A media processing method, apparatus, device, and readable storage medium

Abstract

The application discloses a media processing method, device and equipment and a readable storage medium. The method comprises the following steps: obtaining configuration parameters and access parameters of media data, the configuration parameters being parameters generated when the media data is configured for delivery, and the access parameters being parameters generated after the media data is accessed during the delivery process; constructing a monitoring link for the media data according to the configuration parameters and the access parameters; obtaining media content of the media data by accessing the monitoring link; and performing risk identification processing on the media data based on the media content to obtain a risk identification result of the media data. According to the application, the risk identification of the media data during delivery can be accurately performed, and the audit quality of the media data is improved.

Description

Technical Field

[0001] This application relates to the field of computer technology, and in particular to a media processing method, apparatus, device, and readable storage medium. Background Technology

[0002] With the rapid development of computer technology, more and more users are publishing or obtaining information through applications on electronic devices. Users who create media data can use applications on electronic devices to deliver media data.

[0003] In related technologies, user-uploaded media data requires prior manual review before the media data's placement materials are determined, and whether or not the media data will be distributed is then decided based on the review results. However, this review method relies solely on the media materials uploaded by the creator of the media data. Creators can easily tamper with the materials presented on the landing page during the distribution process. For example, creators can easily adjust the page text content, insert exaggerated promotional content, etc., making the media data risky and unsuitable for distribution to other users. Therefore, a solution that can accurately identify the risks of media data during distribution is urgently needed. Summary of the Invention

[0004] This application provides a media processing method, apparatus, device, and readable storage medium, which can accurately identify risks in media data during delivery and improve the quality of media data review.

[0005] This application provides a media processing method, including:

[0006] Obtain the configuration parameters and access parameters of media data. The configuration parameters are the parameters generated when the media data is configured for delivery, and the access parameters are the parameters generated after the media data is accessed during the delivery process.

[0007] Based on the configuration and access parameters, a monitoring link is built for the media data;

[0008] Media content that obtains media data by accessing monitoring links;

[0009] Based on media content, risk identification processing is performed on media data to obtain risk identification results for media data.

[0010] One embodiment of this application provides a media processing apparatus, including:

[0011] The parameter acquisition module is used to acquire the configuration parameters and access parameters of media data. The configuration parameters are the parameters generated when the media data is configured for delivery, and the access parameters are the parameters generated after the media data is accessed during the delivery process.

[0012] The link building module is used to build monitoring links for media data based on configuration parameters and access parameters;

[0013] The link access module is used to obtain media content by accessing monitoring links;

[0014] The risk identification module is used to perform risk identification processing on media data based on media content, and obtain the risk identification results of the media data.

[0015] In one embodiment, the link building module constructs monitoring links for media data based on configuration parameters and access parameters, including:

[0016] Media access scenarios for acquiring media data;

[0017] If the scenario type of the media access scenario is a specified type, then the specified device parameters of the media data in the media access scenario are obtained, and the configuration parameters, access parameters and the specified device parameters of the media data in the media access scenario are assembled to obtain the monitoring link of the media data.

[0018] If the media access scenario is of the normal type, the configuration parameters and access parameters of the media data are assembled to obtain the monitoring link of the media data.

[0019] In one embodiment, the risk identification result of the media data includes the risk type to which the media data belongs;

[0020] The risk identification module, based on media content, performs risk identification processing on media data to obtain the specific implementation methods for the risk identification results of media data, including:

[0021] Retrieve pre-configured exception elements;

[0022] Detect media content based on abnormal elements;

[0023] If the media content contains anomalous elements, the risk type to which the anomalous elements belong is determined as the risk type to which the media data belongs.

[0024] If the media content does not contain any abnormal elements, then the risk type of the media data is determined to be risk-free.

[0025] In one embodiment, the risk identification result of the media data includes the risk type to which the media data belongs;

[0026] The risk identification module, based on media content, performs risk identification processing on media data to obtain the specific implementation methods for the risk identification results of media data, including:

[0027] Content analysis is performed on media content to obtain its multimodal characteristics;

[0028] Based on the multimodal characteristics of media content, risk identification processing is performed on media data to obtain the risk type to which the media data belongs.

[0029] In one embodiment, the risk identification module performs content analysis on media content to obtain a specific implementation of the multimodal features of the media content, including:

[0030] Get the pre-configured N modal types; N is a positive integer;

[0031] The media content is captured and processed according to each modality type to obtain the modal capture content of the media content under each modality type;

[0032] Feature extraction processing is performed on the modal acquisition content of media content under each modal type to obtain the modal features of media content under each modal type;

[0033] The modal features of media content under N modal types are fused to obtain the multimodal features of the media content.

[0034] In one embodiment, the risk identification module performs risk identification processing on media data based on the multimodal characteristics of media content to obtain the specific implementation of the risk type to which the media data belongs, including:

[0035] Media configuration characteristics for acquiring media data;

[0036] By fusing the multimodal features of media content with the features of media configuration, we obtain the integrated delivery features of media data;

[0037] Based on the characteristics of integrated delivery, a risk identification model is invoked. The risk identification model is used to process media data for risk identification and to obtain the risk type of the media data.

[0038] In one embodiment, the risk identification module performs risk identification processing on media data using a risk identification model to obtain the specific implementation method of the risk type to which the media data belongs, including:

[0039] The probability distribution is obtained by calculating and analyzing the characteristics of integrated delivery through the model parameters of the risk identification model. The probability distribution includes M predicted probabilities, each of which is the probability that the media data predicted by the risk identification model belongs to a candidate risk type. M is a positive integer.

[0040] Obtain the maximum predicted probability from the probability distribution;

[0041] If the maximum predicted probability is greater than the probability threshold, then the candidate risk type indicated by the maximum predicted probability is determined as the risk type to which the media data belongs.

[0042] If the maximum predicted probability is less than the probability threshold, the risk type of the media data will be determined as the risk-free type.

[0043] In one embodiment, the risk identification model includes M expert networks, each expert network corresponding to a candidate risk type;

[0044] The risk identification module calculates and analyzes the characteristics of the fusion delivery using the model parameters of the risk identification model to obtain the specific implementation method of the probability distribution, including:

[0045] The fusion delivery features are input into each expert network, and each expert network calculates and analyzes the fusion delivery features to obtain the predicted probability of each expert network for the corresponding candidate risk type.

[0046] The probability distribution is obtained by combining the predicted probabilities of M expert networks for their respective candidate risk types.

[0047] In one embodiment, the apparatus further includes a model training module.

[0048] The model training module is used to acquire sample media data, as well as the sample multimodal features and sample media configuration features corresponding to the sample media data;

[0049] The model training module is also used to fuse the multimodal features of the sample media data with the configuration features of the sample media data to obtain the sample fusion delivery features of the sample media data.

[0050] The model training module is also used to call the initial risk identification model based on the sample fusion and delivery features, and to perform risk identification processing on the sample media data through the initial risk identification model to obtain the predicted risk type of the sample media data.

[0051] The model training module is also used to obtain the risk type labels corresponding to the sample media data, and to train the initial risk identification model based on the risk type labels corresponding to the sample media data and the predicted risk type, so as to obtain the risk identification model.

[0052] In one embodiment, the model training module trains an initial risk identification model based on the risk type labels and predicted risk types corresponding to the sample media data, resulting in a specific implementation of the risk identification model, including:

[0053] Calculate the error loss between the risk type label corresponding to the sample media data and the predicted risk type;

[0054] The initial model parameters of the initial risk identification model are adjusted and optimized according to the error loss between the risk type label and the predicted risk type to obtain the model parameters;

[0055] The initial risk identification model, which includes model parameters, is determined as the risk identification model.

[0056] In one embodiment, after the risk identification module performs risk identification processing on the media data based on the media content and obtains the risk identification result of the media data, the risk identification module is further specifically used for:

[0057] The time period for acquiring media data;

[0058] If the delivery time falls within a specified time period, then obtain the review policy corresponding to that specified time period;

[0059] The risk identification results of media data are detected according to the review strategy corresponding to the specified time period, and the risk identification results after media data detection are obtained.

[0060] One embodiment of this application provides a computer device, including: a processor and a memory;

[0061] The memory stores a computer program, which, when executed by a processor, causes the processor to perform the methods described in the embodiments of this application.

[0062] One aspect of this application provides a computer-readable storage medium storing a computer program, which includes program instructions. When executed by a processor, the program instructions perform the methods described in this application.

[0063] One aspect of this application provides a computer program product comprising a computer program stored in a computer-readable storage medium. A processor of a computer device reads the computer program from the computer-readable storage medium and executes the computer program, causing the computer device to perform the method provided in one aspect of the embodiments of this application.

[0064] This application provides a scheme for risk identification of media data during delivery, which can accurately identify whether the media data during delivery is risky, thereby enhancing the quality of media data review. Specifically, for media data during delivery, this application can obtain the configuration parameters generated during delivery configuration and the access parameters generated after being accessed during delivery. Based on these configuration parameters and access parameters, this application can construct a monitoring link for this media data to monitor its media content. Since this monitoring link monitors the media content of the media data, it can reflect the actual content of the landing page of the media data. Thus, by accessing this monitoring link, the media content of this media data (the actual delivery content during delivery) can be accurately obtained. Based on this actual media content, risk identification processing can be performed on this media data to obtain the risk identification result of this media data. As can be seen, this application constructs a link for monitoring the content delivered by media data, which can be accessed automatically to obtain the actual content delivered by the media data. Then, risk identification processing can be performed on the obtained actual content delivered to determine the risk identification result of the media data. In this way, not only can the media data in delivery be reviewed to improve the quality of media data review, but also no manual processing is required throughout the process, which can effectively ensure the efficiency of reviewing media data in delivery. Attached Figure Description

[0065] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0066] Figure 1 This is a schematic diagram of the architecture of a media data risk identification system provided in an exemplary embodiment of this application;

[0067] Figure 2a This is a schematic diagram illustrating a scenario of a user accessing advertising data, provided in an embodiment of this application.

[0068] Figure 2b This is a schematic diagram illustrating a scenario for risk identification of advertising data provided in an embodiment of this application;

[0069] Figure 3 This is a schematic flowchart of a media processing method provided by an exemplary embodiment of this application;

[0070] Figure 4 This is a schematic diagram illustrating a process for risk identification and processing of media data based on media data, as provided in an embodiment of this application.

[0071] Figure 5 This is a schematic diagram of a logical architecture for risk identification processing of media data based on media content, provided in an embodiment of this application.

[0072] Figure 6 This is a schematic diagram of the logical architecture of a solution provided in an embodiment of this application;

[0073] Figure 7 This is a schematic diagram of the structure of a media processing device provided in an embodiment of this application;

[0074] Figure 8 This is a schematic diagram of the structure of a computer device provided in an embodiment of this application. Detailed Implementation

[0075] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0076] The media data in this application may refer to advertising data. For ease of understanding, the relevant concepts of advertising data involved in this application will be explained below.

[0077] Advertiser: The creator or distributor of advertising data.

[0078] Ad monitoring links: These are specific links generated by the platform that delivers ad data to monitor the effectiveness of ad delivery. These links typically include information such as user click times, advertiser accounts, ad placement (the specific location where the ad is displayed on the platform, e.g., video interstitials, pop-ups, search results pages), and callback information (data feedback returned to the advertiser or ad platform during ad display, clicks, or conversions. This feedback usually includes key metrics such as ad impressions, clicks, and conversions, helping advertisers monitor ad performance in real time and adjust ad content). Ad monitoring links allow you to differentiate between different ads, ad placements, and user behavior related to the ads.

[0079] Macro replacement: Dynamically replaces parameters in the ad delivery link. Macro replacement can be understood as replacing some custom variables in the delivery link with specific values. For example, suppose "__ADGROUP_ID__" is a placeholder (macro) in the delivery link. When a user actually visits this delivery link (e.g., clicks), the "ADGROUP_ID" field will be replaced with the actual data generated after the visit.

[0080] In practical applications, advertisers can place their advertising data on various media platforms. Different users can view the advertising data placement links on these media platforms. Users can access the advertising data by visiting these placement links (e.g., clicking on the placement link). After visiting the placement link, users can view the landing page of the advertising data (the landing page is the page that users enter after clicking on the advertising data. This page provides detailed advertising content such as the advertised product and brand, which is used to guide users to perform actions such as filling in information, downloading applications, and purchasing products). From the landing page, users can view the content of the advertising data placement (which can be understood as the content presented on the landing page, or the advertising material, such as detailed content of the advertised product and brand). In traditional technology, advertising data undergoes manual review before being deployed. This review checks for non-compliant content (e.g., non-compliant text, products, images, audio, etc.). The decision to deploy the advertising data is based on the review results. Approved data is deployed, while rejected data is returned to the advertiser for modification until all content complies with regulations.

[0081] However, traditional methods of pre-auditing advertising data rely solely on reviewing the ad creatives. Advertisers can easily tamper with the landing page content during ad delivery, such as adjusting text or inserting exaggerated claims. This renders the ad data risky and unsuitable for user viewing. Further auditing during the campaign requires manual, periodic access to check the landing page content. Given the massive amounts of ad data on media platforms, manual auditing is extremely resource-intensive and inefficient. Furthermore, manual auditing is prone to omissions and false positives, resulting in low audit quality and negatively impacting the user experience.

[0082] To improve the quality of reviewing advertising data during campaigns, this application provides a scheme for risk identification of advertising data during campaigns, which can accurately identify whether there are risks in the advertising data. Since advertising data is usually presented in different media formats (e.g., video, image, audio, text, etc.), it can also be considered a type of media data, and will be referred to as media data thereafter. The risk identification scheme for media data during campaigns involved in this application can include at least the following four consecutive steps: 1. First, for any media data to be tested, the configuration parameters and access parameters of this media data can be obtained; where configuration parameters refer to the parameters generated during the configuration of media data deployment. It should be understood that for any media data, before deployment, it needs to be configured to determine the media ID, the ID of the media creation object (e.g., the advertiser), the geographic targeting of the media data, the deployment time, and generate a deployment link for this media data. This information can be considered as the basic information that needs to be configured and determined before the media data is deployed, and can be understood as... The media data delivery configuration parameters are referred to as configuration parameters in this application; access parameters can refer to the parameters generated after a user accesses the media data after it has been delivered. For any media data, once it is accessed (e.g., clicked) after delivery, corresponding access parameters will be generated, such as access time (e.g., click time), attribution ID, delivery location, callback information, etc. These parameters that are generated with the access behavior can be considered as media data delivery access parameters, which are referred to as access parameters in this application; 2. After obtaining the media data configuration parameters and access parameters, a monitoring link can be built for this media data. This monitoring link is used to monitor the media content of the media data. In fact, this monitoring link can be understood as a page link to the landing page of the media data. Different users will enter the landing page after accessing the media data delivery link, and the landing page will present the media content of the media data to the users (the delivery content of the media data may include detailed information such as the media delivery target (delivery product), media text, media audio, etc.). This monitoring link can be understood as a page link to the landing page, used to reflect the media content; 3. After the monitoring link is constructed, this application can simulate access to it, that is, the machine can access it without human clicking. The media content of the media data can be obtained by accessing this monitoring link; 4. After obtaining the media content of the media data, risk identification processing can be performed on the media data based on this media content, and finally the risk identification result of the media data can be obtained.

[0083] As can be seen, in the process of delivering media data, this application embodiment can automatically construct a monitoring link for monitoring the media content of the media data. Then, by automatically accessing this monitoring link, the actual delivered media content can be obtained. This eliminates the need for manual access to the media data to obtain the true delivered content. Based on the media content, this application can automatically perform risk identification processing on the media data to determine whether there is a risk of delivering content that does not comply with regulations. The entire process can accurately identify risks in the delivered media data without human intervention, determining whether it contains non-compliant content. The quality and efficiency of media data review can be greatly improved, and the quality of media data delivery will also increase accordingly, thereby enhancing the user experience.

[0084] The risk identification scheme for media data during delivery provided in this application embodiment can be applied to application scenarios that require review of media content, including but not limited to: short video push scenarios, video playback scenarios (scenarios that can be used to watch TV series, movies, and variety shows), and game scenarios. Wherein:

[0085] Short video push scenarios refer to situations where video data is continuously pushed to users. Users can request updates to display the next video by performing operations such as fetching video data (e.g., swiping the video display interface on the terminal device). Therefore, in short video push scenarios, users can continuously refresh and browse different video data by repeatedly performing these fetching operations.

[0086] Video playback scenarios can refer to the scenario where a user watches a TV series, movie, or variety show on a video playback platform.

[0087] In summary, the risk identification solution for media data during delivery provided in this application can automatically construct monitoring links for media data, allowing real-time access to the actual media content being delivered without manual clicking, and then identifying risks based on the actual delivered media content. The entire process is automated, requiring no manual operation, and effectively improves business coverage to a certain extent (e.g., expanding applicable scenarios).

[0088] It should be noted that the above-mentioned application scenarios are merely examples and do not limit the application scenarios applicable to the risk identification scheme for media data in delivery provided in this application embodiment.

[0089] Furthermore, the dynamic adjustment scheme for media encoding parameters provided in this application embodiment can be executed by a computer device, which may include a terminal or a server, and may also include both a terminal and a server. To facilitate understanding of the scheme for risk identification of media data provided in this application embodiment, the following is in conjunction with... Figure 1 The risk identification system for media data shown illustrates the application scenarios involved in the embodiments of this application; wherein, Figure 1 This is a schematic diagram of the architecture of a media data risk identification system provided in an exemplary embodiment of this application, such as... Figure 1 As shown, the system includes a terminal 101 and a server 102; wherein:

[0090] 1) Terminal 101 may include the terminal device used by the user. Of course, depending on the application scenario and field in which this solution is applied, the terminal provided in this application embodiment may vary. Terminal devices may include, but are not limited to: smartphones (such as smartphones running the Android system, or smartphones running the Internetworking Operating System (IOS), tablet computers, portable personal computers, mobile Internet devices (MIDs), in-vehicle devices, head-mounted smart home devices, and smart voice interaction devices, etc. This application embodiment does not limit the type of terminal device, which is stated here.

[0091] For example, in a short video push scenario, the terminal device can be a smartphone; that is, in this implementation, the solution provided in this application embodiment can be deployed on the smartphone; when a user uses the short video push application on the smartphone, the smartphone obtains the advertising data pushed to the user while the user is playing video data, and builds a monitoring link for this advertising data; then the smartphone obtains the advertising content delivered to the user by accessing this monitoring link, and performs risk identification processing on the advertising data based on the actual delivered advertising content, obtaining the risk identification result of the advertising data. As another example, in a smart vehicle scenario, the application deploying the solution provided in this application embodiment is an in-vehicle application; the type of the in-vehicle application may include, but is not limited to, music, video, or games.

[0092] An application can refer to a computer program designed to perform one or more specific tasks. Classifying applications according to different dimensions (such as how they run and their functions) can yield different types of the same application. For example, based on how they run, applications can include, but are not limited to: clients installed on a terminal, small programs that can be used without downloading and installation (as subroutines of the client), and World Wide Web (Web) applications opened through a browser. Another example is based on their functional type, which can include, but is not limited to: instant messaging (IM) applications, interactive content applications, audio applications, or video applications. Instant messaging applications refer to applications that facilitate instant communication and social interaction over the internet. These applications can include, but are not limited to, applications with communication functions, map applications with interactive functions, and game applications. Interactive content applications are applications capable of content interaction, such as sharing platforms, personal spaces, and news applications. Audio applications are applications that implement audio functions over the internet. Audio applications can include, but are not limited to, music applications with music playback and editing capabilities, radio applications with radio playback capabilities, or live streaming applications with live streaming capabilities. A video application is an application that can play videos. Video applications can include, but are not limited to, applications that have short videos (videos are often short in length, such as a few seconds or a few minutes) (such as short video push applications), applications that have long videos (such as videos with a longer playback time, such as movies or TV series), and so on.

[0093] Of course, the solutions provided in this application can be deployed directly on a device (such as a smartphone) or outside of an application, as described above, or they can be deployed on a device or application in the form of a plugin. This application does not limit the carrier of the deployment solution.

[0094] 2) Server 102 can be a server corresponding to the terminal, used to interact with the terminal to provide computing and application service support. Specifically, this server is a backend server corresponding to the application deployed in the terminal, used to interact with the terminal to provide computing and application services for the application. Server 102 can be a standalone physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDN), and big data and artificial intelligence platforms.

[0095] The terminal 101 and the server 102 can be directly or indirectly connected via wired or wireless communication, and this application does not impose any limitations on this. Furthermore, this application does not limit the number of terminals and servers; Figure 1 The number of terminals 101 and servers 102 is only a single example. In actual applications, multiple servers may be distributed in a distributed manner. This is just a brief explanation.

[0096] Based on the above-described solution and system architecture, the following points should also be noted:

[0097] ①The embodiments of this application mentioned above Figure 1 The system shown is for the purpose of more clearly illustrating the technical solutions of the embodiments of this application and does not constitute a limitation on the technical solutions provided by the embodiments of this application. Those skilled in the art will understand that with the evolution of system architecture and the emergence of new business scenarios, the technical solutions provided by the embodiments of this application are also applicable to similar technical problems. For example, the above description uses the example of the execution subject "computer device" of this application including a terminal and a server, that is, the terminal and server jointly executing the solution provided by the embodiments of this application, to introduce one application scenario of this solution; it should be understood that in practical applications, the computer device can also be a terminal or a server, that is, it supports the terminal or server alone executing the solution provided by the embodiments of this application.

[0098] ② This application embodiment supports the use of models with risk identification capabilities (such as Mixed Expert Models (MoE) comprising multiple expert networks) to implement the above-described scheme. Specifically, this application can call this model to perform risk identification processing on media data. This model can be trained using machine learning (such as reinforcement learning) to make the model's output (i.e., the predicted risk identification result) increasingly accurate. The trained model can be deployed on a computer device; thus, when the computer device needs to perform risk identification processing on certain media data, it can directly call the model. Wherein, if the computer device used to execute the scheme provided in this application embodiment is a terminal, then the model can be deployed on the terminal. If the computer device used to execute the scheme provided in this application embodiment is a server, then the model is deployed on the server; in this case, the terminal used by the user transmits the media data to be predicted to the server for risk identification processing of the media data.

[0099] ③ The data collection and processing in this application embodiment should strictly comply with the requirements of relevant laws and regulations. Obtaining personal information requires the knowledge or consent of the individual (or a legal basis for information acquisition), and subsequent data use and processing should be carried out within the scope of laws, regulations, and the authorization of the personal information subject. For example, when this application embodiment is applied to specific products or technologies, such as obtaining the user's currently playing media data, the user's permission or consent is required, and the collection, use, and processing of related data must comply with the relevant laws, regulations, and standards of the relevant regions.

[0100] For a better understanding of the application scenarios based on the above-described solution, please refer to [the relevant documentation / reference]. Figure 2a and Figure 2b ,in, Figure 2a This is a schematic diagram illustrating a scenario of a user accessing advertising data, provided in an embodiment of this application. For example... Figure 2a As shown, when user A uses a short video push application deployed on a terminal (assuming user A's terminal is terminal device 100a), the application can continuously push new video data to user A. User A can request the display of new video data by performing a fetch operation. Figure 2aAs shown, in the video display interface 2001 of the terminal, the video data currently exposed to the user is video data 200a (that is, the video data currently being played by user a is video data 200a, the creator of video data 200a is an object named "Lala Loves Dancing", and the video text of video data 200a is "Teach you a dance in three minutes"). The video display interface 2001 displays like, comment, and share controls, which can be used by user a to like, comment on, and share video data 200a. In addition, user a can request a refresh of the video display interface 2001 by performing a swipe operation to display new video data (video data that user a has not played or is currently playing).

[0101] like Figure 2a As shown, in the video display interface 2001, a placement link 2P for advertisement data M is also displayed (this placement link includes the text content "Shopping | New Product Sales Boost with Free Plush Toy"). While watching video data 200a, user a can also see this placement link. User a can access the landing page of advertisement data M by triggering (e.g., clicking) this placement link, where the detailed advertisement content of advertisement data M can be displayed. For example, as... Figure 2a As shown, after user a clicks on this ad link 2P, terminal device 100a can display the landing page 2002 of ad data M. This landing page 2002 can display the detailed ad content of ad data M (for example, such as...). Figure 2a The advertisement displays images of the advertised product, available styles, prices, etc. User A can purchase any style of this advertised product by triggering the "Buy Now" control in landing page 2002.

[0102] Therefore, it is evident that without reviewing the content of advertising data, users are likely to encounter ads with exaggerated claims or false advertising, thus degrading their viewing experience. Traditional methods of pre-screening advertising data rely solely on reviewing the ad creatives, making it extremely easy for advertisers to tamper with the landing page content during ad delivery. This renders the advertising data risky and unsuitable for user viewing. To improve the quality of advertising data review during delivery, this application provides a risk identification scheme for advertising data during delivery, accurately identifying whether any risks exist. For clarity, please refer to [link to relevant documentation]. Figure 2b , Figure 2b This is a schematic diagram illustrating a scenario for risk identification of advertising data provided in an embodiment of this application. For example... Figure 2b As shown, combined with Figure 2aIn the corresponding embodiment, for the advertising data M displayed in the video display interface 2001, a monitoring link can be automatically constructed for it. This monitoring link is used to monitor the actual content of the advertising data M. In fact, the monitoring link can be understood as a page link of the landing page of the advertising data M (for example, the landing page 2002 mentioned above). After the monitoring link is constructed, the terminal device 100a used by user a can access the monitoring link (for example, simulate the user clicking the monitoring link). After accessing, the actual content of the advertising data M can be obtained. Based on the actual advertising content, the risk of the advertising data M can be identified to obtain the risk identification result of the advertising data M.

[0103] If the risk identification result indicates that this advertising data M is risky, it means that the advertising content actually delivered by this advertising data M contains content that does not comply with regulations. In this case, the delivery of this advertising data M can be suspended, and the advertiser can modify the advertising content of this advertising data M before delivering the modified advertising data M.

[0104] As can be seen, in the process of delivering media data, this application embodiment can automatically construct a monitoring link for monitoring the media content of the media data. Then, by automatically accessing this monitoring link, the actual delivered media content can be obtained. This eliminates the need for manual access to the media data to obtain the true delivered content. Based on the media content, this application can automatically perform risk identification processing on the media data to determine whether there is a risk of delivering content that does not comply with regulations. The entire process can accurately identify risks in the delivered media data without human intervention, determining whether it contains non-compliant content. The quality and efficiency of media data review can be greatly improved, and the quality of media data delivery will also increase accordingly, thereby enhancing the user experience.

[0105] Based on the above-described solutions and application scenarios, this application proposes a more detailed media processing method. The media processing method proposed in this application will be described in detail below with reference to the accompanying drawings.

[0106] Please see Figure 3 , Figure 3 This is a flowchart illustrating a media processing method provided in an exemplary embodiment of this application. The flowchart may refer to the process of a risk identification scheme for media data during delivery provided in this embodiment. The media processing method can be executed by a computer device in the aforementioned system, such as a terminal and / or a server; the media processing method may include at least the following steps S301-S304:

[0107] Step S301: Obtain the configuration parameters and access parameters of the media data. The configuration parameters are the parameters generated when the media data is configured for delivery, and the access parameters are the parameters generated after the media data is accessed during the delivery process.

[0108] In this application, before any media data is delivered, it needs to be configured to determine the media identifier (a unique identifier for this media data, such as media ID or media name), the object identifier of the media creation object (e.g., advertiser) (a unique identifier for this media creation object, such as object ID or object name), the targeted delivery region, delivery time, delivery duration, and delivery link (accessing this link leads to the media data's landing page). This information can be considered as some basic information that needs to be configured before the media data is delivered, and can be understood as the media data's delivery configuration parameters. This application refers to them as configuration parameters, meaning that configuration parameters can refer to the parameters generated when configuring the media data for delivery. After the media data is delivered, users can access the media data (e.g., ...). Users can click to access media data to view its specific content. For any given media data, once it is accessed after being delivered, corresponding access information is generated. This access information includes, for example, the user's access time (e.g., click time), attribution data (in advertising, attribution is a means of analyzing and determining the degree of influence of advertising touchpoints on conversion behavior. In advertising marketing, users typically experience multiple advertising touchpoints, such as clicking on ads, browsing web pages, receiving emails, etc. The goal of advertising attribution is to identify which advertising touchpoints play a decisive role in the user's final conversion behavior), ad placement, callback information, etc. This information, which is generated only after the access behavior occurs, can be considered as the access parameters generated after the media data is delivered. This application refers to them as access parameters, that is, access parameters refer to the parameters generated after the media data is accessed during the delivery process.

[0109] In fact, based on the monitoring link described above, the monitoring link for advertising data refers to a specific link generated by the media platform (placement platform) of the advertising data to monitor the performance of the advertising data. This monitoring link includes parameters such as access (e.g., click) time, advertiser account, placement, and callback information. These parameters are usually replaced with specific field values ​​after the user actually accesses the advertising data. This application, in order to enable computer devices to obtain the actual advertising content delivered by the advertising data without the user manually accessing the advertising data, constructs a monitoring link and automatically performs macro replacement on the monitoring link. Thus, the constructed monitoring link is equivalent to the landing page link of the advertising data. Subsequently, the computer device can accurately obtain the actual advertising content delivered by the advertising data by simulating a user accessing this monitoring link. In the process of constructing the monitoring link, a series of parameters that make up the monitoring link are needed. The above configuration parameters can be understood as some basic configuration parameters of the advertising data, and the access parameters can be understood as the macro replacement parameters obtained after the actual click.

[0110] Step S302: Based on the configuration parameters and access parameters, construct a monitoring link for the media data.

[0111] In this application, after obtaining the configuration parameters and access parameters used to construct the monitoring link for media data, the monitoring link can be constructed for the media data based on these configuration parameters and access parameters. The specific implementation process may include, but is not limited to, the following: First, the media access scenario of the media data can be obtained; if the scenario type of the media access scenario is a specified type, the specified device parameters of the media data under this media access scenario can be obtained, and the configuration parameters, access parameters, and specified device parameters of the media data under the media access scenario can be assembled to obtain the monitoring link of the media data; if the scenario type of the media access scenario is a normal type, the configuration parameters and access parameters of the media data can be assembled to obtain the monitoring link of the media data. Different media data require specific parameters for different devices under certain specific access scenarios. For example, in the access scenario of a certain media data on a certain social platform, the parameters for iOS devices will have an additional height parameter. That is to say, for this media data, when accessed on this social platform using an iOS device, its monitoring link will have an additional height parameter.

[0112] For media access scenarios with different specific device parameters on iOS or Android devices, this application can pre-configure and define these media access scenarios as specified access scenarios, and set their scenario type as a specified type (e.g., a type that requires specific device parameters). In this way, for any media data, when a monitoring link needs to be built for it, the media access scenario of this media data can be obtained first. If this media access scenario belongs to the specified access scenario, its scenario type will also belong to the pre-configured specified type. Then, the specified device parameters of the media data under this specified access scenario can be directly obtained. Subsequently, the above configuration parameters, access parameters, and the specified device parameters can be assembled (the assembly method is, for example, splicing) to obtain the monitoring link of this media data. If the media access scenario of the media data does not belong to the specified access scenario, but belongs to a normal access scenario, then the scenario type of this media data can be considered to be a normal type. In this case, it is not necessary to introduce the specified device parameters to build the monitoring link. The above configuration parameters and access parameters can be directly assembled to obtain the monitoring link of the media data.

[0113] Step S303: Obtain media content from media data by accessing the monitoring link.

[0114] In this application, after establishing a monitoring link for the media data, the computer device can access this monitoring link to obtain the media content actually delivered (i.e., the delivered content). Since the computer device simulates a real user accessing this monitoring link, to improve the authenticity of the access, the computer device can simulate accessing this monitoring link through a real device (such as a terminal device) according to the targeted delivery region and delivery time information of the media data. For example, the computer device can set the access region information to the region indicated by the targeted delivery region, and access the monitoring link through a real device within the time period indicated by the delivery time. In this way, the computer device can obtain the actual delivered media content within the region indicated by the targeted delivery region and the time period indicated by the delivery time.

[0115] Step S304: Based on the media content, perform risk identification processing on the media data to obtain the risk identification results of the media data.

[0116] In this application, after acquiring the media content actually delivered, the computer device can perform risk identification processing on the media data based on this media content, ultimately obtaining the risk identification result of the media data. For example, this application can identify the risk type of the media data based on its media content. This risk type can include specific risk types (such as false advertising, fabricated materials, exaggerated advertising, etc.) and risk-free types. Based on the risk type of the media data, it can be determined whether the risk identification result of the media data is a risky result or a risk-free result. In other words, the risk identification result of media data can include the risk type to which the media data belongs. The specific implementation process of risk identification processing of media data based on media content to obtain the risk identification result of media data can be as follows: First, pre-configured abnormal elements can be obtained. These abnormal elements can refer to elements of different modal types of pre-configured abnormalities. For example, abnormal elements can refer to text data, audio data, image data, etc. that violate the criteria. The media content can be detected according to the abnormal elements to determine whether the media content includes these abnormal elements. If the media content includes abnormal elements, the risk type to which the abnormal elements belong can be determined as the risk type to which the media data belongs. For example, if the media content includes text data that violates the criteria, and the risk type to which this text data belongs is exaggerated propaganda, then the risk type to which this media data belongs can be considered exaggerated propaganda. Correspondingly, if the media content does not include abnormal elements, the risk type to which the media data belongs can be determined as risk-free.

[0117] Of course, there are other ways to identify the risk type of media data based on media content than those described above. The methods described above are just examples. For example, another way to identify the risk type of media data based on media content is to obtain the text content and media product content in the media content; if the product described in the text content does not match the media product presented in the media content, then the risk type of the media data can be determined to be the false material type.

[0118] As can be seen from the above, the entire process of risk identification and processing of media data during the campaign can be carried out efficiently and accurately without human intervention, thanks to the automatically constructed monitoring links.

[0119] This application provides a scheme for risk identification of media data during delivery, which can accurately identify whether the media data during delivery is risky, thereby enhancing the quality of media data review. Specifically, for media data during delivery, this application can obtain the configuration parameters generated during delivery configuration and the access parameters generated after being accessed during delivery. Based on these configuration parameters and access parameters, this application can construct a monitoring link for this media data to monitor its media content. Since this monitoring link monitors the media content of the media data, it can reflect the actual content of the landing page of the media data. Thus, by accessing this monitoring link, the media content of this media data (the actual delivery content during delivery) can be accurately obtained. Based on this actual media content, risk identification processing can be performed on this media data to obtain the risk identification result of this media data. As can be seen, this application constructs a link for monitoring the content delivered by media data, which can be accessed automatically to obtain the actual content delivered by the media data. Then, risk identification processing can be performed on the obtained actual content delivered to determine the risk identification result of the media data. In this way, not only can the media data in delivery be reviewed to improve the quality of media data review, but also no manual processing is required throughout the process, which can effectively ensure the efficiency of reviewing media data in delivery.

[0120] Furthermore, in the process of risk identification processing of media data based on media content, in order to better improve the accuracy of the risk type identification results, this application can obtain the multimodal features of the media data, and then perform risk identification processing based on the multimodal features of the media data. Compared with the method of risk identification based on the surface content of the media data, the method of risk identification processing based on deep multimodal features can analyze the media data from different dimensions and modalities, and the obtained risk type identification results can be more accurate. For ease of understanding, the process of risk identification processing of media data based on media content provided in the embodiments of this application will be described in detail below with reference to the accompanying drawings.

[0121] Please see Figure 4 , Figure 4 This application provides a schematic diagram of a process for risk identification processing of media data based on media content. This process can refer to a process for risk identification processing of media data based on media content to obtain the risk type of the media data, which can correspond to the above. Figure 3 The corresponding embodiment describes the process of performing risk identification processing on media data based on media content to obtain the risk identification result of the media data. For example... Figure 4As shown, the process may include at least steps S401-S402:

[0122] Step S401: Perform content analysis on the media content to obtain the multimodal features of the media content.

[0123] Specifically, after acquiring the media content, content analysis can be performed on this media content. This content analysis can refer to analyzing the media content under different modal types. Through content analysis of different modal types, the multimodal characteristics of the media content can be obtained. The specific implementation process may include, but is not limited to, the following: First, N (N is a positive integer) pre-configured modal types can be acquired. In this application, modal types can refer to different media types (or media presentation formats). For example, media types can include video, audio, image, and text types, and modal types can also include video, audio, image, and text types. After acquiring various modal types, content acquisition processing can be performed on the media content according to each modal type to obtain the modal acquisition content of the media content under each modal type. For example, this application can perform content acquisition processing on the media content under each modal type. The application can capture video data from media content to obtain video data on the landing page. This video data can be considered as the modal capture content of the media content under the video type. Conversely, by taking a screenshot of a frame from the video data, the modal capture content of the media content under the image type can be obtained. Similarly, the application can capture audio data from media content to obtain audio data on the landing page. This audio data can be considered as the modal capture content of the media content under the audio type. Furthermore, the application can capture text data from media content to obtain all text data on the landing page. This text data can be considered as the modal capture content of the media content under the text type. When collecting text from media content, appropriate character recognition technologies can be used. For example, Optical Character Recognition (OCR) technology can be used to recognize the text data presented in the media content, and the recognized text data can be used as the modal collection content of the media content in the text type. Alternatively, Automatic Speech Recognition (ASR) technology can be used to convert the audio data of the media content into speech data, which can also be used as the modal collection content of the media content in the text type.

[0124] Furthermore, after acquiring the modal acquisition content of media content under various modal types, feature extraction processing can be performed on the modal acquisition content of media content under each modal type to obtain the modal features of media content under each modal type. Here, different or the same artificial intelligence models can be called to extract the modal features of various modal acquisition content. For example, audio feature extraction models (e.g., MFCC, STFT, MFE models, etc.) can be used to extract the modal features of media content under audio type; three-dimensional convolutional neural networks can be used to extract the modal features of media content under video type; and transformer models can be used to extract the modal features of media content under text type. This application does not limit the specific method of feature extraction processing.

[0125] After obtaining the modal features of media content under various modal types, the modal features of media content under N modal types can be fused together to obtain the multimodal features of the media content. One example of this fusion method is splicing.

[0126] Step S402: Based on the multimodal characteristics of media content, perform risk identification processing on media data to obtain the risk type to which the media data belongs.

[0127] Specifically, after obtaining the multimodal features of media content, risk identification processing can be performed on the media data based on these features to obtain the risk type of the media data. The specific implementation process may include, but is not limited to, the following: First, the media configuration features of the media data can be obtained. The media configuration features of the media data refer to some configuration features of the media data before it is delivered, which may include: the descriptive information of the media data, the media type to which the media data belongs (e.g., food, beauty, comedy, fashion, etc.), the object identifier of the media creation object, the media identifier, and other basic configuration features. Further, the multimodal features of the media content can be fused with the media configuration features (the fusion method here is, for example, splicing), thereby obtaining a fused feature, which can be referred to as the fused delivery feature in this application.

[0128] After determining the characteristics of integrated delivery, a risk identification model can be invoked based on these characteristics. This model can then be used to identify risks in the media data, thereby determining the risk type to which the media data belongs. The risk identification model here can be a hybrid expert model (MoE), which may include multiple expert networks. Each expert network corresponds to a candidate risk type (the candidate risk type can be a pre-configured risk type, such as exaggerated advertising, fabricated materials, or text violations). Each expert network is dedicated to analyzing the probability that a particular media data belongs to its corresponding candidate risk type. Therefore, by using multiple expert networks, accurate analysis and identification of different candidate risk types can be achieved under the same input feature system. Specifically, taking the example of M (where M is a positive integer) pre-configured candidate risk types, there will also be M expert networks. Each expert network corresponds to one candidate risk type, meaning the risk identification model will include M expert networks. In this case, the specific implementation process for identifying the risk type of media data through the risk identification model can include, but is not limited to, the following: First, the fusion delivery features can be input into the risk identification model. Through the model parameters of the risk identification model, the received fusion delivery features can be calculated and analyzed to obtain a probability distribution. This probability distribution will include M prediction probabilities, where each prediction probability refers to the probability that the media data, as predicted by the risk identification model, belongs to a candidate risk type. It should be understood that the fusion delivery features will be input into each expert network, and each expert network will calculate and analyze the fusion delivery features based on its own network parameters to obtain the prediction probability of this expert network for the corresponding candidate risk type (i.e., the probability that the risk type of the media data predicted by the expert network is the corresponding candidate risk type). Then, the prediction probabilities of all expert networks for the corresponding candidate risk types are combined and aggregated to obtain a probability distribution.

[0129] Furthermore, this probability distribution can be used to determine the risk type of the media data. For example, the maximum predicted probability can be obtained from the probability distribution. If this maximum predicted probability is greater than a probability threshold (this probability threshold can be determined based on a manually defined risk identification strategy; if the strategy dictates that a candidate risk type can only be identified as a risk type of media data when the predicted probability is greater than 90%, then this probability threshold can be 90%), then the candidate risk type indicated by the maximum predicted probability can be identified as the risk type of the media data. If the maximum predicted probability is less than this probability threshold, then the risk type of the media data can be identified as a risk-free type. In the process of determining the risk type of media data using this probability distribution, a manually defined risk identification strategy can be followed. For example, the manually defined risk identification strategy may require identifying all risk types of media data, not just one, and any candidate risk type whose predicted probability exceeds the probability threshold of 60% can be considered a risk type of media data. After obtaining the probability distribution, various predicted probabilities greater than the probability threshold of 60% can be obtained, and the candidate risk types corresponding to these predicted probabilities will all be identified as the risk types of the media data. In other words, this application can support the combination of multimodal risk identification models and risk identification strategies to more accurately and reasonably determine the risk type of media data.

[0130] In addition to employing a risk identification model to process media data for risk identification, this application can also set different prohibited content for specific time periods. Media data may include such content during normal time periods, but it is not allowed during specific time periods. That is, this application can configure certain time periods as designated time periods (e.g., statutory holidays) and configure corresponding review strategies for these designated time periods (e.g., configuring corresponding prohibited content (such as strong keywords), and the media content is not allowed to include such prohibited content during the designated time periods). After the risk identification model identifies the risk identification results of the media data, this application can re-examine the risk identification results of the media data based on the review strategies corresponding to the designated time periods. This allows determination of whether the media data includes such prohibited content during the designated time periods. Specifically, this application can obtain the media data's delivery time period (i.e., the time period during which the media data is delivered); if this delivery time period falls within a designated time period, the review strategy corresponding to the designated time period can be obtained; the risk identification results of the media data can be detected according to the review strategy corresponding to the designated time period, thus obtaining the risk identification results after the media data detection. Of course, the review strategy corresponding to a specified time period can also be deployed into the risk identification model, which will then perform risk identification processing on the media content of the media data. This way, there is no need to re-examine the risk identification results obtained by the model, and the risk identification results output by the model can be used as the final reliable risk identification results.

[0131] It should be understood that after determining the risk identification result of media data, if the risk identification result indicates that the media data is at risk, then the delivery of the media data can be suspended, and the media creator of the media data can be warned to modify and improve the media data to make it risk-free; if the risk identification result indicates that the media data is not at risk, then the media data can continue to be delivered and accessed and viewed by different users.

[0132] It is worth noting that this application uses a risk identification model to identify the risk type of media data, so the model output is particularly important. To improve the accuracy of the risk identification model's output, this application can use machine learning to train the risk identification model end-to-end. The training process of the risk identification model can include at least the following: First, sample media data can be obtained. This sample media data can refer to any media data used as a training sample for the model. The multimodal features corresponding to this sample media data can be obtained (the method of obtaining multimodal features can be referred to in the above description of obtaining multimodal features of media data; here, the multimodal features corresponding to the sample media data can be referred to as sample multimodal features) and sample media configuration features (sample media configuration features are also the media configuration features of the sample media data; the method of obtaining them can be referred to in the above description of obtaining media configuration features of media data); after obtaining the multimodal features and sample media configuration features corresponding to the sample media data, the sample media... The multimodal features of the sample data and the configuration features of the sample media are fused (e.g., spliced) to obtain the sample fusion delivery features of the sample media data. Then, based on the sample fusion delivery features, the initial risk identification model (the risk identification model that has not yet been trained) can be invoked. The initial risk identification model can be used to perform risk identification processing on the sample media data and obtain the predicted risk type of the sample media data. Furthermore, the risk type label corresponding to the sample media data can be obtained (that is, the actual risk type to which the sample media data belongs, which can be pre-labeled). Based on the risk type label and the predicted risk type of the sample media data, the initial risk identification model can be trained to obtain the risk identification model.

[0133] The specific implementation process of training an initial risk identification model based on the risk type labels and predicted risk types corresponding to sample media data can include, but is not limited to, the following: First, a loss function (e.g., cross-entropy loss function) can be used to calculate the error loss between the risk type labels and predicted risk types corresponding to the sample media data; the initial model parameters of the initial risk identification model can be adjusted and optimized according to the error loss between the risk type labels and predicted risk types to obtain model parameters; further, the initial risk identification model including the model parameters can be determined as the risk identification model.

[0134] This application provides a scheme for risk identification of media data during delivery, which can accurately identify whether the media data during delivery is risky, thereby enhancing the quality of media data review. Specifically, for media data during delivery, this application can obtain the configuration parameters generated during delivery configuration and the access parameters generated after being accessed during delivery. Based on these configuration parameters and access parameters, this application can construct a monitoring link for this media data to monitor its media content. Since this monitoring link monitors the media content of the media data, it can reflect the actual content of the landing page of the media data. Thus, by accessing this monitoring link, the media content of this media data (the actual delivery content during delivery) can be accurately obtained. Based on this actual media content, risk identification processing can be performed on this media data to obtain the risk identification result of this media data. By constructing links to monitor the content being delivered in media data, access to these links can be automated to obtain the actual content being delivered. Then, risk identification processing can be performed on this obtained content to determine the risk assessment results. This not only enables the review of media data during delivery, improving the quality of review, but also eliminates the need for manual processing, effectively ensuring the efficiency of media data review. Furthermore, in the process of risk identification processing based on the obtained actual content, this application extracts modal features of the media content under different modal types, then fuses them to obtain multimodal features, and performs risk identification processing based on these multimodal features. This allows for one-stop monitoring and identification of various risk types. Simultaneously, by combining different risk identification strategies, this application can achieve a combined manual and model-based review system, which can meet the review requirements of intelligent models to a certain extent, while also allowing for the embedding of a manual review engine to improve review efficiency and risk control.

[0135] To better understand the logic of this application for risk identification processing of media data based on media content, please also refer to... Figure 5 , Figure 5 This is a schematic diagram of a logical architecture for risk identification processing of media data based on media content, provided in an embodiment of this application. For example... Figure 5 As shown, this logical architecture can include at least the following components: content acquisition component, feature extraction component, feature fusion component, and hybrid expert model. The functions implemented by each component will be briefly described below:

[0136] Content Acquisition Component: The content acquisition component is used to acquire media content from media data. It mainly acquires media content in different modal types (e.g., video content, image content, text content, audio content, etc.).

[0137] Feature extraction component: The feature extraction component is used to perform feature extraction processing on different modal acquisition content to obtain the modal features of media content under different modal types.

[0138] Feature fusion component: The feature fusion component is used to fuse (e.g., splice) the modal features of media content under different modal types, so as to obtain the multimodal features of media content under multiple modal types; the feature fusion component is also used to fuse this multimodal feature with the media configuration features of media data, and finally obtain the fused delivery features of media data.

[0139] Hybrid expert model: The hybrid expert model uses multiple internal expert networks to calculate and analyze the integrated delivery features, so as to identify different candidate risk types under the same input feature feature system, and finally output the risk type to which the media data belongs.

[0140] In this application embodiment, compared to the traditional review of advertising data for ad placement materials, this application, based on the construction of monitoring links for advertising data, can perform real-time access monitoring of advertising data during placement, thereby achieving front-end risk prevention. The page opened by accessing the monitoring link is the actual placement effect of the advertising data, and the risk type in the placement process can be identified based on the advertising content on the opened page, and such risk can be isolated and blocked. Moreover, in the process of identifying the risk type of advertising data based on advertising content, this application supports the use of a multimodal risk identification model, which can combine the multimodal features of advertising content in the advertising data to achieve one-stop monitoring and identification of risky content. In addition, this application, combined with a strategy engine, supports a review system that combines human and model review, which can meet the review foundation of intelligent models to a certain extent, while also allowing human strategy engine to be embedded to improve review efficiency and risk control.

[0141] Furthermore, for ease of understanding the overall logic of the solution provided in this application, please also refer to... Figure 6 , Figure 6 This is a schematic diagram of the logical architecture of a solution provided in an embodiment of this application. For example... Figure 6 As shown, this logical architecture can include at least the following components: a deployment platform, a simulated access monitoring component, a multimodal auditing model, and a strategy engine. The functions implemented by each component will be briefly described below:

[0142] The ad delivery platform receives ad data from advertisers to obtain the ad creatives they want to use. Advertisers can create new ad delivery requests for specific ad data within the platform.

[0143] Simulated Access Monitoring Component: This component creates a monitoring link for an ongoing ad campaign by an advertiser, enabling real-time monitoring of the ad's actual delivery content. Specifically, the component includes a link building subcomponent, a device pool, and a location simulation subcomponent. The link building subcomponent retrieves the ad's configuration parameters (such as the campaign link, advertiser ID, ad ID, targeted region, and campaign time) and access parameters (which can be understood as macro substitution parameters, such as access time, attribution data, placement location, and callback information). Based on these parameters, a monitoring link is built to monitor the ad's actual delivery content and performance in real time.

[0144] Location Simulation Subcomponent and Device Pool: After identifying the monitoring link for advertising data, computer devices can invoke the location simulation subcomponent and device pool based on the targeted delivery region and time of the advertising data. Through real devices in the device pool, the system simulates real users accessing the monitoring link within that delivery region and time period, thereby obtaining the actual advertising content delivered. Based on this advertising content, content collection can be performed to obtain modal content across different modal types.

[0145] Multimodal review model: The multimodal review model can be understood as the risk identification model mentioned above. Through the multimodal review model, the content collected in different modalities can be adjusted and extracted to obtain the modal features of advertising data under different modal types. Then, the multimodal review model can fuse the modal features of advertising data under different modal types to obtain the multimodal features of advertising data. Based on these multimodal features, the risk type of advertising data can be identified by combining the strategy engine (which can be understood as the risk identification strategy mentioned above, and may include a variety of custom risk control strategies) and output the risk identification results of advertising data.

[0146] In this application embodiment, compared to the traditional review of advertising data for ad placement materials, this application, based on the construction of monitoring links for advertising data, can perform real-time access monitoring of advertising data during placement, thereby achieving front-end risk prevention. The page opened by accessing the monitoring link is the actual placement effect of the advertising data, and the risk type in the placement process can be identified based on the advertising content on the opened page, and such risk can be isolated and blocked. Moreover, in the process of identifying the risk type of advertising data based on advertising content, this application supports the use of a multimodal risk identification model, which can combine the multimodal features of advertising content in the advertising data to achieve one-stop monitoring and identification of risky content. In addition, this application, combined with a strategy engine, supports a review system that combines human and model review, which can meet the review foundation of intelligent models to a certain extent, while also allowing human strategy engine to be embedded to improve review efficiency and risk control.

[0147] Further, please see Figure 7 , Figure 7 This is a schematic diagram of the structure of a media processing device provided in an embodiment of this application. The media processing device can be a computer program (including program code) running on a computer device; for example, the media processing device is an application software. The media processing device can be used to execute... Figure 3 The method shown. (As illustrated) Figure 7 As shown, the media processing device 1 may include: a parameter acquisition module 11, a link construction module 12, a link access module 13, and a risk identification module 14.

[0148] The parameter acquisition module 11 is used to acquire the configuration parameters and access parameters of the media data. The configuration parameters are the parameters generated when the media data is configured for delivery, and the access parameters are the parameters generated after the media data is accessed during the delivery process.

[0149] Link building module 12 is used to build monitoring links for media data based on configuration parameters and access parameters;

[0150] Link access module 13 is used to obtain media content by accessing monitoring links;

[0151] Risk identification module 14 is used to perform risk identification processing on media data based on media content, and obtain the risk identification results of media data.

[0152] The specific implementation methods of the parameter acquisition module 11, link construction module 12, link access module 13, and risk identification module 14 can be found in the above description. Figure 3 The descriptions of steps S301-S304 in the corresponding embodiments will not be repeated here.

[0153] In one embodiment, the link building module 12 constructs monitoring links for media data based on configuration parameters and access parameters, including:

[0154] Media access scenarios for acquiring media data;

[0155] If the scenario type of the media access scenario is a specified type, then the specified device parameters of the media data in the media access scenario are obtained, and the configuration parameters, access parameters and the specified device parameters of the media data in the media access scenario are assembled to obtain the monitoring link of the media data.

[0156] If the media access scenario is of the normal type, the configuration parameters and access parameters of the media data are assembled to obtain the monitoring link of the media data.

[0157] In one embodiment, the risk identification result of the media data includes the risk type to which the media data belongs;

[0158] The risk identification module 14 performs risk identification processing on media data based on media content to obtain the risk identification results of the media data. Specific implementation methods include:

[0159] Retrieve pre-configured exception elements;

[0160] Detect media content based on abnormal elements;

[0161] If the media content contains anomalous elements, the risk type to which the anomalous elements belong is determined as the risk type to which the media data belongs.

[0162] If the media content does not contain any abnormal elements, then the risk type of the media data is determined to be risk-free.

[0163] In one embodiment, the risk identification result of the media data includes the risk type to which the media data belongs;

[0164] The risk identification module 14 performs risk identification processing on media data based on media content to obtain the risk identification results of the media data. Specific implementation methods include:

[0165] Content analysis is performed on media content to obtain its multimodal characteristics;

[0166] Based on the multimodal characteristics of media content, risk identification processing is performed on media data to obtain the risk type to which the media data belongs.

[0167] In one embodiment, the risk identification module 14 performs content analysis on the media content to obtain a specific implementation of the multimodal features of the media content, including:

[0168] Get the pre-configured N modal types; N is a positive integer;

[0169] The media content is captured and processed according to each modality type to obtain the modal capture content of the media content under each modality type;

[0170] Feature extraction processing is performed on the modal acquisition content of media content under each modal type to obtain the modal features of media content under each modal type;

[0171] The modal features of media content under N modal types are fused to obtain the multimodal features of the media content.

[0172] In one embodiment, the risk identification module 14 performs risk identification processing on media data based on the multimodal characteristics of media content to obtain the specific implementation of the risk type to which the media data belongs, including:

[0173] Media configuration characteristics for acquiring media data;

[0174] By fusing the multimodal features of media content with the features of media configuration, we obtain the integrated delivery features of media data;

[0175] Based on the characteristics of integrated delivery, a risk identification model is invoked. The risk identification model is used to process media data for risk identification and to obtain the risk type of the media data.

[0176] In one embodiment, the risk identification module 14 performs risk identification processing on media data through a risk identification model to obtain the specific implementation method of the risk type to which the media data belongs, including:

[0177] The probability distribution is obtained by calculating and analyzing the characteristics of integrated delivery through the model parameters of the risk identification model. The probability distribution includes M predicted probabilities, each of which is the probability that the media data predicted by the risk identification model belongs to a candidate risk type. M is a positive integer.

[0178] Obtain the maximum predicted probability from the probability distribution;

[0179] If the maximum predicted probability is greater than the probability threshold, then the candidate risk type indicated by the maximum predicted probability is determined as the risk type to which the media data belongs.

[0180] If the maximum predicted probability is less than the probability threshold, the risk type of the media data will be determined as the risk-free type.

[0181] In one embodiment, the risk identification model includes M expert networks, each expert network corresponding to a candidate risk type;

[0182] Risk identification module 14 calculates and analyzes the fusion delivery features using the model parameters of the risk identification model to obtain the specific implementation method of the probability distribution, including:

[0183] The fusion delivery features are input into each expert network, and each expert network calculates and analyzes the fusion delivery features to obtain the predicted probability of each expert network for the corresponding candidate risk type.

[0184] The probability distribution is obtained by combining the predicted probabilities of M expert networks for their respective candidate risk types.

[0185] In one embodiment, the media processing device 1 further includes a model training module 15.

[0186] Model training module 15 is used to acquire sample media data, as well as the sample multimodal features and sample media configuration features corresponding to the sample media data;

[0187] Model training module 15 is also used to fuse the multimodal features of the sample media data with the configuration features of the sample media data to obtain the sample fusion delivery features of the sample media data.

[0188] Model training module 15 is also used to call the initial risk identification model based on the sample fusion delivery features, and to perform risk identification processing on the sample media data through the initial risk identification model to obtain the predicted risk type of the sample media data.

[0189] The model training module 15 is also used to obtain the risk type labels corresponding to the sample media data, and to train the initial risk identification model based on the risk type labels corresponding to the sample media data and the predicted risk type, so as to obtain the risk identification model.

[0190] For the specific implementation of the model training module 15, please refer to the above. Figure 4 The relevant descriptions in step S402 of the corresponding embodiments will not be repeated here.

[0191] In one embodiment, the model training module 15 trains the initial risk identification model based on the risk type labels and predicted risk types corresponding to the sample media data, thereby obtaining a specific implementation of the risk identification model, including:

[0192] Calculate the error loss between the risk type label corresponding to the sample media data and the predicted risk type;

[0193] The initial model parameters of the initial risk identification model are adjusted and optimized according to the error loss between the risk type label and the predicted risk type to obtain the model parameters;

[0194] The initial risk identification model, which includes model parameters, is determined as the risk identification model.

[0195] In one embodiment, after the risk identification module 14 performs risk identification processing on the media data based on the media content and obtains the risk identification result of the media data, the risk identification module 14 is further specifically used for:

[0196] The time period for acquiring media data;

[0197] If the delivery time falls within a specified time period, then obtain the review policy corresponding to that specified time period;

[0198] The risk identification results of media data are detected according to the review strategy corresponding to the specified time period, and the risk identification results after media data detection are obtained.

[0199] This application provides a scheme for risk identification of media data during delivery, which can accurately identify whether the media data during delivery is risky, thereby enhancing the quality of media data review. Specifically, for media data during delivery, this application can obtain the configuration parameters generated during delivery configuration and the access parameters generated after being accessed during delivery. Based on these configuration parameters and access parameters, this application can construct a monitoring link for this media data to monitor its media content. Since this monitoring link monitors the media content of the media data, it can reflect the actual content of the landing page of the media data. Thus, by accessing this monitoring link, the media content of this media data (the actual delivery content during delivery) can be accurately obtained. Based on this actual media content, risk identification processing can be performed on this media data to obtain the risk identification result of this media data. By constructing links to monitor the content being delivered in media data, access to these links can be automated to obtain the actual content being delivered. Then, risk identification processing can be performed on this obtained content to determine the risk assessment results. This not only enables the review of media data during delivery, improving the quality of review, but also eliminates the need for manual processing, effectively ensuring the efficiency of media data review. Furthermore, in the process of risk identification processing based on the obtained actual content, this application extracts modal features of the media content under different modal types, then fuses them to obtain multimodal features, and performs risk identification processing based on these multimodal features. This allows for one-stop monitoring and identification of various risk types. Simultaneously, by combining different risk identification strategies, this application can achieve a combined manual and model-based review system, which can meet the review requirements of intelligent models to a certain extent, while also allowing for the embedding of a manual review engine to improve review efficiency and risk control.

[0200] Further, please see Figure 8 , Figure 8 This is a schematic diagram of the structure of a computer device provided in an embodiment of this application. Figure 8 As shown, the aforementioned computer device 8000 may include: a processor 8001, a network interface 8004, and a memory 8005. Furthermore, the computer device 8000 also includes: a user interface 8003, and at least one communication bus 8002. The communication bus 8002 is used to enable communication between these components. The user interface 8003 may include a display screen and a keyboard; optionally, the user interface 8003 may also include a standard wired interface or a wireless interface. The network interface 8004 may optionally include a standard wired interface or a wireless interface (such as a Wi-Fi interface). The memory 8005 may be high-speed RAM or non-volatile memory, such as at least one disk storage device. Optionally, the memory 8005 may also be at least one storage device located remotely from the aforementioned processor 8001. Figure 8 As shown, the memory 8005, which is a computer-readable storage medium, may include an operating system, a network communication module, a user interface module, and a device control application program.

[0201] exist Figure 8 In the computer device 8000 shown, the network interface 8004 provides network communication functionality; the user interface 8003 is mainly used to provide an input interface for the user; and the processor 8001 can be used to call the device control application program stored in the memory 8005 to achieve:

[0202] Obtain the configuration parameters and access parameters of media data. The configuration parameters are the parameters generated when the media data is configured for delivery, and the access parameters are the parameters generated after the media data is accessed during the delivery process.

[0203] Based on the configuration and access parameters, a monitoring link is built for the media data;

[0204] Media content that obtains media data by accessing monitoring links;

[0205] Based on media content, risk identification processing is performed on media data to obtain risk identification results for media data.

[0206] In one embodiment, when the processor 8001 constructs a monitoring link for media data based on configuration parameters and access parameters, it specifically performs the following steps:

[0207] Media access scenarios for acquiring media data;

[0208] If the scenario type of the media access scenario is a specified type, then the specified device parameters of the media data in the media access scenario are obtained, and the configuration parameters, access parameters and the specified device parameters of the media data in the media access scenario are assembled to obtain the monitoring link of the media data.

[0209] If the media access scenario is of the normal type, the configuration parameters and access parameters of the media data are assembled to obtain the monitoring link of the media data.

[0210] In one embodiment, the risk identification result of the media data includes the risk type to which the media data belongs;

[0211] When processor 8001 performs risk identification processing on media data based on media content and obtains the risk identification result of the media data, it specifically executes the following steps:

[0212] Retrieve pre-configured exception elements;

[0213] Detect media content based on abnormal elements;

[0214] If the media content contains anomalous elements, the risk type to which the anomalous elements belong is determined as the risk type to which the media data belongs.

[0215] If the media content does not contain any abnormal elements, then the risk type of the media data is determined to be risk-free.

[0216] In one embodiment, the risk identification result of the media data includes the risk type to which the media data belongs;

[0217] When processor 8001 performs risk identification processing on media data based on media content and obtains the risk identification result of the media data, it specifically executes the following steps:

[0218] Content analysis is performed on media content to obtain its multimodal characteristics;

[0219] Based on the multimodal characteristics of media content, risk identification processing is performed on media data to obtain the risk type to which the media data belongs.

[0220] In one embodiment, when the processor 8001 performs content analysis on the media content to obtain the multimodal features of the media content, it specifically executes the following steps:

[0221] Get the pre-configured N modal types; N is a positive integer;

[0222] The media content is captured and processed according to each modality type to obtain the modal capture content of the media content under each modality type;

[0223] Feature extraction processing is performed on the modal acquisition content of media content under each modal type to obtain the modal features of media content under each modal type;

[0224] The modal features of media content under N modal types are fused to obtain the multimodal features of the media content.

[0225] In one embodiment, when the processor 8001 performs risk identification processing on media data based on multimodal features of media content to obtain the risk type of the media data, it specifically executes the following steps:

[0226] Media configuration characteristics for acquiring media data;

[0227] By fusing the multimodal features of media content with the features of media configuration, we obtain the integrated delivery features of media data;

[0228] Based on the characteristics of integrated delivery, a risk identification model is invoked. The risk identification model is used to process media data for risk identification and to obtain the risk type of the media data.

[0229] In one embodiment, when the processor 8001 performs risk identification processing on the media data using a risk identification model to determine the risk type of the media data, it specifically executes the following steps:

[0230] The probability distribution is obtained by calculating and analyzing the characteristics of integrated delivery through the model parameters of the risk identification model. The probability distribution includes M predicted probabilities, each of which is the probability that the media data predicted by the risk identification model belongs to a candidate risk type. M is a positive integer.

[0231] Obtain the maximum predicted probability from the probability distribution;

[0232] If the maximum predicted probability is greater than the probability threshold, then the candidate risk type indicated by the maximum predicted probability is determined as the risk type to which the media data belongs.

[0233] If the maximum predicted probability is less than the probability threshold, the risk type of the media data will be determined as the risk-free type.

[0234] In one embodiment, the risk identification model includes M expert networks, each expert network corresponding to a candidate risk type;

[0235] When processor 8001 performs calculations and analyses on the fused deployment features using the model parameters of the risk identification model to obtain the probability distribution, it specifically executes the following steps:

[0236] The fusion delivery features are input into each expert network, and each expert network calculates and analyzes the fusion delivery features to obtain the predicted probability of each expert network for the corresponding candidate risk type.

[0237] The probability distribution is obtained by combining the predicted probabilities of M expert networks for their respective candidate risk types.

[0238] In one embodiment, the processor 8001 is further configured to perform the following steps:

[0239] Acquire sample media data, as well as the corresponding multimodal features and configuration features of the sample media data;

[0240] The multimodal features of the sample media data are fused with the configuration features of the sample media data to obtain the sample fusion delivery features of the sample media data.

[0241] Based on the sample fusion delivery features, the initial risk identification model is invoked, and the sample media data is processed for risk identification through the initial risk identification model to obtain the predicted risk type of the sample media data.

[0242] Obtain the risk type labels corresponding to the sample media data, and train the initial risk identification model based on the risk type labels and predicted risk types to obtain the risk identification model.

[0243] In one embodiment, when the processor 8001 trains the initial risk identification model based on the risk type labels and predicted risk types corresponding to the sample media data to obtain the risk identification model, it specifically performs the following steps:

[0244] Calculate the error loss between the risk type label corresponding to the sample media data and the predicted risk type;

[0245] The initial model parameters of the initial risk identification model are adjusted and optimized according to the error loss between the risk type label and the predicted risk type to obtain the model parameters;

[0246] The initial risk identification model, which includes model parameters, is determined as the risk identification model.

[0247] In one embodiment, after the processor 8001 performs risk identification processing on the media data based on the media content and obtains the risk identification result of the media data, the processor 8001 is further specifically used to perform the following steps:

[0248] The time period for acquiring media data;

[0249] If the delivery time falls within a specified time period, then obtain the review policy corresponding to that specified time period;

[0250] The risk identification results of media data are detected according to the review strategy corresponding to the specified time period, and the risk identification results after media data detection are obtained.

[0251] It should be understood that the computer device 8000 described in the embodiments of this application can execute the foregoing text. Figures 3 to 4 The description of the media processing method in the corresponding embodiments can also be performed as described above. Figure 7 The description of the media processing device 1 in the corresponding embodiments will not be repeated here. Furthermore, the beneficial effects of using the same method will also not be repeated here.

[0252] Furthermore, it should be noted that this application embodiment also provides a computer-readable storage medium, which stores a computer program executed by the aforementioned data processing computer device 8000. The computer program includes program instructions, and when the processor executes the program instructions, it can execute the aforementioned... Figures 3 to 4 The description of the media processing method in the corresponding embodiments is already provided and will not be repeated here. Furthermore, the beneficial effects of using the same method will also not be repeated. For technical details not disclosed in the computer-readable storage medium embodiments related to this application, please refer to the description of the method embodiments of this application.

[0253] The aforementioned computer-readable storage medium can be an internal storage unit of the media processing apparatus or computer device provided in any of the foregoing embodiments, such as a hard disk or memory of the computer device. The computer-readable storage medium can also be an external storage device of the computer device, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc., provided on the computer device. Furthermore, the computer-readable storage medium may include both internal and external storage units of the computer device. The computer-readable storage medium is used to store the computer program and other programs and data required by the computer device. The computer-readable storage medium can also be used to temporarily store data that has been output or will be output.

[0254] One aspect of this application provides a computer program product comprising a computer program stored in a computer-readable storage medium. A processor of a computer device reads the computer program from the computer-readable storage medium and executes the computer program, causing the computer device to perform the method provided in one aspect of the embodiments of this application.

[0255] The terms "first," "second," etc., in the specification, claims, and drawings of this application are used to distinguish different objects, not to describe a specific order. Furthermore, the term "comprising," and any variations thereof, are intended to cover a non-exclusive inclusion. For example, a process, method, apparatus, product, or device that includes a series of steps or units is not limited to the listed steps or modules, but may optionally include steps or modules not listed, or may optionally include other step units inherent to these processes, methods, apparatuses, products, or devices.

[0256] In this application embodiment, the terms "module" or "unit" refer to a computer program or part of a computer program that has a predetermined function and works with other related parts to achieve a predetermined goal, and can be implemented wholly or partially using software, hardware (such as processing circuitry or memory), or a combination thereof. Similarly, a processor (or multiple processors or memory) can be used to implement one or more modules or units. Furthermore, each module or unit can be part of an overall module or unit that includes the functions of that module or unit.

[0257] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this application.

[0258] The methods and related apparatuses provided in this application are described with reference to the method flowcharts and / or structural diagrams provided in this application. Specifically, each block of the method flowchart and / or structural diagram, as well as combinations of blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing device to create a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing device, generate instructions for implementing the process. Figure 1 A schematic diagram of one or more processes and / or structures. Figure 1 The computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to operate in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1A schematic diagram of one or more processes and / or structures. Figure 1 The functions specified in one or more boxes. These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable apparatus for implementing the process. Figure 1 A process or multiple processes and / or structures illustrate the steps of the functions specified in one or more boxes.

[0259] The above-disclosed embodiments are merely preferred embodiments of this application and should not be construed as limiting the scope of this application. Therefore, any equivalent variations made in accordance with the claims of this application shall still fall within the scope of this application.

Claims

1. A media processing method, characterized in that, The method includes: The configuration parameters and access parameters of the media data are obtained. The configuration parameters are generated when the media data is configured for delivery, and the access parameters are generated after the media data is accessed during the delivery process. Based on the configuration parameters and the access parameters, a monitoring link is constructed for the media data; The media content of the media data is obtained by accessing the monitoring link; Based on the media content, the media data is subjected to risk identification processing to obtain the risk identification result of the media data.

2. The method according to claim 1, characterized in that, The step of constructing a monitoring link for the media data based on the configuration parameters and the access parameters includes: Media access scenarios for acquiring the aforementioned media data; If the scenario type of the media access scenario belongs to a specified type, then the specified device parameters of the media data under the media access scenario are obtained, and the configuration parameters, access parameters and the specified device parameters of the media data under the media access scenario are assembled to obtain the monitoring link of the media data. If the media access scenario is of the normal type, the configuration parameters and access parameters of the media data are assembled to obtain the monitoring link of the media data.

3. The method according to claim 1, characterized in that, The risk identification results of the media data include the risk type to which the media data belongs; The process of performing risk identification processing on the media data based on the media content to obtain the risk identification result of the media data includes: Retrieve pre-configured exception elements; The media content is detected based on the aforementioned abnormal elements; If the media content includes the abnormal element, then the risk type to which the abnormal element belongs is determined as the risk type to which the media data belongs; If the media content does not contain the abnormal element, then the risk type of the media data is determined to be risk-free.

4. The method according to claim 1, characterized in that, The risk identification results of the media data include the risk type to which the media data belongs; The process of performing risk identification processing on the media data based on the media content to obtain the risk identification result of the media data includes: Content analysis is performed on the media content to obtain its multimodal features; Based on the multimodal characteristics of the media content, the media data is subjected to risk identification processing to obtain the risk type to which the media data belongs.

5. The method according to claim 4, characterized in that, The content analysis of the media content to obtain its multimodal features includes: Get the pre-configured N modal types; N is a positive integer; The media content is processed by acquiring content according to each modal type to obtain the modal acquisition content of the media content under each modal type; Feature extraction processing is performed on the modal acquisition content of the media content under each modal type to obtain the modal features of the media content under each modal type; The modal features of the media content under the N modal types are fused to obtain the multimodal features of the media content.

6. The method according to claim 4, characterized in that, The process of performing risk identification processing on the media data based on the multimodal features of the media content to obtain the risk type of the media data includes: Obtain the media configuration characteristics of the media data; The multimodal features of the media content are fused with the media configuration features to obtain the fused delivery features of the media data; Based on the integrated delivery characteristics, a risk identification model is invoked, and the media data is processed by the risk identification model to obtain the risk type to which the media data belongs.

7. The method according to claim 6, characterized in that, The step of performing risk identification processing on the media data through the risk identification model to obtain the risk type of the media data includes: The risk identification model is used to calculate and analyze the integrated delivery features to obtain a probability distribution; the probability distribution includes M predicted probabilities, where each predicted probability is the probability that the media data belongs to a candidate risk type as predicted by the risk identification model; M is a positive integer; Obtain the maximum predicted probability from the probability distribution; If the maximum predicted probability is greater than the probability threshold, then the candidate risk type indicated by the maximum predicted probability is determined as the risk type to which the media data belongs; If the maximum predicted probability is less than the probability threshold, then the risk type of the media data is determined to be a risk-free type.

8. The method according to claim 7, characterized in that, The risk identification model includes M expert networks, each expert network corresponding to a candidate risk type; The step of calculating and analyzing the fused delivery features using the model parameters of the risk identification model to obtain a probability distribution includes: The fusion delivery features are input into each of the expert networks, and each expert network calculates and analyzes the fusion delivery features to obtain the predicted probability of each expert network for the corresponding candidate risk type. The predicted probabilities of the M expert networks for their respective candidate risk types are combined to obtain a probability distribution.

9. The method according to claim 6, characterized in that, The method further includes: Acquire sample media data, as well as the sample multimodal features and sample media configuration features corresponding to the sample media data; The multimodal features of the sample media data and the configuration features of the sample media data are fused together to obtain the sample fusion delivery features of the sample media data. Based on the sample fusion and delivery characteristics, the initial risk identification model is invoked, and the sample media data is processed by the initial risk identification model to obtain the predicted risk type of the sample media data. Obtain the risk type label corresponding to the sample media data, and train the initial risk identification model based on the risk type label corresponding to the sample media data and the predicted risk type to obtain the risk identification model.

10. The method according to claim 9, characterized in that, The step of training the initial risk identification model based on the risk type labels and predicted risk types corresponding to the sample media data to obtain the risk identification model includes: Calculate the error loss between the risk type label corresponding to the sample media data and the predicted risk type; The initial model parameters of the initial risk identification model are adjusted and optimized according to the error loss between the risk type label and the predicted risk type to obtain the model parameters; The initial risk identification model, which includes the model parameters, is determined as the risk identification model.

11. The method according to claim 1, characterized in that, After performing risk identification processing on the media data based on the media content to obtain the risk identification result of the media data, the method further includes: Obtain the time period for the media data to be distributed; If the delivery time period falls within a specified time period, then obtain the review policy corresponding to the specified time period; The risk identification results of the media data are detected according to the review strategy corresponding to the specified time period, and the risk identification results of the media data after detection are obtained.

12. A media processing device, characterized in that, The device includes: The parameter acquisition module is used to acquire the configuration parameters and access parameters of the media data. The configuration parameters are the parameters generated when the media data is configured for delivery, and the access parameters are the parameters generated after the media data is accessed during the delivery process. A link building module is used to build monitoring links for the media data based on the configuration parameters and the access parameters; The link access module is used to obtain the media content of the media data by accessing the monitoring link; The risk identification module is used to perform risk identification processing on the media data based on the media content, and obtain the risk identification result of the media data.

13. A computer device, characterized in that, include: Processor, memory, and network interface; The processor is connected to the memory and the network interface, wherein the network interface is used to provide network communication functions, the memory is used to store computer programs, and the processor is used to call the computer programs to cause the computer device to execute the method according to any one of claims 1-11.

14. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program adapted to be loaded by a processor and to execute the method according to any one of claims 1-11.

15. A computer program product, characterized in that, The computer program product includes a computer program stored in a computer-readable storage medium, the computer program being adapted to be read and executed by a processor to cause a computer device having the processor to perform the method of any one of claims 1-11.

Images