System and method for realizing AIGC question and answer examination

Abstract

The invention relates to a system and method for achieving AIGC question and answer review, and belongs to the technical field of information security, and the system comprises the steps that user equipment redirects all traffic corresponding to an AIGC site access event to an auditing gateway; the audit gateway performs target API check on the access event; if the target API is of the Event Stream type, reading the data stream according to rows and judging whether the event is a question event or not; if the event is a question event, calling a local AI review module to perform sensitive word detection so as to detect whether the question event contains a preset sensitive word or not; if the sensitive word is detected, generating a first response result to inform a user that the question event contains sensitive content; the user equipment is also used for displaying the received response result; the method supports the processing capability of a streaming data scene, supports the combination of rule auditing and AI review, and supports the accurate review of specific traffic.

Description

Technical Field

[0001] The present application relates to a system and method for implementing AIGC question and answer review, and belongs to the field of information security technology. Background Art

[0002] With the rapid development of artificial intelligence technology, artificial intelligence generated content (AIGC) is increasingly used in text, pictures, videos, etc., especially in the field of natural language processing (NLP), where AIGC technology is widely used in question-answering systems. However, current AIGC question-answering systems are usually controlled by third-party AI service providers, including the strategy and implementation of content review. Traditional content review solutions are mostly suitable for static request and response scenarios and cannot effectively handle complex streaming question-answering scenarios. Content review is performed in the following ways:

[0003] 1. Request / response-based content filtering: This method usually relies on the enterprise's network or application server to detect static web pages, file downloads, and other content. The audit process determines whether sensitive information is contained by analyzing the data packets in the HTTP request, including the URL, query parameters, POST data, or file content. If sensitive content is detected, the server will block the request or return a prompt page to inform the user that the content does not comply with regulations.

[0004] 2. Content review based on Deep Packet Inspection (DPI): Some network security products use deep packet inspection technology to conduct in-depth analysis of data packets to identify and block requests containing illegal or sensitive content. These systems usually target common protocols such as HTTP, HTTPS, FTP, and decide whether to allow user access by parsing the request content and context. However, traditional DPI systems mainly review static data packets and cannot perform real-time analysis and intervention on real-time streaming data (such as WebSocket, EventStream). In particular, there are performance bottlenecks and response delays when dealing with long-term connections and large data streams.

[0005] 3. Filtering based on content classification: Some audit systems filter out bad information by classifying content (such as pornography, violence, political sensitivity, etc.). Such systems rely on static text analysis or file type recognition, combined with machine learning technology to achieve content classification judgment. However, this classification technology is mainly applicable to predefined static content or uploaded files, and cannot meet the real-time requirements of streaming content generation.

[0006] In summary, traditional solutions have many limitations, such as:

[0007] 1) Latency and performance bottleneck: In real-time communication scenarios (such as AI-generated questions and answers, online chats, etc.), the transmission speed and continuity of streaming data are extremely high, and any delay in review will seriously affect the user experience. Traditional Internet behavior management systems, proxy servers or firewalls rely on deep packet inspection, content classification and other means, and often need to parse the entire data packet before reviewing. Therefore, they are less efficient when processing long-connected streaming data, which can easily cause system performance bottlenecks.

[0008] 2) Reliance on predefined rules and sensitive words: Traditional review methods mainly rely on static keyword matching and blacklist mechanisms, which lack dynamic adaptability. This method cannot fully review real-time generated content (such as AI-generated Q&A content) because the diversity and unpredictability of AI content generation make it impossible for simple sensitive word matching to cover all potential illegal information. The expression of AI-generated content is varied, resulting in the problem that traditional sensitive word matching may miss some sensitive information or misjudge harmless content.

[0009] 3) Lack of in-depth analysis of streaming content: Traditional content review solutions are mostly designed for static request and response scenarios and cannot adapt to the review needs of complex streaming data (such as WebSocket or EventStream). Streaming data is transmitted in a long and continuous manner. Traditional systems rely on complete request and response packets for content review and cannot process fragmented data streams in real time, resulting in delayed or incomplete review. Summary of the invention

[0010] This application provides a system and method for implementing AIGC question and answer review, which can solve the problem that the existing technology completely relies on AIGC service providers for content review and users cannot customize content review policies. Through the present invention, enterprises can customize and implement content review policies in real time in the scenario of streaming data transmission, and respond and block in a timely manner when sensitive information is found. This application provides the following technical solutions:

[0011] In a first aspect, a system for implementing AIGC question and answer review is provided, the system comprising:

[0012] A user device, configured to receive an access event of an AIGC site; and redirect all traffic corresponding to the access event to an audit gateway;

[0013] The audit gateway is used to establish an SSL connection with the user device when receiving a connection request from the user device; check the target API of the access event; if the check result indicates that the target API is not of Event Stream type, forward the access event to the corresponding AIGC service; if the target API is of EventStream type, read the data stream line by line; for each line of data, determine whether the event is a question event; if it is a question event, call the local AI review module to perform sensitive word detection to detect whether the question event contains preset sensitive words; if sensitive words are detected, generate a first response result to notify the user that the question event contains sensitive content; if no sensitive words are detected, forward the access event to the corresponding AIGC service;

[0014] The AIGC service is used to process the access event sent by the audit network and return a second response result of the access event to the audit gateway when receiving the access event.

[0015] The audit gateway is further configured to return the second response result to the user equipment when receiving the second response result returned by the AIGC service;

[0016] The user equipment is further used to display a received response result, where the response result is the first response result or the second response result.

[0017] Optionally, the first response result is generated by the audit gateway by simulating the second response result generated by the AIGC service.

[0018] Optionally, the user equipment includes a transparent port forwarding module, and the transparent port forwarding module is used to access the AIGC site through a transparent proxy technology.

[0019] Optionally, the sensitive word checking rules of the local AI review module are set based on preset checking requirements, and the sensitive word checking rules support modification.

[0020] Optionally, the audit gateway is further used to:

[0021] When the data stream corresponding to the access event is closed, or there is an error in reading by row, the data review of the AIGC site is stopped.

[0022] In a second aspect, a method for implementing AIGC question and answer review is provided, the method comprising:

[0023] Receiving an access event to an AIGC site through a user device; redirecting all traffic corresponding to the access event to an audit gateway;

[0024] Upon receiving a connection request from the user device, the audit gateway establishes an SSL connection with the user device; checks the target API of the access event; if the check result indicates that the target API is not of Event Stream type, forwards the access event to the corresponding AIGC service; if the target API is of EventStream type, reads the data stream line by line; for each line of data, determines whether the event is a question event; if it is a question event, calls the local AI review module to perform sensitive word detection to detect whether the question event contains preset sensitive words; if sensitive words are detected, generates a first response result to notify the user that the question event contains sensitive content; if no sensitive words are detected, forwards the access event to the corresponding AIGC service;

[0025] When receiving an access event sent by the audit network, the AIGC service processes the access event and returns a second response result of the access event to the audit gateway;

[0026] When receiving the second response result returned by the AIGC service, the audit gateway returns the second response result to the user equipment;

[0027] The received response result is displayed by the user equipment, where the response result is the first response result or the second response result.

[0028] Optionally, the first response result is generated by the audit gateway by simulating the second response result generated by the AIGC service.

[0029] Optionally, the user equipment includes a transparent port forwarding module, and the transparent port forwarding module is used to access the AIGC site through a transparent proxy technology.

[0030] Optionally, the sensitive word checking rules of the local AI review module are set based on preset checking requirements, and the sensitive word checking rules support modification.

[0031] Optionally, the method further comprises:

[0032] When the data stream corresponding to the access event is closed, or there is an error in reading by row, the data review of the AIGC site is stopped.

[0033] The beneficial effects of this application include:

[0034] 1. Support processing capabilities for streaming data scenarios: Traditional content review systems are mainly designed for static HTTP request and response scenarios, and cannot effectively process streaming data such as WebSocket and EventStream. The present invention is specifically optimized and expanded for these streaming data scenarios, providing real-time sensitive word detection and response processing functions. When the system detects that the streaming data contains sensitive information, the present invention can respond immediately, block sensitive questions in real time, and support forged AI-generated content (AIGC) as answers. This capability ensures that the system will not continue to transmit sensitive question and answer data, thereby effectively enhancing the security and compliance of the content.

[0035] 2. Support the combination of rule review and AI review: The present invention not only supports traditional rule review mechanisms, such as keyword-based static matching and blacklist management, but also introduces advanced AI technology to combat the review of AI-generated content. AI-generated content is highly variable and diverse, and traditional static matching methods are often difficult to cope with. Through AI technology, the present invention can analyze and understand AI-generated questions and answers in real time, identify possible sensitive information, and thus ensure the comprehensiveness and accuracy of content review.

[0036] 3. Support accurate review of specific traffic: Traditional content review solutions usually rely on full network traffic audits. Equipment is deployed at the network exit and all traffic is analyzed through deep packet inspection. This type of solution processes full network traffic, which is prone to high resource consumption and delays, especially when processing streaming data, where performance bottlenecks are obvious. In addition, these solutions lack specificity and are unable to review only specific site traffic, resulting in low efficiency. The present invention uses transparent proxy technology to only proxy specific traffic, reducing performance loss and improving the flexibility and efficiency of the system.

[0037] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application and implement it according to the contents of the specification, the following is a detailed description of the preferred embodiments of the present application in conjunction with the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 This is a block diagram of a system for implementing AIGC question and answer review provided by an embodiment of the present application.

[0039] Figure 2 This is a flowchart of a method for implementing AIGC question and answer review provided by an embodiment of the present application. DETAILED DESCRIPTION

[0040] The specific implementation methods of the present application are further described in detail below in conjunction with the accompanying drawings and examples. The following examples are used to illustrate the present application but are not intended to limit the scope of the present application.

[0041] In the existing technology, the AIGC question-and-answer system is controlled and reviewed by a third-party AI service provider. User companies cannot customize and control the content review in the question-and-answer system, especially when sensitive information is involved, which poses certain risks to some companies with specific content security requirements.

[0042] In addition, AIGC's Q&A stream usually transmits data through WebSocket or Event Stream. Since data is transmitted in real time in streaming form, the complexity of content review is greatly increased. Most existing solutions require AIGC service providers to implement content review by themselves, and enterprises cannot flexibly customize content review policies. This model of relying on service providers brings about the problem that content review standards cannot meet the needs of enterprises, especially when enterprises need to comply with strict compliance requirements.

[0043] Therefore, enterprises urgently need a solution that can customize content review, rather than relying entirely on the review strategy of AIGC service providers. Existing technical solutions cannot provide flexible detection and real-time response to sensitive words, especially during the transmission of streaming data.

[0044] Based on the above technical problems, this application provides a system for implementing AIGC question and answer review, especially a transparent proxy gateway for sensitive word filtering. The system forwards the user's network request to the audit gateway through the transparent port forwarding module of the client to perform content audit and sensitive word filtering. The system includes components such as the audit gateway and AIGC service, which combines the sensitive word library and artificial intelligence review mechanism to realize the full audit of user access services.

[0045] Below, the system for implementing AIGC question and answer review provided by this application is introduced in detail.

[0046] Figure 1 This is a block diagram of a system for implementing AIGC question and answer review provided by an embodiment of the present application. Figure 1 It can be seen that the system includes: user equipment 110, audit gateway 120, and AIGC130 service.

[0047] The user equipment 110 supports the user to access the AIGC 130 site. The user equipment 110 includes a transparent port forwarding module, which is used to access the AIGC 130 site through a transparent proxy technology.

[0048] In this embodiment, the user device 110 is used to receive the access event of the AIGC 130 site, and redirect all traffic corresponding to the access event to the audit gateway 120. In this way, all traffic is redirected to the audit gateway 120, ensuring that the user request is monitored without being detected, so as to realize the audit and analysis of the data flow.

[0049] The audit gateway 120 is used to establish a secure SSL connection with the user device 110 upon receiving a connection request from the user device 110. In this way, the security of data transmission can be ensured.

[0050] The audit gateway 120 is also used to check the target API of the access event to confirm whether the request complies with the preset audit rules.

[0051] For non-Event Stream requests: If the check result indicates that the target API is not of Event Stream type, the audit gateway 120 forwards the access event to the corresponding AIGC130 service. Accordingly, the AIGC130 service is used to process the access event upon receiving the access event sent by the audit network, and return the second response result of the access event to the audit gateway 120; the audit gateway 120 is also used to return the second response result to the user device 110 upon receiving the second response result returned by the AIGC130 service, thus realizing a complete request-response link.

[0052] For Event Stream requests: If the target API is of Event Stream type, the data stream is read line by line, and each line of data is parsed step by step.

[0053] The audit gateway 120 analyzes each line of data step by step, including: conducting a sensitive word review on each line of data. Specifically, the audit gateway 120 determines whether the event is a user's question event. If it is a question event, the audit gateway 120 will call the local AI review module to detect sensitive words. When conducting a sensitive word review, the audit gateway 120 will analyze the question content to determine whether it contains any sensitive words.

[0054] If a sensitive word is detected, the audit gateway 120 generates a first response result to notify the user that the question event contains sensitive content. The first response result is generated by the audit gateway 120 by simulating the second response result generated by the AIGC 130 service.

[0055] If no sensitive words are detected, the audit gateway 120 will continue to forward the access event to the corresponding AIGC 130 service for normal processing.

[0056] Optionally, the sensitive word checking rules of the local AI review module are set based on preset checking requirements, and the sensitive word checking rules support modification.

[0057] The audit gateway 120 is also used to stop the data review of the AIGC 130 site when the data stream corresponding to the access event is closed or there is an error in the row reading.

[0058] The user equipment 110 is further configured to display a received response result, where the response result is the first response result or the second response result.

[0059] In summary, the system for implementing AIGC question-and-answer review provided in this embodiment overcomes the limitations of traditional content review systems in processing streaming data scenarios, which is mainly reflected in the following aspects:

[0060] 1. Support processing capabilities for streaming data scenarios: Traditional content review systems are mainly designed for static HTTP request and response scenarios, and cannot effectively process streaming data such as WebSocket and EventStream. The present invention is specifically optimized and expanded for these streaming data scenarios, providing real-time sensitive word detection and response processing functions. When the system detects that the streaming data contains sensitive information, the present invention can respond immediately, block sensitive questions in real time, and support forged AI-generated content (AIGC) as answers. This capability ensures that the system will not continue to transmit sensitive question and answer data, thereby effectively enhancing the security and compliance of the content.

[0061] 2. Support the combination of rule review and AI review: The present invention not only supports traditional rule review mechanisms, such as keyword-based static matching and blacklist management, but also introduces advanced AI technology to combat the review of AI-generated content. AI-generated content is highly variable and diverse, and traditional static matching methods are often difficult to cope with. Through AI technology, the present invention can analyze and understand AI-generated questions and answers in real time, identify possible sensitive information, and thus ensure the comprehensiveness and accuracy of content review.

[0062] 3. Support accurate review of specific traffic: Traditional content review solutions usually rely on full network traffic audits. Equipment is deployed at the network exit and all traffic is analyzed through deep packet inspection. This type of solution processes full network traffic, which is prone to high resource consumption and delays, especially when processing streaming data, where performance bottlenecks are obvious. In addition, these solutions lack specificity and are unable to review only specific site traffic, resulting in low efficiency. The present invention uses transparent proxy technology to only proxy specific traffic, reducing performance loss and improving the flexibility and efficiency of the system.

[0063] Figure 2 This is a flow chart of a method for implementing AIGC question-and-answer review provided by an embodiment of the present application. This embodiment is described by taking the method used in a controller of an electric compressor as an example. The method includes at least the following steps:

[0064] Step 201: receiving an access event of an AIGC site through a user device; redirecting all traffic corresponding to the access event to an audit gateway;

[0065] The user equipment includes a transparent port forwarding module, and the transparent port forwarding module is used to access the AIGC site through a transparent proxy technology.

[0066] Step 202: upon receiving a connection request from the user device, the audit gateway establishes an SSL connection with the user device; checks the target API for the access event; if the check result indicates that the target API is not of Event Stream type, forwards the access event to the corresponding AIGC service; if the target API is of Event Stream type, reads the data stream line by line; for each line of data, determines whether the event is a question event; if it is a question event, calls the local AI review module to perform sensitive word detection to detect whether the question event contains preset sensitive words; if sensitive words are detected, generates a first response result to notify the user that the question event contains sensitive content; if no sensitive words are detected, forwards the access event to the corresponding AIGC service;

[0067] Among them, the first response result is generated by the audit gateway through simulating the second response result generated by the AIGC service.

[0068] The sensitive word checking rules of the local AI review module are set based on preset checking requirements, and the sensitive word checking rules can be modified.

[0069] Optionally, when the data stream corresponding to the access event is closed, or there is an error in reading by row, the data review of the AIGC site is stopped.

[0070] Step 203: When receiving an access event sent by the audit network, the AIGC service processes the access event and returns a second response result of the access event to the audit gateway;

[0071] Step 204: upon receiving the second response result returned by the AIGC service, the audit gateway returns the second response result to the user equipment;

[0072] Step 205: Display the received response result through the user equipment, where the response result is the first response result or the second response result.

[0073] In summary, the method for implementing AIGC question and answer review provided in this embodiment detects sensitive information in streaming data and forges AI-generated content (AIGC) as answers when detected. This capability ensures that the system does not continue to transmit sensitive question and answer data, thereby effectively enhancing the security and compliance of the content.

[0074] In addition, through AI technology, it is possible to analyze and understand AI-generated questions and answers in real time, identify possible sensitive information, and thus ensure the comprehensiveness and accuracy of content review.

[0075] In addition, through transparent proxy technology, only specific traffic is proxied, reducing performance loss and improving the flexibility and efficiency of the system.

[0076] The technical features of the above-described embodiments may be arbitrarily combined. To make the description concise, not all possible combinations of the technical features in the above-described embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0077] The above-mentioned embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the patent of the present application shall be subject to the attached claims.

Claims

1. A system for implementing AIGC question-answer review, characterized in that: The system comprises: A user device, configured to receive an access event of an AIGC site; and redirect all traffic corresponding to the access event to an audit gateway; The audit gateway is used to establish an SSL connection with the user device when receiving a connection request from the user device; check the target API of the access event; if the check result indicates that the target API is not of EventStream type, forward the access event to the corresponding AIGC service; if the target API is of Event Stream type, read the data stream line by line; for each line of data, determine whether the event is a question event; if it is a question event, call the local AI review module to perform sensitive word detection to detect whether the question event contains preset sensitive words; if sensitive words are detected, generate a first response result to notify the user that the question event contains sensitive content; if no sensitive words are detected, forward the access event to the corresponding AIGC service; The AIGC service is used to process the access event sent by the audit network and return a second response result of the access event to the audit gateway when receiving the access event. The audit gateway is further configured to return the second response result to the user equipment when receiving the second response result returned by the AIGC service; The user equipment is further used to display a received response result, where the response result is the first response result or the second response result.

2. The system according to claim 1, characterized in that The first response result is generated by the audit gateway by simulating the second response result generated by the AIGC service.

3. The system according to claim 1, characterized in that The user equipment includes a transparent port forwarding module, and the transparent port forwarding module is used to access the AIGC site through a transparent proxy technology.

4. The system according to claim 1, characterized in that The sensitive word checking rules of the local AI review module are set based on preset checking requirements, and the sensitive word checking rules support modification.

5. The system according to claim 1, characterized in that The audit gateway is also used for: When the data stream corresponding to the access event is closed, or there is an error in reading by row, the data review of the AIGC site is stopped.

6. A method for implementing AIGC question and answer review, characterized in that: The method comprises: Receiving an access event to an AIGC site through a user device; redirecting all traffic corresponding to the access event to an audit gateway; Upon receiving a connection request from the user device, the audit gateway establishes an SSL connection with the user device; checks the target API of the access event; if the check result indicates that the target API is not of EventStream type, forwards the access event to the corresponding AIGC service; if the target API is of Event Stream type, reads the data stream line by line; for each line of data, determines whether the event is a question event; if it is a question event, calls the local AI review module to perform sensitive word detection to detect whether the question event contains preset sensitive words; if sensitive words are detected, generates a first response result to notify the user that the question event contains sensitive content; if no sensitive words are detected, forwards the access event to the corresponding AIGC service; When receiving an access event sent by the audit network, the AIGC service processes the access event and returns a second response result of the access event to the audit gateway; When receiving the second response result returned by the AIGC service, the audit gateway returns the second response result to the user equipment; The received response result is displayed by the user equipment, where the response result is the first response result or the second response result.

7. The method according to claim 6, characterized in that The first response result is generated by the audit gateway by simulating the second response result generated by the AIGC service.

8. The method according to claim 6, characterized in that The user equipment includes a transparent port forwarding module, and the transparent port forwarding module is used to access the AIGC site through a transparent proxy technology.

9. The method according to claim 6, characterized in that The sensitive word checking rules of the local AI review module are set based on preset checking requirements, and the sensitive word checking rules support modification.

10. The method according to claim 6, characterized in that The method further comprises: When the data stream corresponding to the access event is closed, or there is an error in reading by row, the data review of the AIGC site is stopped.

Images