Digital publication content auditing system based on artificial intelligence

Abstract

The invention is suitable for the technical field of digital publication content auditing, and provides a digital publication content auditing system based on artificial intelligence, which comprises a feedback judgment module, a conflict analysis module, a focusing degree calculation module, a correction coefficient generation module and a display correction module, according to the method, the abnormal advertisement display condition can be automatically identified and corrected according to the characteristic that the attention of the hearing-impaired people is easily interfered when the hearing-impaired people watch the advertisement video, and before the advertisement video is put again, the display conditions such as the limb movement amplitude and the expression exaggeration amplitude of the promotion personnel in the video are quantitatively reduced according to the correction coefficient; therefore, the excessive exaggeration performance is weakened; according to the invention, troubles caused by exaggerated expressions and actions of promotion personnel in the advertisement video to hearing-impaired people can be effectively avoided, the problem that the attention of the hearing-impaired people is distracted when the hearing-impaired people watch advertisements is relieved, and the ability of the hearing-impaired people to acquire price and product key information is improved, so that the watching experience is improved, and the purchase conversion rate of the advertisement video in the hearing-impaired people is improved.

Description

Technical Field

[0001] This invention belongs to the field of digital publication content review technology, and in particular relates to an artificial intelligence-based digital publication content review system. Background Technology

[0002] With the rapid popularization of digital publications, video advertising has become an important business promotion method. Advertisers typically use salespeople's language, facial expressions, body language, and video effects to enhance the appeal of their ads. However, most existing ad designs are geared towards general users and do not fully consider the viewing characteristics of hearing-impaired individuals. Hearing-impaired people primarily rely on vision to understand advertising content. When faced with exaggerated expressions and movements from salespeople, their attention is easily drawn to the excessive performance, and they may become confused about the meaning conveyed by these exaggerated expressions, forcing them to make additional analyses and judgments. This can not only easily trigger their resistance to advertising videos but may also cause them to overlook key information such as price or product details, thus directly affecting their purchasing decisions.

[0003] Existing video ad review systems primarily focus on compliance checks, such as identifying the presence of sensitive content, inappropriate visuals, or offensive language, or simply conducting statistical analysis of ad effectiveness based on view counts and click-through rates. Some research attempts to improve ad recommendations through emotion recognition, but this remains limited to the general audience. Current solutions do not address the needs of the hearing impaired, failing to quantitatively assess and correct the interference of exaggerated displays in ads with information retrieval, taking into account their viewing behavior and interaction characteristics.

[0004] Existing technologies cannot effectively identify resistance from hearing-impaired individuals when watching advertisements due to overly exaggerated presentations; nor can they adjust the presentation based on the viewing characteristics of hearing-impaired individuals before re-running the advertisement video, resulting in a low purchase conversion rate for advertisement videos among hearing-impaired individuals. Summary of the Invention

[0005] The purpose of this invention is to provide an artificial intelligence-based digital publication content review system, which aims to solve the problems mentioned in the background art.

[0006] This invention is implemented as follows: an artificial intelligence-based digital publication content review system, the system comprising: The feedback determination module is used to determine whether the reduction in positive feedback of the target video compared to the control video in the hearing-impaired population exceeds a preset threshold. If so, the historical viewing and interaction data of the target video and the control video are obtained. The resistance analysis module is used to analyze the historical viewing and interaction data to determine whether the resistance value of hearing-impaired people to the display of advertisements in the target video exceeds a preset value. If so, the target video is determined to be an abnormal video. The focus calculation module is used to calculate the first price focus of hearing-impaired people when watching abnormal videos and the second price focus when watching comparison videos, based on the historical viewing data and interaction data. The correction coefficient generation module is used to obtain a correction coefficient based on the difference between the first price focus and the second price focus, as well as the conflict value. The display correction module is used to quantitatively reduce the display of advertisements in abnormal videos based on the correction coefficient during subsequent review of abnormal videos.

[0007] As a further limitation of the technical solution of this embodiment of the invention, the feedback determination module specifically includes: The first indicator calculation unit is used to select several videos with the same background and a purchase conversion rate higher than the standard value as control videos, obtain the purchase conversion rate and positive interaction behavior index of hearing-impaired people when watching the control videos, and calculate the first positive feedback indicator accordingly. The second indicator calculation unit is used to obtain the purchase conversion rate and positive interaction behavior index of hearing-impaired people when watching target videos, and to calculate the second positive feedback indicator accordingly. A positive feedback comparison unit is used to quantify the reduction ratio of the second positive feedback index compared with the first positive feedback index, and compare the reduction ratio with a preset threshold. The feedback difference determination unit is used to obtain the historical viewing and interaction data of the target video when it is determined that the reduction ratio is greater than a preset threshold.

[0008] As a further limitation of the technical solution of this embodiment of the invention, the conflict analysis module specifically includes: The comment analysis unit is used to parse the historical viewing and interaction data, obtain negative comment data on the advertising display of the target video by hearing-impaired people, and determine the ratio of negative comments in all comments, and set the ratio as the first objection factor; An emotion detection unit is used to capture and calculate the ratio of the duration of negative facial emotional state characteristics of hearing-impaired individuals when watching advertising content in a target video to the total duration of watching advertising content in the target video, and this ratio is set as the second aversion factor. The factor fusion unit is used to weight and fuse the first and second conflicting factors to obtain the conflict value of the target video; The conflict determination unit is used to obtain a preset value and determine whether the conflict value exceeds the preset value. If so, the target video is determined to be an abnormal video.

[0009] As a further limitation of the technical solution of this embodiment of the invention, the focus calculation module specifically includes: The gaze behavior acquisition unit is used to capture the click frequency and gaze duration of hearing-impaired people on the price tag area related to advertisements in the screen while watching target videos and abnormal videos, respectively. The focus calculation unit is used to calculate the first price focus and the second price focus of the hearing-impaired person when watching abnormal videos and control videos, respectively, based on the click frequency and gaze duration.

[0010] As a further limitation of the technical solution of this embodiment of the invention, the correction coefficient generation module specifically includes: The difference calculation unit is used to calculate the decrease in the first price focus compared to the second price focus, and the increase in the resistance value corresponding to the abnormal video compared to the preset value. The correction coefficient fusion unit is used to weight and fuse the absolute values ​​of the reduction and increase to calculate the correction coefficient.

[0011] As a further limitation of the technical solution of the embodiments of the present invention, the modification module specifically includes: The display reduction unit is used to quantitatively reduce the advertising display of the abnormal video based on a correction coefficient before the subsequent abnormal video is released, and generate a corrected video. The video publishing unit is used to deliver and publish the corrected video.

[0012] As a further limitation of the technical solution of this invention embodiment, the advertising display includes the brightness of the advertising area and the body language of the advertising salesperson; wherein: The brightness of the advertising area is reduced based on a correction factor to weaken the prominence of the image. The body language of the advertising salesperson is reduced based on a correction coefficient to weaken the overly exaggerated presentation effect.

[0013] Compared with existing technologies, this invention provides an AI-based digital publication content review system that can automatically identify and correct abnormal advertising displays, addressing the issue of easily distracted attention for hearing-impaired viewers. The system quantifies resistance by comparing the positive feedback differences between the target video and a control video, combined with negative comments and facial emotional states from hearing-impaired viewers. Simultaneously, it calculates changes in price focus by collecting click frequency and gaze duration. Furthermore, it weights and fuses the decrease in price focus with the increase in resistance to generate a correction coefficient. Before the advertising video is re-run, the system uses this correction coefficient to quantitatively reduce aspects such as brightness, the salesperson's body language, and exaggerated facial expressions, thereby mitigating overly exaggerated presentations. Through these methods, this invention effectively avoids the distress caused to hearing-impaired viewers by exaggerated expressions and movements of salespeople in advertising videos, alleviating the problem of distracted attention while watching advertisements, improving their ability to grasp key price and product information, thus enhancing the viewing experience and increasing the purchase conversion rate of advertising videos among hearing-impaired viewers. Attached Figure Description

[0014] Figure 1 An application architecture diagram of an AI-based digital publication content review system provided in this embodiment of the invention; Figure 2 A structural block diagram of the feedback judgment module in an artificial intelligence-based digital publication content review system provided in this embodiment of the invention; Figure 3 A structural block diagram of the conflict analysis module in an AI-based digital publication content review system provided in this embodiment of the invention; Figure 4 A structural block diagram of the focus calculation module in an artificial intelligence-based digital publication content review system provided in this embodiment of the invention; Figure 5 A structural block diagram of the correction coefficient generation module in an AI-based digital publication content review system provided in this embodiment of the invention; Figure 6 This is a structural block diagram of the correction module in the AI-based digital publication content review system provided in an embodiment of the present invention. Detailed Implementation

[0015] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0016] Figure 1The diagram illustrates the application architecture of an AI-based digital publication content review system provided in an embodiment of the present invention.

[0017] Specifically, an artificial intelligence-based digital publication content review system includes the following steps: The feedback determination module 100 is used to determine whether the reduction in positive feedback of the target video compared to the control video among hearing-impaired people exceeds a preset threshold. If so, the historical viewing and interaction data of the target video are obtained.

[0018] Specifically, Figure 2 The diagram shows the structural block diagram of the feedback judgment module in an AI-based digital publication content review system.

[0019] The feedback determination module 100 specifically includes: The first indicator calculation unit 101 is used to select several videos with the same background and a purchase conversion rate higher than the standard value as control videos, obtain the purchase conversion rate and positive interaction behavior index of hearing-impaired people when watching the control videos, and calculate the first positive feedback indicator accordingly. The second indicator calculation unit 102 is used to obtain the purchase conversion rate and positive interaction behavior index of hearing-impaired people when watching target videos, and calculate the second positive feedback indicator accordingly. The positive feedback comparison unit 103 is used to quantify the reduction ratio of the second positive feedback index compared with the first positive feedback index, and compare the reduction ratio with a preset threshold. The feedback difference determination unit 104 is used to obtain the historical viewing and interaction data of the target video when it is determined that the reduction ratio is greater than a preset threshold.

[0020] In this embodiment of the invention, several videos with the same background can refer to video advertisements published by the same advertiser for promoting the same type of product, or video advertisements placed on the same platform, in the same placement scenario, and at similar time periods. Preferably, "same background" is limited to advertisements of the same type of product from the same advertiser to ensure that the comparison video and the target video are highly comparable; preferably, "same background" can also be extended to advertisements of the same type of product from different advertisers, thereby enriching the comparison sample; preferably, the duration of the advertisement content of the videos with the same background is limited to a preset time range to ensure the consistency of the comparison. Several videos are at least two videos to ensure the stability of the comparison results; in another way, the number of comparison videos can be flexibly determined according to computing resources and data volume.

[0021] Purchase conversion rate can be defined as the proportion of hearing-impaired individuals who make a purchase after watching an advertising video, out of the total number of hearing-impaired viewers. In one alternative embodiment, the purchase behavior can be either a direct order or potential purchase behaviors such as adding items to the shopping cart or clicking a purchase link. Conversion rate statistics can be based on transaction data from the advertising platform, third-party statistical tools, or sales feedback data provided by the advertiser.

[0022] Standard values ​​are used to filter videos with high conversion rates to ensure the representativeness of the comparison videos. Standard values ​​can be the industry average conversion rate calculated by the platform based on historical big data, or a target conversion rate threshold set by the advertiser, such as greater than 10%, 15%, or 20%. In an adaptive approach, standard values ​​can be dynamically updated, for example, based on the mean or median conversion rates of similar product ads over the past three months.

[0023] The positive interaction behavior index can be constructed based on one or more of the following interaction data: likes, favorites, reposts, active comments, dwell time, etc. In one approach, the positive interaction behavior index can be obtained through interaction data provided by a statistical platform. In another approach, sentiment analysis of comments or bullet screens can be performed using natural language processing technology to filter out content containing positive evaluations. In a preferred approach, the positive interaction behavior index can be obtained through a weighted fusion method, for example: E = a1 × Like Rate + a2 × Favorite Rate + a3 × Share Rate + a4 × Positive Comment Ratio + a5 × Full View Rate, where a1, a2, a3, a4, and a5 are weight parameters that can be set through historical data analysis, machine learning model training, or expert experience. In another implementation, the positive interaction behavior index can be automatically generated by a deep learning model, for example, using a multimodal neural network to simultaneously input interaction behavior data and comment sentiment features, outputting a normalized positive interaction score.

[0024] The system calculates a first positive feedback index and a second positive feedback index based on a weighted fusion of purchase conversion rate and positive interaction behavior index. The weighting fusion method can be linear weighting, normalization, or a non-linear combination based on a machine learning model. In one approach, the first positive feedback index and the second positive feedback index can be expressed as: I1 = α × Conversion Rate + β × Positive Interaction Behavior Index α and β are weight parameters that can be obtained through experiments or training with historical data.

[0025] In another approach, the first positive feedback metric and the second positive feedback metric can also be automatically generated by the deep learning model based on the input purchase conversion rate and positive interaction behavior.

[0026] The reduction in the second positive feedback index compared to the first positive feedback index can be the difference between the two; alternatively, it can be the difference between the two positive feedback indexes divided by the first positive feedback index. This method can eliminate the influence of differences in the size of different control groups on the results. In a further approach, the reduction can also consider both absolute and relative values ​​to form a comprehensive evaluation index.

[0027] A preset threshold is used to determine whether the decrease in positive feedback for a target video is significant. The threshold can be a fixed value, such as 5%, 10%, or 15%, set by the platform operator or advertiser based on experience. The threshold can also be a dynamic value, for example, set based on the average decrease plus the standard deviation of historical advertising data, ensuring that the threshold can be adjusted according to industry trends and time. The significance of the threshold is: when the decrease does not exceed the threshold, it indicates that the change in positive feedback for the target video is within the normal fluctuation range; when the decrease exceeds the threshold, it indicates that the decrease in positive feedback for the target video is abnormal, and further analysis is needed to determine if there are any inappropriate advertising display effects.

[0028] Historical viewing data may include: total video viewing time, average viewing time, complete viewing rate, bounce rate, and the age / geographical distribution of the viewers. Interaction data may include: likes, favorites, shares, comments, bullet comments, and comment sentiment analysis results (proportion of positive comments, proportion of negative comments). In one approach, the data is directly provided by the backend interface of the advertising platform; in another approach, the data can be collected through event tracking technology or SDK and uploaded to the server in real time; in a further approach, comment and bullet comment data can be semantically parsed using Natural Language Processing (NLP) technology to extract negative or positive semantic tags.

[0029] By introducing the calculation of the reduction magnitude and the setting of preset thresholds in the system, this invention can effectively distinguish between normal fluctuations and abnormal drops during the review of advertising videos: when the positive feedback reduction magnitude of the target video does not exceed the preset threshold, it indicates that the feedback change of the advertising video is within the normal fluctuation range, and the system does not need to make further corrections, thereby reducing unnecessary calculations and resource consumption; when the reduction magnitude exceeds the preset threshold, the system triggers the subsequent historical viewing data and interaction data collection and analysis process, which can concentrate review resources on videos with real problems, thereby improving the efficiency and accuracy of the review. In addition, by combining viewing data and interaction data for comprehensive analysis, the system can not only discover the resistance of hearing-impaired people to advertising videos, but also further pinpoint the specific source of resistance, such as: overly exaggerated facial expressions, excessive body movements, mismatch between subtitles and content, and inappropriate brightness of the advertising area. In this way, the system can quantify anomalies while providing clear directions for correction.

[0030] Furthermore, the AI-based digital publication content review system also includes the following steps: The resistance analysis module 200 is used to analyze the historical viewing and interaction data to determine whether the resistance value of the hearing-impaired people to the display of advertisements in the target video exceeds a preset value. If so, the target video is determined to be an abnormal video.

[0031] Specifically, Figure 3 The diagram shows the structure of the conflict analysis module in an AI-based digital publication content review system.

[0032] The conflict analysis module specifically includes: The comment analysis unit 201 is used to analyze the historical viewing and interaction data, obtain negative comment data on the advertising display of the target video by hearing-impaired people, and determine the ratio of negative comments in all comments, and set the ratio as the first objection factor; The emotion detection unit 202 is used to capture and calculate the ratio of the duration of negative facial emotional state characteristics of hearing-impaired people when watching the target video to the total duration of watching the target video, and set this ratio as the second objection factor. The factor fusion unit 203 is used to perform weighted fusion of the first conflicting factor and the second conflicting factor to obtain the conflict value of the target video; The conflict determination unit 204 is used to obtain a preset value and determine whether the conflict value exceeds the preset value. If so, the target video is determined to be an abnormal video.

[0033] In this embodiment of the invention, resistance is quantified through multi-dimensional indicators to avoid misjudgment caused by relying on a single user's feedback; it objectively determines whether the video is abnormal, that is, to identify whether there are problems such as mismatch between facial expressions, actions, colors, subtitles and product content; it provides a reliable basis for subsequent correction coefficient generation and ad display reduction, thereby improving the viewing experience and ad conversion rate of hearing-impaired people.

[0034] Advertising display conditions may include: the salesperson's facial expressions (such as an exaggerated smile, raised eyebrows, or open mouth); the salesperson's body language (such as exaggerated gestures or repeated tapping of the table); the brightness, color saturation, and contrast of the advertising area; the matching degree between the subtitle content and the visual presentation in the video; the volume, rhythm, and pitch of the background music; and the frequency of camera cuts and the proportion of close-up shots.

[0035] The scope of negative comments in the comment analysis unit can include semantics such as "exaggerated," "false," "glaring," "noisy," and "incomprehensible." The methods for obtaining the primary aversion factor include: directly statistically analyzing keyword matching results in comments and bullet comments to obtain negative comments; converting comment text into vector representations based on Natural Language Processing (NLP) technology and using sentiment analysis models to determine their emotional polarity; and dividing the number of negative comments by the total number of comments to obtain the negative comment ratio, which is used as the primary aversion factor.

[0036] The conflict value can be obtained using a linear weighted method: R = α × F1 + β × F2 In this model, F1 represents the ratio of negative comments, F2 represents the ratio of negative facial emotions, and α and β are weighting coefficients. Alternatively, multiple factors can be input into a machine learning model (such as logistic regression, random forest, or neural networks), and a single resistance value can be output. This fused resistance value reflects both subjective user feedback and objective physiological responses, providing a more comprehensive picture of the degree of resistance that hearing-impaired individuals experience to advertising.

[0037] The preset values ​​can be fixed thresholds, such as 0.2 or 0.3, meaning that when negative feedback exceeds 20% or 30%, it is considered abnormal; they can also be dynamic values, such as those based on the average resistance value plus variance of historical comparison videos; or differentiated thresholds can be adopted according to different advertising industries (food, cosmetics, home appliances). This ensures that the system can automatically filter out videos that evoke significantly negative experiences among hearing-impaired individuals and proceed to the subsequent correction stage.

[0038] Furthermore, the AI-based digital publication content review system also includes the following steps: The focus calculation module 300 is used to calculate the first price focus of the hearing-impaired person when watching abnormal videos and the second price focus when watching comparison videos, based on the historical viewing data and interaction data.

[0039] Specifically, Figure 4 The diagram shows the structural block diagram of the focus calculation module in an AI-based digital publication content review system.

[0040] The focus calculation module specifically includes the following steps: The gaze behavior acquisition unit 301 is used to capture the click frequency and gaze duration of hearing-impaired people on the price tag area related to advertisements in the screen while watching the target video and abnormal video, respectively. The focus calculation unit 302 is used to calculate the first price focus and the second price focus of the hearing-impaired person when watching abnormal videos and control videos, respectively, based on the click frequency and gaze duration.

[0041] In the scenario of advertising videos, price information is often an important reference factor for deaf people to make purchase decisions. If the expressions and actions of the salesperson are overly exaggerated and the presentation of price information in the video is not prominent enough, deaf people may be distracted and miss or ignore key information.

[0042] Therefore, in the embodiments of the present invention, the degree of attention of deaf people to price information is quantified by focusing, to verify whether the video display method causes the user's attention to deviate; by comparing the differences between abnormal videos and control videos, a basis is provided for generating correction coefficients; situations where the advertising performance does not match the price information are identified as auxiliary evidence for judging whether there are unreasonable advertising display situations in the video.

[0043] The price label area is the area in the advertising video where product prices, discount information or promotional activities are presented. The price label area can be the area of the price text explicitly marked in the video; the price label area can be automatically located through an image recognition algorithm, such as identifying the subtitle area where keywords such as "￥", "yuan", "discount", "special offer" are located; the price label area can also include graphical price identifiers (such as price stickers, promotional logos, etc.); by analyzing the gaze and interaction in this area, it can be accurately evaluated whether deaf people can obtain price information.

[0044] In the gaze behavior acquisition unit, when a user is interested in a certain area, they usually click to zoom in, view more details or trigger relevant interactions. By counting the click frequency, the attractiveness and attention of price information to the user can be reflected. Deaf people mainly rely on visual information when watching advertisements. If their gaze stays in the price label area for a long time, it means that the price information has attracted attention. The gaze duration can be obtained through an eye movement tracking device, a front camera + face key point detection algorithm or the distribution of big data click heat zones. The click frequency and gaze duration respectively characterize the user's attention from the two perspectives of active interaction and passive vision. The dual indicators can more comprehensively reflect the focus of deaf people on price information.

[0045] In the focus calculation unit, in one approach, price focus can be represented by a weighted sum of click frequency and gaze duration, where the weight coefficients for click frequency and gaze duration can be obtained based on experimental results or data training. In another approach, click frequency and gaze duration can be normalized to the same dimension, added together, and then divided by the total viewing time to obtain a proportionalized index. In a further approach, multimodal data such as clicks and gaze duration can be input using a deep learning model (such as an attention mechanism network) to output a normalized price focus score. If the first price focus is significantly lower than the second price focus, it indicates that the abnormal video weakens the attractiveness of the price information, possibly because the expressions, actions, or visual effects are too prominent, causing users to ignore the price information. The focus difference provides a quantitative basis for the correction coefficient, thereby adjusting the video display effect in subsequent steps.

[0046] Furthermore, the AI-based digital publication content review system also includes the following steps: The correction coefficient generation module 400 is used to obtain a correction coefficient based on the difference between the first price focus and the second price focus, as well as the conflict value.

[0047] Specifically, Figure 5 The diagram shows the structural block diagram of the correction coefficient generation module in an AI-based digital publication content review system.

[0048] The correction coefficient generation module specifically includes: The difference calculation unit 401 is used to calculate the decrease in the first price focus compared to the second price focus, and the increase in the resistance value corresponding to the abnormal video compared to the preset value. The correction coefficient fusion unit 402 is used to weight and fuse the absolute value of the reduction magnitude with the absolute value of the increase magnitude to calculate the correction coefficient.

[0049] In this embodiment of the invention, the correction coefficient serves as a key parameter for quantifying and reducing subsequent ad display. It reflects the degree of adverse impact of abnormal videos on the hearing-impaired population. Through the correction coefficient, abnormal behaviors of two different dimensions—user attention shift and emotional resistance—can be transformed into a unified numerical indicator. The combination of the two can more comprehensively depict the negative impact of abnormal videos on the hearing-impaired population, thereby making the correction coefficient more accurate and representative. Before the ad video is published, the system can automatically determine the extent of reduction required based on the correction coefficient, thereby accurately adjusting the overly exaggerated display effect. The correction coefficient avoids misjudgment caused by relying on a single indicator, improving the reliability of anomaly identification and correction.

[0050] Based on this, the correction coefficient can be used as an input parameter for subsequent display reduction steps to automatically adjust dimensions such as facial expression range, body movement range, screen brightness, color saturation, and subtitle dynamic effects in the video; or it can be fed back to the advertiser in the form of annotations to guide manual modifications.

[0051] Specifically, if the price focus of hearing-impaired individuals in the abnormal video is significantly lower than that in the control video, it indicates that their attention is distracted by the exaggerated presentation by the salesperson, leading to a decrease in their focus on price information; this decrease can quantify the degree to which the advertising performance inhibits the acquisition of key information. If the resistance value is higher than a preset threshold, it indicates that the advertisement has triggered stronger discomfort or rejection among hearing-impaired individuals; this increase reflects the negative impact of the advertising presentation method on the overall viewing experience.

[0052] In the correction coefficient fusion unit, the correction coefficient can be expressed as a weighted sum of the reduction and the increase. In another approach, a nonlinear fusion method can be used, such as combining the reduction and the increase through logistic regression or a neural network, to output the correction coefficient.

[0053] Furthermore, the AI-based digital publication content review system also includes the following steps: The display correction module 500 is used to quantitatively reduce the advertising display of abnormal videos based on the correction coefficient during subsequent review of abnormal videos.

[0054] Specifically, Figure 6 The diagram shows the structure of the correction module in an AI-based digital publication content review system.

[0055] The display correction module specifically includes: The display reduction unit 501 is used to quantitatively reduce the advertising display of the abnormal video according to the correction coefficient before the subsequent abnormal video is released, and generate a corrected video. The video publishing unit 502 is used to deliver and publish the corrected video.

[0056] In actual ad video delivery, the same ad is often run multiple times. This invention identifies abnormal videos and generates correction coefficients after the first round or previous runs through preliminary steps. Before the abnormal video is run again, the display correction module performs quantitative reduction, ensuring that the display level of the abnormal video matches the experience of the target audience (especially the hearing impaired). In this embodiment, an operable reduction range is given based on real audience data (such as price focus differences, increased resistance), and targeted correction is completed before the abnormal video is rerun, reducing bounce and negative reviews caused by negative experiences and improving purchase conversion rates. In this embodiment, correction is only performed on videos determined to be "abnormal," avoiding over-processing of normal videos.

[0057] Furthermore, the ad impressions that can be quantified and reduced in abnormal videos include: Image layer: brightness, color saturation, contrast, close-up ratio, image switching frequency, flickering or motion effect frequency, subtitle dynamic range, screen ratio or visible duration of price tags; People layer: facial expression range, body movement range, frequency of repetition of movements, proportion of close-ups or extreme close-ups; Sound layer: speech rate, background music rhythm and intensity (even when targeting the hearing impaired, the platform can also make slight adjustments for the hearing audience).

[0058] In this embodiment of the invention, each dimension is abstracted as a vector F = [F1, ..., F2]. m (e.g., brightness, saturation, close-up ratio, facial expression range, etc.), set dimension weights w=[w1, ..., w m (This can be obtained from platform experience or training), with correction coefficients K∈[0,1] (or constrained to this interval through linear / piecewise mapping). Common forms of quantitative reduction in ad display include: Proportional reduction; differential reduction; segmented reduction.

[0059] It should be understood that although the steps in the structural block diagrams of the various embodiments of the present invention are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the various embodiments may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least a portion of the sub-steps or stages of other steps.

[0060] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments described above. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0061] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0062] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of this patent should be determined by the appended claims.

[0063] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A digital publication content review system based on artificial intelligence, characterized in that: The system includes: The feedback determination module is used to determine whether the reduction in positive feedback of the target video compared to the control video in the hearing-impaired population exceeds a preset threshold. If so, the historical viewing and interaction data of the target video and the control video are obtained. The resistance analysis module is used to analyze the historical viewing and interaction data to determine whether the resistance value of hearing-impaired people to the display of advertisements in the target video exceeds a preset value. If so, the target video is determined to be an abnormal video. The focus calculation module is used to calculate the first price focus of hearing-impaired people when watching abnormal videos and the second price focus when watching comparison videos, based on the historical viewing data and interaction data. The correction coefficient generation module is used to obtain a correction coefficient based on the difference between the first price focus and the second price focus, as well as the conflict value. The display correction module is used to quantitatively reduce the display of advertisements in abnormal videos based on the correction coefficient during subsequent review of abnormal videos.

2. The AI-based digital publication content review system according to claim 1, characterized in that, The feedback determination module specifically includes: The first indicator calculation unit is used to select several videos with the same background and a purchase conversion rate higher than the standard value as control videos, obtain the purchase conversion rate and positive interaction behavior index of hearing-impaired people when watching the control videos, and calculate the first positive feedback indicator accordingly. The second indicator calculation unit is used to obtain the purchase conversion rate and positive interaction behavior index of hearing-impaired people when watching target videos, and to calculate the second positive feedback indicator accordingly. A positive feedback comparison unit is used to quantify the reduction ratio of the second positive feedback index compared with the first positive feedback index, and compare the reduction ratio with a preset threshold. The feedback difference determination unit is used to obtain the historical viewing and interaction data of the target video when it is determined that the reduction ratio is greater than a preset threshold.

3. The AI-based digital publication content review system according to claim 2, characterized in that, The first positive feedback index and the second positive feedback index can be expressed as: I1 = α × Conversion Rate + β × Positive Interaction Behavior Index Where α and β are weighting parameters.

4. The AI-based digital publication content review system according to claim 2, characterized in that, The first positive feedback metric and the second positive feedback metric are automatically generated by the deep learning model based on the input purchase conversion rate and positive interaction behavior.

5. The AI-based digital publication content review system according to claim 1, characterized in that, The conflict analysis module specifically includes: The comment analysis unit is used to parse the historical viewing and interaction data, obtain negative comment data on the advertising display of the target video by hearing-impaired people, and determine the ratio of negative comments in all comments, and set the ratio as the first objection factor; An emotion detection unit is used to capture and calculate the ratio of the duration of negative facial emotional state characteristics of hearing-impaired people when watching advertising content in the target video to the total duration of watching advertising content in the target video, and this ratio is set as the second aversion factor. The factor fusion unit is used to weight and fuse the first and second conflicting factors to obtain the conflict value of the target video; The conflict determination unit is used to obtain a preset value and determine whether the conflict value exceeds the preset value. If so, the target video is determined to be an abnormal video.

6. The AI-based digital publication content review system according to claim 1, characterized in that, The focus calculation module specifically includes: The gaze behavior acquisition unit is used to capture the click frequency and gaze duration of hearing-impaired people on the price tag area related to advertisements in the screen while watching target videos and abnormal videos, respectively. The focus calculation unit is used to calculate the first price focus and the second price focus of the hearing-impaired person when watching abnormal videos and control videos, respectively, based on the click frequency and gaze duration.

7. The AI-based digital publication content review system according to claim 5, characterized in that, The correction coefficient generation module specifically includes: The difference calculation unit is used to calculate the decrease in the first price focus compared to the second price focus, and the increase in the resistance value corresponding to the abnormal video compared to the preset value. The correction coefficient fusion unit is used to weight and fuse the absolute values ​​of the reduction and increase to calculate the correction coefficient.

8. The AI-based digital publication content review system according to claim 1, characterized in that, The display correction module specifically includes: The display reduction unit is used to quantitatively reduce the advertising display of the abnormal video based on a correction coefficient before the subsequent abnormal video is released, and generate a corrected video. The video publishing unit is used to deliver and publish the corrected video.

9. The digital publication content review system based on artificial intelligence according to claim 8, characterized in that, The advertising display includes reducing the brightness of the advertising area based on a correction factor.

10. The AI-based digital publication content review system according to claim 8, characterized in that, The advertising display process reduces the range of body language of the advertising salesperson based on a correction factor.

Images