A video processing method and system based on a large language model

Abstract

The application discloses a video processing method and system based on a large language model, and the method comprises the following steps: according to a video processing task instruction input by a user, combining private domain data in a vector database and video processing tool set information to output related prompt words; and then according to the prompt words, generating a video processing task execution plan by using a large language model and executing the video processing task execution plan to obtain a processed video file. The application overcomes the problem that traditional video processing tools are difficult to cooperate and parameter adjustment is time-consuming and laborious, and meanwhile, the utilization rate of private domain information is improved, and a complete closed loop from demand processing, task execution to data feedback optimization is realized.

Description

Technical Field

[0001] The present invention relates to the field of video processing, and in particular to a video processing method and system based on a large language model. Background Art

[0002] Amid the explosive growth in video content creation and distribution, video processing needs are becoming increasingly diverse and complex, encompassing, but not limited to, video transcoding, video enhancement, video restoration, and video super-resolution. Traditional video processing methods face numerous technical bottlenecks: 1. The diverse nature of video processing requirements requires manual understanding and time-consuming adjustments to the various parameters of video processing tools to meet these diverse needs. 2. A wide variety of video processing tools with diverse functions are available to meet these diverse needs. Traditional video processing technologies are highly targeted and excel at addressing single video issues. However, in reality, improving video quality, for example, may involve multiple processing steps, including video enhancement, denoising, and sharpening. Some of these issues may even require conflicting processing principles, necessitating a balanced approach across multiple processing methods to ensure optimal results. However, the lack of intelligent collaboration between individual video processing tools makes it difficult to meet the demands of complex scenarios. 3. Enterprises accumulate large amounts of private data related to video processing (such as historical transcoding cases, software parameter manuals, and industry standard documents). Due to a lack of effective data processing and management methods, it is difficult to quickly extract key information to guide new video processing tasks, resulting in low utilization of private data. 4. Traditional video processing tools lack effective feedback mechanisms, making them unable to optimize based on actual video processing results and user feedback, making them difficult to adapt to technological developments and changing user needs.

[0003] On the other hand, with the advancement of AI technology, large language models (LLMs), as large-scale neural network models, have demonstrated impressive performance in processing human language. However, their application in real-world tasks still faces challenges. Furthermore, while large language models utilize extensive training data, they are also highly dependent on that data. When applied to specialized fields, these models are limited by specialized knowledge and lack in-depth domain knowledge. This results in limited usability in these fields. In order to utilize existing large language models in specialized fields, specialized and complex fine-tuning training is required. In the field of video processing, relevant companies and institutions have accumulated vast amounts of specialized, highly confidential, and private data related to video processing through their production, operations, and research. However, model training presents a challenge: while training data is typically de-identified, the model may retain certain information, which can then be leaked during text generation. Consequently, the application of large language models in specialized fields presents an information security dilemma. Summary of the Invention

[0004] Since the existing methods have the above-mentioned problems, the embodiments of the present invention propose a video processing method and system based on a large language model.

[0005] Specifically, the embodiments of the present invention provide the following technical solutions:

[0006] In a first aspect, an embodiment of the present invention provides a video processing method based on a large language model, comprising:

[0007] Use natural language input for video processing task instructions.

[0008] The task instruction is semantically parsed and vectorized into a query vector.

[0009] A similarity search is performed in a vector database according to the query vector.

[0010] The vector database stores private domain knowledge information data related to the video processing field that has been vectorized in advance, including: video processing operation guide, video processing tool parameter description, video pre-processing tool description, historical test data, industry technical specifications, and video processing experience summary.

[0011] The query vector is semantically associated with the similarity search results, combined with the video processing tool set information, integrated according to the preset prompt word template, and the prompt word context is output.

[0012] The video processing toolset information includes but is not limited to functional attributes, input and output interface specifications, calling functions, calling protocols, and performance indicators of the video processing tools.

[0013] A large language model is used to perform semantic analysis on the context of the prompt word, generate and execute a video processing task execution plan, and output a processed video file.

[0014] On the other hand, an embodiment of the present invention provides a video processing system based on a large language model, including:

[0015] The user interaction module is used to input video processing task instructions using natural language.

[0016] The vectorization module is used to perform semantic analysis on the task instruction and vectorize it into a query vector.

[0017] The data retrieval module is used to perform similarity search in a vector database according to the query vector.

[0018] The vector database stores private domain knowledge information data related to the video processing field after vectorization processing.

[0019] The prompt word generation module is used to semantically associate the query vector with the similarity search results, combine the video processing tool set information, integrate them according to the preset prompt word template, and output the prompt word context.

[0020] The video processing tool management module is used to register and uniformly manage video processing tools, and to define detailed information related to video processing tools.

[0021] The intelligent processing module is used to use a large language model to perform semantic analysis on the context of the prompt word, generate a video processing task execution plan and execute it, and output the processed video file.

[0022] It can be seen from the above technical solution that the embodiments of the present invention have the following beneficial effects: on the one hand, the use of large language model artificial intelligence technology for video processing greatly reduces the time cost of manual parameter adjustment and improves the accuracy of video processing; the technical solution provided by the present invention also realizes the unified management and intelligent scheduling of multiple video processing tools, and automatically selects the most suitable one or more tools and the most suitable parameter configuration for video processing tasks; on the other hand, the vectorization model is used to vectorize the professional knowledge in this field, combined with the vector database, while eliminating the need for professional training of existing large language models, it also avoids the information security problems that are currently prevalent in large models, and improves the utilization of corporate private domain professional data; finally, the video processing results are used to self-learn and optimize the entire video processing system, realizing a complete closed loop of the video processing system from demand processing, task execution to data feedback optimization. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 This is a flowchart of a video processing method based on a large language model provided by one embodiment of the present invention.

[0024] Figure 2 This is a flowchart of a video processing method based on a large language model provided by another embodiment of the present invention.

[0025] Figure 3 This is a structural diagram of a large language model-based video processing system provided by one embodiment of the present invention.

[0026] Figure 4 It is a structural diagram of a large language model-based video processing system provided by another embodiment of the present invention. DETAILED DESCRIPTION

[0027] The following embodiments of the present invention are further described in conjunction with the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solutions of the present invention and are not intended to limit the scope of protection of the present invention.

[0028] Figure 1FIG. 1 is a flow chart showing a video processing method based on a large language model according to an embodiment of the present invention. Figure 1 As shown, the video processing method based on the large speech model provided by one embodiment of the present invention specifically includes the following contents:

[0029] Step 1: Use natural language to input video processing task instructions.

[0030] It should be noted that the embodiments of the present invention do not limit the specific task instructions for video processing. According to actual needs, the video processing task can be any related task related to the field of video processing, including but not limited to: video encoding tasks, video denoising tasks, image enhancement tasks, image super-resolution tasks, and video interpolation tasks.

[0031] For example, a task instruction using natural language is as follows: transcode a video file to reduce its volume by 30% while ensuring that the subjective quality index remains unchanged.

[0032] For example, a task instruction in natural language is as follows: remove mosaic noise from an image so that the VMAF score of the denoised image reaches above 90 points.

[0033] Exemplarily, a task instruction in natural language is used: super-resolution processing is performed on a video file with a resolution of 720P, and a video file with a resolution of 1080P is output.

[0034] An example of a task using natural language instructions is to encode the video into a format suitable for posting on social media platforms and highlight the color expression of the video.

[0035] Step 2: semantically parse the task instruction and vectorize it into a query vector.

[0036] Step 3: perform similarity search in a vector database based on the query vector.

[0037] Among them, the vector database stores private domain knowledge information data related to the video processing field that has been vectorized in advance, including but not limited to video processing operation guides, video processing tool parameter descriptions, video pre-processing tool descriptions, historical test data, industry technical specifications, and video processing experience summaries.

[0038] It should be noted that the embodiment of the present invention does not limit the vectorization algorithm of private domain knowledge information data, and vectorization is performed according to actual needs.

[0039] For example, for data in PDF format, the PDF parsing library can be used to extract text content, and semantic analysis can be performed through a BERT-based pre-trained model to convert operation steps, parameter descriptions, and other content into semantic vectors.

[0040] For example, for data in XLS format, you can use the Pandas library to read the data and convert each row of records into a feature vector through a custom data mapping function.

[0041] Furthermore, in order to improve search efficiency and accuracy, hierarchical storage can be performed according to data type and importance, and a multi-level index can be established.

[0042] Step 4: semantically associate the query vector with the similarity search results, combine them with the video processing tool set information, integrate them according to the preset prompt word template, and output the prompt word context.

[0043] The video processing tool set is composed of various video processing tools and is managed in a unified manner. The video processing tool set information includes but is not limited to: functional attributes, input and output interface specifications, calling functions, calling protocols, and performance indicators of each video processing tool.

[0044] It is understandable that the embodiment of the present invention does not limit which video processing tools are specifically included in the video processing tool set, and corresponding video processing tools can be integrated and registered according to actual needs.

[0045] It should also be noted that the embodiment of the present invention does not limit the specific form of the prompt word template, and it can be set according to actual usage scenarios.

[0046] Exemplarily, the prompt word template may be:

[0047] Video processing requirements input by the user: {video processing task instructions described by the user in natural language}.

[0048] Related data retrieved from the vector database: 1. {search result 1}; 2. {search result 2}; ...; n. {search result n}.

[0049] For example, the video processing requirements input by the user are: {encode the video into a format suitable for publishing on social media platforms and highlight the color expression of the video}.

[0050] Relevant data retrieved from the vector database: 1. {Related knowledge: Social media platforms (such as TikTok and WeChat Video) recommend the MP4 video format, 1080×1920 resolution, and 30fps frame rate; video color expression can be improved by adjusting color correction parameters (such as saturation, contrast, and hue); 2. {Video processing tools: video encoding tools, video color processing tools}.

[0051] Furthermore, in order to more intuitively display the correlation between the retrieval results and the video processing task instructions, the retrieved data may be arranged in descending order of relevance.

[0052] Step 5: Use a large language model to perform semantic analysis on the context of the prompt word, generate and execute a video processing task execution plan, and output the processed video file.

[0053] It should be noted that the embodiments of the present invention are not limited to a specific large language model, and can be any large language model, including but not limited to Transformer, MoE, hybrid architecture, etc. The large language model has the ability to acquire knowledge such as basic concepts and terminology related to the field of video processing.

[0054] Specifically, the large language model receives the integrated context of the prompt word and performs in-depth semantic analysis and logical reasoning based on its powerful natural language understanding and generation capabilities. Based on the video processing task instructions, it selects appropriate tools from the video processing toolset and generates a detailed task execution plan that includes the calling sequence, calling methods, parameter configuration, data flow, and more.

[0055] For example, the task execution plan includes but is not limited to the following:

[0056] Task steps: List the specific steps to complete the video processing requirements.

[0057] Video processing tool usage: Indicate the video tool used in each step and its calling method.

[0058] Parameter settings: parameter settings of the called video tool and related parameter settings of each model.

[0059] Data dependency: Describe the data source and processing method required for each step.

[0060] Expected output: Describes the results obtained after each step is completed.

[0061] The task execution plan may be one or more feasible solutions, listed in descending order of priority.

[0062] Exemplarily, the video processing tool calling method is as follows:

[0063] {Tool 1 name}: {Tool 1 function description}, the calling method is {Tool 1 calling parameters or instructions}; {Tool 2 name}: {Tool 2 function description}, the calling method is {Tool 2 calling parameters or instructions}; ...; {Tool n name}: {Tool n function description}, the calling method is {Tool n calling parameters or instructions}.

[0064] For example, for the task instruction "encode the video into a format suitable for publishing on social media platforms and highlight the color expression of the video", the video processing tools that need to be called and the calling method are: {Video encoding tool}: {supports MP4 format conversion, and can adjust resolution, frame rate, and color correction parameters}, the calling method is {VideoTranscoder, parameters are passed in JSON format}; {Video color processing tool}: {specially used for video color optimization}, the calling method is {VideoProcessor, parameters are passed in JSON format}.

[0065] Furthermore, in order to improve security, the task execution plan may be presented to the user before executing the task execution plan, requesting the user to authorize execution.

[0066] Furthermore, in order to ensure smooth collaboration between video processing tools, the tool call status and data transmission status can be monitored during task execution. If any abnormality occurs, the tool can be reselected or the parameters can be adjusted.

[0067] Further, such as Figure 2 As shown, in another embodiment, after the processed video file is output, a video quality test may be performed on the processed video file to generate a test report, and the test report may be vectorized and then updated into the vector database.

[0068] The test report includes but is not limited to: task instructions, video tool parameter settings (such as resolution, frame rate, bit rate), video quality scores (such as PSNR, VMAF), video processing performance indicators (such as processing time, resource utilization), user evaluation and other information.

[0069] The vectorization of the test report and re-updating the vector database includes:

[0070] The test reports are classified according to the video processing task types and labeled with corresponding task type labels.

[0071] It should be noted that the embodiment of the present invention does not limit the specific classification of the video processing task types, and can be classified according to actual conditions, for example: video encoding tasks, video denoising tasks, image enhancement tasks, image super-resolution tasks, and video interpolation tasks.

[0072] The video quality score is mapped to the correlation between the task execution plan of the current video processing task and the current task category and a correlation label is added.

[0073] The higher the video quality score, the higher the correlation, and the lower the video quality score, the lower the correlation; the video quality score is a comprehensive score of the processed video file and / or a single dimension score.

[0074] The content of the test report and the corresponding label are vectorized and updated into a vector database.

[0075] The vectorized test report content is used as new private domain knowledge information data, and compared and analyzed with historical data in the vector database to improve the video processing task capability, forming a complete closed loop from demand processing, task execution to data feedback optimization.

[0076] For example, if it is found that the current encoding task takes too long while improving color performance, the complexity of the color processing algorithm can be appropriately reduced to shorten the encoding time through an algorithm based on reinforcement learning.

[0077] Figure 3 An architecture diagram of a video processing system based on a large language model according to an embodiment of the present application is shown. As shown in the figure, Figure 3 The video processing system based on a large language model according to an embodiment of the present application specifically includes:

[0078] A user interaction module for inputting video processing task instructions using natural language.

[0079] It should be noted that the present embodiment does not limit the specific task instructions for video processing, and according to actual needs, the video processing task can be any related task in the field of video processing, including but not limited to: video encoding task, video denoising task, image enhancement task, image super-resolution task, video frame interpolation task.

[0080] For example, the task instruction using natural language is: transcoding a video file, reducing the volume by 30% while keeping the subjective quality index unchanged.

[0081] For example, the task instruction using natural language is: remove the mosaic noise in the picture, and make the VMAF score of the denoised picture reach more than 90 points.

[0082] For example, the task instruction using natural language is: super-resolution processing on a video file with a resolution of 720P, outputting a video file with a resolution of 1080P.

[0083] For example, the task instruction using natural language is: encode the video into a format suitable for publishing on social media platforms, and highlight the color performance of the video.

[0084] A vectorization module for semantic analysis of the task instruction and vectorization of the task instruction into a query vector.

[0085] A data retrieval module for similarity search in the vector database according to the query vector.

[0086] Among them, the vector database stores private domain knowledge information data related to the video processing field that has been vectorized in advance, including but not limited to video processing operation guides, video processing tool parameter descriptions, video pre-processing tool descriptions, historical test data, industry technical specifications, and video processing experience summaries.

[0087] It should be noted that the embodiment of the present invention does not limit the vectorization algorithm of private domain knowledge information data, and vectorization is performed according to actual needs.

[0088] For example, for data in PDF format, the PDF parsing library can be used to extract text content, and semantic analysis can be performed through a BERT-based pre-trained model to convert operation steps, parameter descriptions, and other content into semantic vectors.

[0089] For example, for data in XLS format, you can use the Pandas library to read the data and convert each row of records into a feature vector through a custom data mapping function.

[0090] Furthermore, in order to improve search efficiency and accuracy, hierarchical storage can be performed according to data type and importance, and a multi-level index can be established.

[0091] The prompt word generation module is used to semantically associate the query vector with the similarity search results, combine the video processing tool set information, integrate them according to the preset prompt word template, and output the prompt word context.

[0092] It should be noted that the embodiment of the present invention does not limit the specific form of the prompt word template, and it can be set according to actual usage scenarios.

[0093] Exemplarily, the prompt word template may be:

[0094] Video processing requirements input by the user: {video processing task instructions described by the user in natural language}.

[0095] Related data retrieved from the vector database: 1. {search result 1}; 2. {search result 2}; ...; n. {search result n}.

[0096] For example, the video processing requirements input by the user are: {encode the video into a format suitable for publishing on social media platforms and highlight the color expression of the video}.

[0097] Relevant data retrieved from the vector database: 1. {Related knowledge: Social media platforms (such as TikTok and WeChat Video) recommend the MP4 video format, 1080×1920 resolution, and 30fps frame rate; video color expression can be improved by adjusting color correction parameters (such as saturation, contrast, and hue); 2. {Video processing tools: video encoding tools, video color processing tools}.

[0098] Furthermore, in order to more intuitively display the correlation between the retrieval results and the video processing task instructions, the retrieved data may be arranged in descending order of relevance.

[0099] The video processing tool management module is used to register and uniformly manage video processing tools, and to define detailed information related to the video processing tool set.

[0100] The information related to the video processing toolset includes but is not limited to: functional attributes, input and output interface specifications, calling functions, calling protocols and performance indicators of each video processing tool.

[0101] It is understandable that the embodiment of the present invention does not limit which video processing tools are specifically included in the video processing tool set, and corresponding video processing tools can be integrated and registered according to actual needs.

[0102] The intelligent processing module is used to use a large language model to perform semantic analysis on the context of the prompt word, generate a video processing task execution plan and execute it, and output the processed video file.

[0103] It should be noted that the embodiments of the present invention are not limited to a specific large language model, and can be any large language model, including but not limited to Transformer, MoE, hybrid architecture, etc. The large language model has the ability to acquire knowledge such as basic concepts and terminology related to the field of video processing.

[0104] Specifically, the large language model receives the integrated context of the prompt word and performs in-depth semantic analysis and logical reasoning based on its powerful natural language understanding and generation capabilities. Based on the video processing task instructions, it selects appropriate tools from the video processing toolset and generates a detailed task execution plan that includes the calling sequence, calling methods, parameter configuration, data flow, and more.

[0105] For example, the task execution plan includes but is not limited to the following:

[0106] Task steps: List the specific steps to complete the video processing requirements.

[0107] Video processing tool usage: Indicate the video tool used in each step and its calling method.

[0108] Parameter settings: parameter settings of the called video tool and related parameter settings of each model.

[0109] Data dependency: Describe the data source and processing method required for each step.

[0110] Expected output: Describes the results obtained after each step is completed.

[0111] The task execution plan may be one or more feasible solutions, listed in descending order of priority.

[0112] Exemplarily, the video processing tool calling method is as follows:

[0113] {Tool 1 name}: {Tool 1 function description}, the calling method is {Tool 1 calling parameters or instructions}; {Tool 2 name}: {Tool 2 function description}, the calling method is {Tool 2 calling parameters or instructions}; ...; {Tool n name}: {Tool n function description}, the calling method is {Tool n calling parameters or instructions}.

[0114] For example, for the task instruction "encode the video into a format suitable for publishing on social media platforms and highlight the color expression of the video", the video processing tools that need to be called and the calling method are: {Video encoding tool}: {supports MP4 format conversion, and can adjust resolution, frame rate, and color correction parameters}, the calling method is {VideoTranscoder, parameters are passed in JSON format}; {Video color processing tool}: {specially used for video color optimization}, the calling method is {VideoProcessor, parameters are passed in JSON format}.

[0115] Furthermore, in order to improve security, the task execution plan may be presented to the user before executing the task execution plan, requesting the user to authorize execution.

[0116] Furthermore, in order to ensure smooth collaboration between video processing tools, the tool call status and data transmission status can be monitored during task execution. If any abnormality occurs, the tool can be reselected or the parameters can be adjusted.

[0117] Further, such as Figure 4 As shown, in another embodiment, after the intelligent processing module, a data updating module is also included for performing a video quality test on the processed video file, generating a test report, and vectorizing the test report and re-updating it into the vector database.

[0118] The test report includes but is not limited to: task instructions, video tool parameter settings (such as resolution, frame rate, bit rate), video quality scores (such as PSNR, VMAF), video processing performance indicators (such as processing time, resource utilization), user evaluation and other information.

[0119] The vectorization of the test report and re-updating the vector database includes:

[0120] The test reports are classified according to the video processing task types and labeled with corresponding task type labels.

[0121] The video quality score is mapped to the correlation between the task execution plan of the current video processing task and the current task category and a correlation label is added.

[0122] The higher the video quality score, the higher the correlation, and the lower the video quality score, the lower the correlation; the video quality score is a comprehensive score of the processed video file and / or a single dimension score.

[0123] The content of the test report and the corresponding labels are vectorized and updated in the vector database.

[0124] The vectorized test report content is used as new private domain knowledge information data and compared with the historical data in the vector database to improve the ability of video processing tasks and form a complete closed loop from demand processing, task execution to data feedback optimization.

[0125] For example, if it is found that the current encoding task improves color expression while the encoding time is too long, through the reinforcement learning-based algorithm, it can be analyzed that the complexity of the color processing algorithm can be appropriately reduced to shorten the encoding time.

[0126] The term "and / or" in the embodiment of the present invention describes the association relationship of associated objects, indicating that three relationships may exist. For example, A and / or B can represent three situations: A exists alone, A and B exist at the same time, and B exists alone.

[0127] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A video processing method based on a large language model, characterized in that: include: Use natural language input for video processing task instructions; Performing semantic parsing on the task instruction and vectorizing it into a query vector; Performing a similarity search in a vector database based on the query vector; The vector database stores private domain knowledge information data related to the video processing field that has been vectorized in advance, including: video processing operation guides, video processing tool parameter descriptions, video pre-processing tool descriptions, historical test data, industry technical specifications, and video processing experience summaries; Semantically associating the query vector with the similarity search results, combining them with the video processing tool set information, integrating them according to a preset prompt word template, and outputting the prompt word context; The video processing toolset is composed of various video processing tools and is managed in a unified manner, and defines the relevant information of the video processing tools in detail; Using a large language model to perform semantic analysis on the context of the prompt word, generate and execute a video processing task execution plan, and output a processed video file; Perform a video quality test on the processed video file to generate a test report, and then vectorize the test report and re-update it into the vector database.

2. The video processing method based on a large language model according to claim 1, characterized in that The video processing tool set information includes: functional attributes, input and output interface specifications, calling functions, calling protocols and performance indicators of each video processing tool.

3. The video processing method based on a large language model according to claim 1, characterized in that The vectorization of the test report and re-updating the vector database includes: Classify the test reports according to the video processing task type and label them with corresponding task type labels; Map the video quality score to the correlation between the task execution plan of the current video processing task and the current task category and add a correlation label; Wherein, the higher the video quality score, the higher the relevance, and the lower the video quality score, the lower the relevance; the video quality score is a comprehensive score of the processed video file and / or a single dimension score; The content of the test report and the corresponding labels are vectorized and updated in the vector database.

4. A video processing system based on a large language model, characterized in that: include: A user interaction module for inputting video processing task instructions using natural language; A vectorization module, configured to perform semantic parsing on the task instruction and vectorize the instruction into a query vector; A data retrieval module, configured to perform similarity search in a vector database based on the query vector; The vector database stores private domain knowledge information data related to the video processing field that has been vectorized in advance, including: video processing operation guides, video processing tool parameter descriptions, video pre-processing tool descriptions, historical test data, industry technical specifications, and video processing experience summaries; A prompt word generation module is used to semantically associate the query vector with the similarity search results, combine them with the video processing tool set information, integrate them according to the preset prompt word template, and output the prompt word context; The video processing tool management module is used to register and uniformly manage video processing tools, and to define detailed information related to video processing tools; An intelligent processing module, configured to perform semantic analysis on the context of the prompt word using a large language model, generate and execute a video processing task execution plan, and output a processed video file; The data updating module is used to perform a video quality test on the processed video file, generate a test report, and vectorize the test report and re-update it into the vector database.

5. The video processing system based on a large language model according to claim 4, characterized in that The video processing tool set information includes: functional attributes, input and output interface specifications, calling functions, calling protocols and performance indicators of each video processing tool.

6. The video processing system based on a large language model according to claim 4, characterized in that The vectorization of the test report and re-updating the vector database includes: Classify the test reports according to the video processing task type and label them with corresponding task type labels; Map the video quality score to the correlation between the task execution plan of the current video processing task and the current task category and label it with the correlation; Wherein, the higher the video quality score, the higher the relevance, and the lower the video quality score, the lower the relevance; the video quality score is a comprehensive score of the processed video file and / or a single dimension score; The content of the test report and the corresponding labels are vectorized and updated in the vector database.

Images