A method and system for labeling advertising material data based on AI

Through the AI-based creative data labeling method, the problem of inefficiency of traditional manual labeling is solved, automated labeling and intelligent decision-making are realized, and the efficiency of advertising delivery and return on investment are improved.

CN118761812BActive Publication Date: 2024-11-22HUNAN TRYINE TECH CO LTD
View PDF 5 Cites 0 Cited by

Patent Information

Application Number
CN202411253799.5
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-09-09
Publication Date
2024-11-22
Estimated Expiration
2044-09-09

AI Technical Summary

Technical Problem

Traditional creative data labeling relies on manual labor, is inefficient and expensive, making it difficult to achieve efficient automation.

Method used

Using AI-based creative data labeling method, we can obtain and standardize the process of creative content data, perform automated labeling and reliability scoring, combine manual re-labeling, and use multiple models for label verification and optimization to realize automated labeling and intelligent decision-making.

Benefits of technology

It improves the efficiency and accuracy of creative data labeling, reduces advertising delivery costs, improves the return on investment, and optimizes the accuracy and effectiveness of advertising marketing.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118761812B_ABST
    Figure CN118761812B_ABST
Patent Text Reader

Abstract

The present invention relates to the field of advertising management technology, and in particular to an advertising material data annotation method and system based on AI. The method comprises the following steps: obtaining advertising material content data, and performing standardization processing to obtain advertising material content standardized data; performing reliability scoring on the advertising material content standardized data to obtain annotation reliability scoring data; performing secondary annotation operations based on the annotation reliability scoring data to obtain advertising material content annotation update data; obtaining the advertising delivery data corresponding to the advertising material content data, and extracting delivery effects based on the advertising delivery data to obtain advertising delivery effect data; annotating based on the advertising delivery effect data to obtain advertising delivery effect annotation data, and performing annotation consistency processing on the advertising delivery effect annotation data and the advertising material content annotation update data to obtain advertising material annotation data. The present invention can automatically annotate advertising material content.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of advertising management, and in particular to an AI-based advertising material data labeling method and system. Background Art

[0002] Traditional advertising creative data labeling mainly relies on manual work, which requires a lot of time and resources, including the design, analysis and labeling of tag libraries, which is inefficient and costly. Artificial Intelligence (AI) is a field of computer science dedicated to developing systems and programs that can imitate human intelligent behavior. The core goal of artificial intelligence is to enable computer systems to have capabilities similar to human thinking, learning, problem solving and decision making. How to combine artificial intelligence and advertising creative data labeling has become a problem. Summary of the invention

[0003] In order to solve the above technical problems, the present invention proposes an AI-based advertising material data labeling method and system to solve at least one of the above technical problems.

[0004] The present application provides an advertising material data annotation method based on AI, the method comprising:

[0005] S1. Acquire advertising material content data, and perform standardization processing on the advertising material content data to obtain standardized advertising material content data;

[0006] S2. Automatically label the standardized data of the advertising material content to obtain advertising material content labeling data; and perform reliability scoring on the advertising material content labeling data to obtain labeling reliability scoring data;

[0007] S3, performing a secondary annotation operation on the advertising material content annotation data according to the annotation reliability score data to obtain advertising material content annotation update data;

[0008] S4, obtaining advertisement delivery data corresponding to the advertisement material content data, and extracting delivery effects according to the advertisement delivery data to obtain advertisement delivery effect data;

[0009] S5. Label the advertising material content data according to the advertising delivery effect data to obtain advertising delivery effect labeling data, and perform labeling consistency processing on the advertising delivery effect labeling data and the advertising material content labeling update data to obtain advertising material labeling data.

[0010] The present invention uses locally uploaded materials and advertising materials obtained through the media platform to collect advertising material content, including data such as pictures, videos, and text, and performs preprocessing to ensure data quality, laying the foundation for subsequent automatic labeling and model training. Data preprocessing includes data cleaning of advertising material content, converting all text materials into a unified encoding format, ensuring that the formats of image, audio, and video materials are consistent, obtaining cleaned data, formatting the cleaned data, and standardizing text, images, and videos. According to the content of the advertising material, multiple models are used to automatically label text data, image data, and audio data, and the reliability of the automatic labeling content is scored. The labeling results of multiple models can be verified and supplemented with each other to improve the accuracy and reliability of the labeling. The labels with low reliability scores are manually re-labeled, and the results of manual re-labeling are fed back to the advertising content labeling model for optimization. Through manual re-labeling, errors in the model prediction results can be corrected, and high-quality labeling data can be provided to help the model learn and improve, and improve the labeling accuracy and reliability of the model. For the advertising delivery data of the media platform, we can realize the automatic labeling of advertising information and delivery effects. Through the in-depth analysis of the advertising creative data by the model, it can automatically identify the actual effect of the advertisement, obtain the advertising data for annotation, and obtain the advertising information for annotation. The data from different media platforms and different formats are standardized. This step ensures the consistency and comparability of the data, and provides a basis for subsequent automatic annotation and intelligent decision-making. Through such an automated annotation process, a scientific decision-making basis can be provided for the implementation of intelligent advertising delivery. For example, according to the advertising creatives with high exposure, high click volume, and high conversion rate in the annotation results, these excellent performing ads can be delivered first, thereby improving the effectiveness and efficiency of advertising delivery, reducing unnecessary advertising costs, and thus improving the return on investment of advertising. Not only does it optimize the advertising delivery process, but it also enhances the accuracy and effect of advertising marketing. By using multiple models of automated annotation, manual re-annotation, collection and preprocessing of advertising information and delivery data, automated annotation of advertising information and delivery effects, and intelligent advertising delivery decisions, a comprehensive analysis and optimization of advertising creative data is achieved, thereby improving the effect and efficiency of advertising delivery, reducing advertising costs, and improving the return on investment of advertising.

[0011] Optionally, the present application further provides an AI-based advertising material data labeling system, which is used to execute the AI-based advertising material data labeling method as described above, and the AI-based advertising material data labeling system includes:

[0012] The advertisement material content standardization module is used to obtain advertisement material content data and perform standardization processing on the advertisement material content data to obtain advertisement material content standardized data;

[0013] The marking reliability scoring module is used to automatically mark the standardized data of the advertising material content to obtain the advertising material content marking data; and to perform reliability scoring on the advertising material content marking data to obtain the marking reliability scoring data;

[0014] A secondary annotation operation module is used to perform secondary annotation operation on the advertisement material content annotation data according to the annotation reliability score data to obtain advertisement material content annotation update data;

[0015] The advertisement delivery effect extraction module is used to obtain the advertisement delivery data corresponding to the advertisement material content data, and extract the delivery effect according to the advertisement delivery data to obtain the advertisement delivery effect data;

[0016] The advertising material annotation consistency processing module is used to annotate the advertising material content data according to the advertising delivery effect data to obtain the advertising delivery effect annotation data, and to perform annotation consistency processing on the advertising delivery effect annotation data and the advertising material content annotation update data to obtain the advertising material annotation data.

[0017] The purpose of the present invention is to realize the automatic labeling of advertising information and delivery effects. Through the in-depth analysis of the advertising material data by the model, the actual delivery effect of the advertisement can be automatically identified, the advertising data can be obtained for labeling, and the advertising information can be obtained for labeling. The data from different media platforms and different formats are standardized in format. This step ensures the consistency and comparability of the data, and provides a basis for subsequent automatic labeling and intelligent decision-making. Through such an automated labeling process, a scientific decision-making basis can be provided for the execution of intelligent advertising delivery. For example, according to the advertising materials with high exposure, high click-through rate, and high conversion rate in the labeling results, these excellent performing advertisements can be placed first, thereby improving the effectiveness and efficiency of advertising delivery, reducing unnecessary advertising costs, and thus improving the return on investment of advertising. Such automated labeling not only optimizes the advertising delivery process, but also enhances the accuracy and effectiveness of advertising marketing. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Other features, objects and advantages of the present application will become more apparent by reading the detailed description of non-limiting implementations made with reference to the following drawings:

[0019] Figure 1 A flowchart showing a method for labeling advertising material data based on AI according to an embodiment of the present invention is shown;

[0020] Figure 2 A flowchart showing a method for standardizing content of advertising materials according to an embodiment of the present invention is provided;

[0021] Figure 3A flowchart showing a method for scoring reliability of a labeling according to an embodiment of the present invention is shown;

[0022] Figure 4 A flowchart showing a method for updating advertisement material content annotation according to an embodiment of the present invention is provided;

[0023] Figure 5 A flowchart showing a method for processing consistency of advertisement material marking according to an embodiment of the present invention is shown;

[0024] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0025] The technical method of the present invention is described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by technicians in this field without creative work are within the scope of protection of the present invention.

[0026] In addition, the drawings are only schematic illustrations of the present invention and are not necessarily drawn to scale. The functional entities can be implemented in software form, or in one or more hardware modules or integrated circuits, or in different network and / or processor methods and / or microcontroller methods.

[0027] It should be understood that, although the terms "first", "second", etc. may be used herein to describe various units, these units should not be limited by these terms. These terms are used only to distinguish one unit from another unit. For example, without departing from the scope of the exemplary embodiments, the first unit may be referred to as the second unit, and similarly the second unit may be referred to as the first unit. The term "and / or" used herein includes any and all combinations of one or more of the listed associated items.

[0028] To achieve the above purpose, a labeling model for generating advertisement content includes the following steps:

[0029] STEP1. Obtain advertising content, collect and pre-process data;

[0030] STEP 2: Use multiple models to automatically annotate text data, image data, and audio data based on the content of the advertising material, and score the reliability of the automated annotation content;

[0031] STEP 3: Manually re-annotate annotations with low reliability scores and use this as feedback to optimize the ad content annotation model;

[0032] STEP 4. Obtain the advertising information and delivery data of the advertising materials, collect and pre-process them;

[0033] STEP 5. Automatically label the advertising information and delivery effects based on the delivery data of the advertising materials.

[0034] STEP1 Details

[0035] STEP 11. Obtain advertising material content through your own material library or the material library of the synchronized media platform, and record the original information of the material;

[0036] STEP 12. Perform data cleaning on the content of the advertising materials, convert all text materials into a unified coding format, ensure that the formats of image, audio and video materials are consistent, and obtain cleaned data;

[0037] STEP 13. Format the cleaned data, standardize text, images, and videos.

[0038] STEP11Details

[0039] STEP 111. Determine the source of advertising material content, including local material library, online advertising platform, and materials provided by customers;

[0040] STEP 112. Identify and classify the types of advertising materials, including text, images, audio and video;

[0041] STEP 113. Collect and record metadata of advertising materials, such as material type, material resolution, release date, target market, advertiser information, creative team, etc.

[0042] STEP13 Details

[0043] STEP 131. Standardize text, unify the capitalization and punctuation of text, and remove or replace special characters;

[0044] STEP132: Image standardization: adjust the image resolution, color space, etc. to meet the requirements of the annotation system;

[0045] STEP 133: Video standardization: extract video frames into multiple images, convert audio analysis into a unified text format, and parse the script content through the audio recognition model, including script timestamp, sound text, male and female voices, and voice age.

[0046] STEP 2 Details

[0047] STEP 21. Use natural language processing technology to annotate the text data of the advertising material content and perform reliability scoring. If the advertising material content has no text data, skip this step.

[0048] STEP 22: Label the image data of the advertising material content and perform reliability scoring. If the advertising material content has no image data, skip this step.

[0049] STEP 23: Label the video data (including image data, audio data, and audio text data) of the advertising material content and perform reliability scoring. If the advertising material content has no video data, skip this step.

[0050] STEP21Details

[0051] STEP 211. Extract keywords and mark the keywords directly related to the product or service in the copy, including product content, product selling points, and product pain points; mark professional terms in specific industries or fields;

[0052] STEP 212: Emotional tendency labeling: label whether the emotion expressed in the copy is positive or negative; label the intensity of the emotional expression, with x being the weakest and y being the strongest;

[0053] STEP 213. Theme classification: According to the purpose and content of the advertisement, classify it into user acquisition, incentive conversion, user retention, user recall, and user communication; mark the industry to which the advertisement copy belongs, such as education, health, technology, fashion, etc.

[0054] STEP 214: Target audience labeling: label the age range of the target audience; label whether the copy is mainly targeted at men, women, or neutral genders; label the target audience’s socioeconomic background, such as students, middle class, high-income groups, etc.

[0055] STEP 215: Language style annotation: mark whether the language style of the copy is formal or informal; mark the tone of the copy, such as appeal, authority, friendliness, humor;

[0056] STEP 216. Mark the promotional information, mark the discount type and range mentioned in the copy; mark the preferential activities mentioned in the copy, such as buy one get one free, limited time discount, etc.

[0057] STEP 217. Call to Action (CTA) annotation: Annotate the call to action in the copy, such as "Buy Now" or "Learn More". Annotate the position of the CTA in the copy, scoring the beginning as x and the end as y.

[0058] STEP 218: Grammar and sentence structure annotation. Annotate the sentence types in the copy, such as declarative sentences, interrogative sentences, imperative sentences, exclamatory sentences, interrogative sentences, rhetorical questions, and imperative sentences; Annotate the grammatical complexity of the sentences, with a scoring system. The higher the complexity, the higher the score.

[0059] STEP219: Use multiple models to annotate the annotation contents of STEP211 to STEP218, export the results of multiple models, and use the reliability verification model to give a score for each annotation content of each model. The larger the score, the more reliable it is.

[0060] STEP22Details

[0061] STEP 221, target detection, using deep learning models to identify and locate target objects in images;

[0062] STEP 222: Scene recognition: by analyzing the image content, determine the scene type in the image, such as outdoor, indoor, sports, etc.

[0063] STEP 223: Color analysis: analyze the color distribution, hue, saturation, etc. of the image to extract color features;

[0064] STEP 224: Image quality assessment: assess the clarity, contrast, noise, etc. of the image to determine the quality of the image;

[0065] STEP 225, style recognition, identify the style of the image, such as modern, retro, art, etc.;

[0066] STEP 226: Brand identification: if the image contains a brand logo, such as a logo, then perform brand identification;

[0067] STEP 227, sentiment analysis, by analyzing elements in the image, such as character expressions, scene atmosphere, etc., to determine the emotions expressed by the image;

[0068] STEP 228: Image description, generating image description text for understanding and indexing image content;

[0069] STEP229. Use multiple models to annotate the annotation contents of STEP221 to STEP228, export the results of multiple models, and use the reliability verification model to give a score for each annotation content of each model. The larger the score, the more reliable it is.

[0070] STEP 23 Details

[0071] STEP231, annotate the audio text data in the video through STEP21;

[0072] STEP 232, mark the video capture image content through STEP 22;

[0073] STEP 233. Perform scene conversion analysis on the captured images of the video, such as single scene and multiple scenes; and character interaction analysis, such as single-person scene and multi-person dialogue;

[0074] STEP 234: Perform sentiment analysis on the audio data in the video to determine whether the text is positive or negative, and mark the intensity of the emotional expression.

[0075] STEP 3 Details

[0076] STEP31: Manually verify and re-annotate the labeled data with reliability lower than z score;

[0077] STEP 32: Analyze the differences between manual re-labeling results and model labeling results, and identify the types and causes of model prediction errors;

[0078] STEP 33: Adjust model parameters, such as learning rate, regularization strength, etc., according to the error type analysis results to optimize model performance;

[0079] STEP 34. Record the manually corrected labeled data and mark the data as manual with a reliability score of y.

[0080] STEP 35. Return the manually corrected annotation data to the advertising content annotation model for learning to improve the accuracy of advertising content annotation.

[0081] STEP 4 Details

[0082] STEP 41. Obtain advertising information from the media platform, including promotion goals, optimization goals, marketing scenarios, advertising types, delivery types, keywords, bidding types, matching types, related products, advertising location categories, advertising delivery locations, regional matching, population gender, population age, behavioral interests, interest words, etc.

[0083] STEP 42. Obtain advertising creative data from media platforms, including exposure, clicks, conversions, conversion rate, cost, ROI, etc.

[0084] STEP 43: Standardize the formats of data from different media platforms and in different formats.

[0085] The present invention obtains the advertising material delivery content through the media platform and records the original information of the material, including promotion goals, optimization goals, marketing scenarios, etc., to ensure that the collected information is comprehensive and detailed. The media platform advertising material data is obtained, including exposure, click volume, conversion volume, conversion rate, cost, ROI, etc.

[0086] See also Figures 1 to 5 , the present application provides an advertising material data labeling method based on AI, the method comprising:

[0087] S1. Acquire advertising material content data, and perform standardization processing on the advertising material content data to obtain standardized advertising material content data;

[0088] The present invention provides an embodiment in which various types of advertising materials are collected, including images, videos, texts, and audio files. These raw data are compiled into a centralized data set. All collected data are converted into a unified format. For example, all image files are converted to a specific resolution, videos are converted to a consistent frame rate and codec, text data is converted to a standard encoding, and audio data is converted to a standard bit rate and sampling rate. The advertising materials are decomposed into their core components. For example, images are decomposed into pixels, text is decomposed into sentences, and audio is decomposed into waveform data. This decomposition method facilitates subsequent analysis. The parsed data is adjusted to conform to a standard range or distribution. For example, pixel values ​​are normalized to between 0 and 1, text vectors are scaled, and audio amplitudes are standardized. Relevant metadata (such as timestamps, locations, formats, etc.) are attached to each content segment.

[0089] S2. Automatically label the standardized data of the advertising material content to obtain advertising material content labeling data; and perform reliability scoring on the advertising material content labeling data to obtain labeling reliability scoring data;

[0090] The present invention provides an embodiment in which automated technology is used to annotate different aspects of standardized data. For example, objects in images are identified, emotions or themes in text are marked, and emotions or themes in audio clips are annotated. Contextual analysis is performed on preliminary annotations to refine labels. For example, the context of text or image content is considered to improve the accuracy of annotations. Patterns in data are identified to indicate common themes or repeated elements, such as identifying brand logos in images or common phrases in text. A scoring mechanism is implemented to evaluate the reliability of each annotation, such as checking label consistency of similar content, cross-referencing with known data sets, or evaluating the confidence level of the annotation process. Reliability scores are aggregated to generate a reliability profile for each annotated content.

[0091] S3, performing a secondary annotation operation on the advertising material content annotation data according to the annotation reliability score data to obtain advertising material content annotation update data;

[0092] The present invention provides an embodiment in which reliability scores are analyzed to identify areas where preliminary annotations may be weak or inaccurate, with a focus on low-scoring annotations. For content with low reliability scores, more detailed annotations are performed, such as using a more refined approach, reviewing more context, or even introducing manual supervision. The new annotations are compared with the preliminary annotations to identify differences, and an assessment is made of which annotation is more accurate based on the improved reliability index. The updated annotations are re-scored, with a focus on corrected or refined labels to ensure that the most accurate labels are prioritized. The original and updated annotations are integrated to ensure that any conflicting annotations are resolved and the final annotation set is reliable. The result of this step is updated and more accurate advertising creative content annotation data with improved reliability.

[0093] S4, obtaining advertisement delivery data corresponding to the advertisement material content data, and extracting delivery effects according to the advertisement delivery data to obtain advertisement delivery effect data;

[0094] The present invention provides an embodiment in which standardized and annotated advertising content is matched with corresponding advertising delivery data (such as click-through rate, viewing time, conversion rate, etc.). Advertising delivery data is analyzed to identify user engagement patterns. For example, evaluate which content elements (such as color, text style, sound) are most likely to drive user action. Segmenting delivery performance data according to different criteria (such as demographics, time, and ad location) helps identify the best performance of content under specific conditions. The segmented delivery performance data is associated with content annotation features to find out which annotations are most predictive of success. Trend analysis is performed to discover long-term patterns or changing trends in advertising effects, which will provide guidance for future content creation.

[0095] S5. Label the advertising material content data according to the advertising delivery effect data to obtain advertising delivery effect labeling data, and perform labeling consistency processing on the advertising delivery effect labeling data and the advertising material content labeling update data to obtain advertising material labeling data.

[0096] The present invention provides an embodiment in which the content is re-annotated using the advertising delivery performance data, this time focusing on the features or elements that directly contribute to the observed performance. This may include highlighting specific phrases, images, or sound elements. The new performance-based annotations are integrated with the updated annotations in step S3 to ensure that the annotations are consistent and the delivery data is accurately reflected in the labels. A consistency check is performed on the entire data set to ensure that the annotations are uniform and reliable, such as checking whether the annotations of similar content are consistent, regardless of their subtle differences. Final adjustments are made to the annotations to resolve any inconsistencies or anomalies found in the consistency check.

[0097] Optionally, S1 includes:

[0098] S11. Acquire advertising material content through the own material library or the synchronous media platform material library, and record the original information of the material to obtain advertising material content data;

[0099] The present invention provides an embodiment, first, the system obtains the required advertising material content by accessing the own material library or the material library of the synchronous media platform. The own material library usually contains advertising materials produced or collected internally by the enterprise, while the media platform material library comes from the external media platform. If the synchronous media platform material library is selected, the system needs to connect to the corresponding API interface to obtain the latest advertising material content from the platform, including calling the platform interface to obtain the material list and downloading the selected material content. While obtaining the material, the system needs to record the original information of the material, such as the creation time of the material, the source platform, the material type (picture, video, text, etc.), format, size, resolution, duration, etc. This original information is very important for subsequent processing and analysis. All the acquired advertising material content and its original information will be stored in a dedicated database for subsequent cleaning and formatting operations. Each advertising material content is assigned a unique identifier (ID) for subsequent tracking and processing.

[0100] S12, cleaning the advertising material content data to obtain advertising material content cleaning data;

[0101] The present invention provides an embodiment in which, first, the system scans the advertising material content data and removes any duplicate or redundant data, including deleting duplicate images or video files, or removing duplicate sentences in the text. The system detects possible anomalies in the data, such as damaged files, incomplete audio and video content, data in unsupported formats, etc. These abnormal data will be marked and removed from the data set. The system filters the advertising material content according to predefined rules. For example, remove video files that do not meet specific resolution or duration requirements, or delete low-quality image files. Clean the metadata of the advertising material to ensure that all metadata fields are complete and accurate. For example, correct incorrect timestamps, complete missing format information, or unify metadata formats from different sources. Verify the cleaned advertising material content data to ensure that all data meets the cleaning standards and no important content is omitted.

[0102] S13. Format the cleaned data of the advertising material content to obtain standardized data of the advertising material content.

[0103] The present invention provides an embodiment in which the cleaned data is converted into a uniform format according to the type of advertising material content. For example, all image files are converted to the same resolution and image format, video files are converted to a consistent frame rate and encoding format, and text files are converted to a uniform character encoding format. The image and video materials are resized to ensure that they meet predefined size specifications. For example, all pictures are adjusted to a standard aspect ratio, and videos are clipped to a specified time length. The encoding of audio and video files is standardized to ensure that all files use the same encoding standard and bit rate. This helps to maintain data consistency and availability during subsequent processing. The formats of all metadata fields are unified to ensure that they are consistent between different types of materials. For example, the date format is unified to YYYY-MM-DD, and the file size is expressed in KB. The formatted advertising material content is verified to ensure that the data integrity is not affected during the format conversion and adjustment process. Any problems that occur during the formatting process need to be corrected in a timely manner.

[0104] In the present invention, the content of advertising materials is obtained through the own material library or the material library of the synchronous media platform, and the original information of the materials is recorded, including the material type, source, release time, etc., to ensure that the collected information is comprehensive and detailed. Data cleaning processing is performed. This link removes irrelevant or erroneous data, such as removing duplicate materials, correcting erroneous information, etc., so as to improve the accuracy and reliability of the data. The format of the cleaned data is standardized to ensure that the data conforms to the predetermined structure and format, such as converting all text materials into a unified encoding format, ensuring that the formats of images, audio and video materials are consistent, and obtaining cleaned data. The cleaned data is formatted, text is standardized, images are standardized, and videos are standardized. For example, text materials are segmented, parts of speech are tagged, etc., image materials are unified in size, formatted, etc., video materials are framed, audio is extracted, etc., for subsequent processing and analysis. By acquiring, recording, cleaning and formatting the content of advertising materials, a solid, clear and consistent data foundation is provided for advertising material data labeling. Data cleaning ensures data quality, reduces errors and deviations in subsequent labeling, and improves the accuracy of data labeling. Data format standardization improves the efficiency of data processing, ensures consistency between different data sources, and facilitates integrated analysis and model training.

[0105] Optionally, S11 includes:

[0106] Get the source data of the advertising creative content;

[0107] In one embodiment of the present invention, first, the system identifies and connects to multiple sources of advertising material content. These sources may include internal material libraries of enterprises, external media platforms, advertising delivery platforms, etc. Each source has its own specific API interface or data access method.

[0108] The system obtains the source data of the creative content by accessing these identified data sources. These source data may include the URL of the file, media ID, creator information, release time, etc. The system needs to ensure that the connection with each data source is secure and stable.

[0109] Integrate the creative content source data collected from multiple sources, eliminate duplicate data, and ensure that each creative content has a unique source identifier.

[0110] During the collection process, the system will detect and process any anomalies in the source data, such as invalid URLs, inaccessible media IDs, etc. Abnormal data will be marked and the operator will be notified for subsequent processing.

[0111] Classify the advertising material type according to the advertising material content source data to obtain advertising material type classification data, wherein the advertising material type classification data includes advertising text type classification data, advertising image type classification data, advertising audio type classification data and advertising video type classification data;

[0112] In one embodiment of the present invention, the system identifies the type of each material according to the source data of the advertising material content. The type identification can be achieved by parsing the file extension, reading the file header information, or by the classification label of the data source.

[0113] The system applies predefined classification rules to classify material content into four categories: text, image, audio and video. The predefined classification rules are formulated based on the file format and content characteristics (such as text length, image resolution, audio duration, etc.).

[0114] For creatives that contain multiple types of elements, such as videos that contain audio and images, the system marks the material as multi-type material and classifies it into relevant types to ensure that all type information is recorded.

[0115] The system will verify the classification results to ensure the accuracy of the classification. For materials that cannot be automatically classified, the system will mark them and send them to manual for secondary confirmation.

[0116] Each type of material data will be assigned a corresponding classification label to facilitate subsequent processing and analysis.

[0117] Metadata is collected and recorded for the advertising material type classification data to obtain the advertising material content data.

[0118] The present invention provides an embodiment in which metadata is extracted from each classified advertising material content. The metadata includes the creation time, modification time, format, size, resolution, duration, bit rate, etc. of the file. This information is usually stored in the header or additional data of the file, and the system can extract relevant information by parsing these areas.

[0119] Analyze the creative content itself to extract more useful metadata, such as the color space information of images, the frame rate and encoding method of videos, the sampling rate and number of channels of audio, the language and character encoding of text, etc.

[0120] Associate the extracted metadata with the creative content and its source data, ensuring all relevant information is fully documented in a single data structure.

[0121] The collected metadata is verified to check for missing or incorrect information. If problems are found, the system will try to re-extract or mark abnormal metadata for manual review.

[0122] In the present invention, the diversity of data is ensured by determining the source of the content of the advertising material. The local material library provides the advertising materials accumulated by the enterprise itself, while the online advertising platform provides a wealth of industry materials, and the materials provided by the customer reflect the needs of a specific market. By integrating materials from different sources, the present invention can build a comprehensive and detailed material library to provide diversified choices for advertising delivery. Secondly, the present invention lays a foundation for further processing and analysis of advertising materials by identifying and classifying the types of advertising materials, including text, images, audio and video. These different types of materials have different characteristics and application scenarios. Therefore, the identification and classification of material types help enterprises better understand the characteristics and delivery effects of advertising materials. In addition, the present invention also collects and records metadata of advertising materials, such as material type, material resolution, release date, target market, advertiser information, creative team, etc. These metadata are of great significance for the analysis and optimization of advertising materials. Metadata can help enterprises understand the background information of materials, such as delivery time, target audience, creative team, etc., so as to guide the formulation and optimization of advertising delivery strategies. Through the above steps, the present invention ensures that the collected information is comprehensive and detailed, and provides strong support for the further processing and analysis of advertising material data.

[0123] Optionally, the advertising material content cleaning data includes advertising text cleaning data, advertising image cleaning data, and advertising video cleaning data, and S13 includes:

[0124] Performing text standardization on the advertisement text cleaning data to obtain advertisement text standardized data;

[0125] In one embodiment of the present invention, first, the system converts all advertisement text cleaning data into a unified character encoding format (eg, UTF-8) to ensure consistency in subsequent processing. Texts in different encoding formats are converted to solve problems caused by encoding differences.

[0126] Perform text normalization operations, including lowercase conversion, removal of extra spaces and punctuation, expansion of abbreviations to full names, standardization of date and number formats, etc. These operations ensure the consistency and readability of text data.

[0127] Perform language detection on the ad text to identify the language of the text. For multilingual text, the system will choose to keep or translate the text in a specific language based on predefined rules (such as the main language of the target market).

[0128] By removing common meaningless words in the text (such as stop words), irrelevant words in advertisements, and redundant information, the system reduces the noise in the text data and improves the quality of the text.

[0129] The standardized text content is reformatted according to the format requirements of the advertising content, for example, ensuring that the paragraph structure is clear and the text length meets the predetermined standards.

[0130] Performing image standardization on the advertising image cleaning data to obtain advertising image standardized data;

[0131] In one embodiment of the present invention, the advertising image cleaning data is resized, and all images are uniformly scaled to a predefined resolution to ensure a consistent visual effect. This includes scaling, cropping, and padding operations to ensure that the display effects of the images on different platforms are consistent.

[0132] Convert the color space of the image to a standard color space (such as sRGB) to avoid display problems caused by color space differences. The system ensures that all images use the same color standard so that consistent colors can be displayed on different devices.

[0133] De-noising is performed on images to remove impurities and noise in the image and enhance the clarity of the image. De-noising can use a variety of techniques, such as smoothing filtering and edge enhancement.

[0134] Convert all image files to a unified file format (such as JPEG or PNG) and ensure that the image compression ratio and quality are within acceptable ranges. The system selects appropriate format and quality parameters based on the image usage scenario.

[0135] Perform geometric correction of images, such as rotation, mirroring, and calibration of image alignment, to ensure that the image content is in the correct direction and proportion on the display device.

[0136] Perform video standardization on the advertising video cleaning data to obtain advertising video standardized data;

[0137] In one embodiment of the present invention, the resolution of the advertising video cleaning data is adjusted to unify the output resolution of the video, which may involve scaling and cropping the video frame to make the video meet the standard resolution specification.

[0138] Adjust the frame rate of the video so that all videos have the same frame rate (for example, 30fps) to ensure smoothness and consistency during playback.

[0139] Convert the video encoding format to a unified standard format (such as H.264) and adjust the bit rate to balance video quality and file size. Ensure that all videos can be decoded and played smoothly on various playback devices.

[0140] The audio part of the video is standardized, including unifying the audio sampling rate, number of channels and bit rate to ensure consistency in sound quality. In addition, the system will filter noise and balance the audio volume.

[0141] Adjust the length of the video according to the requirements of the ad playback, including speeding up or slowing down the video to ensure that the video length meets the scheduled ad playback time.

[0142] The standardized data of the advertisement text, the standardized data of the advertisement image and the standardized data of the advertisement video are integrated to obtain the standardized data of the advertisement material content.

[0143] In one embodiment of the present invention, standardized data of advertisement text, standardized data of advertisement image and standardized data of advertisement video are merged. The system integrates different types of standardized data into a unified data set according to the usage scenario of the advertisement material.

[0144] Ensure that the merged data is synchronized on the timeline, especially for video and audio content, to ensure their timing is consistent to provide a consistent user experience.

[0145] Perform consistency check on the merged data to ensure that all data formats, resolutions, encoding methods, etc. are completely unified without any omissions or conflicts.

[0146] Regenerate metadata for the integrated data set, including new file size, format, duration, and other information, to ensure that all metadata is consistent with the standardized content.

[0147] Perform final verification on the standardized creative content data to ensure that all content complies with predefined standards and can be used normally on various advertising delivery platforms.

[0148] In the process of data cleaning and formatting in the present invention, a series of standardized methods are used to process advertising materials to ensure the quality of data and processing efficiency. First, the text materials are segmented and tagged with parts of speech, which are basic steps in natural language processing and help subsequent text analysis and content understanding. Word segmentation is to divide long texts into phrases and words, while part-of-speech tagging determines the part of speech of each word, such as noun, verb, adjective, etc. For image materials, the present invention ensures that all image materials have the same resolution and format through size unification and format conversion. Size unification can ensure that images can be displayed consistently in different analysis tools and models, while format conversion ensures that images can be correctly read and processed by different systems and devices.

[0149] Optionally, the video normalization includes:

[0150] Perform video frame extraction according to the advertisement video cleaning data to obtain video frame extraction data;

[0151] The present invention provides an embodiment in which the system first analyzes the original frame rate of the advertising video to determine a suitable frame extraction frequency. Depending on the dynamic level of the advertising content and the analysis requirements, the system may choose fixed interval frame extraction or content-based dynamic frame extraction. After determining the frame extraction frequency, the system extracts key frames or continuous frames from the advertising video. Each frame is extracted as a separate image file to ensure that important visual information in the video is retained. The extracted frames are quality checked to ensure that the clarity and brightness of each frame meet the standards. For blurred or poorly illuminated frames, the system will mark and re-extract or repair them through image enhancement technology. All extracted frames will be stored in an ordered directory, and each frame will be accompanied by a timestamp and frame number to ensure that the original position in the video can be accurately located during subsequent processing.

[0152] Perform audio conversion according to the video frame extraction data to obtain video frame extraction audio conversion data;

[0153] The present invention provides an embodiment in which an audio stream is separated from an original video file corresponding to video frame data to ensure that the audio and video content are synchronized in time. The separated audio stream is converted into a standardized audio format, such as a unified sampling rate and encoding method, to ensure that it can be played consistently on different devices. According to the timestamp of the video frame extraction, the audio data is aligned to ensure that the audio content of each frame is consistent with the video content. This step is the key to ensuring the synchronization of sound and picture. The audio data is subjected to noise reduction and equalization processing to improve the sound quality and make it clearer and more natural. The system automatically detects and removes background noise, and adjusts the volume to balance the audio effects of different scenes. The processed audio data is merged with the corresponding frame data to generate video frame data after audio conversion.

[0154] The script content is parsed according to the video frame extraction and audio conversion data to obtain the standardized data of the advertising video.

[0155] The present invention provides an embodiment in which any embedded video subtitles or accompanying script content are extracted from the video frame audio conversion data. The system converts the audio into text by recognizing the subtitle text or using speech recognition technology. The extracted text content is matched with the video frame to ensure that the text content of each frame is consistent with the video scene. The extracted text is semantically analyzed to identify key information in the advertising video, such as brand name, product features, advertising slogans, etc. The system uses natural language processing technology to understand and mark this key information. Based on the parsed script content, the system classifies the video content into different scenes or chapters, such as product introductions, user reviews, purchase tips, etc. Each scene is marked and classified to facilitate subsequent editing and delivery.

[0156] For video materials in the present invention, the present invention converts the video content into a series of static images and audio files through frame extraction and audio extraction, which helps to conduct a more in-depth analysis of the video content. Frame extraction can decompose the video content into key frames, while audio extraction can independently analyze the audio content in the video, such as music, narration and environmental noise, all of which can affect the effectiveness of the advertisement. By acquiring, recording, cleaning and formatting the content of advertising materials, the present invention provides a solid, clear and consistent data foundation for advertising material data labeling. Data cleaning ensures the quality of the data, reduces errors and deviations in subsequent labeling, and improves the accuracy of data labeling. Data format standardization improves the efficiency of data processing, ensures consistency between different data sources, and facilitates integrated analysis and model training.

[0157] Optionally, the advertisement material content annotation data includes advertisement text annotation data, advertisement image annotation data, and advertisement video annotation data, and the annotation reliability scoring data includes text annotation reliability scoring data, image annotation reliability scoring data, and video annotation reliability scoring data, and S2 includes:

[0158] S21, performing text data annotation on the standardized data of the advertisement text to obtain advertisement text annotation data, and performing text reliability scoring on the advertisement text annotation data to obtain text annotation reliability scoring data;

[0159] In one embodiment provided by the present invention, the system first parses the standardized data of the advertising text, extracts the key features of the text, such as sentence structure, keywords, themes, etc., to provide a basis for subsequent annotation. Based on predefined annotation rules, the advertising text is automatically annotated. The annotated content includes brand name, product characteristics, advertising theme, emotional tendency, etc. The system automatically identifies and marks this information by analyzing the text content. During the annotation process, the system considers the context of the text to improve the accuracy of the annotation. For example, for polysemous words or phrases with multiple interpretations, the system determines their exact meaning by analyzing the context. The preliminary text annotation results are verified to check whether there are obvious errors or omissions. The system automatically detects and corrects annotation errors according to the rules or by comparing the known correctly annotated data. The annotated text data is scored for reliability. The system assigns a reliability score to each annotation based on factors such as the accuracy, consistency, and contextual logic of the annotation. The score directly reflects the credibility of the annotation results.

[0160] S22, performing image data annotation on the standardized data of the advertisement image to obtain advertisement image annotation data, and performing image reliability scoring on the advertisement text annotation data to obtain image annotation reliability scoring data;

[0161] In one embodiment provided by the present invention, the system extracts features from standardized data of advertising images and identifies key elements in the images, such as brand logos, product images, color themes, characters, etc. These features will serve as the basis for annotation. Based on the extracted features, the system automatically annotates the images. The annotation content may include object categories, brand names, product models, advertising themes, etc. in the images. The system automatically marks these contents using visual recognition technology. During the annotation process, the system associates different image regions. For example, the product image is associated with the brand logo and annotated as part of the same advertising content. This helps to improve the overall accuracy of the annotation. The system performs a consistency check on the annotated image data to ensure that the same type of objects are consistently annotated in different images. Based on factors such as the accuracy, regional association, and consistency of the image annotation, the system assigns a reliability score to each image annotation. The scoring results will be used to evaluate the quality of the annotation.

[0162] S23, performing video data labeling on the standardized data of the advertising video to obtain advertising video labeling data, and performing video reliability scoring on the advertising video labeling data to obtain video labeling reliability scoring data;

[0163] The present invention provides an embodiment in which the system performs content segmentation on the standardized data of the advertising video, and divides the video into multiple segments, each of which represents an independent scene or chapter. This allows for more accurate annotation. The key frames in each video segment are analyzed to extract important elements in the frame, such as text subtitles, characters, product displays, etc. The system uses these elements as the basis for annotation. Based on the frame analysis results, the system automatically annotates the video content. The annotated content may include scene descriptions, character dialogues, subtitle content, advertising targets, etc. The system automatically generates annotations by analyzing the content of each video segment. During the video annotation process, the system associates the video frame with multimodal data such as audio and text to ensure the comprehensiveness and accuracy of the annotation. For example, the subtitle content is associated with the corresponding scene image for annotation. The system scores the reliability of the video annotation results, considering factors including the accuracy of the annotation, the consistency between frames, the correlation of multimodal data, etc. The score will be used to measure the credibility of the video annotation.

[0164] The text data annotation includes:

[0165] Extract keywords from the standardized data of the advertisement text to obtain keyword data of the advertisement text;

[0166] The present invention provides an embodiment in which the system first performs word segmentation processing on the standardized data of the advertising text, and divides the text into multiple words or phrases to facilitate the extraction of keywords. The granularity of word segmentation may be adjusted according to the complexity of the advertising content. The system automatically identifies the keywords in the advertising text by counting word frequency, inverse document frequency or other methods based on text features. Keywords are usually important words that reflect the theme of the advertisement, product characteristics, brand name, etc. The keywords that are initially identified are filtered to remove irrelevant or low-frequency words, such as common stop words or terms that are irrelevant to the advertising content, to ensure the accuracy and relevance of the keywords. The system assigns weights to the retained keywords and determines the weight of each keyword based on its importance and frequency of occurrence in the text. Keywords with higher weights will be used preferentially in subsequent annotation processes.

[0167] Perform sentiment orientation annotation on the keyword data of the advertisement text to obtain sentiment orientation annotation data;

[0168] The present invention provides an embodiment in which the system identifies emotion-related words in the keyword data of the advertising text, such as "like", "surprise", "anger", etc. These words are usually the main signs of emotional tendencies. The identified emotional words are classified to determine the emotional tendency of each keyword, such as positive, negative or neutral. During the classification process, the system will combine contextual information to avoid misclassification of emotional words. The system further analyzes the intensity of each emotional word, judges the intensity of its emotional expression through factors such as the modifiers of the word and the frequency of occurrence, and assigns an intensity score to its emotional tendency. The classification results of the emotional words are combined with the intensity score to generate the final emotional tendency annotation data. These data will reflect the overall emotional atmosphere of the advertising text.

[0169] Performing topic classification and annotation according to the keyword data of the advertisement text to obtain topic classification and annotation data, wherein the advertisement text topic classification and annotation data includes advertisement text purpose data and advertisement text industry data;

[0170] In one embodiment of the present invention, the system identifies words related to the advertisement theme in the advertisement text keyword data, such as "promotion", "new product", "environmental protection", etc. These keywords usually reflect the main purpose or content direction of the advertisement.

[0171] Based on the identified keywords, the system classifies the ad text into different subject categories. These categories may include brand promotion, product introduction, event promotion, etc., and the specific classification is set according to the actual content of the advertisement. Based on the subject classification, the purpose of the ad text is further refined, such as whether the ad is to increase brand awareness, promote sales, or promote an event. The system analyzes the correlation between text keywords and the purpose of the advertisement for precise labeling. Based on the industry-related keywords in the ad text, the system classifies the advertisement into specific industry fields, such as "technology", "food and beverage", "clothing", etc., to ensure that the industry orientation of the ad theme is clear.

[0172] Performing target audience annotation on the keyword data of the advertisement text to obtain target audience annotation data;

[0173] In one embodiment of the present invention, the system analyzes keyword data of the advertisement text and identifies characteristic words related to the target audience, such as "young people", "business people", "family users", etc. These words usually indicate the audience positioning of the advertisement. Based on the identified characteristic words, the system classifies the target audience of the advertisement, such as by age, gender, occupation, interest, etc. These classifications help the advertisement to be more accurately conveyed to specific audience groups.

[0174] Performing language style annotation on the advertisement text keyword data to obtain language style annotation data;

[0175] In one embodiment provided by the present invention, the system analyzes keyword data of the advertisement text to identify characteristic words and expressions related to the language style, such as "professional", "humorous", "direct", etc. These characteristics generally reflect the language style positioning of the advertisement. Based on the identified language characteristics, the system classifies the language style of the advertisement text, such as formal, informal, lively, serious, etc., to ensure that the style is consistent with the overall tonality of the advertisement. The system further analyzes the tone of the text, such as whether it is persuasive, friendly or urgent, which helps to more accurately determine the language style of the advertisement.

[0176] Performing promotion information tagging on the advertisement text keyword data to obtain promotion information tagging data;

[0177] In one embodiment of the present invention, the system identifies promotion-related words in the keyword data of the advertisement text, such as "discount", "limited time offer", "buy one get one free", etc. These words usually directly indicate the promotion information in the advertisement. The system classifies the identified promotion keywords, such as discount promotion, gift promotion, limited time promotion, etc., and marks the specific promotion type in the advertisement.

[0178] Calling for action annotation is performed on the keyword data of the advertisement text to obtain call for action annotation data;

[0179] In one embodiment of the present invention, the system identifies words related to calls to action in the keyword data of the advertisement text, such as "buy now", "contact us", "download the app", etc. These words are usually used to guide users to take specific actions. Action type classification: The system classifies the identified action words, such as purchase action, contact action, registration action, etc., and marks the specific action type advocated in the advertisement.

[0180] Performing grammar and sentence structure annotation on the standardized data of the advertisement text to obtain grammar and sentence structure annotation data;

[0181] The present invention provides an embodiment in which the system performs grammatical structure analysis on the standardized data of the advertising text, identifies the grammatical components in the text, such as subject, predicate, object, etc., and ensures the correctness of the sentence structure. Sentence classification: Based on the analysis results, the system classifies the sentences, such as declarative sentences, interrogative sentences, exclamatory sentences, etc., and marks the sentence types used in the advertising text.

[0182] Generate advertisement text annotation by using sentiment tendency annotation data, topic classification annotation data, target audience annotation data, language style annotation data, promotion information annotation data, call to action annotation data, and grammar and sentence structure annotation data to obtain advertisement text annotation data;

[0183] The present invention provides an embodiment in which sentiment tendency annotation data, topic classification annotation data, target audience annotation data, language style annotation data, promotional information annotation data, call for action annotation data, and grammar and sentence structure annotation data are integrated into a structured data to obtain advertising text annotation data.

[0184] The image data annotation includes:

[0185] Performing target detection on the standardized data of the advertising image to obtain target detection data;

[0186] In one embodiment provided by the present invention, the system analyzes the standardized data of the advertising image and identifies the key target areas in the image. These target areas may include important elements such as products, people, brand logos, etc. The system locates these target areas by analyzing pixel density, color distribution, and shape characteristics. After identifying the target areas, the system classifies them and determines the category of each target, such as "person", "product", "logo", etc., by analyzing the shape, size and texture characteristics of the target. The system assigns a label to each detected target area, annotates its category and location, and presents it in the image in the form of a bounding box or mask.

[0187] Perform scene recognition based on the target detection data and the standardized data of the advertising image to obtain scene recognition data;

[0188] In one embodiment of the present invention, the system uses target detection data to analyze scene elements in standardized data of advertising images, such as background, environment, spatial layout, etc. Based on the extracted scene elements, the system classifies the image into different scene types, such as "indoor", "outdoor", "urban", "natural", etc. The classification is based on the visual features and element combinations of the scene. The system assigns labels to the identified scene types and annotates them in the image data to ensure that the scene recognition results can be used for further analysis of the advertising content.

[0189] Performing color analysis on standardized data of advertisement images to obtain advertisement color data;

[0190] In one embodiment of the present invention, the system performs color extraction on standardized data of an advertising image and identifies the primary and secondary colors in the image. The extracted colors are usually represented in the form of a color histogram or a color ratio. The system matches the extracted colors with a predefined color library, classifies specific color types, such as "red", "blue", "green", etc., and analyzes whether the combination of these colors meets the expected style or brand requirements of the advertisement. The system further analyzes the overall color tone of the image, such as cool tones, warm tones, soft tones, etc., and marks the color emotional tendency of the image.

[0191] Performing image quality assessment on the standardized data of the advertisement image to obtain advertisement image quality assessment data;

[0192] The present invention provides an embodiment in which the system checks the resolution of standardized data of advertising images to ensure that the image clarity meets the advertising standards. If necessary, the system will mark low-resolution images and recommend reprocessing or replacement. The system analyzes the noise situation of the image, detects whether there is obvious image noise or distortion, and evaluates the impact of noise on image quality. The system analyzes the contrast and brightness of the image to ensure that the visual effect of the image meets the requirements of advertising display. Images that are too dark or too bright will be marked as poor quality. The system assigns a quality score to each advertising image, taking into account factors such as resolution, noise, contrast, and brightness. Images with low scores will be recommended for quality improvement.

[0193] Performing style recognition on the standardized data of the advertising image to obtain style recognition data;

[0194] In one embodiment of the present invention, the system analyzes the visual features in the standardized data of the advertising image and extracts elements related to the style, such as composition, color, light and shadow effects, etc. These features are used to define the visual style of the image. The system matches the extracted style features with predefined style categories, such as "modern", "retro", "minimalist", etc., to determine the visual style type of the image.

[0195] Performing brand recognition on the standardized data of the advertising image to obtain brand recognition data;

[0196] In one embodiment of the present invention, the system identifies any brand logo or trademark in the standardized data of the advertising image. By analyzing the shape, color and text content in the image, the system is able to locate the specific location of the brand logo. The system matches the detected brand logo with the brand database to determine the specific brand displayed in the advertisement, including matching analysis of the shape, color and text of the brand logo.

[0197] Perform sentiment analysis on standardized data of advertising images to obtain image sentiment bias data;

[0198] In one embodiment provided by the present invention, the system analyzes visual elements in the standardized data of advertising images, such as character expressions, scene atmosphere, color matching, etc., and extracts features related to emotions. Based on the extracted emotional features, the system classifies the emotional tendency of the image, such as "happy", "melancholy", "nervous", etc., and marks the emotional atmosphere conveyed by the advertising image. The system further analyzes the intensity of the emotional expression, determines whether the emotional expression of the image is strong, mild or neutral, and assigns an intensity score to the emotional tendency.

[0199] Generate image description for standardized data of advertisement images to obtain image text description data;

[0200] In one embodiment of the present invention, the system summarizes the main content in the standardized data of the advertising image and generates a brief image description, including a description of the main objects, scenes and activities shown in the image. The system converts the summarized image content into natural language and generates a text description of the image suitable for human reading. The text description should be concise and clear and accurately reflect the core content of the image. The system checks whether the generated image description is diverse, avoids repeated or identical description methods, and ensures that the description can clearly distinguish different advertising images.

[0201] Generate image data annotation based on target detection data, scene recognition data, advertising color data, advertising image quality assessment data, style recognition data, brand recognition data, image sentiment bias data, and image text description data to obtain advertising image annotation data;

[0202] The present invention provides an embodiment in which target detection data, scene recognition data, advertising color data, advertising image quality assessment data, style recognition data, brand recognition data, image emotion bias data and image text description data are integrated to form a complete advertising image annotation data set.

[0203] The video data annotation includes:

[0204] Performing text data annotation on the audio text data in the standardized data of the advertising video to obtain audio text annotation data;

[0205] In one embodiment provided by the present invention, the steps in the text data annotation provided by the present invention are used to process the audio text data in the standardized data of the advertising video.

[0206] Performing image data annotation on video image data in the advertisement video standardized data to obtain video image annotation data;

[0207] The present invention provides an embodiment in which the steps in the image data annotation provided by the present invention are used to process the audio text data in the standardized data of the advertising video.

[0208] Performing scene conversion annotation and character interaction annotation on the video image annotation data to obtain scene conversion annotation data and character interaction annotation data;

[0209] The present invention provides an embodiment in which the system analyzes continuous frames in the video image annotation data to detect scene transition points in the video, such as the moment of switching from one environment to another. Each scene transition point will be annotated and the scene type before and after the transition will be recorded. The system further analyzes the character actions and interactions in the video image annotation data, and annotates the types of interactions between the characters, such as dialogue, cooperation, conflict, etc. These interaction annotations help understand the relationships between characters and the development of the plot in the advertisement. The system generates scene transition annotation data, which annotates all scene switching points in the advertising video and the types of scenes before and after them. The system generates character interaction annotation data, which annotates the types and methods of interactions between characters in the advertising video.

[0210] The audio text annotation data, the scene transition annotation data and the character interaction annotation data are used to generate advertisement video annotations to obtain advertisement video annotation data.

[0211] The present invention provides an embodiment in which audio text annotation data, scene transition annotation data, and character interaction annotation data are integrated to form a complete advertisement video annotation data set.

[0212] In the process of automatic labeling of advertising material data in the present invention, the present invention adopts a series of advanced technologies and models to process different types of data. First, for text data, the present invention uses natural language processing technology, including part-of-speech tagging, sentiment analysis, entity recognition, etc., to label the content of advertising materials. Through these technologies, key information in the advertising copy, such as product name, promotional information, brand slogan, etc., can be extracted, and reliability scores are given to these labeling results to evaluate the accuracy of labeling. If there is no text data in the content of the advertising material, the text data labeling step is directly skipped. Secondly, for image data, the present invention uses computer vision technology, including target detection, image segmentation, image sentiment analysis, etc., to label the content of advertising materials. These technologies can identify target objects, scenes, characters, etc. in the image, and reliability scores are given to these labeling results to evaluate the accuracy of labeling. Finally, for video data, the present invention uses speech recognition and audio processing technology, including speech-to-text, speech sentiment analysis, etc., to label the content of advertising materials. These technologies can identify key information in audio, such as advertising slogans, narration, etc., and reliability scores are given to these labeling results to evaluate the accuracy of labeling. By using multiple models to automatically annotate text data, image data, and audio data, and scoring the reliability of the annotation results, the present invention can mutually verify and supplement these annotation results, thereby improving the accuracy and reliability of the annotation. This multimodal annotation method can fully understand the content and characteristics of advertising materials, and provide strong data support for subsequent advertising delivery and effect analysis.

[0213] Optionally, S3 includes:

[0214] S31, when it is determined that the annotation reliability score data is less than or equal to the preset annotation reliability score threshold data, manually annotating the standardized data of the advertising material content to obtain the advertising material content annotation update data;

[0215] The present invention provides an embodiment in which the system first compares the reliability score of the advertising material content annotation data with a preset scoring threshold. If the score is lower than or equal to the threshold, the system will determine that the annotation may have a large deviation or inaccuracy and require manual intervention. When it is determined that manual annotation is required, the system will assign the corresponding advertising material content standardized data to the manual annotation team. The system automatically generates a labeling task list, including annotation items that need to be checked in detail and parts that may have problems. The manual annotators conduct a detailed review and annotation correction of the advertising material content according to the task list provided by the system. The manual annotation process may include re-identifying the target, correcting the sentiment analysis, adjusting the scene recognition results, etc.

[0216] S32, performing differential analysis on the preliminary annotation update data and the advertisement material content annotation data to obtain annotation update differential data to perform an annotation parameter update operation;

[0217] The present invention provides an embodiment in which the system compares the preliminary annotation update data with the original advertising material content annotation data item by item to identify the differences between the two. The comparison content includes target detection results, scene recognition results, sentiment analysis results, etc. The system classifies the annotation differences into different types, such as text differences, image differences, sentiment differences, etc. Each difference type is further refined to facilitate subsequent annotation parameter updates. The system analyzes the degree of impact of each difference and determines its impact on the overall annotation quality. Larger differences will result in major adjustments to the annotation results, while smaller differences may only require fine-tuning of the annotation parameters. The system generates annotation update differentiation data, which records in detail all differences between the preliminary annotation update data and the original annotation data, and provides a basis for subsequent annotation parameter update operations (such as sending to the machine learning model used in several previous recognition steps for parameter learning or iteration to enhance the ability of advertising annotation).

[0218] S33: When it is determined that the annotation reliability score data is greater than the preset annotation reliability score threshold data, the advertisement material content annotation data is determined as advertisement material content annotation update data.

[0219] In one embodiment of the present invention, the system checks again whether the annotation reliability score data is higher than a preset score threshold. If the score is higher than the threshold, the system determines that the accuracy and reliability of the current annotation data have met the requirements and no further correction is required.

[0220] The present invention uses manual re-labeling to tune the labeling model, which is an iterative process of repetition and continuous optimization. In this process, we first manually check the labeled data with reliability lower than z points and re-label these data. After completing the manual re-labeling, analyze the difference between the manual re-labeling results and the model labeling results, identify the type and cause of the model prediction error, and thus provide a basis for adjusting the model parameters. By deeply analyzing these differences, we can better understand the weaknesses of the model and lay the foundation for improving its performance. Next, it is necessary to adjust the model parameters such as learning rate and regularization strength according to the error type analysis results. The adjustment of these parameters is aimed at optimizing the performance of the model so that it can be more accurate when dealing with similar problems. This process requires continuous trial and verification to achieve the best results. We record the manually re-corrected labeled data and mark them as "manual", and the reliability score is set to y, which helps to classify and manage the data in the future and ensure that high-quality data is fully utilized. Finally, the manually re-corrected labeled data is returned to the advertising content labeling model for learning, which helps to improve the accuracy of advertising content labeling, thereby providing strong data support for the optimization of advertising delivery effects and personalized recommendations. Through this iterative process, the model is continuously adjusted and improved to effectively enhance its performance and generalization capabilities, thus better serving the development of the advertising business.

[0221] Optionally, S4 includes:

[0222] S41, obtaining advertisement delivery data corresponding to the advertisement material content data, wherein the advertisement delivery data includes media platform advertisement data and media platform advertisement material delivery data;

[0223] In one embodiment of the present invention, the system first connects to the advertising data interface of each media platform to obtain data related to advertising delivery. These interfaces may include API calls, data synchronization or batch downloads. The system collects advertising data from the media platform, including the number of ad impressions, clicks, user interaction behaviors (such as likes, comments, shares, etc.), conversions, etc. The system further collects delivery data related to advertising materials, such as the display location, display time, delivery budget, target audience segmentation, etc. of the advertising materials.

[0224] S42: Calculate the delivery media conversion rate based on the advertisement delivery data to obtain advertisement delivery effect data.

[0225] The present invention provides an embodiment in which the system matches the advertising delivery data with the advertising material content data to ensure that each delivery data is associated with the corresponding advertising material, such as data matching in multiple dimensions such as advertising ID, display time, and media platform. The system calculates the basic display effect of the advertisement based on data such as the number of displays, the number of clicks, and interactive behaviors. The calculation content includes key indicators such as click-through rate (CTR), interaction rate, and conversion rate. The system further calculates the conversion rate of the media platform, specifically including the conversion rate from advertising display to user action (such as purchase, registration, and download), such as analyzing the various data performances of the advertisement and combining the user behavior patterns of different media platforms. The system integrates various conversion rate data, evaluates the overall effect of advertising delivery, and generates advertising delivery effect data.

[0226] In the present invention, the content of advertising material delivery is obtained through the media platform, and the original information of the material is recorded, including promotion goals, optimization goals, marketing scenarios, etc., to ensure that the collected information is comprehensive and detailed. The advertising material data of the media platform is obtained, including exposure, click-through volume, conversion volume, conversion rate, cost, ROI, etc. Data cleaning processing is performed, and this link removes irrelevant or erroneous data, such as unifying the data format of multiple platforms, correcting erroneous information, etc., thereby improving the accuracy and reliability of the data. The format of the cleaned data is standardized to ensure that the data conforms to the predetermined structure and format, such as converting all advertising information and advertising data into a unified encoding format to ensure consistent format and obtain cleaned data. By acquiring, recording, cleaning and formatting the advertising information and advertising data of the advertising material, a solid, clear and consistent data foundation is provided for the labeling of the advertising material data. Data cleaning ensures data quality, reduces errors and deviations in subsequent labeling, and improves the accuracy of data labeling. Data format standardization improves the efficiency of data processing, ensures consistency between different data sources, and facilitates integrated analysis and model training.

[0227] Optionally, S5 includes:

[0228] S51, annotating the advertising material content data according to the advertising delivery effect data to obtain advertising delivery effect annotated data;

[0229] In one embodiment of the present invention, the system first analyzes the advertising delivery effect data to extract key indicators related to the advertising material content, such as click-through rate, conversion rate, user interaction, etc. Based on the analyzed delivery effect data, the system annotates the advertising material content data.

[0230] S52, constructing a similarity matrix according to the advertisement delivery effect annotation data and the advertisement material content annotation update data to obtain advertisement delivery similarity matrix data;

[0231] In one embodiment of the present invention, a system extracts relevant features from advertisement delivery effect annotation data and advertisement material content annotation update data, including keywords, colors, emotional tendencies, user behaviors, etc. Based on the extracted features, the system calculates the similarity between the advertisement delivery effect annotation data and the advertisement material content annotation update data, such as calculations in multiple dimensions, such as text similarity, visual similarity, emotional similarity, etc. The calculated similarity results are represented in a matrix form to construct an advertisement delivery similarity matrix. Each element of the matrix represents the similarity between two advertisement materials or effect annotations.

[0232] S53, performing conflict detection according to the advertisement placement similarity matrix data to obtain conflict detection data;

[0233] In one embodiment of the present invention, the system presets a set of conflict detection rules, which may include similarity thresholds, content conflict conditions, performance conflict conditions, etc., for determining whether there is a conflict between the contents of advertising materials. The system performs conflict detection according to the set rules based on the advertising delivery similarity matrix data. The system will identify possible content conflicts, performance conflicts or other inconsistencies. The system classifies and prioritizes the detected conflicts, and marks conflicts that need to be handled immediately and minor conflicts that can be ignored according to the type and severity of the conflict. The output of this step is conflict detection data, which marks the conflicts between the advertising delivery effect annotation data and the advertising material content annotation update data.

[0234] S54: Perform context dependency analysis and consistency verification on the advertisement delivery effect annotation data and the advertisement material content annotation update data according to the conflict detection data to obtain the advertisement material annotation data.

[0235] In one embodiment of the present invention, the system extracts context information from the advertising material content and delivery effect annotation data, which may include the context of the advertising content, the delivery time period, the audience group characteristics, etc. Based on the context information, the system analyzes the dependency between the advertising material and the delivery effect, and determines the relevance and consistency between different annotation data.

[0236] The present invention accurately reflects the performance of advertising materials in actual delivery, making delivery effect analysis more accurate and providing data support for further optimization of advertising strategies. By performing similarity calculations on the advertising delivery effect annotation data and the advertising material content annotation update data, a deep association between advertising materials and delivery effects is formed. Through the conflict detection in step S53, the method can identify the contradictions or inconsistencies between the advertising delivery effect annotation data and the advertising material content annotation update data. The context dependency analysis and consistency verification in step S54 ensure that the annotation results are consistent with the context of the advertising content by performing in-depth context analysis on the advertising delivery effect annotation data and the advertising material content annotation update data.

[0237] Optionally, the present application further provides an AI-based advertising material data labeling system, which is used to execute the AI-based advertising material data labeling method as described above, and the AI-based advertising material data labeling system includes:

[0238] The advertisement material content standardization module is used to obtain advertisement material content data and perform standardization processing on the advertisement material content data to obtain advertisement material content standardized data;

[0239] The marking reliability scoring module is used to automatically mark the standardized data of the advertising material content to obtain the advertising material content marking data; and to perform reliability scoring on the advertising material content marking data to obtain the marking reliability scoring data;

[0240] A secondary annotation operation module is used to perform secondary annotation operation on the advertisement material content annotation data according to the annotation reliability score data to obtain advertisement material content annotation update data;

[0241] The advertisement delivery effect extraction module is used to obtain the advertisement delivery data corresponding to the advertisement material content data, and extract the delivery effect according to the advertisement delivery data to obtain the advertisement delivery effect data;

[0242] The advertising material annotation consistency processing module is used to annotate the advertising material content data according to the advertising delivery effect data to obtain the advertising delivery effect annotation data, and to perform annotation consistency processing on the advertising delivery effect annotation data and the advertising material content annotation update data to obtain the advertising material annotation data.

[0243] Therefore, from any point of view, the embodiments should be regarded as illustrative and non-restrictive, and the scope of the present invention is limited by the attached application documents rather than the above description, and it is intended that all changes falling within the meaning and scope of equivalent elements of the application documents are included in the present invention.

[0244] The above description is only a specific embodiment of the present invention, so that those skilled in the art can understand or implement the present invention. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but should conform to the widest scope consistent with the principles and novel features invented herein.

Claims

1. An AI-based advertising material data annotation method, characterized in that: The method comprises: S1. Acquire advertising material content data, and perform standardization processing on the advertising material content data to obtain standardized advertising material content data; S2. Automatically label the standardized data of the advertising material content to obtain advertising material content labeling data; and perform reliability scoring on the advertising material content labeling data to obtain labeling reliability scoring data; S3, performing a secondary annotation operation on the advertising material content annotation data according to the annotation reliability score data to obtain advertising material content annotation update data; S4, obtaining advertisement delivery data corresponding to the advertisement material content data, and extracting delivery effects according to the advertisement delivery data to obtain advertisement delivery effect data; S5. Annotate the advertising material content data according to the advertising delivery effect data to obtain advertising delivery effect annotated data, and perform annotated consistency processing on the advertising delivery effect annotated data and the advertising material content annotated update data to obtain advertising material annotated data; S1 includes: S11. Acquire advertising material content through the own material library or the synchronous media platform material library, and record the original information of the material to obtain advertising material content data; S12, cleaning the advertising material content data to obtain advertising material content cleaning data; S13, formatting the cleaned data of the advertising material content to obtain standardized data of the advertising material content; Among them, S11 includes: Get the source data of the advertising creative content; Classify the advertising material type according to the advertising material content source data to obtain advertising material type classification data, wherein the advertising material type classification data includes advertising text type classification data, advertising image type classification data, advertising audio type classification data and advertising video type classification data; Collect and record metadata of advertising material type classification data to obtain advertising material content data; The advertising material content cleaning data includes advertising text cleaning data, advertising image cleaning data and advertising video cleaning data, and S13 includes: Performing text standardization on the advertisement text cleaning data to obtain advertisement text standardized data; Performing image standardization on the advertising image cleaning data to obtain advertising image standardized data; Perform video standardization on the advertising video cleaning data to obtain advertising video standardized data; Integrate the standardized data of advertisement text, the standardized data of advertisement image and the standardized data of advertisement video to obtain the standardized data of advertisement material content; Video standardization includes: Perform video frame extraction according to the advertisement video cleaning data to obtain video frame extraction data; Perform audio conversion according to the video frame extraction data to obtain video frame extraction audio conversion data; Parse the script content based on the video frame extraction and audio conversion data to obtain the standardized data of the advertising video; The advertising material content annotation data includes advertising text annotation data, advertising image annotation data, and advertising video annotation data, and the annotation reliability score data includes text annotation reliability score data, image annotation reliability score data, and video annotation reliability score data. S2 includes: S21, performing text data annotation on the standardized data of the advertisement text to obtain advertisement text annotation data, and performing text reliability scoring on the advertisement text annotation data to obtain text annotation reliability scoring data; S22, performing image data annotation on the standardized data of the advertisement image to obtain advertisement image annotation data, and performing image reliability scoring on the advertisement text annotation data to obtain image annotation reliability scoring data; S23, performing video data labeling on the standardized data of the advertising video to obtain advertising video labeling data, and performing video reliability scoring on the advertising video labeling data to obtain video labeling reliability scoring data; The text data annotation includes: Extract keywords from the standardized data of the advertisement text to obtain keyword data of the advertisement text; Perform sentiment orientation annotation on the keyword data of the advertisement text to obtain sentiment orientation annotation data; Performing topic classification and annotation according to the keyword data of the advertisement text to obtain topic classification and annotation data, wherein the advertisement text topic classification and annotation data includes advertisement text purpose data and advertisement text industry data; Performing target audience annotation on the keyword data of the advertisement text to obtain target audience annotation data; Performing language style annotation on the advertisement text keyword data to obtain language style annotation data; Performing promotion information tagging on the advertisement text keyword data to obtain promotion information tagging data; Calling for action annotation is performed on the keyword data of the advertisement text to obtain call for action annotation data; Performing grammar and sentence structure annotation on the standardized data of the advertisement text to obtain grammar and sentence structure annotation data; Generate advertisement text annotation by using sentiment tendency annotation data, topic classification annotation data, target audience annotation data, language style annotation data, promotion information annotation data, call to action annotation data, and grammar and sentence structure annotation data to obtain advertisement text annotation data; The image data annotation includes: Performing target detection on the standardized data of the advertising image to obtain target detection data; Perform scene recognition based on the target detection data and the standardized data of the advertising image to obtain scene recognition data; Performing color analysis on standardized data of advertisement images to obtain advertisement color data; Performing image quality assessment on the standardized data of the advertisement image to obtain advertisement image quality assessment data; Performing style recognition on the standardized data of the advertising image to obtain style recognition data; Performing brand recognition on the standardized data of the advertising image to obtain brand recognition data; Perform sentiment analysis on standardized data of advertising images to obtain image sentiment bias data; Generate image description for standardized data of advertisement images to obtain image text description data; Generate image data annotation based on target detection data, scene recognition data, advertising color data, advertising image quality assessment data, style recognition data, brand recognition data, image sentiment bias data, and image text description data to obtain advertising image annotation data; The video data annotation includes: Performing text data annotation on the audio text data in the standardized data of the advertising video to obtain audio text annotation data; Performing image data annotation on video image data in the advertisement video standardized data to obtain video image annotation data; Performing scene conversion annotation and character interaction annotation on the video image annotation data to obtain scene conversion annotation data and character interaction annotation data; The audio text annotation data, the scene transition annotation data, and the character interaction annotation data are used to generate advertisement video annotations to obtain advertisement video annotation data; S3 includes: S31, when it is determined that the annotation reliability score data is less than or equal to the preset annotation reliability score threshold data, manually annotating the standardized data of the advertising material content to obtain the advertising material content annotation update data; S32, performing differential analysis on the preliminary annotation update data and the advertisement material content annotation data to obtain annotation update differential data to perform an annotation parameter update operation; S33: when it is determined that the annotation reliability score data is greater than the preset annotation reliability score threshold data, the advertisement material content annotation data is determined as advertisement material content annotation update data; S4 includes: S41, obtaining advertisement delivery data corresponding to the advertisement material content data, wherein the advertisement delivery data includes media platform advertisement data and media platform advertisement material delivery data; S42, calculating the delivery media conversion rate according to the advertisement delivery data to obtain advertisement delivery effect data; S5 includes: S51, annotating the advertising material content data according to the advertising delivery effect data to obtain advertising delivery effect annotated data; S52, constructing a similarity matrix according to the advertisement delivery effect annotation data and the advertisement material content annotation update data to obtain advertisement delivery similarity matrix data; S53, performing conflict detection according to the advertisement placement similarity matrix data to obtain conflict detection data; S54: Perform context dependency analysis and consistency verification on the advertisement delivery effect annotation data and the advertisement material content annotation update data according to the conflict detection data to obtain the advertisement material annotation data.

2. An AI-based advertising material data annotation system, characterized in that: Used to execute the AI-based advertising material data labeling method according to claim 1, the AI-based advertising material data labeling system comprises: The advertisement material content standardization module is used to obtain advertisement material content data and perform standardization processing on the advertisement material content data to obtain advertisement material content standardized data; The marking reliability scoring module is used to automatically mark the standardized data of the advertising material content to obtain the advertising material content marking data; and to perform reliability scoring on the advertising material content marking data to obtain the marking reliability scoring data; A secondary annotation operation module is used to perform secondary annotation operation on the advertisement material content annotation data according to the annotation reliability score data to obtain advertisement material content annotation update data; The advertisement delivery effect extraction module is used to obtain the advertisement delivery data corresponding to the advertisement material content data, and extract the delivery effect according to the advertisement delivery data to obtain the advertisement delivery effect data; The advertising material annotation consistency processing module is used to annotate the advertising material content data according to the advertising delivery effect data to obtain the advertising delivery effect annotation data, and to perform annotation consistency processing on the advertising delivery effect annotation data and the advertising material content annotation update data to obtain the advertising material annotation data.

Citation Information

Patent Citations

  • Delivery material mining method and device, equipment and storage medium

    CN117114772A

  • AI-driven advertisement originality optimization method and system

    CN117829911A

  • Advertisement preview analysis method and device, equipment and storage medium

    CN118134567A

  • Systems and Methods for Continuous Analysis and Procurement of Advertisement Campaigns

    US20160071162A1

  • Method and system to assess image advertisements

    US20230162228A1