Artificial intelligence-based internet advertisement active defense method and system

Abstract

The application discloses an active defense method and system for internet advertisement based on artificial intelligence, relates to the technical field of artificial intelligence, and comprises the following steps: obtaining an original advertisement material set corresponding to an internet advertisement putting request, performing multi-modal information extraction, correlation arrangement and structured coding processing on the multi-type advertisement contents in the original advertisement material set, and generating a structured advertisement material data packet; performing correlation processing on the visual content features, text semantic relationships, audio semantic features and page element correlation features in the structured advertisement material data packet, constructing an advertisement content time sequence correlation relationship, and determining an advertisement risk state result according to the advertisement content time sequence correlation relationship. The application realizes active defense processing of risk analysis, risk positioning, risk adjustment and safe putting of internet advertisement contents.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to a proactive defense method and system for internet advertising based on artificial intelligence. Background Technology

[0002] With the development of internet advertising technology, internet advertising targeting minors is gradually showing a trend towards multimedia, intelligence, and precise targeting. Existing internet advertising supervision technologies for minors typically employ methods such as keyword detection, image recognition, content classification models, and manual review to conduct security reviews of advertising texts, images, videos, and other content in order to identify potentially misleading, deceptive, or inappropriate information for minors. With the application of artificial intelligence technology, some advertising review systems have begun to combine natural language processing, computer vision, and other technologies to comprehensively analyze multiple types of information in advertising materials, thereby improving the efficiency of advertising content security detection.

[0003] Existing internet advertising review technologies primarily detect single risk elements, making it difficult to analyze the formation process of advertising risks by combining the dynamic changes between visual content features, text semantic relationships, audio semantic features, and page element association features. Furthermore, they lack proactive defense methods that can locate risk factors based on risk status results and adjust and verify risky content. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides an artificial intelligence-based proactive defense method for internet advertising, which solves the problems of insufficient analysis of the advertising risk formation process and insufficient ability to locate and adjust risk factors.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] In a first aspect, the present invention provides an artificial intelligence-based proactive defense method for internet advertising, which includes: obtaining a set of original advertising materials corresponding to an internet advertising delivery request; performing multimodal information extraction, association and organization and structured encoding on the multi-type advertising content in the original advertising material set; and generating a structured advertising material data package.

[0008] The visual content features, text semantic relationships, audio semantic features, and page element association features in the structured advertising creative data package are correlated to construct the temporal association relationship of advertising content, and the advertising risk status result is determined based on the temporal association relationship of advertising content.

[0009] Based on the advertising risk status results and the temporal correlation of advertising content, risk correlation positioning is performed on the advertising content features in the structured advertising material data package to generate advertising risk factor positioning results;

[0010] Based on the results of the advertising risk factor positioning, identify the risk factors in the advertising content, perform adjustment operations on the advertising content corresponding to the risk factors, and generate adjusted advertising materials;

[0011] Perform security verification on the adjusted advertising creatives, generate a verified secure advertising version, and execute advertising delivery control operations based on the advertising risk status results and the verified secure advertising version.

[0012] As a preferred embodiment of the AI-based proactive defense method for internet advertising described in this invention, the step of obtaining the set of original advertising materials corresponding to the internet advertising delivery request includes:

[0013] The original set of advertising creatives includes video advertising creatives, image advertising creatives, text advertising creatives, audio advertising creatives, and advertising page creatives.

[0014] As a preferred embodiment of the AI-based proactive defense method for internet advertising described in this invention, the step of performing multimodal information extraction, association organization, and structured encoding on multi-type advertising content in the original advertising material set to generate a structured advertising material data package includes:

[0015] Multimodal information is extracted from various types of advertising content in the original set of advertising materials. Advertising content information is obtained through image recognition, text parsing, speech conversion and page parsing respectively, and a multimodal advertising content set is generated. The multimodal information includes image content information, text content information, speech content information and page content information.

[0016] The system performs association matching on different types of ad content in the multimodal ad content set, organizes the associations based on the correspondence between different types of ad content, and generates an associated ad content set.

[0017] The ad content in the associated ad content set is encoded in a unified format, and the association relationship is written into the encoded information to generate a structured ad creative data package.

[0018] As a preferred embodiment of the AI-based proactive defense method for internet advertising described in this invention, the step of associating visual content features, text semantic relationships, audio semantic features, and page element association features in the structured advertising material data package to construct the temporal association relationship of advertising content includes:

[0019] The visual content features, text semantic relationships, audio semantic features, and page element association features in the structured advertising material data package are synchronized according to the advertising playback time. The video image expression, text expression, audio expression, and page display expression at the same time position are combined to generate advertising time-related content.

[0020] Perform state change analysis on consecutive time positions of content in the time-related content of advertisements, and generate the state change relationship of advertisement content based on the changes in the display object, display status and display behavior of the advertisement in the preceding and following time positions;

[0021] Based on the direction of state change in the state change relationship of the advertising content, the continuously changing time-related content of the advertising content is connected to generate the state transition relationship of the advertising content.

[0022] Based on the state transition order in the ad content state transition relationship and the ad playback time relationship, the ad content state transition relationship is arranged to generate the ad content evolution path;

[0023] The temporal relationship of advertising content is generated by encoding the direction of state changes, the order of state transitions, and the relationship of advertising playback time in the evolution path of advertising content.

[0024] As a preferred embodiment of the AI-based proactive defense method for internet advertising described in this invention, the step of determining the advertising risk status result based on the temporal correlation of advertising content includes:

[0025] The state elements of continuous ad content states in the temporal relationship of ad content are decomposed, and the expression object, expression action and expression result in each ad content state are separated to generate ad content state description information;

[0026] Based on the changing relationships between the expressive object, expressive action, and expressive result in the state description information of the advertising content, calculate the state transition difference between adjacent advertising content states and generate advertising content state transition difference information.

[0027] Based on the differences in the state transition of advertising content, the direction of transition between consecutive advertising content states is determined. The differences in the state transition of advertising content that change from product information expression to user influence expression are marked as risk state transition information, and risk state transition information is generated.

[0028] Based on the distribution position of risk state transition information in the temporal correlation of the advertising content, the continuous relationship between multiple risk state transition information is connected to generate a risk state evolution chain.

[0029] Based on the number of risk state transition information in the risk state evolution chain, the direction of risk state transition, and the temporal correlation of the advertising content corresponding to the risk state evolution chain, the advertising risk state result is generated.

[0030] As a preferred embodiment of the AI-based proactive defense method for internet advertising described in this invention, the step of performing risk correlation positioning on the advertising content features in the structured advertising material data package based on the advertising risk status results and the temporal correlation of advertising content to generate advertising risk factor positioning results includes:

[0031] The process of advertising content changes in the temporal relationship of advertising content is reversed to locate the time position and content change node that leads to the change of advertising risk status, and generate risk association position;

[0032] Based on the risk-related location, the visual content features, text semantic relationships, audio semantic features, and page element association features in the structured advertising material data package are matched and associated. The visual content features, text semantic relationships, audio semantic features, and page element association features located at the risk-related location are associated with the advertising risk status change process to generate risk-related feature relationships.

[0033] The risk-affecting content is located from visual content features, text semantic relationships, audio semantic features, and page element association features. The advertising content features, risk formation locations, and risk status change relationships corresponding to the risk-affecting content are then linked and organized to generate advertising risk factor location results.

[0034] As a preferred embodiment of the AI-based proactive defense method for internet advertising described in this invention, the step of determining risk factors in the advertising content based on the advertising risk factor positioning results, performing adjustment operations on the advertising content corresponding to the risk factors, and generating adjusted advertising materials includes:

[0035] Based on the location of risk factors in the advertising risk factor positioning results, the visual content features, text semantic relationships, audio semantic features, and page element association features in the risk expression factors are broken down according to the advertising expression process. The changes in the expression state before and after the risk expression factors are determined, and the risk expression migration relationship is generated.

[0036] We conduct a change contribution analysis on the visual content features, text semantic relationships, audio semantic features, and page element association features in the risk expression factors to identify the key changes that cause the risk expression migration and generate risk adjustment objects.

[0037] Based on the risk adjustment object, the expression relationship of the risk adjustment object in the risk expression factors is replaced. The risk adjustment object is replaced with safe expression content that maintains the continuity of the expression state before and after, and the visual content features, text semantic relationships, audio semantic features and page element association features are reconnected to generate the adjusted advertising material.

[0038] As a preferred embodiment of the AI-based proactive defense method for internet advertising described in this invention, the step of performing security verification on the adjusted advertising materials and generating a verified secure advertising version includes:

[0039] Based on the changes in the advertising content in the adjusted advertising creative, the changes in the positions of risk factors in the risk factor positioning results of the adjusted advertising creative are compared with those in the risk factor positioning results to determine the advertising content change status corresponding to the risk factor position and generate the risk change status.

[0040] The risk path is traced back to the temporal change process of the advertising content in the adjusted advertising creative to determine whether the adjusted advertising creative still develops along the risk change direction corresponding to the original advertising risk status result, and a safety verification status is generated.

[0041] Based on the security verification status, the security status of the adjusted ad creatives is confirmed. The adjusted ad creatives that do not show any changes in risk status and maintain the continuity of the ad content are identified as the verified secure ad versions.

[0042] As a preferred embodiment of the AI-based proactive defense method for internet advertising described in this invention, the step of performing advertising delivery control operations based on the advertising risk status result and the verified secure advertising version includes:

[0043] Based on the risk status change information in the advertising risk status results and the security verification status in the verified secure advertising version, establish the delivery status relationship between the advertising risk status results and the verified secure advertising version, and generate the advertising delivery status relationship.

[0044] Based on the changes in risk status in the ad delivery status relationship and the security status of the verified secure ad version, the ad delivery method is adjusted to generate ad delivery control results. The delivery control operation is then executed on the verified secure ad version to generate ad delivery execution results.

[0045] Secondly, the present invention provides an artificial intelligence-based proactive defense system for internet advertising, including a processing module that obtains a set of original advertising materials corresponding to an internet advertising delivery request, performs multimodal information extraction, association and organization and structured encoding on the multi-type advertising content in the original advertising material set, and generates a structured advertising material data package.

[0046] The association module performs association processing on the visual content features, text semantic relationships, audio semantic features, and page element association features in the structured advertising material data package, constructs the temporal association relationship of advertising content, and determines the advertising risk status result based on the temporal association relationship of advertising content;

[0047] The positioning module performs risk correlation positioning on the features of advertising content in the structured advertising material data package based on the advertising risk status results and the temporal correlation of advertising content, and generates advertising risk factor positioning results.

[0048] The adjustment module determines the risk factors in the advertising content based on the advertising risk factor positioning results, performs adjustment operations on the advertising content corresponding to the risk factors, and generates adjusted advertising materials.

[0049] The execution module performs security verification on the adjusted advertising creatives, generates a verified secure advertising version, and executes advertising delivery control operations based on the advertising risk status results and the verified secure advertising version.

[0050] The beneficial effects of this invention are as follows: By performing multimodal information processing on various types of advertising materials corresponding to internet advertising placement requests, a structured advertising material data package is constructed, which includes visual content features, text semantic relationships, audio semantic features, and page element association features. Furthermore, a temporal correlation relationship of advertising content is constructed based on the state changes of advertising content over time. The advertising risk status result is determined by analyzing the advertising content state transition process. Based on the advertising risk status result, the risk formation process of the advertising content temporal correlation relationship is traced back to locate the advertising content factors that lead to changes in risk status, generating advertising risk factor location results. Local adjustments are then made to the advertising content corresponding to the risk factors, generating adjusted advertising materials. By verifying the risk path of the adjusted advertising materials, a verified secure advertising version is generated. Finally, advertising placement control operations are executed in conjunction with the advertising risk status result, achieving proactive defense processing for risk analysis, risk location, risk adjustment, and secure placement of internet advertising content. Attached Figure Description

[0051] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. 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.

[0052] Figure 1 This is a flowchart of an AI-based proactive defense method for internet advertising.

[0053] Figure 2This is a schematic diagram of an AI-based proactive defense system for internet advertising. Detailed Implementation

[0054] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0055] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0056] Secondly, the term "one embodiment" or "example" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the invention. The appearance of an embodiment in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that mutually excludes other embodiments.

[0057] Reference Figures 1-2 This is one embodiment of the present invention, which provides an artificial intelligence-based proactive defense method against internet advertising, comprising the following steps:

[0058] S1. Obtain the original set of advertising materials corresponding to the internet advertising placement request, extract multimodal information, associate and organize the various types of advertising content in the original set of advertising materials, and perform structured encoding to generate a structured advertising material data package.

[0059] S1.1 The original set of advertising materials includes video advertising materials, image advertising materials, text advertising materials, audio advertising materials, and advertising page materials.

[0060] Furthermore, based on the ad identification information, material call identification information, and ad version information in the internet ad delivery request, the ad resources to be delivered in the ad storage terminal are indexed to obtain the original ad material set corresponding to the current delivery request. The original ad material set is not a single form of ad file, but a comprehensive ad data set composed of different media formats, including video ad materials, image ad materials, text ad materials, audio ad materials, and ad page materials. Video ad materials are used to record continuous screen information during ad playback, image ad materials are used to record static display content, text ad materials are used to record ad titles, descriptions, and promotional copy information, audio ad materials are used to record ad voice content and sound expression information, and ad page materials are used to record the layout structure, display components, and interactive element information in the ad delivery page.

[0061] S1.2. Multimodal information extraction is performed on the various types of advertising content in the original set of advertising materials. Advertising content information is obtained through image recognition, text parsing, speech conversion and page parsing, respectively, to generate a multimodal advertising content set. The multimodal information includes image content information, text content information, speech content information and page content information.

[0062] Furthermore, the video advertising materials are extracted frame by frame according to the video time sequence, or the image data in the image advertising materials is directly read; the acquired image data is subjected to target region detection to determine the area where the main object in the image is located; the category, positional relationship and state change information of the main object are feature encoded to generate image content information including image object information, object positional relationship and image scene information.

[0063] The system reads the text from the advertising title, description, and promotional statements, converting the text content into a character sequence. It then performs word segmentation, entity recognition, and semantic relationship analysis on the character sequence to determine the product objects, promotional behaviors, descriptive attributes, and expressive relationships in the text, generating text content information that includes text objects, text semantic relationships, and text expression information.

[0064] The audio signal of the advertisement audio file is collected and the continuous audio signal is converted into processable speech feature data. Then, the speech recognition model is used to perform acoustic feature matching on the speech feature data to convert the speech content in the audio into corresponding text information. Combined with the speech time position, the converted text content is time-stamped to generate speech content information including speech text content, speech time position and speech expression information.

[0065] Read the page structure data from the ad page to obtain the text area, image area, button component, and navigation element; establish the page structure relationship based on the positional and calling relationships between page elements, determine the display order and interaction relationship between different page elements, and generate page content information including page element information, page layout relationship, and page interaction relationship.

[0066] S1.3. Perform association matching on different types of advertising content in the multimodal advertising content set, organize the associations according to the correspondence between different types of advertising content, and generate an associated advertising content set.

[0067] Furthermore, time markers, object markers, location markers, and semantic markers are extracted from various types of advertising content information to form content association features for association judgment. For image content information, the displayed object and display process in the image are determined by reading the area where the image object is located and the corresponding time position; for text content information, the text description object and corresponding expression content are determined through text semantic analysis; for audio content information, the expression position corresponding to the audio segment is determined by the time marker in the audio conversion result; for page content information, the correspondence between page components and the displayed advertising content is determined by the positional relationship of page elements.

[0068] Specifically, the system performs association matching based on the correspondence between different types of advertising content. The time correspondence during ad playback is used as the first association condition. Video footage, audio content, text content, and page element information within the same display time range are initially bound. Further matching is performed based on object consistency across different modalities. Product objects identified through image content are matched with product names in text content; consistency is judged between descriptive objects in audio content and semantic objects in text content; and positional correspondence is determined between display areas in page elements and image display areas. When different types of advertising content satisfy at least one of the association conditions—time correspondence, object correspondence, or display relationship—the corresponding advertising content information is linked. Image content, text content, audio content, and page content with corresponding relationships are organized into the same advertising content unit, and the association path, association type, and association order between each content unit are recorded, forming a set of associated advertising content.

[0069] S1.4. Perform unified format encoding on the advertising content in the associated advertising content set, and write the association relationship into the encoding information to generate a structured advertising material data package.

[0070] Furthermore, data format conversion is performed on different types of data in the associated advertising content set. Image content information, text content information, audio content information, and page content information are converted into a unified data description format. For image content information, image object category, object location, scene information, and corresponding time location are converted into structured fields. For text content information, text objects, semantic relationships, and expressed content are converted into semantic fields. For audio content information, audio-to-text conversion, audio time location, and audio expression features are converted into audio fields. For page content information, page element type, element location, and call relationships between elements are converted into page structure fields. A corresponding data identifier is generated for each advertising content unit, and the advertising content type, content location, time information, and feature information are arranged according to a unified field structure to form standardized advertising content encoding information.

[0071] Specifically, an association description field is added to each advertising content encoding unit to record other modal content units corresponding to that advertising content. The image content encoding information records the corresponding text encoding identifier, voice encoding identifier, and page element encoding identifier; the text content encoding information records the corresponding image display position and voice expression position. The association identifier records the connection relationship between different advertising contents, so that the encoded data still maintains the temporal relationship, object relationship, and display relationship between the original advertising contents. The advertising content encoding data that has completed unified encoding processing and contains association information is integrated to form a structured advertising material data package.

[0072] S2. Perform association processing on the visual content features, text semantic relationships, audio semantic features, and page element association features in the structured advertising material data package to construct the temporal association relationship of advertising content, and determine the advertising risk status result based on the temporal association relationship of advertising content.

[0073] S2.1 Synchronize the visual content features, text semantic relationships, audio semantic features, and page element association features in the structured advertising material data package according to the advertising playback time, and combine the video image expression, text expression, audio expression, and page display expression at the same time position to generate advertising time-related content.

[0074] Furthermore, time stamp information corresponding to various types of advertising content is read from the structured advertising material data package. Time information in visual content features is obtained through the playback position corresponding to the video frame sequence, used to indicate the time stage of a particular scene's content appearance. Time information in text semantic relationships is obtained through the display position or occurrence interval of the text in the advertisement, used to indicate the expression time corresponding to the text content. Time information in audio semantic features is obtained through the start and end positions of the speech segments generated during speech recognition, used to indicate the playback time corresponding to the speech content. Time information in page element association features is obtained through the page element loading time, display time, and interaction trigger time, used to indicate the time stage of page content participating in advertising expression. A unified advertising timeline is established based on the time stamp information in different modal data, using the overall advertising playback process as the time benchmark. Visual content features, text semantic relationships, audio semantic features, and page element association features are mapped to the same time coordinate. For video advertising content, the time position of each scene's content is determined according to the time sequence corresponding to the video frames. For text content, the text expression range is determined according to the time interval of the text's appearance or disappearance. For audio content, the audio expression position is determined according to the time range corresponding to the speech segments. For page content, the page expression position is determined according to the page element display or interaction time.

[0075] Specifically, based on a unified timeline, different types of advertising content within the same time frame are combined. According to the overlap of time positions, video footage, text, audio, and page display within the same time period are linked. When a product is displayed in a video frame at a certain time position, the corresponding product description text, audio introduction, and product display components on the page are bound together to form a complete advertising time-related content unit. For content without a direct correspondence, supplementary connections are made based on temporal proximity and object association to maintain the integrity of the continuous advertising process. Multiple advertising time-related content units corresponding to different time positions are arranged according to the advertising playback order, forming advertising time-related content that includes chronological order, multimodal expression, and content association relationships.

[0076] S2.2 Analyze the state changes of the content in the continuous time position of the advertisement time-related content, and generate the state change relationship of the advertisement content based on the changes of the advertisement display object, display status and display behavior in the time position before and after.

[0077] Furthermore, state information is extracted from each time-related content unit in the advertising time-related content. The video, text, audio, and page display expressions corresponding to each time position are converted into unified state description information. The advertising display object is obtained by identifying the products, services, people, scenes, or page elements appearing at the current time position. The advertising display state is obtained by analyzing the performance of the display object at the current time position, such as whether the object appears, changes, or is emphasized. The advertising display behavior is obtained by analyzing the changes in the object's actions over continuous time to obtain product display, function demonstration, user operation, and page guidance.

[0078] Specifically, the system compares the advertising status information corresponding to adjacent time positions to determine whether the displayed object, display status, and display behavior have changed. When a change in the displayed object is detected, the object migration relationship is determined based on the object category and object relationship in the preceding and following time positions. When a change in the display status is detected, the direction of the status change is determined based on changes in the object display method, prominence, and content. When a change in display behavior is detected, the behavior transformation relationship is determined based on the advertising behavior flow. If the previous time position only contained product appearance display, and the subsequent time position transforms into the user usage process, then the advertising display object has changed from static product display to usage scenario display. If the text expression changes from product introduction to purchase guidance, and the page elements change from information display areas to operation buttons, then the advertising display behavior has changed from information delivery to interactive guidance.

[0079] S2.3. Based on the direction of state change in the state change relationship of the advertising content, connect the continuously changing advertising time-related content to generate the advertising content state transition relationship.

[0080] Furthermore, each state change node in the relationship of ad content state changes is analyzed to obtain the state change direction information between adjacent ad time-related content. This state change direction information includes the change direction of the ad display object, the change direction of the ad display state, and the change direction of the ad display behavior. The change direction of the ad display object is determined by comparing the object category, object quantity, and object relationships in adjacent time positions; the change direction of the ad display state is determined by comparing the object display format, prominence, and expressive attributes; and the change direction of the ad display behavior is determined by comparing the ad action flow and user interaction behavior.

[0081] Specifically, the connection of consecutive advertising time-related content is judged based on the direction of state change. The advertising time-related content corresponding to adjacent time positions is taken as the previous state node and the next state node, respectively. The state change relationship between the two is established when the advertising content in the next time position can be logically continued from the advertising content in the previous time position. If the display object in the previous state node is a product information display and the display object in the next state node changes to a product function display, then a state transition connection is established based on the consistency relationship of objects and the continuation relationship of expression. When the text content changes from product description to function description, and the video screen changes from a static product display to a usage process display, the transition relationship between the two state nodes is strengthened. The state connection relationships between multiple consecutive advertising time-related content are combined. The state nodes with the association direction are connected according to the order of advertisement playback. Each advertising time-related content is taken as a state node, and the state change direction is taken as a state connection edge. The cause information of the state change, the change content information, and the corresponding time information are written into the connection relationship to form a directional state transition structure. All advertising time-related content that meets the state change direction association conditions are connected to generate advertising content state transition relationships.

[0082] S2.4. Arrange the state transition order in the state transition relationship of the advertising content and the relationship of advertising playback time to generate the advertising content evolution path.

[0083] Furthermore, the state identifier, state transition direction, and time information corresponding to each state node are read from the state transition relationship of the advertisement content. The state identifier is used to distinguish different advertisement expression states; the state transition direction is used to indicate the change relationship between states; and the time information is used to indicate the position of the state node during the advertisement playback process. The time ordering relationship of each state node is determined according to the advertisement playback time relationship. The time position corresponding to each state node is sorted, with the state node with earlier playback time as the predecessor node and the state node with later playback time as the successor node. The existence of a valid transition path between adjacent state nodes is determined according to the state transition relationship. When two state nodes satisfy both a time continuity relationship and a state transition relationship, the transition connection is retained; when there is a time sequence but no state change relationship, it is treated as an independent content node.

[0084] Specifically, the state nodes are arranged according to the state transition sequence, with the initial advertising expression state as the starting node of the path. Subsequent state nodes are connected step by step along the state transition direction to form an advertising content evolution sequence. The change direction, reason for change, and occurrence time corresponding to each state transition are synchronously recorded in the path node. This makes the advertising content evolution path not only reflect the time sequence but also the state change logic in the advertising expression process. The advertising content evolution path is generated based on the state node arrangement result. The advertising content evolution path includes a set of state nodes, state transition sequence, state change direction, and corresponding time relationship, which is used to describe the complete evolution process of advertising content from the beginning of display to the end of expression.

[0085] S2.5. Encode the advertising content temporal relationship based on the direction of state change, the order of state transition, and the relationship of advertising playback time in the evolution path of the advertising content.

[0086] Furthermore, the path node information in the advertising content evolution path is read to obtain the state identifier, node position, state change direction, and time information corresponding to each state node. A state direction description field is established based on the change type between state nodes. When the advertising display object remains the same but the display method changes, it is encoded as an object maintaining state change; when the advertising display object is transformed, it is encoded as an object migration state; when the advertising display behavior changes from information display to interactive operation, it is encoded as a behavior change state.

[0087] Specifically, based on the node connection relationships in the advertising content evolution path, a sequence identifier is generated for each state node, and the connection relationship between the current state node and the preceding and following state nodes is recorded. The initial state node is mapped to the path starting point, and subsequent state nodes are sequentially associated according to the migration direction. The predecessor and successor nodes corresponding to each state transition are recorded, so that the encoding result can restore the complete state transition process. The advertising playback time relationship is encoded, and the time position corresponding to each state node is converted into a time description field. The time interval relationship and the occurrence order relationship between state nodes are recorded, so that the system can determine the time stage of each state change. The state change direction encoding information, state migration order encoding information, and time relationship encoding information are combined to generate the advertising content temporal association relationship.

[0088] S2.6. Decompose the state elements of the continuous ad content states in the temporal relationship of ad content, separate the expression object, expression action and expression result in each ad content state, and generate ad content state description information.

[0089] Furthermore, the expressive object in each advertising content state is determined. By extracting object information from the multimodal content in the state node, the main object in the video frame, the display object in the image, the entity object in the text, the target object in the audio description, and the related object in the page are uniformly classified. When the same or corresponding objects exist in different modalities, they are merged according to the object name, object category, and object association to determine the expressive object in the current advertising state. When the product display area in the video frame, the product name in the text, and the product details component in the page all point to the same product, they are merged into the same expressive object, and the corresponding object source information is recorded.

[0090] Specifically, based on textual semantic relationships, visual behavioral change information, and page interaction information, the behavior performed by the object in the current state is identified. Action words in the text content are determined through semantic analysis, such as actions to introduce, showcase, recommend, and guide. Actions in the video content are determined through continuous frame changes, such as visual actions to showcase products, demonstrate functions, and user usage. Actions in the page content are determined through element interaction relationships, such as page behaviors like clicking, jumping, and submitting. The action information obtained from different modalities is associated and fused to form the expressive action information corresponding to the current advertising state.

[0091] Based on the relationship between the object of expression and the action of expression, the output results generated by the advertising state are described. When the object of expression is a product and the action of expression is a function display, the corresponding result is that the user obtains product function information; when the object of expression is a product page and the action of expression is a click guide, the corresponding result is that a page interaction occurs. The expression result is obtained by analyzing the text semantic goals, page interaction goals, and subsequent state change directions in the advertising state. The object of expression, the action of expression, and the result of expression in each advertising content state are combined to form advertising content state description information. The advertising content state description information includes a state identifier, object of expression information, action of expression information, result of expression information, and corresponding time position, which is used to describe the complete expression state of the advertisement in the current time stage.

[0092] S2.7. Based on the changing relationships between the expression object, expression action, and expression result in the advertising content state description information, calculate the state transition difference between adjacent advertising content states and generate advertising content state transition difference information.

[0093] Furthermore, information on the expressive object, expressive action, and expressive result corresponding to adjacent states is obtained from the state description information of continuous advertising content. A previous state description set and a next state description set are established respectively. The relationship between the expressive objects changing between adjacent states is analyzed. The expressive objects in the previous state are matched with those in the next state. By comparing object categories, object attributes, and object associations, it is determined whether the object has changed. When the object categories in two states are the same, but the object attributes or display range have changed, it is determined that the object state has changed. When a new expressive object appears in the next state, it is determined that the object has been added. When an object in the previous state disappears in the next state, it is determined that the object has exited. If the object in the previous state is the product itself, and the object in the next state adds a user scenario, then the expressive object has been expanded from a single product object to an object associated with both the product and the user scenario.

[0094] Specifically, the analysis focuses on the relationship between changes in expressive actions between adjacent states. By comparing the action descriptions in two states, it determines whether the advertising expression behavior has shifted. If the expressive action in the previous state is an introduction or explanation, while the expressive action in the next state is an operation guidance, then the action has shifted from information delivery to behavior guidance. If the expressive action has shifted from function display to effect promise, then the direction of the expressive action has shifted. By comparing the action type, action object, and action goal, action change information is generated. Based on the expression result information corresponding to the two states, it is determined whether the user response or page behavior that the advertisement hopes to generate has changed. If the result corresponding to the previous state is that the user learns about product information, while the result corresponding to the next state is that the user performs a purchase operation, then the expression result has shifted from information acquisition to behavior execution. The changes in the expression object, expression action, and expression result are integrated to generate a state migration difference calculation result. A state migration difference description is established based on the change relationship between the three state elements, including the type of change element, the direction of change, the location of change, and the corresponding time relationship, forming the state migration difference information of the advertising content.

[0095] Let the status description information of two adjacent advertisements be as follows:

[0096] ;

[0097] ;

[0098] in: For the first The object expressed in the status description information of each advertisement content. For the first The expressive actions in the status description information of each advertisement content. For the first The expression results in the status description information of each advertisement content This is a description of the status of the advertisement content. For the first The status description information of the advertisement content corresponding to each time location. For the first The status description information of the advertisement content corresponding to each time location. For the first The object expressed in the status description information of each advertisement content. For the first The expressive actions in the status description information of each advertisement content. For the first The expression results in the state description information of each ad content. Define the ad content state transition difference information:

[0099] ;

[0100] in, For differences in the state transition information of advertising content, The degree of continuity between adjacent ad content states. This represents the range of state changes between adjacent ad content states.

[0101] Calculation of state continuity:

[0102] ;

[0103] in, To express the object by the first The status description information of the first advertisement content is sent to the first The continuous relationship of changes in the status description information of each advertisement content. To express the continuous relationship of action changes, To express the continuous relationship of changes in results.

[0104] Calculation of state change amplitude:

[0105] ;

[0106] in, The magnitude of the state change. To express the direction of change of the result, To express the direction of change of the object.

[0107] S2.8. Based on the advertising content state transition difference information, determine the transition direction between consecutive advertising content states, mark the advertising content state transition difference information that changes from product information expression to user influence expression as risk state transition information, and generate risk state transition information.

[0108] Furthermore, based on the differences in advertising content state transitions, the direction of transition between consecutive advertising content states is determined to ascertain whether the advertising expression process gradually shifts from normal product information expression to an expression state that influences user behavior or psychology. Information on changes in the expression object, expression action, and expression result between adjacent advertising states is obtained from the advertising content state transition differences. Based on the description of the expression object in the preceding and following advertising states, it is determined whether the subject of focus of the advertising content has changed. When the advertising state consistently describes the product itself, such as changes in product structure, functional attributes, or usage methods, this change in object is marked as the direction of product information expression. When the object in the advertising state gradually expands from the product itself to user behavior, user emotions, user decisions, or user results, it is determined that the expression object begins to migrate towards user influence. If the expression object in the previous state was product functionality, and the expression object in the subsequent state transforms into the user's social evaluation or behavioral feedback after use, then the direction of object expression has changed.

[0109] Specifically, the direction of changes in expressive actions during the advertising state transition process is judged. This is done by analyzing the action type, action goal, and target audience within the expressive action information. When the expressive action primarily explains product attributes, demonstrates product functions, or provides usage information, it is identified as a product information expressive action. When the expressive action changes to stimulate user judgment, guide user action, or reinforce user psychological expectations, it is identified as a user-influence expressive action. A change in advertising action from demonstrating product functions to emphasizing immediate decision-making indicates a shift in expressive action from information transmission to user influence. The direction of changes in expressive results during the advertising state transition process is also judged. By comparing the expressive result information in consecutive states, it is determined whether the final advertising goal has changed. When the expressive result is that the user obtains product information or understands product features, it is identified as a product information expressive result. When the expressive result changes to prompt the user to perform a specific behavior, form a specific judgment, or change the user's decision, it is identified as a user-influence expressive result.

[0110] Based on the combined relationships between changes in the object of expression, changes in the action of expression, and changes in the result of expression, a comprehensive judgment is made on the direction of advertising state transition. When the continuous advertising state gradually changes from product object description, product information display, and information acquisition result to user behavior guidance, expression of psychological influence, and behavior execution result, the state transition process is determined to belong to the direction of change from product information expression to user influence expression. The corresponding advertising content state transition difference information is marked as risk state transition information. The state transition difference information that meets the judgment conditions of risk transition direction is organized to generate risk state transition information. The risk state transition information includes the location of the risk transition, the before and after state information, the direction of transition information, the information of the changed elements, and the corresponding time information.

[0111] S2.9. Based on the distribution position of risk state transition information in the temporal correlation of the advertising content, connect the continuous relationship between multiple risk state transition information to generate a risk state evolution chain.

[0112] Furthermore, based on risk state transition information, the positional relationship of multiple risk state transition nodes in the temporal correlation of advertising content is analyzed. A risk state evolution chain is established according to the temporal order, state transition relationship, and consistency of risk direction between nodes. The time position, state node identifier, and preceding and following state transition relationship corresponding to each risk state transition information are read from the temporal correlation of advertising content to determine the distribution position of each risk state transition information in the entire advertising expression process. The continuity relationship between multiple risk state transition information is judged. Taking each risk state transition information as a risk node, the temporal relationship and state transition relationship between adjacent risk nodes are obtained. When two risk nodes have a sequential order in the advertising playback time, and the latter risk node can continue to develop from the risk expression direction corresponding to the former risk node, a connection relationship between risk nodes is established.

[0113] The continuity between risk nodes is determined by multiple conditions. Based on the temporal continuity, it is determined whether two risk nodes are in the same advertising expression stage. For example, if the previous risk node occurs in the product description to user influence stage and the next risk node occurs in the user influence reinforcement stage, then the two belong to a continuous change process. Based on the state transition direction, it is determined whether two risk nodes have the same risk evolution direction. For example, if both nodes show a transition from information description to behavior guidance, then it is determined that the two have a consistent risk direction. Based on the intermediate state node, it is determined whether there is an expression continuity relationship between two risk nodes.

[0114] The risk state transition information that meets the connection conditions is arranged in the order of the advertisement playback time, and a directed connection relationship is established. The earlier risk state transition information is used as the preceding risk node, and the subsequent risk state transition information is used as the subsequent risk node. The migration direction, time relationship and changes between the two are recorded through the connection edge.

[0115] S2.10. Generate the advertising risk status result based on the number of risk state transition information, the direction of risk state transition, and the temporal correlation of the advertising content corresponding to the risk state evolution chain.

[0116] Furthermore, the risk state transition information in the risk state evolution chain is analyzed to obtain the node information corresponding to each risk state transition node. This node information includes the number of risk state transitions, the direction of the risk state transition, the time and location of the risk transition, and the corresponding advertising content state migration information. The number of risk state transitions in the risk state evolution chain is statistically analyzed. All risk state transition nodes in the risk state evolution chain are traversed to count the number of risk transition nodes forming a continuous risk evolution relationship. The continuity of risk state changes is determined based on the connection relationships between nodes. When there is only a single risk state transition node in the risk state evolution chain, and this node does not form a continuous migration relationship with subsequent risk change nodes, it is determined that the advertising content only has a local expression change. When multiple risk state transition nodes appear consecutively along the advertising playback time direction and form a complete risk migration path, it is determined that the advertising content has a continuous risk evolution process.

[0117] Specifically, the direction of risk state transition is analyzed. Based on the description of the preceding and following states in each risk state transition information, the relationship between the change in the risk expression direction is obtained. The direction of risk state transition is determined by comparing the expression object, expression action, and expression result in the preceding and following advertising states. When the state change only manifests as supplementing product information, the risk transition direction remains the product information expression direction. When the state change gradually changes from product information description to user behavior guidance, user psychological influence, or user decision reinforcement, the risk state transition direction is determined to be the risk enhancement direction.

[0118] The temporal correlation of advertising content corresponding to the risk state evolution chain is analyzed. Based on the distribution position of the risk state evolution chain in the temporal correlation of advertising content, the stage position of risk change in the advertising expression process is determined, and whether the risk state gradually expands with the advertising playback time is analyzed. The order of risk occurrence is determined by the time identifier corresponding to the risk node, the risk change path is determined by the state transition relationship between risk nodes, and the scope of risk impact is determined by the advertising content state description information corresponding to the risk node.

[0119] The number of risk state transitions is used as the basis for judging risk persistence, the direction of risk state transitions is used as the basis for judging risk change trends, and the temporal correlation corresponding to the risk state evolution chain is used as the basis for judging the risk development process. Information from these three dimensions is then correlated and integrated. When the number of risk state transitions increases, and multiple risk nodes migrate continuously along the same risk direction, while the risk change is concentrated in the later stages of the advertisement's expression, the advertisement's risk state is determined to be developing in an enhancing direction. When the number of risk state transitions is small, and there is no continuous migration relationship between risk nodes, the risk state is determined to remain in a locally changing state, generating an advertisement risk state result. The advertisement risk state result includes a risk state identifier, risk change direction, risk evolution path information, risk correlation time position, and risk state description information, used to represent the risk change state of the current advertisement content during the complete playback process.

[0120] S3. Based on the advertising risk status results and the temporal correlation of advertising content, perform risk correlation positioning on the advertising content features in the structured advertising material data package, and generate advertising risk factor positioning results.

[0121] S3.1. Perform reverse tracking of the advertising content change process in the temporal relationship of advertising content, locate the time position and content change node that leads to the change of advertising risk status, and generate risk association position.

[0122] Furthermore, the direction of risk status change, risk evolution path information, and the time range corresponding to the risk status are obtained from the advertising risk status results. Based on the last risk status node in the risk status evolution chain as the starting position for reverse tracking, the preceding advertising content status nodes are queried level by level in the opposite direction of the status migration, starting from the final risk node in the risk status evolution chain. The status description information, status migration difference information, and time correlation information corresponding to each status node are obtained. By comparing the status migration difference between the current risk status node and the previous status node, it is determined whether the status change has an impact on the formation of the risk status.

[0123] Specifically, when the change information corresponding to a certain state node can explain the subsequent risk state transition, the node is identified as a risk-related node. If the risk state is manifested as a migration from product information expression to user behavior impact expression, the position of the first expression action change that caused the migration is located during the reverse query process. For example, the position where the text expression changes from functional description to behavioral guidance, the position where the video screen changes from product display to user behavior stimulus, or the position where the page element changes from information display to interaction trigger. Based on the identified risk-related node, the corresponding time position is obtained. Through the time identifier information in the temporal association relationship of the advertising content, the playback time stage where the risk change occurred is determined. The time position is then associated with the corresponding video frame position, text appearance position, audio segment position, and page element display position. The time position, content change node, and corresponding state migration information that caused the risk state change are organized to generate the risk-related position.

[0124] S3.2. Based on the risk-related location, perform association matching on the visual content features, text semantic relationships, audio semantic features, and page element association features in the structured advertising material data package. Associate the visual content features, text semantic relationships, audio semantic features, and page element association features located at the risk-related location with the advertising risk status change process to generate risk-related feature relationships.

[0125] Furthermore, the system reads the risk occurrence time and location, risk status node identifier, and corresponding advertising status change information from the risk-related locations. For visual content features, it queries the corresponding video frame or image region information based on the risk time and location to obtain image objects, scene changes, and visual behavior information appearing within that time period. For text semantic relationships, it obtains the text content, semantic objects, and expressive actions appearing within that time period based on the text display time and semantic relationship information. For audio semantic features, it obtains the audio text content and semantic expression information within the corresponding time range based on the audio segment time identifier. For page element association features, it obtains the buttons, links, pop-ups, or other page components appearing at the corresponding time locations based on the page element display time and interaction relationships.

[0126] Specifically, content features, text semantic relationships, audio semantic features, and page element association features are analyzed in relation to risk state transition differences. When a feature change aligns with the direction of a risk state transition, a risk association is established. Enhanced user behavior display in visual content, behavioral guidance in text content, reinforced semantic meaning in audio content, and added operation entry points in page elements indicate that these multimodal features collectively participate in the risk state change. Multimodal features with risk associations are organized, and the source type, time location, content change, and risk impact direction of each feature are recorded to generate risk association feature relationships.

[0127] S3.3 Locate the risk-affecting content in visual content features, text semantic relationships, audio semantic features, and page element association features. Associate and organize the advertising content features, risk formation locations, and risk status change relationships corresponding to the risk-affecting content to generate advertising risk factor positioning results.

[0128] Furthermore, the visual content features, text semantic relationships, audio semantic features, and page element association features in the risk association feature relationships are broken down and analyzed to obtain the content change information corresponding to each feature. Based on the direction of risk status change, the image objects, visual actions, and scene changes in the risk association time location are analyzed. When a certain visual change causes a shift in the direction of advertising expression, the corresponding visual content is identified as the risk-affecting content. If the visual display changes from product introduction to strengthening the display of user behavior results, the area of ​​visual change is located, and the semantic structure of the text corresponding to the risk association location is analyzed to obtain the expression object, action description, and result orientation in the text. When the text expression changes from product description to behavior guidance, result promise, or user impact description, the corresponding text fragment is identified as the risk-affecting content.

[0129] Specifically, based on the speech-to-text conversion and the timing of the speech, the system analyzes whether the speech content participates in changes in risk expression. When the speech content reinforces a certain user impact expression and forms a consistent direction of change with other modal content, the corresponding speech segment is identified as risk-influencing content. Based on the page element type, display position, and interaction relationship, the system determines whether the page component changes the direction of the advertisement's expression. If a page element changes from an information display area to a highly interactive operation area, the corresponding page component is identified as risk-influencing content. The identified risk-influencing content is then correlated and organized with the corresponding advertisement content characteristics, the location of risk formation, and the relationship between risk status changes to form the advertisement risk factor positioning result.

[0130] S4. Based on the results of the advertising risk factor positioning, determine the risk factors in the advertising content, perform adjustment operations on the advertising content corresponding to the risk factors, and generate the adjusted advertising materials.

[0131] S4.1 Based on the location of risk factors in the advertising risk factor positioning results, the visual content features, text semantic relationships, audio semantic features, and page element association features in the risk expression factors are broken down according to the advertising expression process, the changes in the expression state before and after the risk expression factors are determined, and the risk expression migration relationship is generated.

[0132] Furthermore, the risk factor location, risk impact content, multimodal feature sources, and corresponding risk state change relationships are obtained from the advertising risk factor positioning results. The risk factor location includes the video time position, text appearance position, audio segment position, and page element display position where the risk content is located. The risk impact content includes visual content, text content, audio content, and page content that lead to changes in risk state. The video frames corresponding to the risk location are divided into continuous visual state units according to time sequence. By analyzing the changes in image objects, object states, and visual behaviors at different time positions, the process of visual expression state change is determined. The text content is split according to the semantic expression sequence. By analyzing the text semantics, the expression objects, expression actions, and expression results at different stages are obtained, and the process of text expression change is determined. The audio content is split according to the time position of the speech segments. By analyzing the speech expression content and semantic changes at different time stages through speech conversion results, the process of audio expression change is determined. The page elements are split according to the display and interaction process. By using page structure information, the changes in page components, interaction entry points, and display relationships are determined.

[0133] Specifically, the visual, textual, audio, and page expression states in adjacent time periods are compared to determine the direction of change for each type of expression factor. Changes in visual expression state are determined by comparing object display methods, object actions, and scene relationships; changes in textual expression state are determined by comparing semantic objects, expressive actions, and result orientation; changes in audio expression state are determined by comparing voice content and semantic reinforcement relationships; and changes in page expression state are determined by comparing page element types, display positions, and interactive behaviors. The advertising expression state before the risk factor occurs is taken as the initial expression state, and the advertising expression state after the risk factor occurs is taken as the transition expression state. The content, direction, and source of change between the two are recorded. If the former state is a product function description and the latter state is user behavior guidance, a transition relationship from product information expression to user influence expression is established. If video, text, audio, and page elements all change in the same direction simultaneously, this risk expression transition relationship is strengthened, generating a risk expression transition relationship. The risk expression transition relationship includes information on the previous expression state, the subsequent expression state, the direction of state change, the source of the change modality, and the state transition path, used to describe how the risk expression factor transforms from the original advertising expression state to the risk expression state.

[0134] S4.2 Analyze the contribution of changes in visual content features, text semantic relationships, audio semantic features, and page element association features in risk expression factors to identify the key changes that cause risk expression migration and generate risk adjustment objects.

[0135] Furthermore, the pre-expression and post-expression states in the risk expression migration relationship are analyzed to obtain information on changes in visual content, text semantics, audio semantics, and page elements. For visual content features, by comparing changes in image objects, object actions, and scenes before and after risk migration, it is determined which visual changes participated in the formation of risk expression. If the product display method changes from a normal display to an enhanced user result display, then this visual change is considered as candidate risk adjustment content. For text semantic relationships, by comparing the expression objects, action descriptions, and result orientations in the text before and after, key sentences or semantic structures that led to changes in the expression direction are identified. For audio semantic features, by comparing changes in speech content, the degree of semantic enhancement, and the correspondence with visual content, key audio segments affecting risk expression migration are identified. For page element association features, by comparing changes in page components, interaction paths, and user operation entry points, page adjustment factors affecting the formation of risk expression are identified.

[0136] Specifically, each candidate change is analyzed in relation to the risk expression migration relationship to determine whether the change directly drives the advertising status from product information expression to user influence expression. When a change occurs and the risk expression direction changes simultaneously, the content is considered to have a high correlation contribution. If a content change is merely a normal display change and does not alter the risk expression direction, its correlation is reduced. When text, video, and page elements change simultaneously, if the text expression change directly leads to user behavior guidance, while the video change is only used for synchronous display, the text change is identified as the primary risk adjustment target, and the video change as a secondary adjustment target. Visual content, text content, audio content, and page elements with a primary risk contribution relationship are integrated to generate risk adjustment targets. These risk adjustment targets include the content type, content location, change relationship, and corresponding risk expression migration relationship.

[0137] S4.3. Based on the risk adjustment object, the expression relationship of the risk adjustment object in the risk expression factor is replaced. The risk adjustment object is replaced with safe expression content that maintains the continuity of the expression state before and after. The visual content features, text semantic relationships, audio semantic features and page element association features are reconnected to generate the adjusted advertising material.

[0138] Furthermore, based on the content type, location, and corresponding risk migration relationships within the risk adjustment objects, the specific advertising content that needs to be replaced is determined. The advertising state before replacement and the target expression state that needs to be maintained after replacement are analyzed. If the original text is used to introduce product functions, the replaced text must still maintain the product function description attribute. If the original page element is used to display product information, the replaced page element must still maintain the information display relationship and cannot be directly deleted, causing an interruption in the advertising expression. For text-based risk adjustment objects, risk expression statements are replaced through semantic reconstruction, adjusting the original risk-oriented expression to safe semantic content that conforms to the product information expression relationship. For visual risk adjustment objects, risk visual expressions are adjusted to safe visual expressions that maintain object continuity through image content adjustment or visual area replacement. For audio-based risk adjustment objects, audio semantics are kept consistent with safe advertising expressions by regenerating or replacing the audio content. For page element-based risk adjustment objects, the page expression is kept continuous with the overall advertising state by adjusting the display relationship or interaction logic of page components.

[0139] Specifically, the adjusted text content is rematched with the corresponding video footage, the adjusted audio content is reassociated with the corresponding time and location, and the adjusted page elements are reconnected with the ad display status. This restores the complete association structure of visual content features, text semantic relationships, audio semantic features, and page element association features, thus forming the adjusted ad creative.

[0140] S5. Perform security verification on the adjusted advertising materials, generate a verified secure advertising version, and execute advertising delivery control operations based on the advertising risk status results and the verified secure advertising version.

[0141] S5.1 Based on the changes in advertising content in the adjusted advertising materials, compare the changes in the positions of risk factors in the adjusted advertising materials with those in the risk factor positioning results, determine the advertising content change status corresponding to the risk factor positions, and generate the risk change status.

[0142] Furthermore, the risk factor location, risk impact content, risk formation location, and corresponding risk status change relationship are obtained from the advertising risk factor positioning results. At the same time, the adjusted visual content features, text semantic relationships, audio semantic features, and page element association features are obtained from the adjusted advertising materials. Through the time markers, content markers, and association markers in the advertising materials, the video time position, text position, audio segment position, and page element position of the risk content before adjustment are matched with the corresponding positions in the adjusted advertising materials.

[0143] Specifically, for video ad content, the analysis focuses on whether changes have occurred in image objects, display status, and visual behavior by comparing video frame information at corresponding time positions before and after the adjustment; for text ad content, the analysis focuses on whether changes have occurred in the expressed object, expressed action, and expressed result by comparing the text semantic structure before and after the adjustment; for audio ad content, the analysis focuses on whether the speech expression has been adjusted by comparing the speech-to-text conversion and semantic expression relationship before and after the adjustment; and for page element content, the analysis focuses on whether page components have changed by comparing page structure information and element interaction relationships.

[0144] The adjusted content corresponding to the location of risk factors is divided into three states: changed, partially changed, and unchanged. When the content corresponding to a risk factor has been replaced and the adjusted content no longer maintains the original risk migration direction, it is determined to be in a state of risk change elimination. When some risk expression factors have been adjusted, but risk expression relationships still exist, it is determined to be in a state of risk change residue. When the content at the location of a risk factor has not changed effectively, it is determined to be in a state of risk change retention. The changes in the location of risk factors, the changes in the adjusted content, and the direction of risk status changes are correlated and organized to generate a risk change status. The risk change status includes the risk factor location identifier, information on the differences between the content before and after the adjustment, information on the risk expression migration changes, and the result of the risk status change.

[0145] S5.2. Perform risk path backtracking on the temporal change process of the advertising content in the adjusted advertising material, determine whether the adjusted advertising material still develops along the risk change direction corresponding to the original advertising risk status result, and generate a safety verification status.

[0146] Furthermore, based on the adjusted ad playback time information, a new time correspondence is established for various types of ad content, generating an adjusted ad content state sequence. For video content, visual state nodes are obtained through continuous screen state changes; for text content, text expression state nodes are obtained through semantic parsing; for audio content, semantic state nodes are obtained through speech conversion results; and for page content, interactive state nodes are obtained through page structure parsing. The adjusted ad content state sequence is matched with the risk evolution chain in the original ad risk state results. Based on the order of risk nodes in the original risk state evolution chain, the ad state at the corresponding time stage in the adjusted ad is queried, and it is determined whether the adjusted state transition relationship is still consistent with the original risk direction.

[0147] The specific judgment process includes: if the adjusted advertising content still exhibits a state change from product information expression to user-influence expression (e.g., text changes from product description to behavioral guidance, or page changes from information display to a strong operation entry point), then the risk path is determined to still exist; if the direction of state change in the adjusted advertising content changes (e.g., from user-influence expression back to product information description), or the content corresponding to the original risk node is replaced with normal expression content), then the original risk path is determined to be blocked; if some risk states are eliminated, but subsequent risk migration nodes still exist, then the risk path is determined to remain, and a security verification state is generated based on the above path backtracking results. The security verification state includes whether the risk path continues to exist, whether the direction of risk state migration has changed, whether the risk node has been eliminated, and information on the continuity of advertising content expression.

[0148] S5.3. Based on the security verification status, confirm the security status of the adjusted advertising materials, and determine the adjusted advertising materials that have not undergone any changes in risk status and maintain the continuity of advertising content expression as the verified secure advertising version.

[0149] Furthermore, the adjustment of the advertising creative is assessed to determine whether it meets the security confirmation conditions. A comprehensive analysis of the following information is performed, and the risk path verification results are used to determine whether the adjusted advertisement still exhibits a risk state transition relationship. If the adjusted advertisement content does not exhibit the original risk state evolution direction, the risk path verification is considered successful. The risk state change results are used to determine whether new risk changes have occurred at the original risk factor location. If the content corresponding to the original risk factor location has been replaced with a secure expression and no new risk state transition has occurred, the risk factor adjustment is considered effective. The continuity of the advertisement's expression is used to determine whether the adjusted advertisement maintains a complete expression relationship. By analyzing the relationships between visual content, text content, audio content, and page content, the relationships between these elements are analyzed to determine whether the adjusted advertisement has content breaks, semantic inconsistencies, or abnormal interaction relationships. When the adjusted advertisement simultaneously meets the risk path elimination condition, the risk state not being regenerated condition, and the advertisement expression continuity condition, the adjusted advertising creative is determined as a verified secure advertisement version, and secure advertisement version information is generated. The secure advertisement version information includes the adjusted advertising creative, risk factor adjustment records, security verification status, and corresponding advertisement content relationships.

[0150] S5.4 Based on the risk status change information in the advertising risk status results and the security verification status in the verified secure advertising version, establish the placement status relationship between the advertising risk status results and the verified secure advertising version, and generate the advertising placement status relationship.

[0151] Furthermore, based on the advertising risk status results and the verified safe advertising version, a correspondence is established between the original advertising risk status and the adjusted safe advertising version to determine the advertising delivery status corresponding to different risk statuses. Risk status change information is obtained from the advertising risk status results, including risk status identifier, risk status change direction, risk evolution path, risk association location, and risk status change process information. Simultaneously, the safety verification status in the verified safe advertising version is obtained, including risk path verification results, risk factor change results, advertising expression continuity status, and safety version identifier information.

[0152] Specifically, the risk status results of the advertisement are matched with the security verification status. Based on the risk status identifier in the risk status results, the risk change process corresponding to the original advertisement is associated with the adjustment process corresponding to the security advertisement version. When a risk status inflection point is recorded in the original advertisement risk status results, the security verification status is used to query whether the content corresponding to the risk point has been eliminated, whether the migration direction has changed, and whether the adjusted advertisement still has the corresponding risk status. The risk status change process is used as the input for the delivery decision, and the security verification result is used as the basis for the selection of the advertisement version. When the risk status result indicates that the advertisement has a risk status migration, but the security verification status shows that the risk path has been eliminated, a placement status relationship of original risk status - safe version placement is established; when the risk status result indicates a high risk, and the security verification status still shows residual risk paths, a placement status relationship of risk status - restricted placement is established; when the risk status result does not show a significant risk change, a placement status relationship of normal status - original advertisement version placement is established. The risk status information, security verification information, corresponding advertisement version information, and placement processing method are associated and encoded to form an advertisement placement status relationship. The advertisement placement status relationship includes risk status nodes, safe version nodes, version correspondence, placement conditions, and placement status change information, which are used to describe the correspondence between advertisement versions and placement strategies under different risk conditions.

[0153] S5.5. Based on the changes in risk status in the advertising placement status relationship and the security status of the verified secure advertising version, adjust the advertising placement method, generate advertising placement control results, execute placement control operations on the verified secure advertising version, and generate advertising placement execution results.

[0154] Furthermore, the risk status changes are assessed as follows: When the risk status result indicates no risk migration and the security verification status meets the security confirmation conditions, the ad delivery status is set to normal delivery, and the verified secure ad version is delivered as usual; when the risk status result indicates a change in risk status, but the security verification status indicates the risk path has been eliminated, the ad delivery status is set to secure version replacement delivery, and the original ad creative is replaced with the verified secure ad version before delivery; when the risk status result indicates a change in risk status, and the security verification status indicates the risk path has not been completely eliminated, the ad delivery status is set to restricted delivery, and the ad is delivered accordingly. When an ad version is paused or enters the processing flow, an ad delivery control result is generated based on the confirmed delivery status. The delivery status, corresponding ad version, delivery channel information, and execution actions are associated to form a delivery control instruction. The delivery control instruction includes delivery version selection information, ad creative call information, delivery status identifier, and delivery restrictions. The verified secure ad version is sent to the ad delivery system, which loads the corresponding ad creative based on the control result and performs version switching, normal delivery, or delivery restriction operations according to the delivery status relationship. At the same time, the ad version identifier, delivery time, delivery status, and execution result information are recorded to form the ad delivery execution result.

[0155] This embodiment also provides an artificial intelligence-based proactive defense system for internet advertising, including: a processing module that obtains the original set of advertising materials corresponding to the internet advertising delivery request, performs multimodal information extraction, association and organization and structured encoding processing on the multi-type advertising content in the original set of advertising materials, and generates a structured advertising material data package;

[0156] The association module performs association processing on the visual content features, text semantic relationships, audio semantic features, and page element association features in the structured advertising material data package, constructs the temporal association relationship of advertising content, and determines the advertising risk status result based on the temporal association relationship of advertising content;

[0157] The positioning module performs risk correlation positioning on the features of advertising content in the structured advertising material data package based on the advertising risk status results and the temporal correlation of advertising content, and generates advertising risk factor positioning results.

[0158] The adjustment module determines the risk factors in the advertising content based on the advertising risk factor positioning results, performs adjustment operations on the advertising content corresponding to the risk factors, and generates adjusted advertising materials.

[0159] The execution module performs security verification on the adjusted advertising creatives, generates a verified secure advertising version, and executes advertising delivery control operations based on the advertising risk status results and the verified secure advertising version.

[0160] In summary, this invention performs multimodal information processing on various types of advertising materials corresponding to internet advertising delivery requests, constructing a structured advertising material data package that includes visual content features, textual semantic relationships, audio semantic features, and page element association features. It also constructs a temporal correlation relationship for advertising content based on the state changes of the advertising content over time, and determines the advertising risk status by analyzing the advertising content state transition process. Based on the advertising risk status results, it traces back the risk formation process of the advertising content temporal correlation relationship, identifies the advertising content factors that lead to changes in risk status, generates advertising risk factor location results, and makes local adjustments to the advertising content corresponding to the risk factors, generating adjusted advertising materials. By verifying the risk path of the adjusted advertising materials, it generates a verified secure advertising version and executes advertising delivery control operations based on the advertising risk status results, thus achieving proactive defense processing for internet advertising content risk analysis, risk location, risk adjustment, and secure delivery.

[0161] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A proactive defense method for internet advertising based on artificial intelligence, characterized in that: include, Obtain the original set of advertising creatives corresponding to the internet advertising placement request, extract multimodal information, associate and organize the various types of advertising content in the original set of advertising creatives, and perform structured encoding to generate a structured advertising creative data package; The visual content features, text semantic relationships, audio semantic features, and page element association features in the structured advertising creative data package are correlated to construct the temporal association relationship of advertising content, and the advertising risk status result is determined based on the temporal association relationship of advertising content. Based on the advertising risk status results and the temporal correlation of advertising content, risk correlation positioning is performed on the advertising content features in the structured advertising material data package to generate advertising risk factor positioning results; Based on the results of the advertising risk factor positioning, identify the risk factors in the advertising content, perform adjustment operations on the advertising content corresponding to the risk factors, and generate adjusted advertising materials; Perform security verification on the adjusted advertising creatives, generate a verified secure advertising version, and execute advertising delivery control operations based on the advertising risk status results and the verified secure advertising version.

2. The method for proactive defense against internet advertising based on artificial intelligence as described in claim 1, characterized in that, The set of original advertising materials corresponding to the internet advertising delivery request includes: The original set of advertising creatives includes video advertising creatives, image advertising creatives, text advertising creatives, audio advertising creatives, and advertising page creatives.

3. The artificial intelligence-based proactive defense method for internet advertising as described in claim 2, characterized in that, The process of extracting, associating, and structurally encoding multimodal information from the original set of advertising materials to generate a structured advertising material data package includes: Multimodal information is extracted from various types of advertising content in the original set of advertising materials. Advertising content information is obtained through image recognition, text parsing, speech conversion and page parsing respectively, and a multimodal advertising content set is generated. The multimodal information includes image content information, text content information, speech content information and page content information. The system performs association matching on different types of ad content in the multimodal ad content set, organizes the associations based on the correspondence between different types of ad content, and generates an associated ad content set. The ad content in the associated ad content set is encoded in a unified format, and the association relationship is written into the encoded information to generate a structured ad creative data package.

4. The artificial intelligence-based proactive defense method for internet advertising as described in claim 3, characterized in that, The process of associating visual content features, text semantic relationships, audio semantic features, and page element association features in the structured advertising material data package to construct the temporal association relationship of advertising content includes: The visual content features, text semantic relationships, audio semantic features, and page element association features in the structured advertising material data package are synchronized according to the advertising playback time. The video image expression, text expression, audio expression, and page display expression at the same time position are combined to generate advertising time-related content. Perform state change analysis on consecutive time positions of content in the time-related content of advertisements, and generate the state change relationship of advertisement content based on the changes in the display object, display status and display behavior of the advertisement in the preceding and following time positions; Based on the direction of state change in the state change relationship of the advertising content, the continuously changing time-related content of the advertising content is connected to generate the state transition relationship of the advertising content. Based on the state transition order in the ad content state transition relationship and the ad playback time relationship, the ad content state transition relationship is arranged to generate the ad content evolution path; The temporal relationship of advertising content is generated by encoding the direction of state changes, the order of state transitions, and the relationship of advertising playback time in the evolution path of advertising content.

5. The artificial intelligence-based proactive defense method for internet advertising as described in claim 4, characterized in that, The determination of advertising risk status based on the temporal correlation of advertising content includes: The state elements of continuous ad content states in the temporal relationship of ad content are decomposed, and the expression object, expression action and expression result in each ad content state are separated to generate ad content state description information; Based on the changing relationships between the expressive object, expressive action, and expressive result in the state description information of the advertising content, calculate the state transition difference between adjacent advertising content states and generate advertising content state transition difference information. Based on the differences in the state transition of advertising content, the direction of transition between consecutive advertising content states is determined. The differences in the state transition of advertising content that change from product information expression to user influence expression are marked as risk state transition information, and risk state transition information is generated. Based on the distribution position of risk state transition information in the temporal correlation of the advertising content, the continuous relationship between multiple risk state transition information is connected to generate a risk state evolution chain. Based on the number of risk state transition information in the risk state evolution chain, the direction of risk state transition, and the temporal correlation of the advertising content corresponding to the risk state evolution chain, the advertising risk state result is generated.

6. The artificial intelligence-based proactive defense method for internet advertising as described in claim 5, characterized in that, The step of performing risk correlation positioning on the advertising content features in the structured advertising material data package based on the advertising risk status results and the temporal correlation of advertising content to generate advertising risk factor positioning results includes: The process of advertising content changes in the temporal relationship of advertising content is reversed to locate the time position and content change node that leads to the change of advertising risk status, and generate risk association position; Based on the risk-related location, the visual content features, text semantic relationships, audio semantic features, and page element association features in the structured advertising material data package are matched and associated. The visual content features, text semantic relationships, audio semantic features, and page element association features located at the risk-related location are associated with the advertising risk status change process to generate risk-related feature relationships. The risk-affecting content is located from visual content features, text semantic relationships, audio semantic features, and page element association features. The advertising content features, risk formation locations, and risk status change relationships corresponding to the risk-affecting content are then linked and organized to generate advertising risk factor location results.

7. The artificial intelligence-based proactive defense method for internet advertising as described in claim 6, characterized in that, The process of determining risk factors in advertising content based on the advertising risk factor positioning results, adjusting the advertising content corresponding to the risk factors, and generating adjusted advertising materials includes: Based on the location of risk factors in the advertising risk factor positioning results, the visual content features, text semantic relationships, audio semantic features, and page element association features in the risk expression factors are broken down according to the advertising expression process. The changes in the expression state before and after the risk expression factors are determined, and the risk expression migration relationship is generated. We conduct a change contribution analysis on the visual content features, text semantic relationships, audio semantic features, and page element association features in the risk expression factors to identify the key changes that cause the risk expression migration and generate risk adjustment objects. Based on the risk adjustment object, the expression relationship of the risk adjustment object in the risk expression factors is replaced. The risk adjustment object is replaced with safe expression content that maintains the continuity of the expression state before and after, and the visual content features, text semantic relationships, audio semantic features and page element association features are reconnected to generate the adjusted advertising material.

8. The artificial intelligence-based proactive defense method for internet advertising as described in claim 7, characterized in that, The step of performing security verification on the adjusted advertising creatives and generating a secure advertising version that has passed verification includes: Based on the changes in the advertising content in the adjusted advertising creative, the changes in the positions of risk factors in the risk factor positioning results of the adjusted advertising creative are compared with those in the risk factor positioning results to determine the advertising content change status corresponding to the risk factor position and generate the risk change status. The risk path is traced back to the temporal change process of the advertising content in the adjusted advertising creative to determine whether the adjusted advertising creative still develops along the risk change direction corresponding to the original advertising risk status result, and a safety verification status is generated. Based on the security verification status, the security status of the adjusted ad creatives is confirmed. The adjusted ad creatives that do not show any changes in risk status and maintain the continuity of the ad content are identified as the verified secure ad versions.

9. The artificial intelligence-based proactive defense method for internet advertising as described in claim 8, characterized in that, The step of performing ad delivery control operations based on the ad risk status results and the verified secure ad version includes: Based on the risk status change information in the advertising risk status results and the security verification status in the verified secure advertising version, establish the delivery status relationship between the advertising risk status results and the verified secure advertising version, and generate the advertising delivery status relationship. Based on the changes in risk status in the ad delivery status relationship and the security status of the verified secure ad version, the ad delivery method is adjusted to generate ad delivery control results. The delivery control operation is then executed on the verified secure ad version to generate ad delivery execution results.

10. An AI-based proactive defense system for internet advertising, based on the AI-based proactive defense method for internet advertising as described in any one of claims 1 to 9, characterized in that, include: The processing module obtains the original set of advertising materials corresponding to the internet advertising placement request, performs multimodal information extraction, association and organization, and structured encoding on the multi-type advertising content in the original set of advertising materials, and generates a structured advertising material data package; The association module performs association processing on the visual content features, text semantic relationships, audio semantic features, and page element association features in the structured advertising material data package, constructs the temporal association relationship of advertising content, and determines the advertising risk status result based on the temporal association relationship of advertising content; The positioning module performs risk correlation positioning on the features of advertising content in the structured advertising material data package based on the advertising risk status results and the temporal correlation of advertising content, and generates advertising risk factor positioning results. The adjustment module determines the risk factors in the advertising content based on the advertising risk factor positioning results, performs adjustment operations on the advertising content corresponding to the risk factors, and generates adjusted advertising materials. The execution module performs security verification on the adjusted advertising creatives, generates a verified secure advertising version, and executes advertising delivery control operations based on the advertising risk status results and the verified secure advertising version.

Images