Media asset intelligent auditing system based on large model
Through a large-model-based media intelligent audit system, deep learning and machine learning technology are used to automatically identify and analyze media resources, solving the problems of low efficiency and high cost of traditional manual audits, and achieving efficient and accurate audit of media resources and Healthy and orderly development.
Patent Information
- Application Number
- CN202410850577.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-06-28
- Publication Date
- 2024-11-15
AI Technical Summary
Traditional manual review of media resources is inefficient and susceptible to subjective factors, resulting in inconsistency and high error rates, and is expensive, making it impossible to cope with the challenges of massive media growth.
Design a large-scale model-based intelligent media audit system, including content recognition module, risk prediction module and intelligent decision-making module, use deep learning technology to automatically identify and analyze media resources, predict potential risks through machine learning algorithms, and automatically formulate audits based on results. Strategy, self-learning and scalability.
It realizes automatic, fast and accurate audit of media resources, reduces the error rate caused by human factors, improves audit efficiency and accuracy, adapts to different media types and needs, reduces manual audit costs, and promotes the healthy media resources. Order development.
Smart Images

Figure CN118965190A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of intelligent review of media assets, and in particular to an intelligent review system of media assets based on a large model. Background Art
[0002] With the rapid development of information technology and the popularization of the Internet, media content has shown an explosive growth trend. From short videos and audio files on social media, to massive graphic reports on online news websites, to high-definition video content on various streaming platforms, the number and types of media resources are increasing. However, this growth has also brought many challenges, one of which is how to efficiently and accurately review these massive media resources.
[0003] Traditional media review methods mainly rely on manual reviewers to review media content one by one, but this method has obvious limitations. First, manual review is inefficient and cannot cope with the growing amount of media content. Second, manual review is easily affected by subjective factors, resulting in inconsistency in review results and a high error rate. Finally, manual review requires a lot of human resources and is costly.
[0004] Therefore, with the continuous development of artificial intelligence technology, especially the breakthrough of deep learning technology, it has become possible to use large models to conduct intelligent review of media resources. Deep learning technology can automatically learn and extract the characteristics of media content, and train efficient review models through large amounts of data, thereby realizing automatic, fast and accurate review of media resources. Summary of the invention
[0005] 1. Technical issues to be solved
[0006] In view of the deficiencies in the prior art, the present invention provides a media asset intelligent review system based on a large model to solve the above problems.
[0007] (II) Technical solution
[0008] To achieve the above-mentioned purpose, the present invention provides the following technical solutions: a media resource intelligent review system based on a large model, comprising: a content recognition module, a risk prediction module and an intelligent decision-making module, wherein the content recognition module is used to recognize various information such as text, images, audio, video, etc. in media resources, and perform in-depth analysis; the risk prediction module uses a machine learning algorithm based on historical data and real-time data to predict and evaluate potential risks in media resources; the intelligent decision-making module automatically formulates review strategies and gives review recommendations or decisions based on content recognition and risk prediction results.
[0009] Preferably, the content recognition module can accurately recognize text, images, audio, video and other information, and perform sentiment analysis and content classification.
[0010] Preferably, the risk prediction module can monitor content updates in media resources in real time and perform risk assessment and early warning in a timely manner.
[0011] Preferably, the intelligent decision-making module can optimize the audit strategy and decision-making model based on the audit history data and feedback information.
[0012] Preferably, the system has self-learning capabilities and can continuously adjust and optimize audit algorithms and models based on feedback information from actual applications.
[0013] Preferably, the system can adapt to different media resource types and audit requirements and has a high degree of scalability and flexibility.
[0014] According to another aspect of an embodiment of the present invention, a method for intelligent review of media assets based on a large model is also provided, comprising the following steps:
[0015] S1. System framework design: Select a suitable large model as the basis, combine it with the deep learning framework to build the system, and construct the core modules of content recognition, risk prediction, and intelligent decision-making;
[0016] S2. Data collection and processing: Collect various types of media resource data, including text, images, audio, and video, and perform data cleaning, labeling, and preprocessing to ensure data quality;
[0017] S3. Model training: Use deep learning technology to train large models so that they can accurately identify and analyze different types of media resources. Use transfer learning methods to accelerate the training process and improve model performance.
[0018] S4. Review rule setting: According to business needs and relevant laws and regulations, set the rules and standards for media resource review, embed the review rules into the system as a reference for the intelligent decision-making module;
[0019] S5. Multimodal data analysis: For text data, use natural language processing (NLP) technology to analyze; for image data, use computer vision (CV) technology to analyze; for audio data, use audio processing technology to analyze; for video data, use a combination of image and audio processing technology to analyze;
[0020] S6. Feature extraction: Extract features from the parsed information to generate feature vectors that represent the content of the media resource. Use convolutional neural networks (CNNs) to extract image features, and use recurrent neural networks (RNNs) or Transformers to extract text and audio features.
[0021] S7. Deep analysis: Use the generated feature vectors to perform deep analysis and identify potential content issues, such as sensitive words, violent images, and inappropriate audio.
[0022] S8. Data collection and preprocessing: Collect historical data and real-time data, including reviewed media resource information, user feedback, and changes in laws and regulations, and preprocess the data, such as denoising and standardization;
[0023] S9. Machine learning model: Select appropriate machine learning algorithms to train risk prediction models and use ensemble learning methods to improve the prediction accuracy of the models;
[0024] S10. Risk Assessment: Perform risk assessment on media resources, generate risk level and risk type information, and use multi-task learning methods to simultaneously predict multiple risk dimensions;
[0025] S11. Audit strategy formulation: Based on the results of content identification and risk prediction, combined with the set audit rules and standards, the audit strategy is automatically formulated, and the audit strategy is formulated using rule engines or decision tree technology;
[0026] S12. Output audit suggestions: Output the audit strategy to system users or managers in the form of suggestions or decisions;
[0027] S13. Manual review and feedback: Users conduct manual review based on suggestions to ensure the accuracy of the review results, and input the feedback information from manual review into the system to continuously optimize and improve the model;
[0028] S14. Content recognition: Text: Use the BERT model to parse text content and extract keywords and sentence features; Image: Use the ResNet model to parse image content and extract image features; Audio: Use the WaveNet model to parse audio content and extract audio features;
[0029] S15. Feature fusion: Fuse the feature vectors of text, image and audio to generate a comprehensive feature vector;
[0030] S16. Risk prediction: The comprehensive feature vector is input into the deep neural network (DNN) model to perform risk prediction. The prediction results include risk level and risk type.
[0031] S17. Intelligent decision-making: formulate audit strategies based on risk prediction results and preset audit rules. If the risk level is high, it is recommended to reject; if the risk level is medium, it is recommended to wait for review;
[0032] S18. Output and feedback: The audit suggestions are output to the system users or managers, who conduct manual review and feed back the review results to the system to optimize the model.
[0033] (III) Beneficial effects
[0034] Compared with the prior art, the present invention provides a media asset intelligent review system based on a large model, which has the following beneficial effects:
[0035] 1. Improved efficiency and accuracy: By utilizing advanced deep learning technology and big data processing capabilities, the media resource intelligent review system of the present invention can realize automatic, fast and accurate review of media resources, which greatly improves the efficiency of the review, while reducing the error rate caused by human factors and ensuring the accuracy of the review.
[0036] 2. Strong adaptability: This system is highly scalable and flexible, and can adapt to different types of media resources and audit requirements. Whether it is different types of media resources such as text, images, audio or video, the system can perform efficient and accurate audits to meet the needs of various audit scenarios.
[0037] 3. Reduce manual review costs: By realizing automated review of media resources, the present invention can greatly reduce the cost of manual review, which not only reduces the pressure of manual review, but also improves the efficiency and quality of review, making the management of media resources more efficient and convenient.
[0038] 4. Promote the healthy and orderly development of media content: The implementation of the present invention can ensure the healthy and orderly development of media content. By accurately identifying potential risks in media resources and giving corresponding review suggestions or decisions, the system can effectively curb the spread of bad information and protect the legitimate rights and interests of users. BRIEF DESCRIPTION OF THE DRAWINGS
[0039] Figure 1 The figure is a schematic diagram of the overall workflow of the media asset intelligent review system based on the big model of the present invention. DETAILED DESCRIPTION
[0040] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only 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 ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0041] It should be understood that in various embodiments of the present invention, the size of the sequence number of each process does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.
[0042] It should be understood that in the present invention, "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or apparatus comprising a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to these processes, methods, products or apparatuses.
[0043] See also Figure 1 The media resource intelligent review system based on the big model uses the powerful data processing and feature extraction capabilities of the big model to achieve efficient and accurate review of massive media resources, meeting the challenges posed to the review system by the current explosive growth of media content. The system has designed multi-dimensional review modules including content identification, risk prediction, and intelligent decision-making. The content identification module can accurately identify various types of information in media resources. The risk prediction module performs risk assessment based on historical data and real-time data, and the intelligent decision-making module automatically formulates review strategies based on the results of the first two modules, realizing a comprehensive and detailed review process.
[0044] System construction and initialization:
[0045] According to the needs of the media resource intelligent review system, a system framework based on large models is designed and built, including core modules such as content identification, risk prediction, and intelligent decision-making. A large amount of media resource data, including text, images, audio, video, etc., is collected as training data. The model is trained using deep learning technology and big data processing capabilities to enable it to accurately identify information in media resources and conduct in-depth analysis. According to business needs and relevant laws and regulations, rules and standards for media resource review are set and embedded in the system as a reference for the intelligent decision-making module.
[0046] Content Identification Module:
[0047] The content recognition module needs to parse the received media resources, extract the text, image, audio, video and other information therein, perform feature extraction on the recognized information, and generate a feature vector representing the content of the media resources.
[0048] Risk prediction module:
[0049] Collect historical data and real-time data, including audited media resource information, user feedback, changes in laws and regulations, etc., and perform preprocessing and feature extraction. Input the processed data into the risk prediction model, and use machine learning algorithms to predict and evaluate potential risks in media resources. The prediction results can include information such as risk level and risk type.
[0050] Intelligent decision-making module:
[0051] The intelligent decision-making module automatically formulates audit strategies based on the results of content identification and risk prediction, combined with the set audit rules and standards. The strategies can include different processing suggestions such as approval, rejection, and pending review. The formulated audit strategies are output to system users or managers in the form of suggestions or decisions. Users can conduct manual reviews based on the suggestions or directly execute the decision results.
[0052] Among them, the content recognition module can accurately identify text, images, audio, video and other information, and perform sentiment analysis and content classification.
[0053] Among them, the risk prediction module can monitor content updates in media resources in real time and conduct risk assessment and early warning in a timely manner.
[0054] Among them, the intelligent decision-making module can optimize the audit strategy and decision-making model based on the audit history data and feedback information.
[0055] Among them, the system has self-learning capabilities and can continuously adjust and optimize audit algorithms and models based on feedback information in actual applications.
[0056] Among them, the system can adapt to different media resource types and audit requirements, and is highly scalable and flexible.
[0057] According to another aspect of an embodiment of the present invention, a method for intelligent review of media assets based on a large model is also provided, comprising the following steps:
[0058] S1. System framework design: Select a suitable large model as the basis, combine it with the deep learning framework to build the system, and construct the core modules of content recognition, risk prediction, and intelligent decision-making;
[0059] S2. Data collection and processing: Collect various types of media resource data, including text, images, audio, and video, and perform data cleaning, labeling, and preprocessing to ensure data quality;
[0060] S3. Model training: Use deep learning technology to train large models so that they can accurately identify and analyze different types of media resources. Use transfer learning methods to accelerate the training process and improve model performance.
[0061] S4. Review rule setting: According to business needs and relevant laws and regulations, set the rules and standards for media resource review, embed the review rules into the system as a reference for the intelligent decision-making module;
[0062] S5. Multimodal data analysis: For text data, use natural language processing (NLP) technology to analyze; for image data, use computer vision (CV) technology to analyze; for audio data, use audio processing technology to analyze; for video data, use a combination of image and audio processing technology to analyze;
[0063] S6. Feature extraction: Extract features from the parsed information to generate feature vectors that represent the content of the media resource. Use convolutional neural networks (CNNs) to extract image features, and use recurrent neural networks (RNNs) or Transformers to extract text and audio features.
[0064] S7. Deep analysis: Use the generated feature vectors to perform deep analysis and identify potential content issues, such as sensitive words, violent images, and inappropriate audio.
[0065] S8. Data collection and preprocessing: Collect historical data and real-time data, including reviewed media resource information, user feedback, and changes in laws and regulations, and preprocess the data, such as denoising and standardization;
[0066] S9. Machine learning model: Select appropriate machine learning algorithms to train risk prediction models and use ensemble learning methods to improve the prediction accuracy of the models;
[0067] S10. Risk Assessment: Perform risk assessment on media resources, generate risk level and risk type information, and use multi-task learning methods to simultaneously predict multiple risk dimensions;
[0068] S11. Audit strategy formulation: Based on the results of content identification and risk prediction, combined with the set audit rules and standards, the audit strategy is automatically formulated, and the audit strategy is formulated using rule engines or decision tree technology;
[0069] S12. Output audit suggestions: Output the audit strategy to system users or managers in the form of suggestions or decisions;
[0070] S13. Manual review and feedback: Users conduct manual review based on suggestions to ensure the accuracy of the review results, and input the feedback information from manual review into the system to continuously optimize and improve the model;
[0071] S14. Content recognition: Text: Use the BERT model to parse text content and extract keywords and sentence features; Image: Use the ResNet model to parse image content and extract image features; Audio: Use the WaveNet model to parse audio content and extract audio features;
[0072] S15. Feature fusion: Fuse the feature vectors of text, image and audio to generate a comprehensive feature vector;
[0073] S16. Risk prediction: The comprehensive feature vector is input into the deep neural network (DNN) model to perform risk prediction. The prediction results include risk level and risk type.
[0074] S17. Intelligent decision-making: formulate audit strategies based on risk prediction results and preset audit rules. If the risk level is high, it is recommended to reject; if the risk level is medium, it is recommended to wait for review;
[0075] S18. Output and feedback: The audit suggestions are output to the system users or managers, who conduct manual review and feed back the review results to the system to optimize the model.
[0076] In summary, the present invention provides a media asset intelligent review system based on a big model and a media asset intelligent review method based on a big model. Based on the big model and combined with the characteristics of media resources, a multi-dimensional review module including content identification, risk prediction, and intelligent decision-making is constructed. The content identification module can accurately identify various information such as text, images, audio, video, etc. in media resources, and conduct in-depth analysis thereof. The risk prediction module uses machine learning algorithms based on historical data and real-time data to predict and evaluate potential risks in media resources. The intelligent decision-making module automatically formulates review strategies based on the results of content identification and risk prediction, and gives review suggestions or decisions, which greatly improves the efficiency of the review, while reducing the error rate caused by human factors and ensuring the accuracy of the review.
[0077] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.
Claims
1. Intelligent media asset review system based on big model, including: The content recognition module, risk prediction module and intelligent decision-making module are characterized in that the content recognition module is used to identify various information such as text, images, audio, video, etc. in media resources and perform in-depth analysis; the risk prediction module uses machine learning algorithms based on historical data and real-time data to predict and evaluate potential risks in media resources; the intelligent decision-making module automatically formulates audit strategies and gives audit recommendations or decisions based on content recognition and risk prediction results.
2. The media asset intelligent review system based on a large model according to claim 1 is characterized in that: The content recognition module can accurately recognize text, images, audio, video and other information, and perform sentiment analysis and content classification.
3. The media asset intelligent review system based on a large model according to claim 1 is characterized in that: The risk prediction module can monitor content updates in media resources in real time and conduct risk assessment and early warning in a timely manner.
4. The media asset intelligent review system based on a large model according to claim 3 is characterized by: The intelligent decision-making module can optimize the audit strategy and decision-making model based on the audit history data and feedback information.
5. The media asset intelligent review system based on a large model according to claim 1 is characterized in that: The system has self-learning capabilities and can continuously adjust and optimize audit algorithms and models based on feedback information from actual applications.
6. The media asset intelligent review system based on a large model according to claim 1 is characterized in that: The system can adapt to different media resource types and audit requirements and has high scalability and flexibility.
7. A media asset intelligent review method based on a large model, characterized in that: The following steps are involved: S1. System framework design: Select a suitable large model as the basis, combine it with the deep learning framework to build the system, and construct the core modules of content recognition, risk prediction, and intelligent decision-making; S2. Data collection and processing: Collect various types of media resource data, including text, images, audio, and video, and perform data cleaning, labeling, and preprocessing to ensure data quality; S3. Model training: Use deep learning technology to train large models so that they can accurately identify and analyze different types of media resources. Use transfer learning methods to accelerate the training process and improve model performance. S4. Review rule setting: According to business needs and relevant laws and regulations, set the rules and standards for media resource review, embed the review rules into the system as a reference for the intelligent decision-making module; S5. Multimodal data analysis: For text data, use natural language processing (NLP) technology to analyze; for image data, use computer vision (CV) technology to analyze; for audio data, use audio processing technology to analyze; for video data, use a combination of image and audio processing technology to analyze; S6. Feature extraction: Extract features from the parsed information to generate feature vectors that represent the content of the media resource. Use convolutional neural networks (CNNs) to extract image features, and use recurrent neural networks (RNNs) or Transformers to extract text and audio features. S7. Deep analysis: Use the generated feature vectors to perform deep analysis and identify potential content issues, such as sensitive words, violent images, and inappropriate audio. S8. Data collection and preprocessing: Collect historical data and real-time data, including reviewed media resource information, user feedback, and changes in laws and regulations, and preprocess the data, such as denoising and standardization; S9. Machine learning model: Select appropriate machine learning algorithms to train risk prediction models and use ensemble learning methods to improve the prediction accuracy of the models; S10. Risk Assessment: Perform risk assessment on media resources, generate risk level and risk type information, and use multi-task learning methods to simultaneously predict multiple risk dimensions; S11. Audit strategy formulation: Based on the results of content identification and risk prediction, combined with the set audit rules and standards, the audit strategy is automatically formulated, and the audit strategy is formulated using rule engines or decision tree technology; S12. Output audit suggestions: Output the audit strategy to system users or managers in the form of suggestions or decisions; S13. Manual review and feedback: Users conduct manual review based on suggestions to ensure the accuracy of the review results, and input the feedback information from manual review into the system to continuously optimize and improve the model; S14. Content recognition: Text: Use the BERT model to parse text content and extract keywords and sentence features; Image: Use the ResNet model to parse image content and extract image features; Audio: Use the WaveNet model to parse audio content and extract audio features; S15. Feature fusion: Fuse the feature vectors of text, image and audio to generate a comprehensive feature vector; S16. Risk prediction: The comprehensive feature vector is input into the deep neural network (DNN) model to perform risk prediction. The prediction results include risk level and risk type. S17. Intelligent decision-making: formulate audit strategies based on risk prediction results and preset audit rules. If the risk level is high, it is recommended to reject; if the risk level is medium, it is recommended to wait for review; S18. Output and feedback: The audit suggestions are output to the system users or managers, who conduct manual review and feed back the review results to the system to optimize the model.
Citation Information
Cited By
Video content risk detection method and device, medium and equipment
CN119274119A
Large model target range construction method and system based on expert model and risk level-to-level management
CN119514505A
Multi-modal content auditing method and system and computer storage medium
CN119577460A
Well lid intelligent operation and maintenance management method and system
CN119648200A
Image-text content auditing method based on multi-modal large model
CN119941157A
