A marketing video management system based on data analysis

Abstract

The application relates to the technical field of marketing video management, and particularly discloses a marketing video management system based on data analysis, which comprises a collection module, music segments, text content, portrait images and non-portrait images in a marketing video are extracted by the collection module; a risk analysis module, the collected music segments, text content, portrait images and non-portrait images are sequentially compared and analyzed with corresponding copyright databases; and a comprehensive analysis module, a comprehensive infringement risk value corresponding to the marketing video is obtained. The marketing video with the infringement risk is further processed in layers, the efficiency and accuracy of infringement screening are improved, the dynamic risk threshold is controlled and adjusted according to historical infringement data of a publisher, the whole-process closed loop from content collection, risk analysis, response disposal to threshold optimization is realized, the infringement detection is comprehensive, the response strategy is flexible, and the resource distribution is reasonable, and the problem of marketing video infringement supervision difficulty is solved.

Description

Technical Field

[0001] This invention relates to the field of marketing video management technology, specifically a marketing video management system based on data analysis. Background Technology

[0002] The existing marketing video management process has many shortcomings in dealing with infringement supervision. For example, when faced with a massive amount of videos, manual review is not only time-consuming and labor-intensive, but also prone to missed or false judgments. System review can only detect infringement of a certain type of content (such as text or images) in the video, and cannot achieve multi-dimensional comprehensive analysis, which is also prone to missed or false judgments. In addition, the infringement risk assessment threshold of the existing system is fixed and cannot be dynamically adjusted according to the video publisher's historical compliance status, resulting in insufficient supervision, while the review of new users is too strict. Summary of the Invention

[0003] The purpose of this invention is to provide a data analysis-based marketing video management system to solve the following technical problems:

[0004] How can we address the challenges of regulating marketing video copyright infringement by conducting multi-dimensional copyright comparison and analysis, combining user feedback confidence levels, and dynamically assessing risks based on historical data related to copyright infringement in marketing videos published by the publisher?

[0005] The objective of this invention can be achieved through the following technical solutions:

[0006] A data analytics-based marketing video management system, the system comprising:

[0007] The data acquisition module extracts music clips, text content, portrait images, and non-portrait images from marketing videos.

[0008] The infringement risk analysis module compares and analyzes the collected music clips, text content, portrait images, and non-portrait images against the corresponding copyright databases in turn; the comprehensive analysis yields the overall infringement risk value corresponding to the marketing video.

[0009] The human feedback module collects and analyzes relevant negative user feedback data for marketing videos, and generates the confidence score of user feedback for the marketing videos.

[0010] The risk response module compares the overall infringement risk value with a preset infringement risk threshold range. When the overall infringement risk value exceeds the upper limit of the preset infringement risk threshold range, the marketing video is determined to have an infringement risk. When the overall infringement risk value is within the preset infringement risk threshold range, a multi-dimensional infringement risk value is generated based on the user feedback confidence level of the marketing video. If the multi-dimensional infringement risk value exceeds the dynamic risk threshold, the corresponding marketing video is allocated to the manual review pool for manual review of its infringement risk. Marketing videos determined to have an infringement risk are removed from the platform.

[0011] The dynamic adjustment module adjusts the dynamic risk threshold based on the marketing video publisher's historical data on marketing video infringement.

[0012] Furthermore, the process of comparing and analyzing the music clips with the corresponding copyright database includes:

[0013] Extract audio features from music clips in marketing videos, including melody features, rhythm features, harmony features, and timbre features;

[0014] The audio features are converted into audio feature vectors;

[0015] Calculate the first cosine similarity between the audio feature vector and the audio feature vectors of each musical work in the copyright library;

[0016] If any first cosine similarity score is greater than the preset music similarity threshold, the marketing video is deemed to have an infringement risk.

[0017] Furthermore, the process of comparing and analyzing the text content with the corresponding copyright database includes:

[0018] The text content of the marketing video is segmented into words and keywords are extracted.

[0019] The text similarity between the text content and each text work in the copyright database is calculated. The text similarity calculation adopts the TF-IDF algorithm combined with the cosine similarity formula.

[0020] If any text has a similarity score greater than the preset text similarity threshold, the marketing video is deemed to have an infringement risk.

[0021] Furthermore, the process of comparing and analyzing the portrait image with the corresponding copyright library includes:

[0022] Extract facial feature points from portrait images;

[0023] The facial feature points are converted into facial feature vectors;

[0024] Calculate the Euclidean distance between the facial feature vector and the feature vectors of each portrait image in the copyright library;

[0025] If any Euclidean distance is less than the preset portrait similarity threshold, the marketing video is deemed to have an infringement risk.

[0026] Furthermore, the process of comparing and analyzing the non-portrait images with the corresponding copyright database includes:

[0027] Extract visual features from non-portrait images, including color features, texture features, and shape features;

[0028] The visual features are fused into a visual feature vector;

[0029] Calculate the second cosine similarity between the visual feature vector and the feature vectors of each non-portrait image in the copyright library;

[0030] If any second cosine similarity is greater than the preset image similarity threshold, the marketing video is deemed to have an infringement risk.

[0031] Furthermore, the process of obtaining the comprehensive infringement risk value includes:

[0032] The infringement risk value can be obtained by weighting and summing the infringement risk values ​​of music clips, text content, portrait images, and non-portrait images.

[0033] The formula for calculating the infringement risk value is as follows:

[0034]

[0035]

[0036]

[0037]

[0038]

[0039] in, The overall infringement risk value for marketing videos. The copyright infringement risk value for music clips. This represents the risk value for copyright infringement of the text content. The copyright infringement risk value for portrait images. The infringement risk value is for non-portrait images. Let the first cosine similarity between the music fragment and the i-th music work in the copyright database be denoted as . This is a weighting coefficient for the risk of music copyright infringement. Let be the text similarity between the text content and the j-th text work in the copyright database. This is a text infringement risk weighting coefficient. Let be the Euclidean distance between the portrait image and the k-th portrait image in the copyright library. For the risk weighting coefficient of portrait image infringement, Let be the second cosine similarity between the non-portrait image and the l-th non-portrait image in the copyright library. This is a weighting coefficient for the risk of infringement of non-portrait images.

[0040] Furthermore, the process of obtaining the multi-dimensional infringement risk value includes:

[0041] Extract negative user feedback data related to marketing videos, including negative comments and bullet comments related to copyright infringement in marketing videos.

[0042]

[0043] The confidence level of user feedback corresponding to the marketing video is obtained by analyzing and calculating using formula (6). ;

[0044] Where N is the number of items in the user's negative feedback data. , Let p be the number of negative user feedback items. Let be the weighting coefficient of the p-th negative feedback item;

[0045]

[0046] The multi-dimensional infringement risk value is obtained by analyzing and calculating using formula (7). .

[0047] Furthermore, the process of adjusting the preset dynamic risk threshold includes:

[0048] Obtain historical marketing video infringement data from marketing video publishers, including the total number of historically published marketing videos, the number of marketing videos with historical infringement risks, and the number of historically confirmed infringing marketing videos.

[0049]

[0050] The compliance index of the publisher's marketing video is obtained by analyzing and calculating using formula (8). ;

[0051] Where M represents the total number of marketing videos released in history. The number of marketing videos that have historical copyright infringement risks. The comprehensive infringement risk value for the u-th marketing video with infringement risk;

[0052] The preset dynamic threshold is adjusted based on the compliance index S, and the adjustment formula is as follows:

[0053]

[0054] in, For dynamic risk thresholds, This is a preset baseline risk threshold.

[0055] The beneficial effects of this invention are:

[0056] (1) This invention extracts multi-dimensional data from marketing videos through a data acquisition module, analyzes whether the multi-dimensional data infringes on rights through an infringement risk analysis module, and performs tiered response processing on the infringement risk level of marketing videos through a risk analysis module with a tiered processing logic. During the processing, a manual feedback module generates feedback confidence based on the negative feedback data of marketing videos, and further tiers marketing videos with moderate infringement risk, thereby improving the efficiency and accuracy of infringement screening. Furthermore, it controls and adjusts the dynamic risk threshold based on the publisher's historical infringement data, realizing a closed-loop process from content acquisition, risk analysis, response handling to threshold optimization. This makes infringement detection more comprehensive, response strategies more flexible, and resource allocation more reasonable, thus solving the problem of difficult supervision of infringement in marketing videos. Attached Figure Description

[0057] The invention will now be further described with reference to the accompanying drawings.

[0058] Figure 1 This is a schematic diagram of a marketing video management system based on data analysis proposed in this invention. Detailed Implementation

[0059] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0060] Please see Figure 1 As shown, in one embodiment, a data analysis-based marketing video management system is provided, the system comprising:

[0061] The acquisition module uses audio and video analysis technology to separate and extract music clips (such as background music and sound effects), text content (such as subtitles and slogans), portrait images (including images of people's faces), and non-portrait images (such as scene images, icons, illustrations, etc.) from marketing videos.

[0062] The infringement risk analysis module compares and analyzes the collected music clips, text content, portrait images, and non-portrait images against the corresponding copyright libraries (such as music copyright libraries, text works libraries, portrait copyright libraries, and image material libraries); and obtains the comprehensive infringement risk value corresponding to the marketing video through multi-dimensional feature matching algorithms.

[0063] The human feedback module collects negative feedback data from users on marketing videos (such as negative comments), and quantifies the intensity of user feedback on infringement issues through natural language processing (NLP) and data statistical analysis to generate feedback confidence scores.

[0064] The risk response module employs a layered processing logic, comparing the comprehensive infringement risk value with a preset infringement risk threshold range. If the comprehensive infringement risk value exceeds the upper limit of the preset threshold range, the marketing video is considered to have a high infringement risk and is thus deemed to have infringement risk. If the comprehensive infringement risk value falls within the preset threshold range, the marketing video is considered to have a medium infringement risk. Based on user feedback confidence levels, a multi-dimensional infringement risk value is generated. If this multi-dimensional value exceeds a dynamic risk threshold, the corresponding marketing video is allocated to a manual review pool for manual review of its infringement risk. The system dynamically determines whether to include the video in the manual review pool, improving the efficiency and accuracy of infringement screening. Marketing videos deemed to have infringement risk are then removed from the platform.

[0065] The dynamic adjustment module dynamically adjusts the risk threshold based on the historical infringement data of marketing videos published by the marketing video publisher (such as infringement frequency and severity). This allows the system to adapt to the compliance level of different publishers and improves the scientific nature of the threshold setting.

[0066] Through the above technical solution, this embodiment provides a marketing video management system based on data analysis. The system extracts multi-dimensional data from marketing videos through a data acquisition module, analyzes whether the multi-dimensional data infringes on rights through an infringement risk analysis module, and uses a risk analysis module to collect and process layered data to handle the infringement risk of marketing videos in a layered manner. During the processing, a manual feedback module generates feedback confidence scores based on negative feedback data of marketing videos, and further layers marketing videos with moderate infringement risk for further processing, improving the efficiency and accuracy of infringement screening. Furthermore, it controls and adjusts dynamic risk thresholds based on the publisher's historical infringement data, realizing a closed-loop process from content acquisition, risk analysis, response handling to threshold optimization. This ensures comprehensive infringement detection, flexible response strategies, and reasonable resource allocation, solving the problem of difficult supervision of marketing video infringement.

[0067] In one embodiment, the process of comparing and analyzing the music clip with the corresponding copyright library includes:

[0068] Audio features of music clips in marketing videos are extracted. Digital signal processing technology is used to separate and quantify melody features (pitch sequence), rhythm features (beat, tempo), harmony features (chord progression), and timbre features (instrument type, timbre texture) from the music clips, forming a multidimensional feature vector.

[0069] Calculate the first cosine similarity between the audio feature vector and the audio feature vectors of each musical work in the copyright library. The closer the value of the first cosine similarity is to 1, the more similar the audio structures are.

[0070] If any first cosine similarity is greater than a preset music similarity threshold, which can be obtained by pre-setting based on experience, then the marketing video is determined to have an infringement risk, and the marketing video will be taken down.

[0071] The process of comparing and analyzing the text content with the corresponding copyright database includes:

[0072] The text content in the marketing video is segmented into words. NLP technology is used to segment the text content (such as subtitles and narration) into Chinese words, and after removing stop words, core keywords (such as brand name and creative concept words) are extracted.

[0073] The text similarity between the text content and each text work in the copyright database is calculated. The text similarity calculation adopts the TF-IDF algorithm combined with the cosine similarity formula. The specific process is as follows: calculate the frequency of occurrence (TF) of keywords in the text and the inverse document frequency (IDF) in the copyright database, highlight the distinguishing words, convert the text into a TF-IDF vector, and calculate the cosine value of the angle between the vector and the text in the copyright database. The larger the value, the more similar the text structure is.

[0074] If any text has a similarity score greater than a preset text similarity threshold (which can be preset based on experience), then the marketing video is deemed to have an infringement risk and will be taken down.

[0075] The process of comparing and analyzing the portrait image with the corresponding copyright database includes:

[0076] The facial landmarks in the portrait image were extracted by using computer vision techniques (such as OpenCV and FaceNet) to detect facial key points (such as 68 landmarks including the corners of the eyes, the tip of the nose, and the corners of the mouth).

[0077] The facial feature points are converted into facial feature vectors;

[0078] Calculate the Euclidean distance between the facial feature vector and the feature vectors of each portrait image in the copyright library. The smaller the distance, the higher the facial similarity.

[0079] If any Euclidean distance is less than a preset portrait similarity threshold (which can be preset based on experience), the marketing video is determined to have an infringement risk and will be taken down.

[0080] The process of comparing and analyzing the non-portrait images with the corresponding copyright database includes:

[0081] Visual features are extracted from non-portrait images. These visual features include color features, texture features, and shape features. Color features are quantified by color histograms and dominant color distribution. Texture features are extracted by using the gray-level co-occurrence matrix (GLCM) to extract features such as image texture roughness and directionality. Shape features are extracted by using contour detection and Hough transform to extract geometric shape parameters (such as roundness and aspect ratio).

[0082] The visual features are fused into a visual feature vector;

[0083] Calculate the second cosine similarity between the visual feature vector and the feature vectors of each non-portrait image in the copyright library. The closer the value of the second cosine similarity is to 1, the more similar the non-portrait images are.

[0084] If any second cosine similarity is greater than a preset image similarity threshold (which can be preset based on experience), then the marketing video is determined to have an infringement risk, and the marketing video will be taken down.

[0085] The process of obtaining the comprehensive infringement risk value includes:

[0086] The infringement risk value can be obtained by weighting and summing the infringement risk values ​​of music clips, text content, portrait images, and non-portrait images.

[0087] The formula for calculating the infringement risk value is as follows:

[0088]

[0089]

[0090]

[0091]

[0092]

[0093] in, The overall infringement risk value for marketing videos. The copyright infringement risk value for music clips. This represents the risk value for copyright infringement of the text content. The copyright infringement risk value for portrait images. The infringement risk value is for non-portrait images. Let the first cosine similarity between the music fragment and the i-th music work in the copyright database be denoted as . This is a weighting coefficient for the risk of music copyright infringement. Let be the text similarity between the text content and the j-th text work in the copyright database. This is a text infringement risk weighting coefficient. Let be the Euclidean distance between the portrait image and the k-th portrait image in the copyright library. For the risk weighting coefficient of portrait image infringement, Let be the second cosine similarity between the non-portrait image and the l-th non-portrait image in the copyright library. For non-portrait image infringement risk weighting coefficients, , , , It reflects the infringement sensitivity of different content types, and the specific values ​​can be obtained based on experience and preset values.

[0094] Through the above technical solution, this embodiment provides a method for quantitative assessment of infringement risk based on multimodal feature extraction and weighted fusion. The method constructs a unified comprehensive infringement risk assessment system by weighted summation of the infringement risk values ​​of music clips, text content, portrait images, and non-portrait images. This achieves full-element infringement risk scanning of marketing videos, enabling accurate quantification of infringement risks and facilitating subsequent graded handling mechanisms based on comprehensive risk values ​​(direct removal, manual review, or temporary non-processing), thus achieving optimized resource allocation.

[0095] In one embodiment, the process of obtaining the multi-dimensional infringement risk value includes:

[0096] Extract negative user feedback data related to marketing videos to form a multi-source feedback dataset. The negative user feedback data includes negative comments and bullet comments about infringement in marketing videos. Comments or bullet comments that mention keywords such as "infringement" and "plagiarism" are considered negative comments and bullet comments about infringement.

[0097]

[0098] The confidence level of user feedback corresponding to the marketing video is obtained by analyzing and calculating using formula (6). ;

[0099] Where N is the number of items in the user's negative feedback data. , The number of negative user feedback items (item p) is obtained by statistically analyzing the negative user feedback data. The weighting coefficient for the p-th negative feedback item reflects the credibility of the user feedback and can be preset based on experience.

[0100]

[0101] The multi-dimensional infringement risk value is obtained by analyzing and calculating using formula (7). .

[0102] Through the above technical solution, this embodiment provides a multi-dimensional risk assessment method that integrates automated system analysis and user supervision. The method collects and quantifies negative user feedback, transforms subjective evaluations into objective confidence indicators, and cross-validates them with automated infringement detection results, thereby improving the reliability of infringement risk assessment results.

[0103] In one embodiment, the process of adjusting the preset dynamic risk threshold includes:

[0104] Obtain historical marketing video infringement data from marketing video publishers, including the total number of historically published marketing videos, the number of marketing videos with historical infringement risks, and the number of historically confirmed infringing marketing videos.

[0105]

[0106] The compliance index of the publisher's marketing video is obtained by analyzing and calculating using formula (8). ;

[0107] Where M represents the total number of marketing videos published historically, which can be obtained by statistically analyzing the marketing videos published by the publisher. The number of marketing videos with historical copyright infringement risks can be statistically analyzed based on historical infringement records. The comprehensive infringement risk value of the u-th marketing video with infringement risk can be calculated based on the above infringement risk value;

[0108] The preset dynamic threshold is adjusted based on the compliance index S, and the adjustment formula is as follows:

[0109]

[0110] in, For dynamic risk thresholds, The preset baseline risk threshold can be obtained based on experience.

[0111] Through the above technical solution, this embodiment provides a dynamic threshold adjustment method based on the publisher's historical records. The method achieves adaptive adjustment of risk thresholds by quantitatively analyzing the publisher's infringement history, thereby achieving the purpose of optimizing system resource allocation (prioritizing the review of high-risk entities) and personalizing management strategies.

[0112] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the claims of this invention should still fall within the patent coverage of this invention.

Claims

1. A marketing video management system based on data analysis, characterized in that, The system includes: The data acquisition module extracts music clips, text content, portrait images, and non-portrait images from marketing videos. The infringement risk analysis module compares and analyzes the collected music clips, text content, portrait images, and non-portrait images against the corresponding copyright databases in turn; the comprehensive analysis yields the overall infringement risk value corresponding to the marketing video. The human feedback module collects and analyzes relevant negative user feedback data for marketing videos, and generates the confidence score of user feedback for the marketing videos. The risk response module compares the overall infringement risk value with a preset infringement risk threshold range. When the overall infringement risk value exceeds the upper limit of the preset infringement risk threshold range, the marketing video is determined to have an infringement risk. When the overall infringement risk value is within the preset infringement risk threshold range, a multi-dimensional infringement risk value is generated based on the user feedback confidence level of the marketing video. If the multi-dimensional infringement risk value exceeds the dynamic risk threshold, the corresponding marketing video is allocated to the manual review pool for manual review of its infringement risk. Marketing videos determined to have an infringement risk are removed from the platform. The dynamic adjustment module adjusts the dynamic risk threshold based on the marketing video publisher's historical data on marketing video infringement. The process of obtaining the multi-dimensional infringement risk value includes: Extract negative user feedback data related to marketing videos, including negative comments and bullet comments related to copyright infringement in marketing videos. The confidence level of user feedback corresponding to the marketing video is obtained by analyzing and calculating using formula (6). ; Where N is the number of items in the user's negative feedback data. , Let p be the number of negative user feedback items. Let be the weighting coefficient of the p-th negative feedback item; The multi-dimensional infringement risk value is obtained by analyzing and calculating using formula (7). ; in, The overall infringement risk value for marketing videos; The process of adjusting the preset dynamic risk threshold includes: Obtain historical marketing video infringement data from marketing video publishers, including the total number of historically published marketing videos, the number of marketing videos with historical infringement risks, and the number of historically confirmed infringing marketing videos. The compliance index of the publisher's marketing video is obtained by analyzing and calculating using formula (8). ; Where M represents the total number of marketing videos released in history. The number of marketing videos that have historical copyright infringement risks. The comprehensive infringement risk value for the u-th marketing video with infringement risk; The preset dynamic threshold is adjusted based on the compliance index S, and the adjustment formula is as follows: in, For dynamic risk thresholds, This is a preset baseline risk threshold.

2. The marketing video management system based on data analysis according to claim 1, characterized in that, The process of comparing and analyzing the music clips with the corresponding copyright databases includes: Extract audio features from music clips in marketing videos, including melody features, rhythm features, harmony features, and timbre features; The audio features are converted into audio feature vectors; Calculate the first cosine similarity between the audio feature vector and the audio feature vectors of each musical work in the copyright library; If any first cosine similarity score is greater than the preset music similarity threshold, the marketing video is deemed to have an infringement risk.

3. The marketing video management system based on data analysis according to claim 2, characterized in that, The process of comparing and analyzing the text content with the corresponding copyright database includes: The text content of the marketing video is segmented into words and keywords are extracted; The text similarity between the text content and each text work in the copyright database is calculated. The text similarity calculation adopts the TF-IDF algorithm combined with the cosine similarity formula. If any text has a similarity score greater than the preset text similarity threshold, the marketing video is deemed to have an infringement risk.

4. A marketing video management system based on data analysis according to claim 3, characterized in that, The process of comparing and analyzing the portrait image with the corresponding copyright database includes: Extract facial feature points from portrait images; The facial feature points are converted into facial feature vectors; Calculate the Euclidean distance between the facial feature vector and the feature vectors of each portrait image in the copyright library; If any Euclidean distance is less than the preset portrait similarity threshold, the marketing video is deemed to have an infringement risk.

5. A marketing video management system based on data analysis according to claim 4, characterized in that, The process of comparing and analyzing the non-portrait images with the corresponding copyright database includes: Extract visual features from non-portrait images, including color features, texture features, and shape features; The visual features are fused into a visual feature vector; Calculate the second cosine similarity between the visual feature vector and the feature vectors of each non-portrait image in the copyright library; If any second cosine similarity is greater than the preset image similarity threshold, the marketing video is deemed to have an infringement risk.

6. A marketing video management system based on data analysis according to claim 5, characterized in that, The process of obtaining the comprehensive infringement risk value includes: The infringement risk value can be obtained by weighting and summing the infringement risk values ​​of music clips, text content, portrait images, and non-portrait images. The formula for calculating the infringement risk value is as follows: in, The overall infringement risk value for marketing videos. The copyright infringement risk value for music clips. This represents the risk value for copyright infringement of the text content. The copyright infringement risk value for portrait images. The infringement risk value is for non-portrait images. Let the first cosine similarity between the music fragment and the i-th music work in the copyright database be denoted as . This is a weighting coefficient for the risk of music copyright infringement. Let be the text similarity between the text content and the j-th text work in the copyright database. This is a text infringement risk weighting coefficient. Let be the Euclidean distance between the portrait image and the k-th portrait image in the copyright library. This is a weighting coefficient for the risk of portrait image infringement. Let be the second cosine similarity between the non-portrait image and the l-th non-portrait image in the copyright library. This is a weighting coefficient for the risk of infringement of non-portrait images.

Images