Website published information auditing method and device

Abstract

The invention provides a website published information auditing method and device, and the method comprises the steps that a security auditing engine receives an encryption auditing model from a training server, securely loads the encryption auditing model in a trusted execution environment of a hardware security module, requests the hardware security module to decrypt the loaded encryption auditing model, and sends the decrypted encryption auditing model to the training server; the hardware security module is used for loading an encrypted auditing model, so that the hardware security module decrypts the loaded encrypted auditing model to obtain a decrypted auditing model, the encrypted auditing model is obtained by encrypting an auditing model through the hardware security module, and the auditing model is obtained by pre-training a large model through the training server; and performing auditing based on a to-be-published information text of a website by adopting the decryption auditing model, and outputting an auditing result corresponding to the to-be-published information text. According to the method, the intelligence, efficiency, accuracy and data security of website published information auditing can be improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of artificial intelligence and information auditing technology, and in particular to a website published information auditing method and device. BACKGROUND

[0002] The traditional information auditing method is checked by artificial, with the increase of information auditing amount, it is increasingly faced with the problem of inaccurate auditing and low efficiency, and the negative remarks and the risk of information leakage caused by the non-strict control of the auditing personnel are significantly increased. Some auditing methods use the extension program function of the browser to embed different auditing programs into the browser auditing system, and the information field to be audited is compared with the predefined specification for auditing. With the system update or the browser reinstallation operation, this way needs to repeatedly install the plug-in, too many plug-ins and extensions may occupy a lot of resources, resulting in slow browser running and low auditing efficiency. Moreover, the above-mentioned auditing method has poor flexibility, low intelligence level, and the accuracy of simultaneous auditing needs to be improved. SUMMARY

[0003] In view of this, the embodiments of the present application provide a website published information auditing method and device to eliminate or improve one or more defects in the prior art.

[0004] One aspect of the present application provides a website published information auditing method, the method comprising the following steps: The security review engine receives an encrypted auditing model from a training server, securely loads the encrypted auditing model in a trusted execution environment of a hardware security module, and requests the hardware security module to decrypt and load the encrypted auditing model, so that the hardware security module decrypts the loaded encrypted auditing model to obtain a decrypted auditing model, wherein the encrypted auditing model is obtained by encrypting an auditing model by the hardware security module, and the auditing model is obtained by pre-training a large model by the training server; The decrypted auditing model is used to audit the website published information text, and the auditing result corresponding to the published information text is output.

[0005] In some embodiments of the present application, the decrypted auditing model is used to audit the website published information text, and the auditing result corresponding to the published information text is output, comprising: The security review engine requests the training server to obtain an encrypted sample library, so that the training server performs security verification and authorization access based on identity information carried in the request sent by the security review engine, and returns the encrypted sample library and verification and authorization results to the security review engine after verification and authorization are passed and agreed, wherein the encrypted sample library is obtained by encrypting a sample library by the hardware security module, and the sample library includes multiple samples of multiple violation types, each sample including violation information text, a corresponding violation type label and violation judgment logic; The security review engine requests the hardware security module to decrypt the encrypted sample library and sends the verification and authorization results, so that the hardware security module decrypts the encrypted sample library to obtain the information that the security review engine is verified and authorized after obtaining the verification and authorization results. The decrypted review model is used to perform semantic understanding and identify implicit violation content based on the website information text to be published, and output multiple keywords. At least one sample similar in semantics to the multiple keywords is obtained based on the similarity between the multiple keywords and each sample in the decrypted sample library. The decrypted review model is used to output an audit result corresponding to the information text to be published based on the multiple keywords and the at least one sample, and the audit result includes a result of whether the information text to be published is violation information and a violation type and violation judgment logic corresponding to the information text to be published.

[0006] In some embodiments of the present application, before the step of outputting the audit result, the decrypted review model is used to perform audit based on the website information text to be published, and output an audit result corresponding to the information text to be published, which further includes: the decrypted review model is used to sort the at least one sample from high to low based on the semantic similarity between the multiple keywords and the at least one sample, and obtain the sorted at least one sample. Correspondingly, the decrypted review model is used to output an audit result corresponding to the information text to be published based on the multiple keywords and the sorted at least one sample.

[0007] In some embodiments of the present application, the audit model is obtained by the following steps: The training server requests the hardware security module to decrypt the encrypted sample library, so that the hardware security module decrypts the encrypted sample library to obtain a decrypted sample library, wherein the encrypted sample library is obtained by encrypting a sample library by the hardware security module, and the sample library includes multiple samples of multiple violation types, each sample including violation information text, a corresponding violation type label and violation judgment logic. Pre-training the large model based on the decrypted sample library to obtain an audit model.

[0008] In some embodiments of the present application, before the step of pre-training the large model based on the decrypted sample library, the pre-training step further comprises: adding perturbation to each sample in the decrypted sample library to generate each sample conforming to differential privacy to obtain a decrypted sample library conforming to differential privacy; Correspondingly, pre-training the large model based on the decrypted sample library conforming to differential privacy to obtain an audit model.

[0009] In some embodiments of the present application, the violation types include that the text contains sensitive words, the text will cause negative network comments after being published, or the effective period of the text does not conform to the preset audit specification requirements.

[0010] In some embodiments of the present application, the security review engine generates an audit log based on the to-be-published information text and the audit result corresponding to the to-be-published information text, generates a hash value based on the audit log and encrypts the hash value, and stores the audit log through blockchain technology based on the encrypted hash value.

[0011] Another aspect of the present application provides a website information publishing audit device, which comprises a computer device comprising a processor and a memory, the memory storing computer instructions, and the processor being configured to execute the computer instructions stored in the memory, so that the device implements the steps of the foregoing method.

[0012] Another aspect of the present application provides a computer readable storage medium storing a computer program, which is executed by a processor to implement the steps of the foregoing method.

[0013] Another aspect of the present application provides a computer program product comprising computer instructions, which are executed by a processor to implement the steps of the foregoing method.

[0014] The website information publishing audit method and device of the present application can solve the technical problems of poor flexibility, low intelligence level, and low accuracy of simultaneous auditing of the conventional audit method, thereby improving the audit efficiency, intelligence, and accuracy, and enabling the website information to be published accurately and quickly.

[0015] Additional advantages, objects, and features of the present application will be set forth in part in the description which follows, and will in part be apparent to those of ordinary skill in the art upon examination of the following or can be learned from a practice of the present application. The objects and other advantages of the present application can be realized and attained by the structure particularly pointed out in the specification and claims.

[0016] Those skilled in the art will understand that the objects and advantages of the application can be realized and attained by means illustrated in the following detailed description and accompanying drawings. BRIEF DESCRIPTION OF DRAWINGS

[0017] The accompanying drawings, which are included to provide a further understanding of the application and are incorporated in and constitute a part of this application, illustrate embodiments of the application and together with the detailed description serve to explain the principles of the application.

[0018] Figure 1 A flowchart of a website publishing information auditing method according to an embodiment of the application; Figure 2 A flowchart of a website publishing information auditing method according to an embodiment of the application. DETAILED DESCRIPTION

[0019] To make the objects, technical solutions and advantages of the application clearer, the following will further describe the application with reference to the embodiments and drawings. Herein, the illustrative embodiments of the application and their descriptions are used to explain the application, but not to limit the application.

[0020] It should be noted that, in order to avoid the application being obscured by unnecessary details, only the structures and / or processing steps closely related to the solutions according to the application are shown in the drawings, and other details not closely related to the application are omitted.

[0021] It should be emphasized that the terms "comprises / comprising" when used in this specification are taken to specify the presence of stated features, elements, steps or components, but do not preclude the presence or addition of one or more other features, elements, steps, components, or groups thereof.

[0022] It should be noted that, if not specifically stated, the term "connected" herein can not only mean direct connection, but also indirect connection with an intermediate.

[0023] In the following, embodiments of the application will be described with reference to the accompanying drawings. In the drawings, the same reference numerals represent the same or similar components, or the same or similar steps.

[0024] To solve the technical problem of poor flexibility, low intelligence level and low accuracy of simultaneous auditing of the conventional auditing method, the embodiments of the application provide a website publishing information auditing method and device, which can improve the auditing efficiency, intelligence and accuracy, and thus make the website publishing information be accurately and quickly published.

[0025] Figure 1 and Figure 2These are, respectively, a flowchart and a detailed flowchart of a website information posting review method according to an embodiment of the present invention. Figure 1 and Figure 2 As shown, the method includes the following steps: In step S110, the security audit engine receives the encrypted audit model from the training server, securely loads the encrypted audit model in the trusted execution environment of the hardware security module, and requests the hardware security module to decrypt the loaded encrypted audit model so that the hardware security module can decrypt the loaded encrypted audit model to obtain the decrypted audit model. The encrypted audit model is obtained by encrypting the audit model through the hardware security module, and the audit model is obtained by pre-training the large model through the training server.

[0026] Specifically, in the auditing system, the security auditing engine loads the encrypted auditing model into the Trusted Execution Environment (TEE) of the hardware security module through the hsm_api library, thereby creating an isolated computing environment that prevents external parties from stealing or tampering with the data. The security auditing engine requests model decryption from the hardware security module, which then decrypts the encrypted auditing model to obtain the decrypted auditing model.

[0027] Step S120: Using the decryption and review model, review the information text to be published on the website and output the review result corresponding to the information text to be published.

[0028] The security review engine uses a decryption review model to perform security reasoning on the text of information to be published on the website and outputs the review result. The review result includes whether the text of the proposed publication constitutes prohibited content. If the text of the proposed publication does not constitute prohibited content, it is published directly. If the text of the proposed publication does constitute prohibited content, it is sent to a human review queue for further human review. If the further human review also finds the text of the proposed publication to be prohibited content, it is modified to eliminate the violation. The modified text is then reviewed using this method until the text of the proposed publication no longer constitutes prohibited content and meets the publication requirements. Through these steps, the quality and efficiency of the review and publication of information to be published on the website are improved. Furthermore, by conducting the review in a trusted environment and performing encryption and decryption operations, data security during the review process is also enhanced.

[0029] In some embodiments, step S120, which involves using the decryption review model to review the information text to be published on the website and outputting the review result corresponding to the information text to be published, includes the following steps: The security review engine requests the encrypted sample library from the training server, so that the training server can perform security verification and authorization access based on the identity information carried in the request sent by the security review engine. After the verification is successful and the authorization is granted, the encrypted sample library and the verification and authorization results are returned to the security review engine. The encrypted sample library is obtained by encrypting the sample library through the hardware security module. The sample library includes multiple samples of various violation types. Each sample includes violation information text, corresponding violation type label, and violation judgment logic. The security review engine requests the hardware security module to decrypt the encrypted sample library and sends the verification and authorization results. After the hardware security module obtains the information that the security review engine has been verified and authorized based on the verification and authorization results, it decrypts the encrypted sample library to obtain the decrypted sample library. Using the aforementioned decryption and review model, semantic understanding and identification of hidden violations are performed based on the text of the information to be published on the website, and multiple keywords are output; Based on the similarity between the multiple keywords and the samples in the decryption sample library, at least one sample that is semantically similar to the multiple keywords is obtained; Using the decryption review model, the review result corresponding to the proposed information text is output based on the multiple keywords and the at least one sample. The review result includes whether the proposed information text belongs to illegal information, as well as the violation type and violation judgment logic corresponding to the proposed information text.

[0030] In the above steps, the prohibited content in the text of the information is usually information that does not meet the publication requirements. For example, the security classification of the information may be inconsistent with the security classification of the original document, the validity period of the information may not meet the prescribed review standards, the title or body of the information may contain sensitive words, and the information may lead to or trigger negative online comments after publication. It may exist in the form of a compressed file. By calculating similarity and finding at least one sample from the decryption sample library that is similar to the prohibited content in the text of the information to be published, as a reference example for the decryption review model, the accuracy of the model's review of the text of the information to be published can be improved. Specifically, the above similarity can be calculated using methods such as cosine similarity.

[0031] In some other embodiments, before the step of outputting the review result, the step of using the decryption review model to review the information text to be published on the website and outputting the review result corresponding to the information text to be published further includes the following steps: using the decryption review model to sort the at least one sample from high to low based on the semantic similarity between the multiple keywords and the at least one sample, to obtain the sorted at least one sample; Correspondingly, using the decryption and review model, the review result corresponding to the proposed information text is output based on the multiple keywords and at least one sorted sample.

[0032] Based on the above steps, after finding at least one reference example that is similar to the illegal content contained in the text to be published, the decryption review model sorts these reference examples from high to low semantic similarity to the illegal content contained in the text to be published. This allows the example that is most semantically similar to the text to be published to be prioritized during the violation review process, thereby further improving the accuracy of text review and achieving efficient intelligent review.

[0033] In some embodiments, the audit model is pre-trained through the following steps: The training server requests the hardware security module to decrypt the encrypted sample library, so that the hardware security module can decrypt the encrypted sample library to obtain the decrypted sample library. The encrypted sample library is obtained by encrypting the sample library by the hardware security module. The sample library includes multiple samples of various violation types. Each sample includes violation information text, corresponding violation type label, and violation judgment logic. The large model is pre-trained based on the decrypted sample library to obtain the audit model.

[0034] The sample library contains multiple samples obtained by collecting violation information text, corresponding violation type tags, and violation judgment logic from historical review records. These samples are then encrypted and stored using a hardware security module to form an encrypted sample library. A large model is deployed on a private server for pre-training, and this pre-trained model (the review model) is also encrypted using the hardware security module. This creates a training system with encryption capabilities, consisting of the private server and the hardware security module. The large model can be a model such as Deepseek-11m-7b. Through this pre-training process, the trained review model can learn the violation content characteristics of specific violation types in specific business scenarios, as well as the violation judgment standards and logic, thereby improving the accuracy of its information review.

[0035] In other embodiments, the pre-training step further includes, prior to the step of pre-training the large model based on the decrypted sample library: Perturbations are added to each sample in the decryption sample library to generate samples that conform to differential privacy, thus obtaining a decryption sample library that conforms to differential privacy. Correspondingly, the large model is pre-trained based on a decryption sample library that conforms to differential privacy to obtain the audit model.

[0036] In this embodiment, by adding perturbations to the input data of the large training model, the anti-interference capability of the review model can be improved, achieving more accurate information review. In this embodiment, the review model is also encrypted using a hardware security module to obtain an encrypted review model, which is then sent to the security review engine in the review system for information review.

[0037] In some embodiments, the types of violations include text containing sensitive words, text that could lead to negative online comments after publication, or text whose validity period does not meet preset review guidelines. Sensitive words include content related to user privacy, politically sensitive terms, classified information, and misleading or misleading information.

[0038] In some embodiments, the method further includes the following steps: The security review engine generates a review log based on the text of the information to be published and the review result corresponding to the text of the information to be published. It generates a hash value based on the review log and encrypts the hash value. Based on the encrypted hash value, the review log is stored using blockchain technology.

[0039] By storing the audit results and the security audit engine's audit process on the blockchain using the above method, subsequent queries, retrieval, and traceability are facilitated, and log tampering is prevented. This leverages the immutability of the blockchain to achieve traceability of the audit logs. Specifically, the blockchain consists of a series of blocks, each storing a certain number of audit records. These blocks are linked to the previous block via a hash function, with each block containing a hash value pointing to the previous block, forming a continuous chain. This structure ensures data continuity and integrity. Its advantage lies in the fact that once a new audit record is added to a block, it is permanently stored in the system and cannot be tampered with. All audit processes are recorded on the blockchain in real time, and system administrators or authorized users can view these records at any time for auditing and supervision.

[0040] Furthermore, in this embodiment of the invention, through the aforementioned series of encryption and decryption operations and the interaction between the security review engine, training server, and hardware security module, the entire process of model pre-training and review is conducted in a highly confidential environment, ensuring that even if the text contains sensitive data, it will not be leaked to the outside world. Therefore, this method can also achieve efficient and accurate review of information to be published on websites under a highly confidential premise, thereby achieving efficient and accurate website information dissemination, mitigating the risk of negative online comments, and improving data security during the information review process.

[0041] Corresponding to the above method, the present invention also provides a website information posting review device, which includes a computer device, the computer device including a processor and a memory, the memory storing computer instructions, the processor being used to execute the computer instructions stored in the memory, and when the computer instructions are executed by the processor, the device implements the steps of the aforementioned website information posting review method.

[0042] This invention also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of the aforementioned website information posting review method. The computer-readable storage medium can be a tangible storage medium, such as random access memory (RAM), main memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, floppy disks, hard disks, removable storage disks, CD-ROMs, or any other form of storage medium known in the art.

[0043] This invention also provides a computer program product, including computer instructions, which, when executed by a processor, implement the steps of the aforementioned website information review method.

[0044] Those skilled in the art will understand that the exemplary components, systems, and methods described in conjunction with the embodiments disclosed herein can be implemented in hardware, software, or a combination of both. Whether implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this invention. When implemented in hardware, it can be, for example, electronic circuits, application-specific integrated circuits (ASICs), appropriate firmware, plug-ins, function cards, etc. When implemented in software, the elements of this invention are programs or code segments used to perform the desired tasks. The programs or code segments can be stored in a machine-readable medium or transmitted over a transmission medium or communication link via data signals carried in a carrier wave.

[0045] It should be clarified that the present invention is not limited to the specific configurations and processes described above and shown in the figures. For the sake of brevity, detailed descriptions of known methods are omitted here. In the above embodiments, several specific steps are described and shown as examples. However, the method process of the present invention is not limited to the specific steps described and shown. Those skilled in the art can make various changes, modifications, and additions, or change the order of steps, after understanding the spirit of the present invention.

[0046] In this invention, features described and / or illustrated for one embodiment may be used in the same or similar manner in one or more other embodiments, and / or combined with or in place of features of other embodiments.

[0047] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, various modifications and variations of the embodiments of the present invention are possible. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for reviewing information published on a website, characterized in that, The method includes: The security audit engine receives an encrypted audit model from the training server, securely loads the encrypted audit model in the trusted execution environment of the hardware security module, and requests the hardware security module to decrypt the loaded encrypted audit model so that the hardware security module can decrypt the loaded encrypted audit model to obtain a decrypted audit model. The encrypted audit model is obtained by encrypting the audit model by the hardware security module, and the audit model is obtained by pre-training a large model by the training server. Using the aforementioned decryption and review model, the website reviews the text of the information to be published and outputs the review result corresponding to the text of the information to be published.

2. The method according to claim 1, characterized in that, The aforementioned decryption review model is used to review the information text to be published on the website, and the review result corresponding to the information text to be published is output, including: The security review engine requests the encrypted sample library from the training server, so that the training server can perform security verification and authorization access based on the identity information carried in the request sent by the security review engine. After the verification is successful and the authorization is granted, the encrypted sample library and the verification and authorization results are returned to the security review engine. The encrypted sample library is obtained by encrypting the sample library through the hardware security module. The sample library includes multiple samples of various violation types. Each sample includes violation information text, corresponding violation type label, and violation judgment logic. The security review engine requests the hardware security module to decrypt the encrypted sample library and sends the verification and authorization results. After the hardware security module obtains the information that the security review engine has been verified and authorized based on the verification and authorization results, it decrypts the encrypted sample library to obtain the decrypted sample library. Using the aforementioned decryption and review model, semantic understanding and identification of hidden violations are performed based on the text of the information to be published on the website, and multiple keywords are output; Based on the similarity between the multiple keywords and the samples in the decryption sample library, at least one sample that is semantically similar to the multiple keywords is obtained; Using the decryption review model, the review result corresponding to the proposed information text is output based on the multiple keywords and the at least one sample. The review result includes whether the proposed information text belongs to illegal information, as well as the violation type and violation judgment logic corresponding to the proposed information text.

3. The method according to claim 2, characterized in that, Before the step of outputting the review result, the step of using the decryption review model to review the information text to be published on the website and outputting the review result corresponding to the information text to be published also includes: Using the decryption verification model, the at least one sample is sorted from high to low based on the semantic similarity between the multiple keywords and the at least one sample, resulting in the sorted at least one sample; Correspondingly, using the decryption and review model, the review result corresponding to the proposed information text is output based on the multiple keywords and at least one sorted sample.

4. The method according to claim 1, characterized in that, The audit model is pre-trained through the following steps: the training server requests the hardware security module to decrypt the encrypted sample library, so that the hardware security module decrypts the encrypted sample library to obtain the decrypted sample library. The encrypted sample library is obtained by encrypting the sample library through the hardware security module. The sample library includes multiple samples of various violation types. Each sample includes violation information text, corresponding violation type label, and violation judgment logic. The large model is pre-trained based on the decrypted sample library to obtain the audit model.

5. The method according to claim 4, characterized in that, Before the step of pre-training the large model based on the decrypted sample library, the pre-training step also includes: Perturbations are added to each sample in the decryption sample library to generate samples that conform to differential privacy, thus obtaining a decryption sample library that conforms to differential privacy. Correspondingly, the large model is pre-trained based on a decryption sample library that conforms to differential privacy to obtain the audit model.

6. The method according to claim 2 or 4, characterized in that, The types of violations include text containing sensitive words, text that could lead to negative online comments after publication, or text whose validity period does not meet the preset review standards.

7. The method according to claim 1, characterized in that, The method further includes: The security review engine generates a review log based on the text of the information to be published and the review result corresponding to the text of the information to be published. It generates a hash value based on the review log and encrypts the hash value. Based on the encrypted hash value, the review log is stored using blockchain technology.

8. A website information posting review device, comprising a processor, a memory, and computer instructions stored in the memory, characterized in that, The processor is configured to execute the computer instructions, and when the computer instructions are executed, the device implements the steps of the method as described in any one of claims 1 to 7.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the steps of the method as described in any one of claims 1 to 7.

10. A computer program product comprising computer instructions, characterized in that, When executed by a processor, the computer instructions implement the steps of the method according to any one of claims 1 to 7.

Images