Artificial intelligence governance method for bad content in easy-to-confuse scene

Abstract

According to the artificial intelligence treatment method for the bad content in the easy-to-confuse scene, not only is the bad content type of the bad network picture and video detected, but also the scene features in the picture and video are detected, so that whether the network picture and the video are bad can be comprehensively judged according to the bad content type and the scene features, and the user experience is improved. And the recognition accuracy of the bad content in the easy-to-confuse scene is improved.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to an artificial intelligence governance method for bad content in easily confusing scenarios. Background Art

[0002] It is difficult to detect bad content in easily confused scenes, such as distinguishing between wearing a swimsuit while swimming (not sensitive content) and wearing a swimsuit in a live broadcast (sensitive content). Bad network pictures and videos in easily confused scenes are difficult to detect using existing smart detection models for bad network pictures and videos.

[0003] In the process of implementing the present invention, the inventors found that there are at least the following problems in the prior art: the prior art may miss or misjudge based solely on the type of inappropriate content.

[0004] Therefore, the prior art still needs to be improved and developed. Summary of the invention

[0005] Based on this, it is necessary to provide an artificial intelligence governance method for bad content in confusing scenarios to address the defects or shortcomings of existing technologies, combine scenarios and content for comprehensive identification, and improve the recognition rate of bad content in confusing scenarios.

[0006] In a first aspect, an embodiment of the present invention provides an artificial intelligence governance method, the method comprising:

[0007] The steps of using the deep learning model for identifying the type of bad content are as follows: obtaining the network content to be detected as input, and using the output calculated by the deep learning model for identifying the type of bad content as the labeling of the bad content and the bad content type in the network content to be detected;

[0008] The scene recognition deep learning model is used in the following steps: obtaining the network content to be detected as input, and using the output calculated by the scene recognition deep learning model as the scene type of the network content to be detected;

[0009] Comprehensive judgment step: if the scenario type of the network content to be detected belongs to the first scenario type corresponding to the bad content type of the bad content in the annotation of the network content to be detected, the bad content of the network content to be detected and the annotation of its bad content type are returned to the user; if the scenario type of the network content to be detected belongs to the second scenario type corresponding to the bad content type of the bad content in the annotation of the network content to be detected, the bad content and the annotation of its bad content type in the network content to be detected are deleted.

[0010] Preferably, the method further comprises:

[0011] Steps for constructing the bad content scenario knowledge base: obtaining bad content types whose scenarios will cause content types to change and each first scenario type corresponding to the bad content type, adding them to the bad content scenario knowledge base; obtaining bad content types and each second scenario type corresponding to the bad content type, adding them to the bad content scenario knowledge base; the first scenario type corresponding to the bad content type is a scenario in which the bad content type is not allowed to exist; the second scenario type corresponding to the bad content type is a scenario in which the bad content type is allowed to exist;

[0012] The step of acquiring the first and second scene types of the inappropriate content type is as follows: retrieving the first scene type and the second scene type corresponding to the inappropriate content type of the inappropriate content in the network content to be detected from the inappropriate content scene knowledge base.

[0013] Preferably, the method further comprises:

[0014] The first manual review step: if the bad content type of the bad content in the annotation of the network content to be detected exists in the bad content scenario knowledge base, and the scene type of the network content to be detected does not belong to the first scenario type and the second scenario type corresponding to the bad content type of the bad content in the annotation of the network content to be detected, then the bad content in the network content to be detected and the annotation of its bad content type and the scene of the network content to be detected are sent to manual review, and information is sent to remind the manual to determine whether the scene of the network content to be detected is the first scenario type or the second scenario type corresponding to the bad content type of the bad content in the annotation of the network content to be detected; obtain the result of the review, if the network content to be detected If the scenario of the network content to be detected is the first scenario type corresponding to the bad content type of the bad content in the annotation of the bad content to be detected, the scenario of the network content to be detected is added to the bad content scenario knowledge base as the first scenario type corresponding to the bad content type, and the bad content of the network content to be detected and the annotation of its bad content type are returned to the user; if the scenario of the network content to be detected is the second scenario type corresponding to the bad content type of the bad content in the annotation of the network content to be detected, the scenario of the network content to be detected is added to the bad content scenario knowledge base as the second scenario type corresponding to the bad content type, and the annotation of the bad content and its bad content type in the network content to be detected is deleted.

[0015] Preferably, the method further comprises:

[0016] Steps for using the deep learning model for scene and bad content identification: obtain the network content to be detected as input, and use the output calculated by the bad content type identification deep learning model as the label of the bad content in the network content and its bad content type, and the scene type of the network content.

[0017] In a second aspect, an embodiment of the present invention provides an artificial intelligence governance system, the system comprising:

[0018] A module for using a deep learning model for identifying bad content types: obtaining the network content to be detected as input, and using the output calculated by the deep learning model for identifying bad content types as a label for the bad content and its type in the network content to be detected;

[0019] A scene recognition deep learning model using module: obtaining the network content to be detected as input, and using the output calculated by the scene recognition deep learning model as the scene type of the network content to be detected;

[0020] Comprehensive judgment module: If the scenario type of the network content to be detected belongs to the first scenario type corresponding to the bad content type of the bad content in the annotation of the network content to be detected, the bad content of the network content to be detected and the annotation of its bad content type are returned to the user; if the scenario type of the network content to be detected belongs to the second scenario type corresponding to the bad content type of the bad content in the annotation of the network content to be detected, the bad content and the annotation of its bad content type in the network content to be detected are deleted.

[0021] Preferably, the system further comprises:

[0022] Unhealthy content scenario knowledge base construction module: Obtain unhealthy content types whose scenarios will cause content types to change and each first scenario type corresponding to the unhealthy content type, add them to the unhealthy content scenario knowledge base; Obtain unhealthy content types and each second scenario type corresponding to the unhealthy content type, add them to the unhealthy content scenario knowledge base; the first scenario type corresponding to the unhealthy content type is a scenario where the unhealthy content type is not allowed to exist; the second scenario type corresponding to the unhealthy content type is a scenario where the unhealthy content type is allowed to exist;

[0023] The first and second scene type acquisition module of the inappropriate content type is used to retrieve the first scene type and the second scene type corresponding to the inappropriate content type of the inappropriate content in the network content to be detected from the inappropriate content scene knowledge base.

[0024] Preferably, the system further comprises:

[0025] The first manual review module: if the bad content type of the bad content in the annotation of the network content to be detected exists in the bad content scenario knowledge base, and the scene type of the network content to be detected does not belong to the first scenario type and the second scenario type corresponding to the bad content type of the bad content in the annotation of the network content to be detected, then the bad content in the network content to be detected and the annotation of its bad content type and the scene of the network content to be detected are sent to manual review, and information is sent to remind the manual to determine whether the scene of the network content to be detected is the first scenario type or the second scenario type corresponding to the bad content type of the bad content in the annotation of the network content to be detected; obtain the result of the review, if the network content to be detected If the scenario of the network content to be detected is the first scenario type corresponding to the bad content type of the bad content in the annotation of the bad content to be detected, the scenario of the network content to be detected is added to the bad content scenario knowledge base as the first scenario type corresponding to the bad content type, and the bad content of the network content to be detected and the annotation of its bad content type are returned to the user; if the scenario of the network content to be detected is the second scenario type corresponding to the bad content type of the bad content in the annotation of the network content to be detected, the scenario of the network content to be detected is added to the bad content scenario knowledge base as the second scenario type corresponding to the bad content type, and the annotation of the bad content and its bad content type in the network content to be detected is deleted.

[0026] Preferably, the system further comprises:

[0027] The module using the scene and bad content identification deep learning model is as follows: the network content to be detected is obtained as input, and the output calculated by the bad content type identification deep learning model is used as the label of the bad content in the network content and its bad content type, and the scene type of the network content.

[0028] In a third aspect, an embodiment of the present invention provides an artificial intelligence governance device, which includes a module of the system described in any one of the embodiments of the second aspect.

[0029] In a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein the program, when executed by a processor, implements the steps of the method described in any one of the embodiments of the first aspect.

[0030] In a fifth aspect, an embodiment of the present invention provides a robot, comprising a memory, a processor, and an artificial intelligence robot program stored in the memory and executable on the processor, wherein the processor implements the steps of any one of the method described in the first aspect when executing the program.

[0031] The artificial intelligence governance method for bad content in easily confused scenes provided in this embodiment not only detects the bad content type of bad network pictures and videos, but also detects the scene features in the pictures and videos, and can then comprehensively judge whether the bad network pictures and videos are bad based on the bad content type and scene features, thereby improving the recognition accuracy of bad content in easily confused scenes. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 A flowchart of an artificial intelligence governance method provided for one embodiment of the present invention;

[0033] Figure 2 A framework diagram of an artificial intelligence governance system provided for one embodiment of the present invention. DETAILED DESCRIPTION

[0034] The technical solutions in the embodiments of the present invention are described in detail below in conjunction with the embodiments of the present invention.

[0035] Basic Embodiments of the Invention

[0036] First, as Figure 1 As shown, an embodiment of the present invention provides an artificial intelligence governance method, the method comprising the step of using a deep learning model for identifying the type of bad content; the step of using a deep learning model for scene recognition; and the step of comprehensive judgment. In a preferred embodiment, the method further comprises the step of constructing a bad content scene knowledge base; and the step of obtaining the first and second scene types of the bad content type. In a preferred embodiment, the method further comprises a first manual review step. In a preferred embodiment, the method further comprises the step of using a deep learning model for scene and bad content recognition.

[0037] Second, as Figure 2 As shown, an embodiment of the present invention provides an artificial intelligence governance system, the system includes a module for using a deep learning model for identifying bad content types; a module for using a deep learning model for scene recognition; and a comprehensive judgment module. In a preferred embodiment, the system also includes a module for building a bad content scene knowledge base; and a module for acquiring the first and second scene types of bad content types. In a preferred embodiment, the system also includes a first manual review module. In a preferred embodiment, the system also includes a module for using a deep learning model for scene and bad content recognition.

[0038] Preferred embodiments of the present invention

[0039] Combine the deep learning-based artificial intelligence algorithm framework with the expert system. First, use the deep learning model to detect whether there are bad content types in online pictures and videos. If there are bad content types, use the expert system to determine which scenes will be allowed and which scenes will be prohibited. Then use the deep learning model to detect whether there are prohibited scenes in online pictures and videos, and then determine whether the online pictures and videos with bad content types are really bad online pictures and videos.

[0040] Technical effect: Identifying images and videos in easily confused scenes is not only a difficult problem for the intelligent detection model of bad network content, but also a difficult problem faced by artificial intelligence. Solving this problem is not only meaningful for the detection of bad network images and videos, but can also be used as a reference for other artificial intelligence applications.

[0041] 1. Training and testing steps of the deep learning model for identifying the type of bad content: Initialize the deep learning model as a deep learning model for identifying the type of bad content; obtain network content from the training and testing data set as input, obtain the bad content in the network content and the annotation of the bad content type as the expected output, and train and test the deep learning model for identifying the type of bad content;

[0042] 2. Steps for using the deep learning model for identifying the type of bad content: obtaining the network content to be detected as input, and using the output calculated by the deep learning model for identifying the type of bad content as a label for the bad content and its type in the network content to be detected; the type of bad content includes sensitive content;

[0043] 3. Steps for constructing the bad content scenario knowledge base: Obtain bad content types whose scenarios will cause content types to change and each first scenario type corresponding to the bad content types, and add them to the bad content scenario knowledge base; Obtain bad content types and each second scenario type corresponding to the bad content types, and add them to the bad content scenario knowledge base; The first scenario type corresponding to the bad content type is a scenario where the bad content type is not allowed to exist, for example, photos of people wearing swimsuits will be identified as bad content types such as sensitive content, which is not allowed in the first scenario type such as public places and is sensitive content; The second scenario type corresponding to the bad content type is a scenario where the bad content type is allowed to exist, that is, the content of the bad content type will become good content in the corresponding second scenario type; for example, photos of people wearing swimsuits will be identified as bad content types such as sensitive content, but will become non-sensitive content in the second scenario type such as swimming pools;

[0044] 4. The step of obtaining the first and second scene types of the inappropriate content type: retrieving the first and second scene types corresponding to the inappropriate content types of the inappropriate content in the inappropriate content to be detected in the network content from the inappropriate content scene knowledge base; the retrieved first and second scene types may be one or more;

[0045] 5. Scene recognition deep learning model training and testing steps: Initialize the deep learning model as a scene recognition deep learning model; obtain network content from the training and testing data set as input, obtain the scene type of the network content as the expected output, and train and test the scene recognition deep learning model; the scene types include kitchen, sports field, square, swimming pool, river, room, and live studio;

[0046] 6. Steps of using the scene recognition deep learning model: obtaining the network content to be detected as input, and using the output calculated by the scene recognition deep learning model as the scene type of the network content to be detected;

[0047] 7 Comprehensive judgment step: if the scenario type of the network content to be detected belongs to the first scenario type corresponding to the bad content type of the bad content in the annotation of the network content to be detected, the bad content of the network content to be detected and the annotation of the bad content type are returned to the user; if the scenario type of the network content to be detected belongs to the second scenario type corresponding to the bad content type of the bad content in the annotation of the network content to be detected, the bad content and the annotation of the bad content type in the network content to be detected are deleted;

[0048] Technical effect: By extracting the bad types and scene types of bad content through artificial intelligence models, and taking into account the impact of scene types on the bad types of bad content, it can reduce the misjudgment of bad types caused by scenes and improve the accuracy of identifying bad content.

[0049] After step 7,

[0050] 8. A first manual review step: if the undesirable content type of the undesirable content in the annotation of the network content to be detected exists in the undesirable content scenario knowledge base, and the scenario type of the network content to be detected does not belong to the first scenario type and the second scenario type corresponding to the undesirable content type in the annotation of the network content to be detected, then the undesirable content in the network content to be detected and the annotation of its undesirable content type and the scenario of the network content to be detected are sent to manual review, and information is sent to remind the manual to determine whether the scenario of the network content to be detected is the first scenario type or the second scenario type corresponding to the undesirable content type in the annotation of the network content to be detected; obtain the review result, if the network content to be detected If the scenario of the content is the first scenario type corresponding to the bad content type of the bad content in the annotation of the network content to be detected, the scenario of the network content to be detected is added to the bad content scenario knowledge base as the first scenario type corresponding to the bad content type, and the bad content of the network content to be detected and the annotation of its bad content type are returned to the user; if the scenario of the network content to be detected is the second scenario type corresponding to the bad content type of the bad content in the annotation of the network content to be detected, the scenario of the network content to be detected is added to the bad content scenario knowledge base as the second scenario type corresponding to the bad content type, and the annotation of the bad content and its bad content type in the network content to be detected is deleted.

[0051] Technical effect: Adding more scenarios of bad content that are easily affected by scenarios to the knowledge base can improve the ability to judge the types of bad content in different scenarios;

[0052] After step 7,

[0053] 9. Second manual review step: If the bad content type of the bad content in the annotation of the network content to be detected does not exist in the bad content scenario knowledge base (indicating that the bad content type does not belong to a scenario that is easily confused), the bad content of the network content to be detected and the annotation of its bad content type are returned to the user; if the user applies for a review of the returned result, the bad content and the annotation of its bad content type in the network content to be detected and the scenario of the network content to be detected are sent to a manual review, and a message is sent to remind the manual to determine whether the scenario of the network content to be detected causes a change in the content type; the result of the review is obtained. If the scenario of the network content to be detected causes a change in the content type, the bad content type is added to the bad content scenario knowledge base, and the scenario of the network content to be detected is added to the bad content scenario knowledge base as the second scenario type corresponding to the bad content type, and at the same time, the bad content and the annotation of its bad content type in the network content to be detected are deleted, and a message is sent to the user who applied for the review that "your application is successful, and the content is good content."

[0054] Technical effect: Adding more bad types that are easily affected by scenarios to the knowledge base can improve the ability to identify more bad types that are easily affected by scenarios;

[0055] After step 7,

[0056] 10 The third manual review step: If the bad content type of the bad content in the annotation of the network content to be detected obtained by any artificial intelligence model or manual detection receives a user's review application, then the network content to be detected is obtained as input, and the output calculated by the scene recognition deep learning model is used as the scene type of the network content to be detected; if the bad content type of the bad content in the annotation of the network content to be detected exists in the bad content scene knowledge base, and the scene type of the network content to be detected does not belong to the first scene type and the second scene type corresponding to the bad content type of the bad content in the annotation of the network content to be detected, then the bad content in the network content to be detected and the annotation of its bad content type and the scene of the network content to be detected are sent to manual review, and information is sent to remind the manual to determine whether the scene of the network content to be detected is the bad content type of the network content to be detected. The first scene type or the second scene type corresponding to the bad content type of the bad content in the network content annotation; obtaining the result of the review, if the scene of the network content to be detected is the first scene type corresponding to the bad content type of the bad content in the network content annotation to be detected, then the scene of the network content to be detected is added to the bad content scene knowledge base as the first scene type corresponding to the bad content type, and the bad content of the network content to be detected and the annotation of its bad content type are returned to the user; if the scene of the network content to be detected is the second scene type corresponding to the bad content type of the bad content in the network content annotation to be detected, then the scene of the network content to be detected is added to the bad content scene knowledge base as the second scene type corresponding to the bad content type, and the annotation of the bad content and its bad content type in the network content to be detected is deleted.

[0057] Technical effect: The correctness of the results identified by the existing technology can be judged in combination with the knowledge base of the bad content scenario, thereby improving the accuracy of identifying bad content on the basis of the existing technology;

[0058] After step 7,

[0059] 11. Replace the output layers of the deep learning model for identifying the type of inappropriate content and the deep learning model for identifying the scene with a common preset output layer, and use the obtained model as a deep learning model for identifying the scene and inappropriate content; the preset output layer can output the inappropriate content and its type, and the scene type;

[0060] 12. Training and testing steps of the deep learning model for scene and bad content recognition: obtaining network content from a training and testing data set as input, obtaining bad content in the network content and annotations of its type, and scene types of the network content (the scene types of the network content include confusing scenes) as expected outputs, and training and testing the deep learning model for scene and bad content recognition;

[0061] 13. Steps for using the deep learning model for scene and bad content identification: obtaining the network content to be detected as input, and using the output calculated by the bad content type identification deep learning model as the label of the bad content in the network content and its bad content type, and the scene type of the network content;

[0062] Technical effect: Able to identify both bad types and scene types through one model, thereby improving the speed and efficiency of identification.

[0063] The above-mentioned embodiments only express several implementation methods of the present invention, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present invention. It should be pointed out that, for a person of ordinary skill in the art, several modifications and improvements can be made without departing from the concept of the present invention, and these all belong to the protection scope of the present invention. Therefore, the protection scope of the present invention patent shall be subject to the attached claims.

Claims

1. An artificial intelligence governance method, characterized in that: The method comprises: The steps of using the deep learning model for identifying the type of bad content are as follows: obtaining the network content to be detected as input, and using the output calculated by the deep learning model for identifying the type of bad content as the labeling of the bad content and the bad content type in the network content to be detected; The scene recognition deep learning model is used in the following steps: obtaining the network content to be detected as input, and using the output calculated by the scene recognition deep learning model as the scene type of the network content to be detected; Comprehensive judgment step: if the scenario type of the network content to be detected belongs to the first scenario type corresponding to the bad content type of the bad content in the annotation of the network content to be detected, the bad content of the network content to be detected and the annotation of its bad content type are returned to the user; if the scenario type of the network content to be detected belongs to the second scenario type corresponding to the bad content type of the bad content in the annotation of the network content to be detected, the bad content and the annotation of its bad content type in the network content to be detected are deleted.

2. The artificial intelligence governance method according to claim 1, characterized in that: The method further comprises: Steps for constructing the bad content scenario knowledge base: obtaining bad content types whose scenarios will cause content types to change and each first scenario type corresponding to the bad content type, adding them to the bad content scenario knowledge base; obtaining bad content types and each second scenario type corresponding to the bad content type, adding them to the bad content scenario knowledge base; the first scenario type corresponding to the bad content type is a scenario in which the bad content type is not allowed to exist; the second scenario type corresponding to the bad content type is a scenario in which the bad content type is allowed to exist; The step of acquiring the first and second scene types of the inappropriate content type is as follows: retrieving the first scene type and the second scene type corresponding to the inappropriate content type of the inappropriate content in the network content to be detected from the inappropriate content scene knowledge base.

3. The artificial intelligence governance method according to claim 1, characterized in that: The method further comprises: The first manual review step: if the bad content type of the bad content in the annotation of the network content to be detected exists in the bad content scenario knowledge base, and the scene type of the network content to be detected does not belong to the first scenario type and the second scenario type corresponding to the bad content type of the bad content in the annotation of the network content to be detected, then the bad content in the network content to be detected and the annotation of its bad content type and the scene of the network content to be detected are sent to manual review, and information is sent to remind the manual to determine whether the scene of the network content to be detected is the first scenario type or the second scenario type corresponding to the bad content type of the bad content in the annotation of the network content to be detected; obtain the result of the review, if the network content to be detected If the scenario of the network content to be detected is the first scenario type corresponding to the bad content type of the bad content in the annotation of the bad content to be detected, the scenario of the network content to be detected is added to the bad content scenario knowledge base as the first scenario type corresponding to the bad content type, and the bad content of the network content to be detected and the annotation of its bad content type are returned to the user; if the scenario of the network content to be detected is the second scenario type corresponding to the bad content type of the bad content in the annotation of the network content to be detected, the scenario of the network content to be detected is added to the bad content scenario knowledge base as the second scenario type corresponding to the bad content type, and the annotation of the bad content and its bad content type in the network content to be detected is deleted.

4. The artificial intelligence governance method according to claim 1, characterized in that: The method further comprises: Steps for using the deep learning model for scene and bad content identification: obtain the network content to be detected as input, and use the output calculated by the bad content type identification deep learning model as the label of the bad content in the network content and its bad content type, and the scene type of the network content.

5. An artificial intelligence governance system, characterized in that: The system comprises: A module for using a deep learning model for identifying bad content types: obtaining the network content to be detected as input, and using the output calculated by the deep learning model for identifying bad content types as a label for the bad content and its type in the network content to be detected; A scene recognition deep learning model using module: obtaining the network content to be detected as input, and using the output calculated by the scene recognition deep learning model as the scene type of the network content to be detected; Comprehensive judgment module: If the scenario type of the network content to be detected belongs to the first scenario type corresponding to the bad content type of the bad content in the annotation of the network content to be detected, the bad content of the network content to be detected and the annotation of its bad content type are returned to the user; if the scenario type of the network content to be detected belongs to the second scenario type corresponding to the bad content type of the bad content in the annotation of the network content to be detected, the bad content and the annotation of its bad content type in the network content to be detected are deleted.

6. The artificial intelligence governance system according to claim 5, characterized in that: The system further comprises: Unhealthy content scenario knowledge base construction module: Obtain unhealthy content types whose scenarios will cause content types to change and each first scenario type corresponding to the unhealthy content type, add them to the unhealthy content scenario knowledge base; Obtain unhealthy content types and each second scenario type corresponding to the unhealthy content type, add them to the unhealthy content scenario knowledge base; the first scenario type corresponding to the unhealthy content type is a scenario where the unhealthy content type is not allowed to exist; the second scenario type corresponding to the unhealthy content type is a scenario where the unhealthy content type is allowed to exist; The first and second scene type acquisition module of the inappropriate content type is used to retrieve the first scene type and the second scene type corresponding to the inappropriate content type of the inappropriate content in the network content to be detected from the inappropriate content scene knowledge base.

7. The artificial intelligence governance system according to claim 5, characterized in that: The system further comprises: The first manual review module: if the bad content type of the bad content in the annotation of the network content to be detected exists in the bad content scenario knowledge base, and the scene type of the network content to be detected does not belong to the first scenario type and the second scenario type corresponding to the bad content type of the bad content in the annotation of the network content to be detected, then the bad content in the network content to be detected and the annotation of its bad content type and the scene of the network content to be detected are sent to manual review, and information is sent to remind the manual to determine whether the scene of the network content to be detected is the first scenario type or the second scenario type corresponding to the bad content type of the bad content in the annotation of the network content to be detected; obtain the result of the review, if the network content to be detected If the scenario of the network content to be detected is the first scenario type corresponding to the bad content type of the bad content in the annotation of the bad content to be detected, the scenario of the network content to be detected is added to the bad content scenario knowledge base as the first scenario type corresponding to the bad content type, and the bad content of the network content to be detected and the annotation of its bad content type are returned to the user; if the scenario of the network content to be detected is the second scenario type corresponding to the bad content type of the bad content in the annotation of the network content to be detected, the scenario of the network content to be detected is added to the bad content scenario knowledge base as the second scenario type corresponding to the bad content type, and the annotation of the bad content and its bad content type in the network content to be detected is deleted.

8. The artificial intelligence governance system according to claim 5, characterized in that: The system further comprises: The module using the scene and bad content identification deep learning model is as follows: the network content to be detected is obtained as input, and the output calculated by the bad content type identification deep learning model is used as the label of the bad content in the network content and its bad content type, and the scene type of the network content.

9. A robot comprising a memory, a processor, and an artificial intelligence robot program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the steps of the method according to any one of claims 1 to 4 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of the method according to any one of claims 1 to 4 are implemented.

Images