable read-only memory (EEPROM), a compact disc read-only memory (CD-ROM), or other optical disk storage, magnetic disk storage, magnetic tape storage, or any other computer-readable medium capable of carrying or storing data.

[0207] Finally, it should be noted that the content verification method and system based on LLM security protection disclosed in the embodiment of the present invention is only a preferred embodiment of the present invention, which is only used to illustrate the technical solution of the present invention, rather than to limit it; although the present invention has been described in detail with reference to the aforementioned embodiments, ordinary technicians in this field should understand that it is still possible to modify the technical solutions recorded in the aforementioned embodiments, or to make equivalent replacements for some of the technical features therein; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A content verification method based on LLM security protection, characterized in that: The method comprises: Determine the data to be reviewed from all data corresponding to the LLM security protection, perform a preliminary review operation on the data to be reviewed based on a predetermined review mechanism, and obtain a preliminary review result; Performing a data verification operation on the preliminary review result according to a predetermined target analysis algorithm and data review strategy to obtain a data verification result corresponding to the data to be reviewed; According to the data verification result, determining whether the data verification result meets a predetermined protection verification condition; When it is determined that the data verification result does not satisfy the predetermined protection verification condition, determining a target reason why the data verification result does not satisfy the protection verification condition based on the data verification result and the protection verification condition; performing an optimization operation on the data review strategy according to the target reason; The performing of an optimization operation on the data review strategy according to the target reason includes: Determining the review effectiveness of the data review strategy based on the target reason, and judging whether the review effectiveness is greater than or equal to a preset review effectiveness threshold; When it is determined that the review effectiveness is greater than or equal to the preset review effectiveness threshold, at least one first optimization factor matching the key verification factor is determined in a predetermined optimization parameter database based on the target cause, and an optimization operation is performed on the data review strategy based on all of the first optimization factors; When it is determined that the review effectiveness is less than the preset review effectiveness threshold, the review performance indicators in the data review strategy are analyzed based on the target cause to obtain an indicator analysis result, and a reason to be optimized is generated based on the indicator analysis result and the target cause, and at least one second optimization factor matching the key check factor is determined in a predetermined optimization parameter database based on the reason to be optimized, and an optimization operation is performed on the data review strategy based on all the second optimization factors; The step of determining at least one first optimization factor that matches a key verification factor in a predetermined optimization parameter database based on the target reason includes: extracting a first cause keyword of the target cause, and determining, based on the first cause keyword, an optimization factor that matches the first cause keyword in a predetermined optimization parameter database, and determining all optimization factors that match the first cause keyword as first optimization factors that match the key check factor; Furthermore, analyzing the review performance indicators in the data review strategy based on the target cause to obtain an indicator analysis result, and generating a reason to be optimized based on the indicator analysis result and the target cause, including: Extract the second reason keyword of the target reason, and extract the target review performance indicator in the data review strategy, generate an indicator analysis model based on the target review performance indicator, input the second reason keyword into the indicator analysis model, obtain the indicator analysis result, and generate the reason to be optimized based on the indicator analysis result and the second reason keyword.

2. The content verification method based on LLM security protection according to claim 1 is characterized in that: Before performing a data verification operation on the preliminary review result according to a predetermined target analysis algorithm and data review strategy to obtain a data verification result corresponding to the data to be reviewed, the method further includes: Obtaining comprehensive review information, wherein the comprehensive review information includes one or more of language review information, cultural review information, and background review information; performing an information preprocessing operation on the comprehensive review information to obtain preprocessed review information; Determining whether the pre-processed review information meets the preset information review conditions; When it is determined that the pre-processing review information meets the preset information review condition, a data review strategy is generated based on the pre-processing review information and the information pre-processing operation.

3. The content verification method based on LLM security protection according to claim 2 is characterized in that: The method further comprises: When it is determined that the pre-processing review information does not meet the preset information review condition, determining the operation parameters corresponding to the information pre-processing operation, and determining the difference parameters between the pre-processing review information and the information review condition; Based on the operation parameters and the difference parameters, operation adjustment parameters for the information preprocessing operation are generated, and based on the operation adjustment parameters, the information preprocessing operation is updated, and the operation of performing the information preprocessing operation on the comprehensive review information is re-triggered to obtain the preprocessing review information and to trigger the execution of the operation of determining whether the preprocessing review information meets the preset information review conditions.

4. The content verification method based on LLM security protection according to claim 1 is characterized in that: The preliminary review operation is performed on the data to be reviewed based on the predetermined review mechanism to obtain a preliminary review result, including: Generate a target review model based on the predetermined review mechanism and all data corresponding to the LLM security protection; Inputting the data to be reviewed into the target review model to obtain a model review result, wherein the model review result includes a review result after performing a review operation on the data to be reviewed by the review mechanism; Based on the model review result, a preliminary review result corresponding to the data to be reviewed is generated.

5. The content verification method based on LLM security protection according to claim 3 is characterized in that: Before performing the optimization operation on the data review strategy according to the target reason, the method further includes: Obtaining policy feedback information for the data review policy, and determining whether the policy feedback information meets a preset policy optimization condition; When it is determined that the policy feedback information satisfies the preset policy optimization condition, generating feedback optimization parameters for the data review policy based on the policy feedback information; The performing of an optimization operation on the data review strategy according to the target reason includes: An optimization operation is performed on the data review strategy based on the target cause and the feedback optimization parameter.

6. A content verification system based on LLM security protection, characterized in that: The system comprises: A determination module is used to determine the data to be reviewed from all the data corresponding to the LLM security protection; A review module, configured to perform a preliminary review operation on the data to be reviewed based on a predetermined review mechanism to obtain a preliminary review result; A verification module is used to perform a data verification operation on the preliminary review result according to a predetermined target analysis algorithm and data review strategy to obtain a data verification result corresponding to the data to be reviewed; A judgment module, configured to judge whether the data verification result satisfies a predetermined protection verification condition according to the data verification result; The determining module is further configured to, when the judging module determines that the data verification result does not satisfy the predetermined protection verification condition, determine, based on the data verification result and the protection verification condition, a target reason why the data verification result does not satisfy the protection verification condition; an optimization module, configured to perform an optimization operation on the data review strategy according to the target cause; The specific manner in which the optimization module performs the optimization operation on the data review strategy according to the target reason includes: Determining the review effectiveness of the data review strategy based on the target reason, and judging whether the review effectiveness is greater than or equal to a preset review effectiveness threshold; When it is determined that the review effectiveness is greater than or equal to the preset review effectiveness threshold, at least one first optimization factor matching the key verification factor is determined in a predetermined optimization parameter database based on the target cause, and an optimization operation is performed on the data review strategy based on all of the first optimization factors; When it is determined that the review effectiveness is less than the preset review effectiveness threshold, the review performance indicators in the data review strategy are analyzed based on the target cause to obtain an indicator analysis result, and a reason to be optimized is generated based on the indicator analysis result and the target cause, and at least one second optimization factor matching the key check factor is determined in a predetermined optimization parameter database based on the reason to be optimized, and an optimization operation is performed on the data review strategy based on all the second optimization factors; The specific manner in which the optimization module determines at least one first optimization factor matching the key verification factor in a predetermined optimization parameter database based on the target cause includes: extracting a first cause keyword of the target cause, and determining, based on the first cause keyword, an optimization factor that matches the first cause keyword in a predetermined optimization parameter database, and determining all optimization factors that match the first cause keyword as first optimization factors that match the key check factor; Furthermore, the optimization module analyzes the review performance indicators in the data review strategy based on the target cause to obtain an indicator analysis result, and generates a cause to be optimized based on the indicator analysis result and the target cause in a specific manner including: Extract the second reason keyword of the target reason, and extract the target review performance indicator in the data review strategy, generate an indicator analysis model based on the target review performance indicator, input the second reason keyword into the indicator analysis model, obtain the indicator analysis result, and generate the reason to be optimized based on the indicator analysis result and the second reason keyword.

7. A content verification system based on LLM security protection, characterized in that: The system comprises: a memory storing executable program code; a processor coupled to the memory; The processor calls the executable program code stored in the memory to execute the content verification method based on LLM security protection as described in any one of claims 1-5.

8. A computer storage medium, characterized in that The computer storage medium stores computer instructions, and when the computer instructions are called, they are used to execute the content verification method based on LLM security protection as described in any one of claims 1 to 5.

Images