File content auditing method and system for enterprise cloud disk based on deep learning

Abstract

The invention relates to an enterprise cloud disk file content auditing method and system based on a deep learning technology, and the method comprises the steps: triggering a corresponding auditing program according to a file type, and carrying out the analysis of the extracted file content, so as to judge whether the file content contains data in a sensitive lexicon or not; if the content contains the sensitive word, further judging whether the content contains data in the associated word bank or not; and if the content does not contain the data of the associated lexicon or contains the data of the associated lexicon and the type of the content is forbidden, forbidding operation is executed on the file, and a notification is sent to a user. And if the content contains the content and the type is passing, not taking any treatment measures. The invention aims to extract and analyze content from a file to audit the content of the file. The method aims at preventing files related to sensitive information such as pornographic information, violence information and political information from appearing in an enterprise cloud disk, and therefore legal risks and safety problems caused by the files are avoided. The time required for manually checking the file content is shortened, and the labor cost is reduced; information leakage is prevented, and document security is ensured; and the file content auditing accuracy is improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of artificial intelligence, and specifically relates to a file content review method and system for an enterprise cloud disk based on deep learning. Background Art

[0002] With the rapid development of cloud computing and big data technology, more and more companies are beginning to rely on cloud disks to store and manage massive amounts of files and data. However, the files stored in these cloud disks often contain some sensitive information or inappropriate content, the existence of which may bring legal risks and security risks to companies. Traditional file content review methods mainly rely on manual review, but this method has many problems such as slow review speed, low efficiency, prone to errors, and high labor costs. Although some cloud disks have begun to use content review tools for automated review, most of the file content review tools on the market are based on rule matching or simple machine learning algorithms, which have certain limitations in accuracy and adaptability. Therefore, companies urgently need a more intelligent, efficient and accurate file content review method to help companies effectively manage and control the file content in cloud disks and ensure information security.

[0003] With the continuous advancement of artificial intelligence technology, it is now possible to read feature information in images through programs, extract text information from various files (such as PDF, Word, Excel, etc.) through programs, and process video and audio files through programs to extract information from each frame for analysis and processing. These technologies provide a solid technical foundation for creating more advanced and efficient file content review methods.

[0004] At present, there are some file content review tools on the market, but most of these tools use rule-based matching or simple machine learning methods, which have problems such as low accuracy and poor adaptability. Therefore, in order to meet the higher requirements of enterprises for file content review, a more intelligent and accurate file content review method is urgently needed to help enterprises effectively manage and control the file content in cloud disks and ensure information security. Summary of the invention

[0005] In view of the defects of the prior art, the present invention proposes a method and system for enterprise cloud disk file content review based on deep learning. The invention belongs to the field of artificial intelligence, covering image information analysis technology, document content extraction technology, and video / audio decomposition analysis and calculation technology. The present invention is a deep learning method that focuses on extracting content from files and analyzing them to review the content of files. The method aims to prevent the appearance of files involving sensitive information such as pornography, violence, and politics in enterprise cloud disks, thereby avoiding the legal risks and security issues caused by this; reduce the time required for manual review of file content and reduce labor costs; prevent information leakage and ensure document security; improve the accuracy of file content review, enhance learning effects, and summarize errors in manual review to prevent the same errors from being repeated in the future.

[0006] To achieve the above-mentioned object, the present invention provides a first aspect of a file content review method for an enterprise cloud disk based on deep learning, which at least comprises the following steps: Step S1: Upload the file uploaded by the user to the quarantine area first, and then use the file detection program to detect and kill it. If the file contains a virus or a Trojan horse, the file will be deleted and the user will be notified; Step S2: After the check is passed, the corresponding program will be triggered to review the file content according to the file type; Step S3: Determine whether the extracted file content contains data from the sensitive word library; if so, determine whether it contains data from the associated word library; if it does not contain the associated word library or contains the data and the type is blocked, directly block the file and send a message to notify the user; if it contains the data and the type is passed, do not process it.

[0007] Furthermore, in step S2, the file type includes one or more of a picture type, a text type, an audio type, and a video type; after determining the type of the file, different types of files trigger different programs for machine review, and after the review is passed, the file is moved to the storage area.

[0008] Furthermore, the blocking conditions for review content can be adjusted through the management of sensitive word libraries, associated word libraries, and white-name word libraries; The sensitive word library and the management word library are combined to determine whether the sentences containing sensitive words are pornographic, violent, or political, thereby ensuring the accuracy of the document content.

[0009] Furthermore, the content review steps for image files are as follows: Image receiving: receiving images uploaded by users for review; Image preprocessing: preprocess the uploaded images for subsequent processing; Content recognition: Identify and analyze the content of the image; identify the characters, scenes, objects, and possible sensitive content in the image; if sensitive content is identified, directly block the file, intercept the file, enter the file blocking library, and send a message to notify the user; Security detection: Perform security detection on images, including detecting whether they contain pornographic, violent, or terrorist sensitive content; if sensitive content is identified, the file will be directly blocked and a message will be sent to the user; If the image passes the content identification and security detection, the image review is successful.

[0010] Furthermore, the content review steps for text files are as follows: File receiving: receiving files uploaded by users; File extraction: extract the contents of the file; File preprocessing: preprocess the text content, including format removal, word segmentation, and stop word removal; Text review: Determine whether the text content contains data from the sensitive word library. If so, determine whether it contains data from the associated word library. If it does not contain the associated word library or contains the data and the type is blocked, directly block the file, intercept the file, enter the file blocking library, and send a message to notify the user; if it contains the data and the type is passed, no processing will be done; If the text recognition and security detection are passed, the text file audit is successful.

[0011] Furthermore, the content review steps for audio files are as follows: Audio receiving: receiving audio uploaded by users for review; Audio preprocessing: Preprocess the audio, extract sound features, perform semantic analysis and emotion recognition through programs, automatically analyze the audio content, and identify possible illegal content; Audio Transliteration: Use language recognition technology to convert audio into text, and then use text review methods for detection; Audio review: If the review passes, the audio file review is successful.

[0012] Furthermore, the content review steps for video files are as follows: Video receiving: receiving videos uploaded by users for review; Video preprocessing: preprocess the received video and perform video transcoding to adapt to the subsequent analysis and review process; Video decomposition: decompose the preprocessed video into images, sounds and subtitles; Video content analysis: extract key frames from images, and then detect the key frames using the image review method; detect subtitle information using the text review method; and detect audio using the audio review method; Video review: Once the images, sounds and subtitles are reviewed and approved, the video file will be reviewed and approved.

[0013] Furthermore, it also includes an error correction processing mechanism, and its specific method is as follows: If a false positive occurs during the document content review, you can file an appeal against the false positive file. After review by the administrator, you can proceed as follows: Method 1: Only release the file, and subsequent files involving the same sensitive words will still be blocked; Method 2: Process the sensitive words involved in the file, remove them from the sensitive word library and add them to the white-name word library or increase its associated word library to prevent other files involving the words from being mistakenly blocked.

[0014] Furthermore, by implementing the management of the file interception list, it is possible to view and summarize the sensitive words contained in the intercepted files; in-depth analysis of these sensitive words will help to promptly discover and correct erroneous interceptions and remove erroneous sensitive words from the sensitive word library.

[0015] The second aspect of the present invention provides a file content review system for an enterprise cloud disk based on deep learning, the system is used in the above method, and the system at least includes File storage module, used to store various types of files uploaded by users; A file virus and Trojan killing module is used to detect and kill files uploaded by users to identify whether the uploaded files have viruses or Trojans; The video review module is used to pre-process, decompose, analyze and review the video types uploaded by users; The audio review module is used to pre-process, convert and review audio content uploaded by users; This article review module is used to extract files, preprocess files, and review text content uploaded by users; Image review module, used to perform image preprocessing, content recognition and security detection on image types uploaded by users; A sensitive word management module, wherein the sensitive word management module has a built-in sensitive word library and is responsible for detecting whether the content of the uploaded file contains sensitive words included in the sensitive word library; a white-name word library management module, wherein the white-name word library management module has a built-in white-name word library and is responsible for detecting whether the content of the uploaded file contains white-list words included in the white-name word library; an associated word library management module, wherein the white-name word library management module has a built-in associated word library and is responsible for detecting whether the content of the uploaded file contains associated words included in the associated word library; adjusting the blocking conditions of the audited content through the management of the sensitive word library, the associated word library and the white-name word library; combining the sensitive word library with the management word library to determine whether the sentence in which the sensitive word appears involves pornography, violence or politics, thereby providing the accuracy of the file content; The banned file management module intercepts uploaded files involving sensitive information and is responsible for the management of banned files.

[0016] The technical solution adopted by the present invention first temporarily stores the files uploaded by the user in an isolated area, and then uses a file detection program to detect them. If the file contains a virus or a Trojan horse, the file will be automatically deleted and a notification will be sent to the user. After the detection program confirms that it is harmless, it triggers the corresponding review program according to the file type to conduct a detailed inspection of the file content. During the review process, the system will determine one by one whether the file content contains data in the sensitive vocabulary. If sensitive words are found, further analysis is performed to determine whether data in the associated vocabulary is involved. If the file content does not involve the associated vocabulary or contains and is classified as a banned type, the ban measures are immediately executed on the file and the user is notified. If it contains sensitive words but is classified as passed, no measures are taken. The invention has at least the following advantages: 1. Based on the machine review results, the system can count and display the sensitive words and their frequency of occurrence contained in the review files in the management interface.

[0017] 2. The management interface provides effective management functions for sensitive word libraries, white-name word libraries, and associated word libraries to improve the accuracy and timeliness of sensitive words. By managing sensitive word libraries and white-name word libraries, the timeliness and accuracy of these word libraries are enhanced. Combining sensitive word libraries with management word libraries, the system can determine whether sentences containing sensitive words involve inappropriate content, such as pornography, violence, politics, etc., thereby improving the accuracy of file content review. The review is not limited to content review of individual sensitive words, but also combines the associated word library to review the entire sentence content, thereby improving the accuracy of content review.

[0018] 3. A file isolation library is set up to ensure that the files to be reviewed are first stored in the library and are checked and processed before being stored to prevent viruses and other viruses from affecting other files.

[0019] 4. Introduce an error correction mechanism. When an error occurs in the file review, the manual reviewer can confirm whether to process the sensitive words in the review file (i.e. remove them from the sensitive word library and add them to the whitelist library), or add corresponding associated words and associated types to the sensitive words, so as to avoid the same error in future reviews. In the process of manual review of appeal files, an associated word library and processing type are established for sensitive words based on manual operations, effectively avoiding the recurrence of the wrong interception phenomenon in the same scenario. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0021] Figure 1 This is the file content review method process for enterprise cloud disk based on deep learning of the present invention Figure 1 ; Figure 2 This is the file content review method process for enterprise cloud disk based on deep learning of the present invention Figure 2 . DETAILED DESCRIPTION

[0022] Exemplary embodiments will be described in detail herein, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, the same numbers in different drawings represent the same or similar elements unless otherwise indicated. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present invention. Instead, they are merely examples of devices and methods consistent with some aspects of the present invention as detailed in the appended claims.

[0023] like Figure 1 and Figure 2 As shown, the first aspect of this embodiment provides a file content review method for an enterprise cloud disk based on deep learning, which at least includes the following steps: Step S1: Upload the file uploaded by the user to the quarantine area first, and then use the file detection program to detect and kill it. If the file contains a virus or a Trojan horse, the file will be deleted and the user will be notified; Step S2: After the check is passed, the corresponding program will be triggered to review the file content according to the file type; Step S3: Determine whether the extracted file content contains data from the sensitive word library; if so, determine whether it contains data from the associated word library; if it does not contain the associated word library or contains the data and the type is blocked, directly block the file and send a message to notify the user; if it contains the data and the type is passed, do not process it.

[0024] As a preferred implementation, in step S2 of this embodiment, the file type includes one or more of a picture type, a text type, an audio type, and a video type; after determining the type of the file, different types of files trigger different programs for machine review, and after the review is passed, the file is moved to the storage area.

[0025] As a preferred implementation, in this embodiment, the blocking conditions of the audited content are adjusted by managing the sensitive word library, the associated word library and the white-name word library; The sensitive word library and the management word library are combined to determine whether the sentences containing sensitive words are pornographic, violent, or political, thereby ensuring the accuracy of the document content.

[0026] As a preferred implementation, the content review steps for a file type of picture in this embodiment are as follows: Image receiving: receiving images uploaded by users for review; Image preprocessing: Perform preprocessing operations such as format conversion and size adjustment on uploaded images for subsequent processing; Content recognition: Identify and analyze the content of the image; identify the characters, scenes, objects and other elements in the image, as well as possible sensitive content; if sensitive content is identified, directly block the file, intercept the file, enter the file blocking library, and send a message to notify the user; Security detection: Perform security detection on images, including detecting whether they contain sensitive content such as pornography, violence, terrorism, etc. If sensitive content is identified, the file will be directly blocked and a message will be sent to the user; If the image passes the content identification and security detection, the image review is successful.

[0027] As a preferred implementation, the content review steps for a text file in this embodiment are as follows: File receiving: receiving files uploaded by users; File extraction: extract the contents of documents, ppt, pdf and other files; File preprocessing: preprocess the text content, including format removal, word segmentation, and stop word removal; Text review: Determine whether the text content contains data from the sensitive word library. If so, determine whether it contains data from the associated word library. If it does not contain the associated word library or contains the data and the type is blocked, directly block the file, intercept the file, enter the file blocking library, and send a message to notify the user; if it contains the data and the type is passed, no processing will be done; If the text recognition and security detection are passed, the text file audit is successful.

[0028] As a preferred implementation, the content review steps for audio files in this embodiment are as follows: Audio receiving: receiving audio uploaded by users for review; Audio preprocessing: Preprocess the audio, extract sound features, perform semantic analysis and emotion recognition through programs, automatically analyze the audio content, and identify possible illegal content; Audio Transliteration: Use language recognition technology to convert audio into text, and then use text review methods for detection; Audio review: If the review passes, the audio file review is successful.

[0029] As a preferred implementation, the content review steps for a video file in this embodiment are as follows: Video receiving: receiving videos uploaded by users for review; Video preprocessing: preprocess the received video and perform video transcoding to adapt to the subsequent analysis and review process; Video decomposition: decompose the preprocessed video into images, sounds and subtitles; Video content analysis: extract key frames from images, and then detect the key frames using the image review method; detect subtitle information using the text review method; and detect audio using the audio review method; Video review: Once the images, sounds and subtitles are reviewed and approved, the video file will be reviewed and approved.

[0030] In this embodiment, after the file is reviewed and approved, the file is moved from the isolation area to the storage area for storage. If the file is mistakenly blocked by the program, a manual appeal can be made.

[0031] It also includes an error correction mechanism. If a false positive occurs during the document content review, you can appeal against the false positive file and the administrator can review and handle it: Method 1: only release the file, and subsequent files involving the same sensitive words will still be blocked; Method 2: process the sensitive words involved in the file, remove them from the sensitive word library and add them to the white-name word library or increase its associated word library to prevent other files involving the vocabulary from being mistakenly blocked.

[0032] As a preferred implementation, in this embodiment, by implementing the management of the file interception list, the sensitive words contained in the intercepted files can be viewed and summarized; in-depth analysis of these sensitive words helps to timely discover and correct the situation of incorrect interception, and remove the incorrect sensitive words from the sensitive word library. By managing the sensitive word library and the white-name word library, the program can be adjusted to improve the accuracy of content review.

[0033] A second aspect of this embodiment provides a file content review system for an enterprise cloud disk based on deep learning, the system is used in the above method, and the system at least includes File storage module, used to store various types of files uploaded by users; A file virus and Trojan killing module is used to detect and kill files uploaded by users to identify whether the uploaded files have viruses or Trojans; The video review module is used to pre-process, decompose, analyze and review the video types uploaded by users; The audio review module is used to pre-process, convert and review audio content uploaded by users; This article review module is used to extract files, preprocess files, and review text content uploaded by users; Image review module, used to perform image preprocessing, content recognition and security detection on image types uploaded by users; A sensitive word management module, wherein the sensitive word management module has a built-in sensitive word library and is responsible for detecting whether the content of the uploaded file contains sensitive words included in the sensitive word library; a white-name word library management module, wherein the white-name word library management module has a built-in white-name word library and is responsible for detecting whether the content of the uploaded file contains white-list words included in the white-name word library; an associated word library management module, wherein the white-name word library management module has a built-in associated word library and is responsible for detecting whether the content of the uploaded file contains associated words included in the associated word library; adjusting the blocking conditions of the audited content through the management of the sensitive word library, the associated word library and the white-name word library; combining the sensitive word library with the management word library to determine whether the sentence in which the sensitive word appears involves pornography, violence or politics, thereby providing the accuracy of the file content; The banned file management module intercepts uploaded files involving sensitive information and is responsible for the management of banned files.

[0034] Another implementation is also provided, as follows: Customers trigger the content review program by uploading files by right-clicking, dragging, or clicking a button to upload files; the files uploaded by users are first uploaded to the quarantine area, and then checked by the file checker. If the file has a virus or Trojan, the file will be deleted and the user will be notified; after the check is passed, the corresponding program will be triggered to review the file content according to the file type. During the content review, the file will be pre-processed and then the file content will be extracted; the extracted file content will be used to determine whether it contains data from the sensitive word library. If it does, it will be determined whether it contains data from the associated word library. If it does not contain the associated word library or it contains and the type is blocked, the file will be directly blocked and a message will be sent to the user; if it contains and the type is passed, it will not be processed. For example, if there is this sentence in the file: The traces left by the war are bloody, and the buildings destroyed by artillery shells and the bodies of innocent victims are all proof of the cruelty of war. If the sensitive word library contains the word bloody, the associated word library will be searched based on bloody. If it is not found, the file will be blocked and a message will be sent to the user.

[0035] After receiving the notification, the user can file an appeal. If the reviewer finds that the sensitive word is a misjudgment, the user can proceed as follows: If you delete "bloody" from the sensitive word library, the word "bloody" will not be blocked in subsequent files.

[0036] Add a related word to "bloody", the content of the related word is "war", and the type is "pass". If the subsequent file contains the sentence "bloody" and "war", it will not be intercepted.

[0037] Another implementation method is as follows: customers can trigger the file content review process by right-clicking, dragging and dropping, or clicking a button; the files uploaded by users will first be transferred to the isolation area and then detected by the file detection program. If the file is found to contain malware such as viruses or Trojans, it will be automatically deleted and a notification will be sent to the user. After the detection program confirms that it is harmless, it will start the corresponding review program to check the file content according to the file type. During the content review stage, the file will undergo preprocessing and its content will be extracted for analysis. Next, the system will determine whether the extracted content contains words in the sensitive word library. If sensitive words are found, it will further determine whether it contains words in the associated word library. If it does not contain or contains and is marked as a banned type, the file will be banned and a notification will be sent to the user. If it contains and is marked as a pass type, the file will not be restricted. For example, if a sentence like "The traces left by war are bloody. The buildings destroyed by artillery shells and the remains of innocent victims are all proof of the cruelty of war" appears in a file, and the sensitive vocabulary contains the word "bloody", but no related words are found in the associated vocabulary, the file will be blocked and the user will be notified.

[0038] After receiving the ban notification, the user has the right to file an appeal, which will be reviewed by the auditor. If the auditor confirms that the word "bloody" is a misjudgment, the following measures can be taken: Delete bloody from the sensitive word library, and the word bloody will not be blocked in subsequent files. Add an associated word to bloody, and the associated word content is war, and the type is pass. If the word bloody is also included in the sentence war in subsequent files, it will not be blocked.

[0039] The present invention focuses on extracting content from files and analyzing it to review the content of the files. The method provided in this embodiment can prevent the appearance of files involving sensitive information such as pornography, violence, and politics in the enterprise cloud disk, thereby avoiding the legal risks and security issues caused by this; reduce the time required for manual review of file content and reduce labor costs; prevent information leakage and ensure document security; improve the accuracy of file content review, enhance learning effects, and summarize errors in manual review to prevent the same errors from being repeated in the future.

[0040] Although the embodiments of the present invention have been shown and described above, it will be appreciated that the above embodiments are exemplary and are not to be construed as limitations of the present invention. A person skilled in the art may change, modify, substitute and vary the above embodiments within the scope of the present invention.

Claims

1. A file content review method for enterprise cloud disk based on deep learning, characterized by: At least the following steps are included: Step S1: Upload the file uploaded by the user to the quarantine area first, and then use the file detection program to detect and kill it. If the file contains a virus or a Trojan horse, the file will be deleted and the user will be notified; Step S2: After the check is passed, the corresponding program will be triggered to review the file content according to the file type; Step S3: Determine whether the extracted file content contains data from the sensitive word library; if so, determine whether it contains data from the associated word library; if it does not contain the associated word library or contains the data and the type is blocked, directly block the file and send a message to notify the user; if it contains the data and the type is passed, do not process it.

2. The method according to claim 1, characterized in that: In step S2, the file type includes one or more of a picture type, a text type, an audio type, and a video type; after determining the type of the file, different types of files trigger different programs for machine review, and after the review is passed, the file is moved to the storage area.

3. The method according to claim 2, characterized in that: Adjust the blocking conditions of the audited content through the management of sensitive word libraries, associated word libraries, and white-name word libraries; The sensitive word library and the management word library are combined to determine whether the sentences containing sensitive words are pornographic, violent, or political, thereby ensuring the accuracy of the document content.

4. The method according to claim 3, characterized in that: The steps for content review of image files are as follows: Image receiving: receiving images uploaded by users for review; Image preprocessing: preprocess the uploaded images for subsequent processing; Content recognition: Identify and analyze the content of the image; identify the characters, scenes, objects, and possible sensitive content in the image; if sensitive content is identified, directly block the file, intercept the file, enter the file blocking library, and send a message to notify the user; Security detection: Perform security detection on images, including detecting whether they contain pornographic, violent, or terrorist sensitive content; if sensitive content is identified, the file will be directly blocked and a message will be sent to the user; If the image passes the content identification and security detection, the image review is successful.

5. The method according to claim 4, characterized in that: The content review steps for text files are as follows: File receiving: receiving files uploaded by users; File extraction: extract the contents of the file; File preprocessing: preprocess the text content, including format removal, word segmentation, and stop word removal; Text review: Determine whether the text content contains data from the sensitive word library. If so, determine whether it contains data from the associated word library. If it does not contain the associated word library or contains the data and the type is blocked, directly block the file, intercept the file, enter the file blocking library, and send a message to notify the user; if it contains the data and the type is passed, no processing will be done; If the text recognition and security detection are passed, the text file audit is successful.

6. The method according to claim 5, characterized in that: The steps for content review of audio files are as follows: Audio receiving: receiving audio uploaded by users for review; Audio preprocessing: Preprocess the audio, extract sound features, perform semantic analysis and emotion recognition through programs, automatically analyze the audio content, and identify possible illegal content; Audio translation: Use language recognition technology to convert audio into text, and then use the text review method in claim 5 for detection; Audio review: If the review passes, the audio file review is successful.

7. The method according to claim 1, characterized in that: The steps for content review of video files are as follows: Video receiving: receiving videos uploaded by users for review; Video preprocessing: preprocess the received video and perform video transcoding to adapt to the subsequent analysis and review process; Video decomposition: decompose the preprocessed video into images, sounds and subtitles; Video content analysis: extract key frames from images, and then detect the key frames using the image review method in claim 4; detect subtitle information using the text review method in claim 5; and detect audio using the audio review method in claim 6; Video review: Once the images, sounds and subtitles are reviewed and approved, the video file will be reviewed and approved.

8. The method according to claims 1 to 7, characterized in that: It also includes an error correction mechanism, the specific method is as follows: If a false positive occurs during the document content review, you can file an appeal against the false positive file. After review by the administrator, you can proceed as follows: Method 1: Only release the file, and subsequent files involving the same sensitive words will still be blocked; Method 2: Process the sensitive words involved in the file, remove them from the sensitive word library and add them to the white-name word library or increase its associated word library to prevent other files involving the words from being mistakenly blocked.

9. The method according to claim 8, characterized in that: By implementing the management of the file blocking list, you can view and summarize the sensitive words contained in the blocked files; in-depth analysis of these sensitive words will help to promptly discover and correct erroneous interception situations and remove erroneous sensitive words from the sensitive word library.

10. A file content review system for enterprise cloud disk based on deep learning, characterized by: The system is used for the method described in any one of claims 1 to 9, and the system at least comprises File storage module, used to store various types of files uploaded by users; A file virus and Trojan killing module is used to detect and kill files uploaded by users to identify whether the uploaded files have viruses or Trojans; The video review module is used to pre-process, decompose, analyze and review the video types uploaded by users; The audio review module is used to pre-process, convert and review audio content uploaded by users; This article review module is used to extract files, preprocess files, and review text content uploaded by users; Image review module, used to perform image preprocessing, content recognition and security detection on image types uploaded by users; A sensitive word management module, which has a built-in sensitive word library and is responsible for detecting whether the uploaded file content contains sensitive words included in the sensitive word library; A white-name word library management module, wherein the white-name word library is built-in and is responsible for detecting whether the content of the uploaded file contains white-list words included in the white-name word library; an associated word library management module, wherein the white-name word library management module is built-in and is responsible for detecting whether the content of the uploaded file contains associated words included in the associated word library; Adjust the blocking conditions of the audited content through the management of sensitive word libraries, associated word libraries, and white-name word libraries; combine the sensitive word library with the management word library to determine whether the sentences containing sensitive words are pornographic, violent, or political, thereby ensuring the accuracy of the document content; The banned file management module intercepts uploaded files involving sensitive information and is responsible for the management of banned files.

Images