Content management system based on digital media

Abstract

The present invention discloses a content management method based on digital media, comprising: step one: a content-based platform produces professional content by connecting with digital resource partners; step two: hierarchical review management is performed on the generated professional content through a combination mechanism of technical review, manual review and user complaint mechanism; step three: in the first-level review process, the elements to be reviewed in the produced content are defined as content pictures and content texts, and the status of the content pictures is reviewed and managed through a model; step four: for the content review of the content text part, the content text is reviewed by matching and comparing sensitive words and using natural language to conduct semantic analysis of the text; step five: content review management is performed by outputting content pictures and content texts as label categories of uncertain content pictures and uncertain content texts through manual review. The present invention has the characteristic of improving the coupling between content technical review and manual review.

Description

Technical Field

[0001] The present invention relates to the technical field of digital media content management, in particular to a content management system based on digital media. Background Art

[0002] Since digital media data is generated quickly and updated frequently, content products have begun to use the dynamic display mechanism of information flow and artificial intelligence technology to distribute data to users more accurately. The value of the content management system lies mainly in managing the data information of the content library, including additional identification of hot news that needs to be actively pushed and preprocessing of content, labeling of content through artificial intelligence, etc., which lays the foundation for the algorithm team to further distribute content on the content library. However, the content management system tends to serve operators, such as displaying statistical information, identifying current social hot events through crawler technology and natural language processing technology, and facilitating operators to formulate corresponding operation strategies. However, the combination of machine review and human review based on rules and algorithm models requires the rapid iteration of algorithms and models and the shortest possible data processing cycle to ensure high-speed and high-frequency data supply and iterative updates in the content management system. The current content management system uses a management model composed of manual and algorithmic combinations, which has a coupling relationship that requires manual and algorithmic real-time synchronous updates and iterations, and has low performance in data processing and weak iteration capabilities. Therefore, it is necessary to design a digital media-based content management system that can realize high-performance operation of algorithms in content management and improve the high coupling of manual and algorithmic combinations. Summary of the invention

[0003] The object of the present invention is to provide a content management system based on digital media to solve the problems raised in the above background technology.

[0004] In order to solve the above technical problems, the present invention provides the following technical solutions: a content management system based on digital media, including a digital content generation module, a content level review management module, and a manual review management module;

[0005] The digital content generation module is used for the content platform to produce professional content by connecting with digital resource partners;

[0006] The content level review management module is used to conduct level review management on the generated professional content through a combination of technical review, manual review and user complaint mechanism;

[0007] The manual review management module is used to perform content review management on uncertain tag categories through manual review.

[0008] According to the above technical solution, the digital content generation module includes:

[0009] Content category dimension division module is used for content-based platforms to divide dimensions according to the categories of produced content;

[0010] A content production factor definition module is used to define content production factors using the category subject of content production;

[0011] The content resource library construction module is used to construct the content resource library belonging to the platform based on the produced content resources.

[0012] According to the above technical solution, the content level review management module includes:

[0013] Content review building module, which is used by content-based platforms to access third-party technical review interfaces and build their own content review terminals;

[0014] The content image review module is used to review and manage the status of content images through the model;

[0015] A machine learning classification review module is used to classify and review the content images based on machine learning;

[0016] The cross-domain content image review module is used to perform image content review tasks where the source domain and target domain data do not belong to the same domain;

[0017] The content text review module is used to match and compare sensitive words in the content text and use natural language to perform semantic analysis and review of the text.

[0018] According to the above technical solution, the manual review management module includes:

[0019] Manual review trigger module, used to review the content images and content texts that trigger manual review;

[0020] User complaint mechanism feedback module, which allows users to submit complaints about content text and content images through the complaint mechanism;

[0021] The review task regeneration module is used to resubmit for review and generate a re-review task when the threshold of complaint feedback is reached, and the reviewer will re-review this type of content.

[0022] According to the above technical solution, the digital media-based content management system executes the digital media-based content management method including the following steps:

[0023] Step 1: Content platforms produce professional content by connecting with digital resource partners, forming a platform content operation and management circle based on digital media. Professional content producers and digital resource partners analyze the content hierarchy and structural attributes of the content platform and define the form of content cooperation based on the application scenarios of the content platform;

[0024] Step 2: Conduct hierarchical review and management of the professional content produced through a combination of technical review, manual review and user complaint mechanism;

[0025] Step 3: In the first-level review process, define the elements to be reviewed in the produced content as content images and content texts, and review and manage the status of content images through the model;

[0026] Step 4: Review the text of the content by matching sensitive words in the text and using natural language to analyze the text semantics;

[0027] Step 5: Through manual review, the content images and content texts are output as label categories of uncertain content images and uncertain content texts for content review management.

[0028] According to the above technical solution, the steps of producing professional content by connecting with digital resource partners include:

[0029] Content-based platforms divide the dimensions according to the categories of content production, and use the categories of content production to define content production factors. The categories of content production refer to the content style and content characteristics presented by the producers in the professional content they produce. Information quality is set as the first-level indicator of content management, and the comprehensibility of information content and the timeliness of feedback services are used as the second-level indicators of content management. Specific categories are constructed based on the content of content producers. Synchronous content-based platforms associate and bind content producers among digital resource partners with identity support resources such as personal exclusive certification and symbolic identification, and manage and bind content producers to honorary awards such as achievement medals, and build a content resource library belonging to the platform based on the content resources they produce.

[0030] According to the above technical solution, the method for hierarchical review and management of the generated professional content through the combination mechanism is:

[0031] Content-based platforms access the technical review interface of a third party and build their own content review terminals, and use the shielding fonts in the technical review to perform the first-level review and management of content through server verification. The relatively standardized content of the shielding fonts includes content that should be prohibited from being published and transmitted in order to fulfill the platform's responsibilities, and can be personalized and adjusted. Professional content that passes the technical review will be published and disseminated normally. Content that triggers the shielding fonts will be visible only to professional content producers and digital resource partners themselves, and not to other users. When the technical review triggers a sensitive font or suspected sensitive word, the background prompt will be entered for manual review, that is, manual review will be entered, that is, the professional content generated will be judged by the content information standards set by humans to determine whether it violates the regulations and what kind of processing will be performed.

[0032] According to the above technical solution, the step of reviewing and managing the status of the content image through the model includes:

[0033] The content image status in the content platform, i.e., the category label of the content image, is defined. According to the image review standard defined in the platform, it is defined as three different category labels: normal, abnormal, and uncertain. The CNN-based image content review model is used as the main classification model, combined with a small part of the model based on detection or image segmentation for content review. The review of the content images is classified and reviewed according to machine learning. The main classification model is used to take the image i as the input data, and the output data is the probability p of each category x of the image, i.e., p=p θ (x|i), x∈{normal, abnormal, uncertain}, where p θ The classification model with parameter θ is set. This model can be used to solve the problem that the image source domain of content producers on the existing platform is not only for identification and detection on the platform, but also can be used to perform the image content review task. The source domain and target domain data do not belong to the same domain, that is, the data distribution is inconsistent, but the source domain task and the target domain task are both to predict the probability of a category label for the input data, so as to realize the cross-domain image content review of the model, that is, to conduct content review in multiple fields.

[0034] For image input data with multiple source domains and target domain image input data, each domain data is input into its own feature extractor to obtain features, and the parameter calculation amount is reduced by sharing parameters between each feature extractor. Then, the features extracted by the feature extractor are divided into two branches. The first branch obtains the classification result of the domain through a gradient reversal layer and a domain category classifier, realizes adversarial training with the feature extractor, makes the features extracted by the feature extractor closer to the domain invariant features, and obtains the integration weight. The second branch first aligns the features between the various domains through a moment matching module, and then inputs the aligned source domain features into the feature extractors and classifiers corresponding to each source domain to obtain the classification result of each source domain data. For the target domain, the classification results of multiple branches are obtained by inputting into all classifiers corresponding to the source domain, and the classification results of each classifier on the target domain data are integrated using the integration weights obtained by the domain classifier to obtain the final output result; and the output image of the uncertain label category is transmitted to the manual terminal for review, that is, it enters the manual review and is judged by the manual reviewer. The manual reviewer judges whether the generated professional content violates the regulations based on the set content information standards.

[0035] According to the above technical solution, the method of matching and comparing the content text for sensitive words and using natural language to perform semantic analysis and review of the text is as follows:

[0036] Artificial intelligence is used to review massive amounts of content text in the first phase, distinguishing between three content text categories: normal content text, abnormal content text, and uncertain content text. Natural language processing technology is used to process content text. Through text representation and specified review models, the risk word list is used to match sensitive words in the content text with the risk word list through matching rules, and the comparison results between sensitive text and risk words in the database are determined. Problem text is identified, labeled and classified, and corresponding features are generated based on the classification statistical model.

[0037] According to the above technical solution, the method for conducting content review management on uncertain tag categories through manual review includes:

[0038] The content images and text are reviewed for the first time and are labeled as uncertain and uncertain content text categories, triggering manual review to review the results of the technical review, as well as to review some uncertain situations where the technical review cannot produce a high degree of confidence. When content images or content text enter the review flow and have a high number of views in a short period of time, the content that has passed the review will be pushed to users and accept user feedback. Users can make complaints and feedback on content text and content images through the complaint mechanism. When the threshold of the complaint feedback is reached, it will be resubmitted for review to generate a re-review task, and the reviewer will review this type of content again.

[0039] Compared with the prior art, the beneficial effects achieved by the present invention are as follows: the present invention forms a platform content operation and management circle based on digital media by connecting digital resource partners to produce professional content in a content platform. Professional content producers and digital resource partners analyze the content main body hierarchy and structural attributes of the content platform, define the form of content cooperation for the application scenarios of the content platform, and utilize the review management of the combination mechanism. By inputting image data with multiple source domains and target domain image data, each domain data is input into a respective feature extractor to obtain features. By utilizing the shared parameters between each feature extractor, the problem of the image source domain of the content producer in the existing platform is solved, which is not only the recognition and detection problem on the platform, but also the synchronous execution of the source domain and target domain data of the image content review task that do not belong to the same domain, that is, the data distribution is inconsistent but the source domain task and the target domain task are both to predict the probability of a category label for the input data, thereby realizing the cross-domain image content review of the model, conducting content review in multiple fields, and improving the coupling between technical review and manual review in the content review management process. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0041] Figure 1 is a flow chart of the method steps of the present invention;

[0042] Figure 2 It is a schematic diagram of the system module composition of the present invention. DETAILED DESCRIPTION

[0043] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0044] Embodiment 1: Figure 1 This is a flowchart of a digital media-based content management method provided in Embodiment 1 of the present invention. This embodiment can be applied to a digital media-based content management scenario. This method can be executed by a digital media-based content management system provided in this embodiment. Figure 1 As shown, the method specifically comprises the following steps:

[0045] Step 1: Content platforms produce professional content by connecting with digital resource partners, forming a platform content operation and management circle based on digital media. Professional content producers and digital resource partners analyze the content hierarchy and structural attributes of the content platform and define the form of content cooperation based on the application scenarios of the content platform;

[0046] In an embodiment of the present invention, a content-based platform divides the dimensions according to the categories of produced content, and uses the category subjects of produced content to define content production factors. The category subjects of produced content refer to the content style and content characteristics presented by the producers in the production of professional content. Information quality is set as the first-level indicator of content management, and the comprehensibility of information content and the timeliness of feedback services are used as the second-level indicators of content management. Specific categories are constructed according to the content of content producers. The synchronous content-based platform associates and binds content producers among digital resource partners with identity support resources such as personal exclusive authentication and symbolic identification, and manages and binds content producers with honorary awards such as achievement medals, and constructs a content resource library belonging to the platform based on the content resources produced by them.

[0047] Step 2: Conduct hierarchical review and management of the professional content produced through a combination of technical review, manual review and user complaint mechanism;

[0048] In an embodiment of the present invention, a content-based platform accesses a third-party technical review interface and builds its own content review terminal, and uses the shielding character library in the technical review to perform the first-level review management of the content through server verification; wherein the relatively standardized content of the shielding character library includes content that should be prohibited from being published and transmitted in order to fulfill the platform's responsibilities, and can be personalized and adjusted. Professional content that passes the technical review will be published and disseminated normally, and the content that triggers the shielding character library will be published and visible only to professional content producers and digital resource partners themselves, and other users will not be able to see it. When the technical review triggers a sensitive character library and suspected sensitive characters, the background will be prompted for manual review, that is, manual review will be entered, that is, the professional content generated will be judged by the content information standards set by humans to determine whether it violates the regulations and what kind of processing to perform.

[0049] Step 3: In the first-level review process, define the elements to be reviewed in the produced content as content images and content texts, and review and manage the status of content images through the model;

[0050] In an embodiment of the present invention, the content image status in the content platform, i.e., the category label of the content image, is defined, and it is defined as three different category labels of normal, abnormal, and uncertain according to the image review standard defined in the platform. The CNN-based image content review model is used as the main classification model, combined with a small part of the model based on detection or image segmentation for content review, and the review of the content images is classified and reviewed according to machine learning. The main classification model is used to take the image i as input data, and the output data is the probability p of each category x of the image, i.e., p=p θ (x|i), x∈{normal, abnormal, uncertain}, where p θ The classification model with parameter θ is set. This model can be used to solve the problem that the image source domain of content producers on the existing platform is not only for identification and detection on the platform, but also can be used to perform the image content review task. The source domain and target domain data do not belong to the same domain, that is, the data distribution is inconsistent, but the source domain task and the target domain task are both to predict the probability of a category label for the input data, so as to realize the cross-domain image content review of the model, that is, to conduct content review in multiple fields.

[0051] Exemplarily, for image input data with multiple source domains and image input data of target domains, each domain data is input into its own feature extractor to obtain features, and the parameter calculation amount is reduced by sharing parameters between each feature extractor. Then, the features extracted by the feature extractor are divided into two branches. The first branch obtains the classification result of the domain through a gradient reversal layer and a domain category classifier, realizes adversarial training with the feature extractor, makes the features extracted by the feature extractor closer to the domain invariant features, and obtains the integration weight. The second branch first aligns the features between the domains through a moment matching module, and then inputs the aligned source domain features into the feature extractors and classifiers corresponding to each source domain to obtain the classification result of each source domain data. For the target domain, the classification results of multiple branches are obtained by inputting into all classifiers corresponding to the source domain, and the classification results of each classifier on the target domain data are integrated using the integration weights obtained by the domain classifier to obtain the final output result; and the output image of the uncertain label category is transmitted to the manual terminal for review, that is, it enters the manual review and is judged by the manual reviewer. The manual reviewer judges whether the generated professional content violates the regulations based on the set content information standards.

[0052] Step 4: Review the text of the content by matching sensitive words in the text and using natural language to analyze the text semantics;

[0053] In the embodiment of the present invention, artificial intelligence is used to review the massive content text in the first stage, distinguish the three content text categories of normal content text, abnormal content text, and uncertain content text, and use natural language processing technology to process the content text. Through text representation and a specified review model, the risk word list is used to match the sensitive words in the content text with the risk word list through matching rules, and the comparison results between the sensitive text and the risk words in the database are determined, the problematic text is identified, annotated and classified, and corresponding features are generated according to the classification statistical model;

[0054] Utilize auxiliary strategy management of content text review to intervene and assist in the strategy of technical review, fill the review database according to the real-time update of vocabulary, terminology, and Internet hot words, and ensure the accuracy of technical review by configuring keywords for technical review.

[0055] Step 5: Through manual review, the content images and content texts are output as label categories of uncertain content images and uncertain content texts for content review management.

[0056] In an embodiment of the present invention, content images and content text are reviewed for the first time and are labeled as uncertain and content text uncertain, triggering a manual review to review the results of the technical review, as well as to review some uncertain situations where the technical review cannot produce a high degree of confidence. When content images or content text enter the review flow and have a high number of views in a short period of time, the content that has passed the review will be pushed to the user and accept user feedback. The user can make complaint feedback on the content text and content images through the complaint mechanism. When the threshold of the complaint feedback is reached, it will be resubmitted for review to generate a re-review task, and the reviewer will re-review this type of content.

[0057] Embodiment 2: Embodiment 2 of the present invention provides a content management system based on digital media. Figure 2 The module composition diagram of the digital media-based content management system provided in the second embodiment of the present invention is as follows: Figure 2 As shown, the system includes:

[0058] Digital content generation module, used for content-based platforms to produce professional content by connecting with digital resource partners;

[0059] The content-level review management module is used to conduct hierarchical review and management of the generated professional content through a combination of technical review, manual review and user complaint mechanism;

[0060] The manual review management module is used to perform content review management on uncertain tag categories through manual review.

[0061] In some embodiments of the present invention, the digital content generation module includes:

[0062] Content category dimension division module is used for content-based platforms to divide dimensions according to the categories of produced content;

[0063] A content production factor definition module is used to define content production factors using the category subject of content production;

[0064] The content resource library construction module is used to construct the content resource library belonging to the platform based on the produced content resources.

[0065] In some embodiments of the present invention, the content level review management module includes:

[0066] Content review building module, which is used by content-based platforms to access third-party technical review interfaces and build their own content review terminals;

[0067] The content image review module is used to review and manage the status of content images through the model;

[0068] A machine learning classification review module is used to classify and review the content images based on machine learning;

[0069] The cross-domain content image review module is used to perform image content review tasks where the source domain and target domain data do not belong to the same domain;

[0070] The content text review module is used to match and compare sensitive words in the content text and use natural language to perform semantic analysis and review of the text.

[0071] In some embodiments of the present invention, the manual review management module includes:

[0072] Manual review trigger module, used to review the content images and content texts that trigger manual review;

[0073] User complaint mechanism feedback module, which allows users to submit complaints about content text and content images through the complaint mechanism;

[0074] The review task regeneration module is used to resubmit for review and generate a re-review task when the threshold of complaint feedback is reached, and the reviewer will re-review this type of content.

[0075] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0076] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art can still modify the technical solutions described in the aforementioned embodiments or replace some of the technical features therein by equivalents. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A content management system based on digital media, including a digital content generation module, a content level review management module, and a manual review management module, characterized in that: The digital content generation module is used for the content platform to produce professional content by connecting with digital resource partners; The content level review management module is used to conduct level review management on the generated professional content through a combination of technical review, manual review and user complaint mechanism; The manual review management module is used to perform content review management on uncertain tag categories through manual review; The digital content generation module comprises: Content category dimension division module is used for content-based platforms to divide dimensions according to the categories of produced content; A content production factor definition module is used to define content production factors using the category subject of content production; A content resource library construction module is used to construct a content resource library belonging to the platform based on the produced content resources; The content level review management module includes: Content review building module, which is used by content-based platforms to access third-party technical review interfaces and build their own content review terminals; The content image review module is used to review and manage the status of content images through the model; The machine learning classification review module is used to classify and review content images based on machine learning; The cross-domain content image review module is used to perform image content review tasks where the source domain and target domain data do not belong to the same domain; The content text review module is used to match sensitive words in the content text and use natural language to conduct semantic analysis and review of the text; The manual review management module includes: Manual review trigger module, used to review the content images and content texts that trigger manual review; User complaint mechanism feedback module, used for users to submit complaints and feedback on content text and content images through the complaint mechanism; The review task regeneration module is used to resubmit the content for review and generate a re-review task when the threshold of complaint feedback is reached. The reviewer will review the content again. The digital media-based content management system executes a digital media-based content management method, the method comprising the following steps: Step 1: Content platforms produce professional content by connecting with digital resource partners, forming a platform content operation and management circle based on digital media. Professional content producers and digital resource partners analyze the content hierarchy and structural attributes of the content platform and define the form of content cooperation based on the application scenarios of the content platform; Step 2: Conduct hierarchical review and management of the professional content produced through a combination of technical review, manual review and user complaint mechanism; Step 3: In the first-level review process, define the elements to be reviewed in the produced content as content images and content text, and use the model to review and manage the status of the content images; Step 4: Review the text of the content by matching sensitive words in the text and using natural language to analyze the text semantics; Step 5: Through manual review, the content images and content texts are output as the label categories of uncertain content images and uncertain content texts for content review management; The steps of producing professional content by connecting with digital resource partners include: Content-based platforms divide content production into different dimensions according to the categories of content production, and use the main categories of content production to define content production factors. The main categories of content production refer to the content style and content characteristics presented by the producers in the production of professional content. Information quality is set as the first-level indicator of content management, and the comprehensibility of information content and the timeliness of feedback services are used as the second-level indicators of content management. Specific categories are constructed based on the content of content producers. Synchronous content-based platforms associate content producers among digital resource partners with identity support resources of personal exclusive authentication and symbolic identification, and associate content producers with honorary reward resources, and build a content resource library belonging to the platform based on the content resources they produce; The method for conducting hierarchical review and management of the generated professional content through the combination mechanism is as follows: Content-based platforms access the technical review interface of a third party and build their own content review terminals, and use the shielded word library in the technical review to conduct the first-level review and management of content through server verification. The shielded word library contains content that should be prohibited from being published and transmitted in order to fulfill the platform's responsibilities, and personalized adjustments are made. Professional content that passes the technical review is published and disseminated normally. Content that triggers the shielded word library is only visible to professional content producers and digital resource partners themselves, and is not visible to other users. When the technical review triggers the sensitive word library and suspected sensitive words, the background prompts manual review and enters manual review. The content information standards set by humans will determine whether the professional content is in violation of regulations and what kind of treatment to take. The step of reviewing and managing the status of the content image through the model includes: Define the category labels of content images in the content platform, and define them as three different category labels: normal, abnormal, and uncertain according to the image review standards defined in the platform. Use the CNN-based image content review model as the main classification model, combined with a small part of the model based on detection or image segmentation for content review, and classify and review the content images according to machine learning. Use the main classification model to take image i as input data, and output data as the probability of each category x of the image. ,in, To set the parameters A classification model is used to conduct content review in multiple fields; For image input data with multiple source domains and target domain image input data, each domain data is input into its own feature extractor to obtain features. The features extracted by the feature extractor are divided into two branches by using the shared parameters between each feature extractor. The first branch obtains the classification result of the domain through a gradient reversal layer and a domain category classifier, realizes adversarial training with the feature extractor, and obtains integration weights. The second branch first aligns the features between the domains through a moment matching module, and inputs the aligned source domain features into the feature extractors and classifiers corresponding to each source domain to obtain the classification result of each source domain data. For the target domain, the classification results of multiple branches are obtained by inputting them into all classifiers corresponding to the source domain. The classification results of each classifier on the target domain data are integrated using the integration weights obtained by the domain classifier to obtain the final output result. The output image of the uncertain label category is transmitted to the manual terminal for review and processing, and enters the manual review for judgment by the manual reviewer. The manual reviewer judges whether the generated professional content violates the regulations based on the set content information standards.

2. The digital media-based content management system according to claim 1, characterized in that: The method of matching and comparing sensitive words in the content text and using natural language to perform semantic analysis and review of the text is as follows: Use artificial intelligence to review massive amounts of content text in the first phase, distinguish between three types of content text: normal content text, abnormal content text, and uncertain content text, use natural language processing technology to process content text, use text representation and specified review models, use risk word lists to match sensitive words in content text with risk word lists through matching rules, determine the comparison results between sensitive text and risk words in the database, identify problematic text, mark and classify it, and generate corresponding features based on the classification statistical model; Utilize auxiliary strategy management of content text review to intervene and assist in the strategy of technical review, and fill the review database according to the real-time update of vocabulary, terminology, and Internet hot words.

3. The digital media-based content management system according to claim 2, characterized in that: The method for performing content review management on uncertain tag categories through manual review includes: The content images and text are reviewed for the first time and are labeled as uncertain and uncertain content text categories, triggering manual review to review the results of the technical review, as well as to review some uncertain situations where the technical review cannot produce a high degree of confidence. When content images or content text enter the review flow and have views within a short period of time, the content that has passed the review will be pushed to users and accept user feedback. Users make complaints about content text and content images through the complaint mechanism. When the threshold of the complaint feedback is reached, it will be resubmitted for review to generate a re-review task, and the reviewer will review this type of content again.

Images