Short video information stream advertisement immersive lightweight intelligent implantation optimization method and system thereof

Abstract

This invention discloses a method and system for optimizing immersive, lightweight, and intelligent embedding of short video feed ads, belonging to the field of advertising technology. First, it collects short video data and extracts features, while simultaneously capturing user status and outputting real-time user profile data. Then, it filters ads, performs compliance screening and weight matching, generates embedding plans, calibrates time slots, and outputs precise embedding plans. Next, it performs lightweight processing, noise reduction and purification, point-to-point embedding, image blending, and interactive adaptation on the ad materials, outputting the final optimized immersive short video ad product. Finally, it collects and scores the ad delivery results, and iteratively optimizes parameters based on the scores. Based on this, the invention significantly reduces the size of ad materials, substantially lowers device computing power and bandwidth resource consumption, and fundamentally improves the problems of loading delays, frame drops, and stuttering in short video ads.

Description

Technical Field

[0001] This invention relates to the field of advertising technology, specifically to a method and system for optimizing immersive, lightweight, and intelligent embedding of short video feed ads. Background Technology

[0002] With the rapid development of short video platforms, in-feed advertising has become a core carrier of internet marketing. Users' demands for the short video viewing experience continue to rise, and traditional hard-sell ads and pop-up ads are prone to causing user resistance. Immersive, seamless, and low-load lightweight ad placement has become the mainstream development trend in the industry.

[0003] Currently, most ad integration technologies in the industry only achieve basic ad placement functionality, lacking a multi-dimensional collaborative optimization mechanism for video scenes, user status, and ad creatives, and generally suffer from the following problems:

[0004] 1. Advertising creatives are not optimized for lightweighting, resulting in redundant loading and stuttering: Existing ad placements directly use original high-definition creatives without compression or redundancy removal optimizations for the lightweight dissemination scenarios of short video feeds. High-definition creatives consume significant bandwidth and device computing power, easily leading to stuttering and loading delays in short video playback. Furthermore, conventional compression methods are prone to producing image noise and watermark residue, making it impossible to simultaneously achieve lightweight design and image clarity. For example, a 4K ultra-high-definition car advertisement creative (original size approximately 200MB) directly embedded in a 15-second short video will still experience frame drops in a 5G environment, with stuttering rates exceeding 30% on low-end mobile devices; conventionally compressed ad creatives also exhibit noticeable grainy noise.

[0005] 2. Abrupt ad insertion and lack of immersive experience: Existing technologies often employ a fixed-location, fixed-duration forced insertion model, failing to adapt to scene characteristics such as video transitions, blank frames, and color tones. Furthermore, they do not accommodate real-time user interactions like swiping and pausing, leading to conflicts between ad display and user actions, easily causing users to skip or become averse to the ad. For example, a skincare product ad with a solid-color background is forcibly inserted midway through a short food review video. The color tone and content theme clash with the food scene, forcing the ad to remain displayed when the user swipes, resulting in an ad skip rate as high as 65%. Summary of the Invention

[0006] To address the technical issues of un-lightweighted advertising materials, resulting in redundant loading, lag, awkward integration, and a lack of immersive experience, this invention provides the following technical solution:

[0007] A method for optimizing immersive, lightweight, and intelligent integration of short video feed ads, comprising the following steps:

[0008] S1, Full-Domain Data Analysis: Collects short video data and extracts features, while capturing user status and outputting real-time user profile data;

[0009] S2, Intelligent Matching Modeling: Based on real-time user profile data, it filters advertisements, performs compliance screening and weight matching, generates an insertion plan and calibrates the time period, and outputs a precise insertion plan;

[0010] S3, lightweight immersive implantation:

[0011] S31, Lightweight Material Processing: Based on the precise insertion solution, the bitrate of the advertising materials is controlled, redundancy is removed and the encoding is compressed to output lightweight advertising materials;

[0012] S32, Layered Noise Reduction and Purification: Based on lightweight advertising materials, noise reduction filtering, color restoration and watermark removal are performed to output high-definition, clean and lightweight advertising materials;

[0013] S33, Precise Placement: Based on high-definition, clean, and lightweight advertising materials, it detects transition frames and blank frames, and places the advertisement into low-interference locations to complete the initial placement.

[0014] S34, Immersive Fusion Optimization: Based on the short video that has been initially mounted and embedded, adjust the color tone, brightness and transparency of the advertisement to blend it with the original image and output a basic immersive short video advertisement product.

[0015] S35, Weak Interaction Adaptive Configuration: Based on the basic immersive short video ad product, preset sliding, pause, and fast-forward trigger rules, weaken or pause the ad when the user operates, and automatically resume after the operation, outputting the final optimized immersive short video ad product.

[0016] S4, end-to-end dynamic iteration: Based on the final optimized immersive advertising short video product, collect and score the delivery effect, and iteratively optimize parameters based on the score results.

[0017] As a preferred embodiment of the immersive, lightweight, and intelligent implantation optimization method for short video feed ads according to the present invention, the specific steps of S1 are as follows:

[0018] S11, Collect Information Flow Data: Collect information flow data from the entire short video domain and platform distribution data, and output the original information flow dataset of short videos;

[0019] S12, Extracting video features: Based on the original short video information stream dataset, video features are extracted through frame differencing and semantic recognition, and a multi-dimensional feature parameter set of the video is output.

[0020] S13, Capture User Status: Based on the multi-dimensional feature parameter set of video, collect user viewing time, preferences, interactive behavior and consumption attributes, and output real-time user profile data.

[0021] As a preferred embodiment of the immersive, lightweight, and intelligent implantation optimization method for short video feed ads according to the present invention, the specific steps of S2 are as follows:

[0022] S21, Filtering Ads for Matching: Based on real-time user profile data and combined with a multi-dimensional feature parameter set of the video, filter a set of matching candidate ad creatives;

[0023] S22, Risk Control and Compliance Screening: Based on the candidate ad creative set, it connects to the compliance rule library, eliminates non-compliant creatives, and outputs a subset of compliant candidate ads;

[0024] S23, Construct Matching Weights: Based on a subset of compliant candidate ads, construct a matching weight model from the perspectives of scenario fit, user preference, and creative suitability, and output an ad weight scoring table;

[0025] S24, Generate an integration plan: Based on the ad weight scoring table, select the best ad and generate a preliminary integration plan that includes ad creatives, integration time period and location;

[0026] S25, Time-Based Adaptive Calibration: Based on the initial implantation plan, combined with traffic popularity and user activity data, the implantation time period is calibrated to output a precise implantation plan.

[0027] As a preferred embodiment of the immersive, lightweight, and intelligent implantation optimization method for short video feed ads described in this invention, the specific steps of S4 are as follows:

[0028] S41, Collect campaign performance data: Based on the final optimized immersive short video ad, collect user interaction data and output the raw dataset of campaign performance data;

[0029] S42, Quantitative Embedding Score: Based on the original dataset of campaign performance, scores are given from four dimensions: loading efficiency, immersive experience, conversion effect, and compliance, and the performance score results are output.

[0030] S43, Iterative model optimization: Based on the effect scoring results, reverse the weight coefficients, screening thresholds, time period parameters, as well as the compression coefficients, noise reduction parameters, and interaction trigger thresholds, and update the model database.

[0031] A lightweight, intelligent embedded optimization system for immersive short video feed ads includes:

[0032] The full-domain data analysis module collects short video data and extracts features, while capturing user status and outputting real-time user profile data.

[0033] The intelligent matching and modeling module filters ads based on real-time user profile data, performs compliance screening and weight matching, generates an insertion plan and calibrates the time period, and outputs a precise insertion plan.

[0034] The lightweight immersive embedding module includes a material lightweighting processing unit, a hierarchical noise reduction and purification unit, a precise point embedding unit, an immersive blending and optimization unit, and a weak interaction adaptive configuration unit.

[0035] The lightweight material processing unit, based on the precise implantation scheme, performs bitrate control, redundancy removal, and encoding compression on the advertising materials, and outputs lightweight advertising materials.

[0036] The hierarchical noise reduction and purification unit performs noise reduction filtering, color restoration and watermark removal based on lightweight advertising materials, and outputs high-definition, clean and lightweight advertising materials.

[0037] The precise placement unit, based on high-definition, clean, and lightweight advertising materials, detects transition frames and blank frames, and places the advertisement into low-interference locations to complete the initial placement.

[0038] The immersive blending and optimization unit adjusts the color tone, brightness, and transparency of the advertisement based on the short video that has been initially mounted and embedded, blends it with the original image, and outputs a basic immersive advertisement short video product.

[0039] The weak interaction adaptive configuration unit is based on the basic immersive advertising short video product, and presets sliding, pausing, and fast-forward trigger rules. When the user operates, the advertisement is weakened or paused, and automatically resumed after the operation, outputting the final optimized immersive advertising short video product.

[0040] The end-to-end dynamic iteration module collects and scores the campaign performance based on the final optimized immersive short video ad, and iteratively optimizes parameters based on the scoring results.

[0041] As a preferred embodiment of the immersive, lightweight, and intelligent embedded optimization system for short video feed ads described in this invention, the full-domain data parsing module includes:

[0042] The data collection unit collects information flow data from the entire short video information flow and platform distribution data, and outputs the original short video information flow dataset.

[0043] Extract video feature units: Based on the original short video information stream dataset, extract video features through frame differencing and semantic recognition, and output a multi-dimensional video feature parameter set;

[0044] Capture user status units, collect user viewing time, preferences, interactive behaviors and consumption attributes based on video multidimensional feature parameter set, and output real-time user profile data.

[0045] As a preferred embodiment of the immersive, lightweight, and intelligent implantation optimization system for short video feed ads described in this invention, the intelligent matching modeling module includes:

[0046] Filter suitable ad units based on real-time user profile data and combined with multi-dimensional video feature parameter sets to filter a set of matching candidate ad creatives;

[0047] The risk control and compliance screening unit, based on the candidate advertising material set, connects to the compliance rule library, eliminates non-compliant materials, and outputs a subset of compliant candidate advertisements;

[0048] Construct a matching weight unit, and based on a subset of compliant candidate ads, build a matching weight model from scenario fit, user preference, and creative suitability, and output an ad weight score table;

[0049] Generate an integration plan unit, select the best ad based on the ad weight scoring table, and generate a preliminary integration plan that includes ad creative, integration time period and location;

[0050] The time-adaptive calibration unit, based on the initial implantation plan, combines traffic popularity and user activity data to calibrate the implantation time period and output a precise implantation plan.

[0051] As a preferred embodiment of the immersive, lightweight, and intelligent implantation optimization system for short video feed ads described in this invention, the full-link dynamic iteration module includes:

[0052] The campaign performance data collection unit collects user interaction data based on the final optimized immersive short video ad and outputs the raw dataset of campaign performance.

[0053] The quantitative embedding scoring unit scores the campaign based on the original dataset of campaign performance, from four dimensions: loading efficiency, immersive experience, conversion rate, and compliance, and outputs the performance score results.

[0054] The model units are iteratively optimized. Based on the performance score results, the weight coefficients, screening thresholds, time period parameters, compression coefficients, noise reduction parameters, and interaction trigger thresholds are adjusted in reverse, and the model database is updated.

[0055] Compared with existing technologies:

[0056] 1. By combining lossless compression, redundant pixel removal, dynamic bitrate adaptation, and hierarchical noise reduction and purification into an integrated lightweight processing method, the size of advertising materials is greatly reduced, and the consumption of device computing power and bandwidth resources is significantly reduced. This fundamentally improves the problems of loading delay, frame drop, and screen stuttering in short video ads, while also repairing noise and color distortion defects caused by compression, thus balancing the needs of lightweight material loading with high-definition visual display effects.

[0057] 2. Through an immersive optimization strategy that involves precise insertion of low-interference video elements, synchronized blending of color tone parameters, and dynamic adaptation to weak user interaction, the advertising materials are seamlessly integrated with the original video visuals, rhythm, and scenes. This completely eliminates the sense of visual disconnect and display conflict between the advertisement and the native video, avoids user aversion and skipping behavior, and comprehensively enhances the naturalness of short video ad insertion and the immersive viewing experience for users. Attached Figure Description

[0058] Figure 1 This is a schematic diagram of the overall framework of the present invention;

[0059] Figure 2 This is a schematic diagram of the global data parsing module framework of the present invention;

[0060] Figure 3 This is a schematic diagram of the intelligent matching modeling module framework of the present invention;

[0061] Figure 4 This is a schematic diagram of the lightweight immersive implantation module framework of the present invention;

[0062] Figure 5 This is a schematic diagram of the full-link dynamic iteration module framework of the present invention. Detailed Implementation

[0063] To make the objectives, technical solutions, and advantages of the present invention clearer, the embodiments of the present invention will be described in further detail below with reference to the accompanying drawings.

[0064] This invention provides an immersive, lightweight, and intelligent embedding optimization method for short video feed ads. Please refer to [link / reference]. Figures 1-5 This includes the following steps:

[0065] S1, Full-Domain Data Analysis: Collects short video data and extracts features, while capturing user status and outputting real-time user profile data;

[0066] S11, Collect Information Flow Data: Collect full-domain information flow data of the short videos to be launched in real time, including video frame sequences, audio content, video duration, scene tags, content themes, and publishing categories, as well as basic platform traffic distribution data, and output the original information flow dataset of the short videos.

[0067] S12, Extracting Video Features: Based on the original short video information stream dataset, the frame difference algorithm is used to detect video shot transitions and motion intensity; the pre-trained semantic recognition model is used to extract features such as scene category (e.g., food, travel, emotion), main color tone, core objects, and pace, and outputs a multi-dimensional feature parameter set of the video.

[0068] S13, Capture User Status: Based on the multi-dimensional feature parameter set of the video, match user data in the corresponding playback scenario, collect user viewing time, historical browsing preferences (such as ad types such as likes, favorites, and shares), interactive behaviors (skip, clicks, dwell time), and consumption attributes (such as gender, age, and interest tags) in real time, and output real-time user profile data.

[0069] S2, Intelligent Matching Modeling: Based on real-time user profile data, it filters advertisements, performs compliance screening and weight matching, generates an insertion plan and calibrates the time period, and outputs a precise insertion plan;

[0070] S21, Filtering Ads for Matching: Based on real-time user profile data and combined with the multi-dimensional feature parameter set of videos, the ad creative library is traversed to calculate the matching degree between category, style, tone and video scene and user preferences, and a set of candidate ad creatives with a matching degree higher than the preset threshold is selected.

[0071] S22, Risk Control and Compliance Screening: Based on the candidate ad creative set, it connects to the platform's ad placement compliance rule library (including prohibited words, restricted industries, qualification requirements, etc.), verifies the placement qualifications, content compliance, and scenario adaptation permissions of each candidate creative, eliminates illegal, restricted, and incomplete creative materials, and outputs a compliant candidate ad subset.

[0072] S23, Construct Matching Weights: Based on a subset of compliant candidate ads, construct a matching weight model from three dimensions: scenario fit (semantic similarity between ad theme and video scene), user preference (acceptance of this type of ad in users' historical interactions), and material suitability (the degree of matching between ad duration, color tone, and video rhythm). Calculate a comprehensive weight score for each candidate ad and output an ad weight scoring table.

[0073] S24, Generate Insertion Plan: Based on the ad weight scoring table, select the ad creative with the highest weight score, and combine it with video length and visual rhythm to generate a preliminary insertion plan that includes ad creative, initial insertion time, and initial insertion point. Initial insertion points are prioritized in low-interference areas such as video transition frames, blank screens, and content transition nodes.

[0074] S25, Time-Based Adaptive Calibration: Based on the initial placement plan, retrieve real-time traffic and peak user activity data from the platform, and combine this with the short video posting time attributes (such as weekdays / weekends, morning and evening peak hours) to dynamically adjust the start time and display duration of ad placement, avoiding placement during low traffic and low activity periods, and outputting a precise placement plan optimized for the time period.

[0075] S3, Lightweight Immersive Integration: Based on a precise integration solution, advertising materials are lightweighted, noise-reduced and purified, embedded at specific locations, blended and adapted to interactive features, and the final optimized immersive short video advertising product is output.

[0076] S31, Lightweight Creative Processing: Based on a precise integration solution, intelligent optimization is performed on target ad creatives. Adaptive bitrate control technology dynamically allocates bitrate according to video content complexity and device playback capabilities; redundant pixel removal algorithms remove excess data from static background areas in the ad; and high-efficiency video encoding formats (such as HEVC) are used for compression, reducing the ad creative size to less than 20% of its original size, resulting in lightweight ad creatives.

[0077] S32, Layered Noise Reduction and Cleaning: Based on lightweight advertising materials, it addresses issues such as image noise, color distortion, and residual watermarks caused by compression by employing pixel-level adaptive noise reduction filters (such as median filtering combined with edge-preserving filtering) to smooth high-frequency noise in the image; it uses color restoration algorithms to adjust color-shifted areas; and it removes residual watermarks through watermark detection and background filling technology, outputting high-definition, clean, and lightweight advertising materials.

[0078] S33, Precise Placement: Based on high-definition, clean, and lightweight ad creatives, further analysis is conducted to pinpoint the precise locations of low-interference elements in the video. Transition frames are detected using inter-frame difference thresholds, and blank frames or content transition nodes are identified by calculating the pixel change rate. Ad creatives are then overlaid onto these locations, ensuring that the ad's start time matches the video's rhythm, thus completing the initial placement.

[0079] S34, Immersive Fusion Optimization: Based on the short video that has been initially embedded, the overall color tone, average brightness, and contrast distribution of the original video are extracted. The color tone, brightness, and transparency parameters of the advertising material are adjusted simultaneously to make the advertisement and the surrounding original image transition naturally, eliminate the sense of disjointedness, and output a basic immersive advertising short video product.

[0080] S35, Weak Interaction Adaptive Configuration: Based on the existing immersive short video ad, it pre-sets weak interaction trigger rules such as user swiping, pausing, and fast-forwarding. When a user swipes the screen, the ad transparency automatically reduces to 30% and stops playing; when the user pauses the video, the ad pauses and maintains the current frame display; when the user fast-forwards, the ad automatically skips. All interactive operations automatically resume normal ad display within 2 seconds after the user stops interacting, outputting the final optimized immersive short video ad.

[0081] S4, End-to-End Dynamic Iteration: Based on the final optimized immersive advertising short video product, collect and score the delivery effect, and iteratively optimize parameters based on the score results;

[0082] S41, Collection of campaign performance: Based on the final optimized immersive short video ad, after launch, real-time collection of full-link interaction data such as user skips, clicks, dwell time, likes, favorites, comments, and conversions (such as purchases and registrations) is performed, and the raw dataset of campaign performance is output.

[0083] S42, Quantitative Embedding Score: Based on the original dataset of campaign performance, a comprehensive scoring model is constructed from four dimensions: lightweight loading efficiency (indicators of first frame loading time and stuttering rate), immersive experience (indicators of user skip rate and dwell time ratio), ad conversion performance (indicators of click-through rate and conversion rate), and campaign compliance (indicators of whether platform risk control is triggered). The score for each dimension is calculated and weighted to sum the results, and the performance score is output.

[0084] S43, Iterative Model Optimization: Based on the performance scoring results, the weight coefficients of each dimension of the matching weight model, compliance screening thresholds, time period calibration parameters, lightweight compression coefficients, noise reduction parameters, and interaction adaptation trigger thresholds are corrected in reverse. The corrected parameters are then updated in the ad matching and insertion model database to provide more accurate model support for subsequent ad delivery, forming a closed-loop mechanism of continuous iteration.

[0085] A lightweight, intelligent embedded optimization system for immersive short video feed ads includes:

[0086] The full-domain data analysis module collects short video data and extracts features, while capturing user status and outputting real-time user profile data.

[0087] The data collection unit collects real-time information flow data of the short videos to be published, including raw video data such as video frame sequences, audio content, video duration, scene tags, content themes, and publishing categories. At the same time, it collects basic data on platform traffic distribution and outputs raw information flow datasets of short videos.

[0088] Video feature units are extracted based on the original short video information stream dataset. The frame difference algorithm is used to detect video shot transitions and motion intensity. Features such as scene category (e.g., food, travel, emotion), main color tone, core objects, and pace are extracted through a pre-trained semantic recognition model, and a multi-dimensional feature parameter set of the video is output.

[0089] Capture user state units, match user data in the corresponding playback scenario based on the multi-dimensional feature parameter set of the video, and collect user viewing time, historical browsing preferences (such as ad types such as likes, favorites, and shares), interactive behaviors (skip, clicks, dwell time), and consumption attributes (such as gender, age, and interest tags) in real time, and output real-time user profile data.

[0090] The intelligent matching and modeling module filters ads based on real-time user profile data, performs compliance screening and weight matching, generates an insertion plan and calibrates the time period, and outputs a precise insertion plan.

[0091] The system filters and selects suitable ad units based on real-time user profile data and combined with multi-dimensional video feature parameter sets. It traverses the ad creative library, calculates the degree of matching between category, style, tone and video scene and user preferences, and selects a set of candidate ad creatives with a matching degree higher than a preset threshold.

[0092] The risk control and compliance screening unit, based on the candidate advertising material set, connects to the platform's advertising placement compliance rule library (including prohibited words, restricted industries, qualification requirements, etc.), verifies the placement qualifications, content compliance, and scenario adaptability of each candidate material, eliminates materials that violate regulations, are restricted, or lack complete qualifications, and outputs a subset of compliant candidate advertisements.

[0093] Construct a matching weight unit. Based on a subset of compliant candidate ads, build a matching weight model from three dimensions: scenario fit (semantic similarity between ad theme and video scene), user preference (acceptance of this type of ad in users' historical interactions), and material fit (the degree of matching between ad duration, color tone and video rhythm). Calculate a comprehensive weight score for each candidate ad and output an ad weight score table.

[0094] The embedded placement plan unit generates an initial embedded placement plan based on the ad weight scoring table. It selects the ad creative with the highest weight score and combines this with video length and visual pacing to generate the initial embedded placement plan, which includes the ad creative, initial placement time, and initial placement location. Initial placement locations prioritize low-interference areas such as video transition frames, blank screens, and content transition nodes.

[0095] The time-adaptive calibration unit, based on the initial placement plan, retrieves real-time traffic and peak user activity data from the platform, and combines this with the short video posting time attributes (such as weekdays / weekends, morning and evening peak hours) to dynamically adjust the start time and display duration of ad placement, avoiding placement during low traffic and low activity periods, and outputting a precise placement plan optimized for the time period.

[0096] The lightweight immersive integration module, based on a precise integration solution, performs lightweight processing, noise reduction and purification, point-to-point integration, image blending and interactive adaptation on advertising materials, and outputs the final optimized immersive advertising short video product.

[0097] The lightweight creative processing unit intelligently optimizes target ad creatives based on a precise embedding solution. It employs adaptive bitrate control technology to dynamically allocate bitrate according to video content complexity and device playback capabilities; it removes redundant data from static background areas in the ad through a redundant pixel removal algorithm; and it uses efficient video encoding formats (such as HEVC) for compression, reducing the ad creative size to less than 20% of its original size, resulting in lightweight ad creatives.

[0098] The hierarchical noise reduction and purification unit, based on lightweight advertising materials, addresses issues such as image noise, color distortion, and residual watermarks caused by compression. It employs pixel-level adaptive noise reduction filters (such as median filtering combined with edge-preserving filtering) to smooth high-frequency noise in the image; it uses color restoration algorithms to adjust color-shifted areas; and it removes residual watermarks through watermark detection and background filling technology, outputting high-definition, clean, and lightweight advertising materials.

[0099] The precise placement unit, based on high-definition, clean, and lightweight ad creatives, further analyzes the precise locations of low-interference points in the video. It detects transition frames by using inter-frame difference thresholds, identifies blank frames or content transition nodes by calculating the pixel change rate, and overlays the ad creatives onto these locations, ensuring the ad's start time matches the video's rhythm, thus completing the initial placement.

[0100] The immersive integration and optimization unit extracts the overall tone, average brightness, and contrast distribution of the original video footage based on the short video that has been initially embedded. It then simultaneously adjusts the tone, brightness, and transparency parameters of the advertising material to make the advertisement visually transition naturally with the surrounding original footage, eliminating any sense of disjointedness and outputting a basic immersive advertising short video product.

[0101] The weak-interaction adaptive configuration unit, based on the existing immersive short video ad, pre-sets weak-interaction trigger rules such as user swiping, pausing, and fast-forwarding. When a user swipes the screen, the ad transparency automatically reduces to 30% and stops playing; when the user pauses the video, the ad pauses and maintains the current frame display; when the user fast-forwards, the ad is automatically skipped. All interactive operations automatically resume normal ad display within 2 seconds after the user stops interacting, outputting the final optimized immersive short video ad.

[0102] The end-to-end dynamic iteration module collects and scores the campaign performance based on the final optimized immersive short video ad, and iteratively optimizes parameters based on the scoring results.

[0103] The campaign performance data collection unit, based on the final optimized immersive short video ad, collects real-time user interaction data in real time after launch, including skips, clicks, dwell time, likes, favorites, comments, and conversions (such as purchases and registrations), and outputs the raw dataset of campaign performance data.

[0104] The quantitative embedding scoring unit constructs a comprehensive scoring model from four dimensions based on the original dataset of campaign performance: lightweight loading efficiency (indicating first frame loading time and stuttering rate), immersive experience (indicating user skip rate and dwell time ratio), ad conversion performance (indicating click-through rate and conversion rate), and campaign compliance (indicating whether platform risk control is triggered). The score for each dimension is calculated and weighted to sum the results, and the performance score is output.

[0105] The model units are iteratively optimized. Based on the performance scoring results, the weight coefficients of each dimension of the matching weight model, compliance screening thresholds, time-period calibration parameters, lightweight compression coefficients, noise reduction parameters, and interaction adaptation trigger thresholds are corrected in reverse. The corrected parameters are then updated in the ad matching and embedding model database to provide more accurate model support for subsequent ad delivery, forming a closed-loop mechanism of continuous iteration.

[0106] Although the present invention has been described above with reference to embodiments, various modifications can be made and components can be replaced with equivalents without departing from the scope of the invention. In particular, as long as there is no structural conflict, the features in the disclosed embodiments can be combined with each other in any manner. The lack of an exhaustive description of these combinations in this specification is merely for the sake of brevity and resource conservation. Therefore, the present invention is not limited to the specific embodiments disclosed herein, but includes all technical solutions falling within the scope of the claims.

Claims

1. A method for optimizing immersive, lightweight, and intelligent embedding of short video feed ads, characterized in that: Includes the following steps: S1, Full-Domain Data Analysis: Collects short video data and extracts features, while capturing user status and outputting real-time user profile data; S2, Intelligent Matching Modeling: Based on real-time user profile data, it filters advertisements, performs compliance screening and weight matching, generates an insertion plan and calibrates the time period, and outputs a precise insertion plan; S3, lightweight immersive implantation: S31, Lightweight Material Processing: Based on the precise insertion solution, the bitrate of the advertising materials is controlled, redundancy is removed and the encoding is compressed to output lightweight advertising materials; S32, Layered Noise Reduction and Purification: Based on lightweight advertising materials, noise reduction filtering, color restoration and watermark removal are performed to output high-definition, clean and lightweight advertising materials; S33, Precise Placement: Based on high-definition, clean, and lightweight advertising materials, it detects transition frames and blank frames, and places the advertisement into low-interference locations to complete the initial placement. S34, Immersive Fusion Optimization: Based on the short video that has been initially mounted and embedded, adjust the color tone, brightness and transparency of the advertisement to blend it with the original image and output a basic immersive short video advertisement product. S35, Weak Interaction Adaptive Configuration: Based on the basic immersive short video ad product, preset sliding, pause, and fast-forward trigger rules, weaken or pause the ad when the user operates, and automatically resume after the operation, outputting the final optimized immersive short video ad product. S4, end-to-end dynamic iteration: Based on the final optimized immersive advertising short video product, collect and score the delivery effect, and iteratively optimize parameters based on the score results.

2. The immersive, lightweight, intelligent embedding optimization method and system for short video feed ads according to claim 1, characterized in that, The specific steps of S1 are as follows: S11, Collect Information Flow Data: Collect information flow data from the entire short video domain and platform distribution data, and output the original information flow dataset of short videos; S12, Extracting video features: Based on the original short video information stream dataset, video features are extracted through frame differencing and semantic recognition, and a multi-dimensional feature parameter set of the video is output. S13, Capture User Status: Based on the multi-dimensional feature parameter set of video, collect user viewing time, preferences, interactive behavior and consumption attributes, and output real-time user profile data.

3. The immersive, lightweight, and intelligent embedding optimization method for short video feed ads according to claim 1, characterized in that, The specific steps of S2 are as follows: S21, Filtering Ads for Matching: Based on real-time user profile data and combined with a multi-dimensional feature parameter set of the video, filter a set of matching candidate ad creatives; S22, Risk Control and Compliance Screening: Based on the candidate ad creative set, it connects to the compliance rule library, eliminates non-compliant creatives, and outputs a subset of compliant candidate ads; S23, Construct Matching Weights: Based on a subset of compliant candidate ads, construct a matching weight model from the perspectives of scenario fit, user preference, and creative suitability, and output an ad weight scoring table; S24, Generate an integration plan: Based on the ad weight scoring table, select the best ad and generate a preliminary integration plan that includes ad creatives, integration time period and location; S25, Time-Based Adaptive Calibration: Based on the initial implantation plan, combined with traffic popularity and user activity data, the implantation time period is calibrated to output a precise implantation plan.

4. The immersive, lightweight, and intelligent embedding optimization method for short video feed ads according to claim 1, characterized in that, The specific steps of S4 are as follows: S41, Collect campaign performance data: Based on the final optimized immersive short video ad, collect user interaction data and output the raw dataset of campaign performance data; S42, Quantitative Embedding Score: Based on the original dataset of campaign performance, scores are given from four dimensions: loading efficiency, immersive experience, conversion effect, and compliance, and the performance score results are output. S43, Iterative model optimization: Based on the effect scoring results, reverse the weight coefficients, screening thresholds, time period parameters, as well as the compression coefficients, noise reduction parameters, and interaction trigger thresholds, and update the model database.

5. A lightweight, intelligent, immersive embedding optimization system for short video feed ads, characterized in that: include: The full-domain data analysis module collects short video data and extracts features, while capturing user status and outputting real-time user profile data. The intelligent matching and modeling module filters ads based on real-time user profile data, performs compliance screening and weight matching, generates an insertion plan and calibrates the time period, and outputs a precise insertion plan. The lightweight immersive embedding module includes a material lightweighting processing unit, a hierarchical noise reduction and purification unit, a precise point embedding unit, an immersive blending and optimization unit, and a weak interaction adaptive configuration unit. The lightweight material processing unit, based on the precise implantation scheme, performs bitrate control, redundancy removal, and encoding compression on the advertising materials, and outputs lightweight advertising materials. The hierarchical noise reduction and purification unit performs noise reduction filtering, color restoration and watermark removal based on lightweight advertising materials, and outputs high-definition, clean and lightweight advertising materials. The precise placement unit, based on high-definition, clean, and lightweight advertising materials, detects transition frames and blank frames, and places the advertisement into low-interference locations to complete the initial placement. The immersive blending and optimization unit adjusts the color tone, brightness, and transparency of the advertisement based on the short video that has been initially mounted and embedded, blends it with the original image, and outputs a basic immersive advertisement short video product. The weak interaction adaptive configuration unit is based on the basic immersive advertising short video product, and presets sliding, pausing, and fast-forward trigger rules. When the user operates, the advertisement is weakened or paused, and automatically resumed after the operation, outputting the final optimized immersive advertising short video product. The end-to-end dynamic iteration module collects and scores the campaign performance based on the final optimized immersive short video ad, and iteratively optimizes parameters based on the scoring results.

6. The immersive, lightweight, intelligent embedding optimization system for short video feed ads according to claim 5, characterized in that, The global data parsing module includes: The data collection unit collects information flow data from the entire short video information flow and platform distribution data, and outputs the original short video information flow dataset. Extract video feature units: Based on the original short video information stream dataset, extract video features through frame differencing and semantic recognition, and output a multi-dimensional video feature parameter set; Capture user status units, collect user viewing time, preferences, interactive behaviors and consumption attributes based on video multidimensional feature parameter set, and output real-time user profile data.

7. The immersive, lightweight, intelligent implantation optimization system for short video feed ads according to claim 5, characterized in that, The intelligent matching modeling module includes: Filter suitable ad units based on real-time user profile data and combined with multi-dimensional video feature parameter sets to filter a set of matching candidate ad creatives; The risk control and compliance screening unit, based on the candidate advertising material set, connects to the compliance rule library, eliminates non-compliant materials, and outputs a subset of compliant candidate advertisements; Construct a matching weight unit, and based on a subset of compliant candidate ads, build a matching weight model from scenario fit, user preference, and creative suitability, and output an ad weight score table; Generate an integration plan unit, select the best ad based on the ad weight scoring table, and generate a preliminary integration plan that includes ad creative, integration time period and location; The time-adaptive calibration unit, based on the initial implantation plan, combines traffic popularity and user activity data to calibrate the implantation time period and output a precise implantation plan.

8. The immersive, lightweight, intelligent implantation optimization system for short video feed ads according to claim 5, characterized in that, The end-to-end dynamic iteration module includes: The campaign performance data collection unit collects user interaction data based on the final optimized immersive short video ad and outputs the raw dataset of campaign performance. The quantitative embedding scoring unit scores the campaign based on the original dataset of campaign performance, from four dimensions: loading efficiency, immersive experience, conversion rate, and compliance, and outputs the performance score results. The model units are iteratively optimized. Based on the performance score results, the weight coefficients, screening thresholds, time period parameters, compression coefficients, noise reduction parameters, and interaction trigger thresholds are adjusted in reverse, and the model database is updated.

Images