Media content repairing method and device, equipment, storage medium and program product

Abstract

The invention relates to the technical field of computers, and discloses a media content repairing method and device, equipment, a storage medium and a program product, and the method comprises the steps: obtaining target media content which is not successfully published and a corresponding reason for the unsuccessful publication; according to the first media information and the reason of the target media content, the repaired media content is obtained and displayed, the repaired media content is obtained by performing corresponding repair task processing on the target media content based on the target language model and task prompt information, and the task prompt information corresponds to the repair task. The task prompt information is obtained based on the first media information and the reason. The method and the device can solve the problem that the repairing of the media content which is not published successfully consumes long time.

Description

Technical Field

[0001] The present disclosure relates to the field of computer technology, and particularly to a method, apparatus, device, storage medium, and program product for repairing media content. Background Art

[0002] Currently, a large number of media content materials fail to be published every day. Due to various reasons for the failure to publish and the complex structure of the media content materials, which usually include voiceover scripts, pictures, text captions, etc., after receiving the reasons for the failure to publish, users need to check the content of different repair tasks in the media content materials and perform targeted repairs on the error points, resulting in a long time-consuming for repairing the media content. Summary of the Invention

[0003] In view of this, the present disclosure provides a method, apparatus, device, storage medium, and program product for repairing media content to solve the problem of long time-consuming for repairing media content that fails to be published successfully.

[0004] In a first aspect, the present disclosure provides a method for repairing media content, the method comprising:

[0005] Obtaining a target media content that fails to be published successfully and the corresponding reason for the failure to be published;

[0006] Based on the first media information of the target media content and the reason, obtaining and presenting the repaired media content, where the repaired media content is obtained by processing the target media content with corresponding repair tasks based on a target language model and task prompt information, the task prompt information corresponds to the repair task, and the task prompt information is obtained based on the first media information and the reason.

[0007] In a second aspect, the present disclosure provides a device for repairing media content, the device comprising:

[0008] A data acquisition module for obtaining a target media content that fails to be published successfully and the corresponding reason for the failure to be published;

[0009] A data processing module for obtaining and presenting the repaired media content based on the first media information of the target media content and the reason, where the repaired media content is obtained by processing the target media content with corresponding repair tasks based on a target language model and task prompt information, the task prompt information corresponds to the repair task, and the task prompt information is obtained based on the first media information and the reason.

[0010] In a third aspect, the present disclosure provides an electronic device, including: a memory and a processor, which are communicatively connected to each other. The memory stores computer instructions, and the processor executes the computer instructions to perform the above-mentioned method for repairing media content.

[0011] In a fourth aspect, the present disclosure provides a computer-readable storage medium, on which computer instructions are stored, and the computer instructions are used to cause a computer to perform the above-mentioned method for repairing media content.

[0012] In a fifth aspect, the present disclosure provides a computer program product, including computer instructions, and the computer instructions are used to cause a computer to perform the above-mentioned method for repairing media content.

[0013] The method for repairing media content provided by the embodiments of the present disclosure disassembles the repair task of the target media content into repair tasks that can be executed by the target language model. For each repair task, based on the reason for the failure to publish successfully and the first media information corresponding to the repair task, task prompt information is constructed, so that the target language model can quickly find the repair points in each piece of the first media information under the task prompt information and perform targeted repair on them to obtain the repaired media content. Therefore, it can automatically repair the target media content when it fails to be published successfully, without the need for manual search for repair points and repair of the target media content, thereby effectively improving the repair efficiency of the target media content and reducing the repair time. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] In order to more clearly illustrate the specific embodiments of the present disclosure or the technical solutions in the prior art, the following will briefly introduce the drawings required for use in the description of the specific embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present disclosure. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0015] Figure 1 is a flowchart of a method for repairing media content according to an embodiment of the present disclosure;

[0016] Figure 2 is a flowchart of a method for generating repaired media content according to an embodiment of the present disclosure;

[0017] Figure 3 is a schematic diagram of an information prompt template according to an embodiment of the present disclosure;

[0018] Figure 4 is a flowchart of a process for repairing target media content according to an embodiment of the present disclosure;

[0019] Figure 5It is a schematic flowchart of a method for determining a target language model according to an embodiment of the present disclosure;

[0020] Figure 6 It is a schematic diagram of a training process of a target language model according to an embodiment of the present disclosure;

[0021] Figure 7 It is a schematic diagram of repairing image text according to an embodiment of the present disclosure;

[0022] Figure 8 It is a schematic diagram of repairing missing prompt words according to an embodiment of the present disclosure;

[0023] Figure 9 It is a structural block diagram of a media content repair device according to an embodiment of the present disclosure;

[0024] Figure 10 It is a structural block diagram of an electronic device according to an embodiment of the present disclosure. Detailed implementation manners

[0025] To make the objectives, technical solutions, and advantages of the embodiments of the present disclosure clearer, the technical solutions in the embodiments of the present disclosure will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present disclosure. Obviously, the described embodiments are some, but not all, of the embodiments of the present disclosure. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present disclosure without creative efforts shall fall within the protection scope of the present disclosure.

[0026] It can be understood that before using the technical solutions disclosed in the embodiments of the present disclosure, the user should be informed of the repair tasks, usage scopes, usage scenarios, etc. of the personal information involved in the present disclosure and obtain the user's authorization in an appropriate manner according to relevant laws and regulations.

[0027] For example, when responding to a user's active request, a task prompt message is sent to the user to clearly prompt the user that the operation requested by the user will require obtaining and using the user's personal information. Thus, the user can autonomously choose whether to provide personal information to software or hardware such as an electronic device, application program, server, or storage medium that performs the operations of the technical solutions of the present disclosure according to the task prompt message.

[0028] As an optional but non-limiting implementation manner, the way of sending a task prompt message to the user in response to receiving the user's active request can be, for example, in the form of a pop-up window. The task prompt message can be presented in text in the pop-up window. In addition, the pop-up window can also carry a selection control for the user to choose "agree" or "disagree" to provide personal information to the electronic device.

[0029] It can be understood that the above-mentioned notification and the process of obtaining user authorization are only illustrative and do not limit the implementation manner of the present disclosure. Other manners that comply with relevant laws and regulations can also be applied to the implementation manner of the present disclosure.

[0030] It can be understood that the data involved in the present technical solution (including but not limited to the data itself, the acquisition or use of the data) should comply with the requirements of corresponding laws, regulations and related provisions.

[0031] Currently, there are a large number of media content materials that have not been successfully published every day. Since there are many reasons for the failure to publish successfully and the media content materials have a complex structure, usually including voiceover scripts, pictures, text captions, etc., after receiving the reasons for the failure to publish successfully, users need to check the content of different repair tasks in the media content materials and perform targeted repairs on the error points, resulting in a relatively long repair time for the media content.

[0032] In view of this, according to an embodiment of the present disclosure, an embodiment of a method for repairing media content is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although the logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in a different order than here.

[0033] In this embodiment, a method for repairing media content is provided, which can be used in a media platform for managing media content, for example, the client of the media platform. Figure 1 is a flowchart of a method for repairing media content according to an embodiment of the present disclosure, as Figure 1 shown, the process includes the following steps:

[0034] Step S101, obtain the target media content that has not been successfully published and the corresponding reason for the failure to publish successfully.

[0035] Optionally, an evaluation model can be used to evaluate whether the target media content meets the media publication conditions to obtain the publication result of the target media content and the reason for the failure to publish successfully. Or, manually evaluate whether the target media content meets the media publication conditions to obtain the publication result of the target media content and the reason for the failure to publish successfully. In addition, the reason for the failure to publish successfully can also be preset in advance. The manner of obtaining the publication result and the reason for the failure to publish successfully is not limited herein.

[0036] Step S102: Obtain and display the repaired media content according to the first media information and reasons of the target media content. The repaired media content is obtained by performing corresponding repair tasks on the target media content based on the target language model and task prompt information. The task prompt information corresponds to the repair task and is obtained based on the first media information and the reason for the unsuccessful publication.

[0037] Specifically, the first media information of the target media content includes at least one of a script, image text, prompt, and storyboard. Among them, the image text is the text in the image recognized by the Optical Character Recognition (OCR) technology. In addition, the first media information may also include other repairable information, which is not limited here.

[0038] Specifically, the basic repair operations for the media content can be disassembled into repair tasks that the language model can execute. For example, the first repair task for repairing the script, the second repair task for repairing the image text, the third repair task for repairing the missing prompt, the fourth repair task for repairing the storyboard, etc.

[0039] Different prompt information templates can be configured for different repair tasks. Placeholders for the first media information and reasons are configured in the prompt information template. The first media information and reasons can be filled into the corresponding prompt information template through the placeholders, and then the task prompt information for each repair task is obtained.

[0040] The media content repair method provided in this embodiment disassembles the repair task of the target media content into repair tasks that the target language model can execute. For each repair task, based on the reason for the unsuccessful publication and the first media information corresponding to the repair task, task prompt information is constructed, so that the target language model can quickly find the repair points in each first media information under the task prompt information and perform targeted repair on them to obtain the repaired media content. Therefore, it can automatically repair the target media content when it fails to be published successfully, without the need for manual searching for repair points and repairing the target media content, thereby effectively improving the repair efficiency of the target media content and reducing the repair time.

[0041] In some optional embodiments, as Figure 2 shown, the generation method of the repaired media content includes:

[0042] Step S201: Extract information from the target media content to obtain the first media information corresponding to at least one repair task.

[0043] Specifically, according to the repair objects of the repair tasks, such as scripts, image texts, captions, and storyboards, information extraction is performed on the target media content to obtain the first media information corresponding to each repair task.

[0044] For example, when the target media content is a video, if the repair task is the first repair task for repairing the script, the first media information is the script of the video.

[0045] Step S202, obtain the repair requirements corresponding to the repair task.

[0046] Specifically, corresponding repair requirements can be pre-configured for different repair tasks.

[0047] For example, the repair requirements for the script repair task include: 1. Only perform local modifications on the statements that violate the media release conditions, and keep the statements that do not violate them strictly consistent; 2. There may be typos in the script. It is prohibited to modify the typos in the original script, and the output result should be consistent with the typo statements; 3. Ensure that the length of the script before and after modification is the same, avoid deleting or adding a large number of statements, and ensure that the modified statements are basically the same in length before and after modification; 4. Only output the modified script, and it is prohibited to output explanations or other content. Regarding the second point, when using a language model to repair the script, it is very likely to modify the correct words into typos, and the probability of typos in the script is relatively low. Therefore, in order to avoid modifying the script to the wrong words, it is necessary to prohibit modifying the typos in the original script.

[0048] For example, the repair requirements for the image text repair task include: 1. Only select the lemmas that belong to the reasons for unsuccessful publication from the image text list of the target media content. If it is determined that there are no lemmas that belong to the reasons for unsuccessful publication, return an empty list; 2. If there are lemmas that belong to the reasons for unsuccessful publication in the image text list, output the lemmas that belong to the reasons for unsuccessful publication in exactly the same content as the original, and it is prohibited to only output the lemmas that belong to the reasons for unsuccessful publication, nor to change the original content; 3. The output result must be output in the given data format, and it is prohibited to output any explanations or other content other than the given data format.

[0049] Step S203, based on the reason, the first media information, and the repair requirements, obtain the task prompt information corresponding to the repair task.

[0050] Specifically, fuse the corresponding reason, the first media information, and the repair requirements to obtain the task prompt information corresponding to the repair task.

[0051] Step S204, based on the task prompt information and the target language model, process the target media content for the corresponding repair task to obtain the second media information of at least one repair task, so as to obtain the repaired media content.

[0052] Specifically, if the task prompt information is related to the script, the task prompt information is input into the target language model to repair the script of the target media content, and the repaired script is obtained.

[0053] If the task prompt information is related to the image text, the task prompt information is input into the target language model to repair the text in the image of the target media content, and the repaired text is obtained. It should be noted that the repaired text is the text that does not meet the image text publication conditions. For example, license plate information, etc.

[0054] If the task prompt information is related to the prompt, the task prompt information is input into the target language model to repair the missing prompt in the target media content, and the added prompt is obtained. It should be noted that the first media information corresponding to the prompt can be the same as the first media information of the image text, and the missing prompt in the image of the target media content is repaired.

[0055] If the task prompt information is related to the storyboard, the task prompt information is input into the target language model to repair the storyboard in the target media content, and the storyboard that does not meet the storyboard publication conditions is deleted.

[0056] The media content repair method provided in this embodiment disassembles the target media content to obtain the first media information corresponding to at least one repair task. Therefore, the repair task of the target media content can be disassembled into repair tasks that the target language model can execute. Then, for each repair task, based on the reason for the failure to publish successfully, the first media information, and the repair requirements, task prompt information is constructed so that the target language model can quickly find the repair points in each first media information under the task prompt information and perform targeted repairs. Furthermore, the repaired second media information is aggregated to obtain the repaired media content. Therefore, the target media content that fails to be published successfully can be automatically repaired without manually searching for repair points to repair the target media content, effectively improving the repair efficiency of the target media content and reducing the repair time.

[0057] In some alternative embodiments, obtaining the task prompt information corresponding to the repair task based on the reason, the first media information, and the repair requirements in step S203 includes:

[0058] Step a1, obtaining a prompt information template corresponding to the repair task.

[0059] Specifically, placeholders for the first media information, the reason, and the repair requirements are provided in the prompt information template.

[0060] Step a2: Based on the prompt information template, fuse the first media information, the reason, and the repair requirements to obtain task prompt information corresponding to the repair task.

[0061] Specifically, write the reason, the first media information corresponding to the repair task, and the repair requirements into the corresponding placeholders in the prompt information template of the repair task to obtain the task prompt information for each repair task.

[0062] The media content repair method provided in this embodiment provides a corresponding prompt information template for different repair tasks to fuse the first media information, the reason, and the repair requirements corresponding to the repair task. Therefore, it can provide corresponding task prompt information for the target language model for different repair tasks to improve the accuracy of repair task processing.

[0063] In some optional implementation manners, the step of fusing the first media information, the reason, and the repair requirements based on the prompt information template in step a2 to obtain task prompt information corresponding to the repair task includes:

[0064] Step a21: Extract the attribute information of the target media content.

[0065] Specifically, the attribute information includes at least one of the industry and category to which the target media content belongs. Or other attribute information related to the repair of the target media content.

[0066] Exemplarily, when the target media content covers the target product, the attribute information includes the industry and category to which the target product belongs.

[0067] Step a22: Based on the prompt information template, fuse the attribute information, the reason, the first media information, and the repair requirements to obtain task prompt information corresponding to the repair task.

[0068] Furthermore, the prompt information template also includes placeholders for the attribute information. Fill the attribute information, the reason, the first media information corresponding to the repair task, and the repair requirements into the corresponding placeholders in the prompt information template of the repair task to obtain the task prompt information for each repair task.

[0069] The media content repair method provided in this embodiment adds the attribute information of the target media content to the task prompt information. Therefore, it can make the output result of the target language model correspond to the business scenario of the target media content by using the attribute information to improve the repair effect.

[0070] Further, the prompt message template further includes a guiding statement template corresponding to the repair task, and placeholders for attribute information are provided in the guiding statement template. The attribute information is written into the corresponding placeholder of the guiding statement template to obtain a repair guiding statement. Together with the first media information, reason, and repair requirements written in the prompt message template, it constitutes the task prompt information.

[0071] Exemplarily, as Figure 3 shown in the information prompt template, assuming that the target media content is a video and the repair task is used to repair the script, the repair object of the repair task is the script. Taking the guiding statement template of the script as "Please perform partial repair on the script of the target media content according to the reason for unsuccessful release and the repair requirements, so that it can be successfully released. The script with <attribute information> violation of the target media content is:", for example, the extracted attribute information can be written into the guiding statement template. For example, if the attribute information is "The industry is beauty - makeup - facial makeup, and the product name is liquid foundation", then the repair guiding statement can be obtained as "Please perform partial repair on the script of the target media content according to the reason for unsuccessful release and the repair requirements, so that it can be successfully released. The industry of the target media content is beauty - makeup - facial makeup, and the product name is liquid foundation. The script with violation is:", and then combined with the script (i.e., the first media information), reason, and repair requirements of the target media content written in the prompt message template to obtain the task prompt information of the script.

[0072] Exemplarily, as Figure 4 shown, taking the target media content as a video as an example, the overall release process of the target media content includes: when obtaining the target media content that has not been successfully released and the corresponding reason, constructing the task prompt information for each repair task, for example, the task prompt information of the script, the task prompt information of the image text, the task prompt information of the prompt language, and the task prompt information of the storyboard. Then, input the task prompt information of each repair task into the target language model to process the corresponding repair task for the target media content to obtain the repaired media content.

[0073] In some alternative embodiments, as Figure 5 shown, the determination method of the target language model includes:

[0074] Step S301, obtain the media release conditions and the media content samples that have not been successfully released, and the media release conditions include the release conditions corresponding to the repair tasks.

[0075] Specifically, for different repair tasks, the media release conditions can be further divided into release conditions corresponding to each repair task, such as script release conditions, image text release conditions, prompt language release conditions, and storyboard release conditions.

[0076] It should be noted that the media content sample and the target media content belong to the same data type. For example, if the target media content is a video, the media content sample for training is also a video.

[0077] Specifically, an evaluation model can be used to evaluate whether the media content sample meets the release conditions to obtain the release result of the media content sample. Alternatively, manual evaluation can be used to determine whether the media content sample meets the media release conditions to obtain the release result of the media content sample. The method for obtaining the release result of the media content sample is not limited herein.

[0078] Step S302: Adjust the parameters of the initial language model based on the media release conditions to obtain the first language model.

[0079] Specifically, the initial language model is a large language model (LLM).

[0080] Furthermore, the initial language model is the language model of the business scenario to which the target media content belongs. For example, if the target media content is an advertising video, the initial language model is the language model in the advertising business scenario.

[0081] Specifically, add the media release conditions to the ContinuePretrain data of the initial language model and adjust the parameters of the initial language model to obtain the first language model. Therefore, during the pre-training stage (i.e., the stage of adjusting the parameters of the initial language model), the first language model can understand the media release conditions and repair suggestions (such as the reasons for unsuccessful release), making it easier to find content that does not meet the media release conditions from the target media content.

[0082] Step S303: Obtain the successfully released repair samples corresponding to the media content sample, and generate the first sample pair based on the media content sample and the successfully released repair samples.

[0083] Specifically, perform one or more repairs on the media content sample to obtain multiple repair samples, and then select the successfully released repair samples from them. Alternatively, perform one repair on the media content to obtain a repair sample. If the repair sample is not successfully released, continue to repair based on the media content sample or the unsuccessfully released repair sample to obtain more versions of repair samples, and then select the successfully released repair samples from them.

[0084] Furthermore, combine the media content sample and the successfully released repair samples to obtain the first sample pair.

[0085] Step S304: Adjust the parameters of the first language model based on the first sample pair to obtain the second language model.

[0086] Specifically, based on the first sample pair, the first language model is trained by supervised fine-tuning (SFT) to obtain a second language model.

[0087] It should be noted that the purpose of the supervised fine-tuning stage is to construct high-quality repair data; among them, the data access condition in the supervised fine-tuning stage is that the release result obtained by the evaluation model is a successful release and the release result obtained by manual evaluation is a successful release, and it is not the highest-priority unreleased successful repair sample (also known as non-P0 Badcase). Thus, the first language model can learn such successfully released repair samples to improve the online repair ability.

[0088] It should be noted that the reason why the data access condition in the supervised fine-tuning stage is that the release result obtained by the evaluation model is a successful release and the release result obtained by manual evaluation is a successful release is to reduce the misjudgment probability of successfully released repair samples and avoid misjudging repair samples that do not meet the media release conditions as successfully released repair samples, thus affecting the training effect of the language model.

[0089] Of course, when the accuracy of the evaluation result of the evaluation model is relatively high, it is also possible to determine the successfully released repair samples only by the evaluation method of the evaluation model, which is not restricted here.

[0090] Step S305: Obtain the unreleased successful repair samples corresponding to the media content samples, and generate a second sample pair based on the media content samples, the successfully released repair samples, and the unreleased successful repair samples.

[0091] Specifically, perform one or more repairs on the media content samples, then for multiple repair samples, obtain at least one unreleased successful repair sample. Or, perform one repair on the media content to obtain a repair sample. If the repair sample is not successfully released, then use this repair sample as the unreleased successful repair sample. In addition, it is also possible to further repair the repair sample or the media content sample to obtain more versions of repair samples, and obtain other unreleased successful repair samples from them.

[0092] Furthermore, combine the corresponding media content samples, the successfully released repair samples, and the unreleased successful repair samples to obtain a second sample pair.

[0093] Step S306: Adjust the parameters of the second language model based on the second sample pair to obtain a target language model.

[0094] Specifically, based on the second sample pair, direct preference optimization (DPO) is performed on the second language model to obtain the target language model.

[0095] Specifically, a release result obtained by evaluating whether the repaired sample meets the media release conditions by an evaluation model and / or manually can be obtained, and this release result is used as a reward to align the second language model with the reward to obtain the target language model.

[0096] For the media content repair method provided in this embodiment, the parameters of the initial language model are adjusted based on the media release conditions to obtain the first language model. Therefore, the first language model can understand the media release conditions and the reasons for the unsuccessful release, making it easier to find the content that does not meet the media release conditions from the target media content. Then, based on the media content sample and the successfully released repaired sample, a first sample pair is generated to adjust the parameters of the first language model to obtain the second language model. Therefore, the second language model can learn the information of the successfully released repaired sample and improve the media content repair ability. Furthermore, based on the media content sample, the successfully released repaired sample, and the unsuccessfully released repaired sample, a second sample pair is generated to adjust the parameters of the second language model to obtain the target language model. Therefore, the output result of the target language model can be continuously approximated to the successfully released media content, and the release success probability of the media content repaired by the target language model can be improved.

[0097] In some optional implementation manners, obtaining the successfully released repaired sample corresponding to the media content sample in step S303 includes:

[0098] Step b1, repairing the media content sample to obtain a first repaired sample.

[0099] Specifically, at the initial stage of training the language model, other trained language models can be used to repair the media content sample to obtain the first repaired sample. The currently trained language model (such as the first language model) can also be used to repair the media content sample to obtain the first repaired sample.

[0100] Step b2, obtaining the release result of the first repaired sample.

[0101] Specifically, an evaluation model is used to evaluate whether the first repaired sample meets the media release conditions to obtain the release result of the first repaired sample. Alternatively, manual evaluation is used to evaluate whether the first repaired sample meets the media release conditions to obtain the release result of the first repaired sample. Alternatively, an evaluation model and manual evaluation are used together to evaluate whether the first repaired sample meets the media release conditions to obtain the release result of the first repaired sample.

[0102] For example, an evaluation model is used to evaluate whether the first repaired sample meets the media release conditions once to obtain a first release result. Then, when the first release result indicates a successful release, manual evaluation is used to evaluate whether the first repaired sample meets the media release conditions a second time to obtain the release result of the first repaired sample.

[0103] Step b3, if the first repaired sample is not successfully released, the media content sample and / or the first repaired sample are repaired multiple times to obtain multiple second repaired samples.

[0104] Specifically, repeated sampling is performed on the media content sample and / or the first repaired sample, and the repeatedly sampled media content sample and / or the first repaired sample are repaired to obtain different versions of the second repaired samples.

[0105] Specifically, during the process of each sampling and repair, the repair parameters of the language model for repairing the media content sample and the first repaired sample can be fine-tuned to obtain multiple second repaired samples.

[0106] Specifically, in the initial stage of training the language model, other trained language models can be used to perform multiple samplings and repairs on the media content sample and the first repaired sample to obtain different versions of the second repaired samples. It is also possible to use the currently trained language model (such as the first language model) to perform multiple samplings and repairs on the media content sample to obtain different versions of the second repaired samples.

[0107] Step b4, obtain the release results of each second repaired sample.

[0108] Specifically, the method for obtaining the release result of the second repaired sample can refer to the method for obtaining the release result of the first repaired sample, and will not be elaborated here.

[0109] Step b5, if any second repaired sample is successfully released, the second repaired sample is determined as the repaired sample that is successfully released.

[0110] It should be noted that after demonstration, when using a language model to repair media content, during the process of repeated sampling and repair of the media content, among the multiple versions of repaired samples generated, there will basically be a repaired sample that is successfully released. If none of the second repaired samples generated this time are successfully released, the media content sample and / or the first repaired sample can be continuously repaired until a second repaired sample that is successfully released is obtained.

[0111] For the media content repair method provided in this embodiment, if the first repaired sample obtained by repairing the media content sample fails to be successfully published, the media content sample and / or the first repaired sample are repaired multiple times to obtain multiple second repaired samples. Therefore, by repeating the sampling and repair method, a successfully published repaired sample can be automatically obtained, reducing manual intervention.

[0112] In some alternative embodiments, obtaining the successfully published repaired sample corresponding to the media content sample in step S303 further includes:

[0113] Step b6, if the first repaired sample is successfully published, determine the first repaired sample as the successfully published repaired sample.

[0114] For the media content repair method provided in this embodiment, in the case where the first repaired sample is successfully published, directly determine the first repaired sample as the successfully published repaired sample. Therefore, the model training efficiency can be improved.

[0115] In some alternative embodiments, obtaining the unreleased repaired sample corresponding to the media content sample in step S305 includes: obtaining the unreleased repaired sample from the unreleased first repaired sample and / or the unreleased second repaired sample.

[0116] Specifically, the unreleased first repaired sample can be selected as the unreleased repaired sample, and one of the unreleased second repaired samples can be selected as the unreleased repaired sample. Alternatively, one or more of the unreleased first repaired sample and / or the unreleased second repaired sample can be selected as the unreleased repaired sample.

[0117] For the media content repair method provided in this embodiment, the unreleased repaired sample is selected from the unreleased first repaired sample and / or the unreleased second repaired sample as the negative sample for model training. Therefore, the model training data can be enriched.

[0118] Exemplarily, assume that the media content sample is x, the repaired samples obtained by multiple sampling repairs of x form a set of repaired samples {y}, the successfully published repaired sample is y_pass, and the unreleased repaired sample is y_reject. Then, the second sample pair <x, y_pass, y_reject> can be obtained to train the second language model, so that the finally obtained target language model continuously approaches the successfully published media sample, improving the online publication success probability.

[0119] It should be noted that after determining the target language model, whenever a new media content sample is obtained, the target language model is used to repair the media content sample to obtain a third repaired sample. If the third repaired sample is successfully published, the third repaired sample is determined as the repaired sample that has been successfully published. If the third repaired sample is not successfully published, the third repaired sample and / or the new media content sample are repaired multiple times to obtain multiple fourth repaired samples. If any one of the fourth repaired samples is successfully published, the fourth repaired sample is determined as the repaired sample that has been successfully published. And the unreleased repaired samples are determined from the unreleased third repaired samples and the unreleased fourth repaired samples. For example, an unreleased third repaired sample and an unreleased fourth repaired sample are selected as the unreleased repaired samples. Then, third sample pairs are generated from the new media content samples, the corresponding successfully published repaired samples, and the unreleased repaired samples. Based on the third sample pairs, the target language model is optimized.

[0120] Understandably, after determining the target language model, it is not necessary to perform supervised fine-tuning on the target language model, and only direct preference optimization of the target language model is required.

[0121] As an example, such as Figure 6As shown below, the overall training process of the target language model is as follows: The initial language model is trained using media release conditions to obtain the first language model. Then, media content samples that have not been successfully released are obtained, and the media content samples are repaired to obtain the first repaired sample. An evaluation model is used to conduct a first evaluation on whether the first repaired sample meets the media release conditions to obtain the release result of the first evaluation. If the release result of the first evaluation indicates unsuccessful release, it is determined that the first repaired sample has not been successfully released. If the release result of the first evaluation indicates successful release, a second evaluation is manually conducted on whether the first repaired sample meets the media release conditions to obtain the release result of the second evaluation. If the release result of the second evaluation indicates successful release, the first repaired sample is used as the repaired sample that has been successfully released. If the release result of the second evaluation indicates unsuccessful release, it is determined that the first repaired sample has not been successfully released. The media content samples or the first repaired samples that have not been successfully released are repeatedly sampled and repaired using other language models or the language model obtained from the current training to obtain multiple second repaired samples. If any of the second repaired samples is successfully released, the second repaired sample is used as the repaired sample that has been successfully released. The first sample pair is constructed using the media content samples and the repaired samples that have been successfully released, and the first language model is supervised and fine-tuned using the first sample pair to obtain the second language model. And the first repaired samples that have not been successfully released and an unsuccessful second repaired sample are used as the repaired samples that have not been successfully released. The second sample pair is constructed using the media content samples, the repaired samples that have been successfully released, and the repaired samples that have not been successfully released, and the second language model is directly preference optimized using the second sample pair to obtain the target language model. And in the subsequent process, sample pairs for direct preference optimization are continuously constructed (refer to the construction method of the second sample pair) to optimize the target language model.

[0122] In some alternative embodiments, if the target media content includes a script and at least one repair task includes a first repair task for repairing the script, then the processing of the target media content for the corresponding repair task based on the task prompt information and the target language model in step S204 above to obtain the second media information of at least one repair task includes: Processing the script of the target media content for the first repair task based on the target language model and the task prompt information of the first repair task to obtain the second media information of the first repair task.

[0123] Specifically, if the target media content is audio-visual, the target media content includes a script. Therefore, a first repair task for repairing the script can be configured to repair the script in the target media content that does not meet the script release conditions to obtain the repaired script, that is, the second media information of the first repair task.

[0124] For the media content repair method provided in this embodiment, if the target media content contains a script, the target media content is repaired for the script. Therefore, the script in the target media content that does not meet the media release conditions can be repaired to improve the repair accuracy.

[0125] In some alternative embodiments, if the target media content contains an image and at least one repair task includes a second repair task for repairing image text, then in step S204 above, based on the task prompt information and the target language model, processing the target media content for the corresponding repair task to obtain the second media information of at least one repair task includes: based on the target language model and the task prompt information of the second repair task, processing the image text of the target media content for the second repair task to obtain the second media information of the second repair task.

[0126] Specifically, if the target media content is an image or a video composed of multiple frames of images, then the target media content contains an image. Therefore, a second repair task for repairing image text can be configured to repair the image text in the target media content that does not meet the image text release conditions to obtain the repaired image text, that is, the second media information of the second repair task. It should be noted that the repaired image text is the image text that does not meet the media release conditions, and when presenting the repaired media content, the image text that does not meet the media release conditions needs to be masked.

[0127] For the media content repair method provided in this embodiment, if the target media content contains an image, then the target media content is repaired for image text. Therefore, the image text in the target media content that does not meet the media release conditions can be repaired to improve the repair accuracy.

[0128] In some alternative embodiments, at least one repair task includes a third repair task for repairing missing captions. Then, in step S204 above, based on the task prompt information and the target language model, processing the target media content for the corresponding repair task to obtain the second media information of at least one repair task includes: based on the target language model and the task prompt information of the third repair task, processing the target media content for the third repair task to obtain the second media information of the third repair task, and the second media information includes the missing captions.

[0129] Specifically, if the target media content is an image, a video composed of multiple frames of images, or other media content that needs to display captions, then a third repair task for repairing missing captions can be configured to add the missing captions in the target media content to obtain the second media information of the third repair task, so that the repaired media content meets the caption release conditions.

[0130] Exemplarily, taking the target media content as a recommended video for a certain brand as an example, assume that this type of video needs to add prompt 1 and prompt 2. However, only prompt 1 is provided in the target media content, then prompt 2 can be added to the repaired video.

[0131] For the media content repair method provided in this embodiment, if prompt needs to be displayed in the target media content, the missing prompt is added to the target media content to make the repaired media content compliant, so as to improve the repair accuracy.

[0132] In some optional implementation manners, if the target media content includes storyboards and at least one repair task includes a fourth repair task for repairing storyboards, then the processing of the corresponding repair task for the target media content based on the task prompt information and the target language model in step S204 to obtain the second media information of at least one repair task includes: processing the storyboards of the target media content for the fourth repair task based on the target language model and the task prompt information of the fourth repair task to obtain the second media information of the fourth repair task.

[0133] Specifically, if the target media content is a media content with storyboards such as a video, a fourth repair task for repairing storyboards can be configured to delete the storyboards in the target media content that do not meet the storyboard release conditions to obtain the second media information of the fourth repair task.

[0134] For the media content repair method provided in this embodiment, if the target media content includes storyboards, the storyboards of the target media content are repaired. Therefore, the storyboards in the target media content that do not meet the media release conditions can be deleted to improve the repair accuracy.

[0135] In some optional implementation manners, the repaired media content includes a repaired script, and the audio played when the repaired media content is displayed corresponds to the repaired script; and / or, the repaired media content includes target image text that does not meet the media release conditions, and the target image text is covered when the repaired media content is displayed; and / or, the repaired media content newly adds the prompts missing in the target media content; and / or, the repaired media content removes the storyboards that do not meet the media release conditions.

[0136] Specifically, if the target media content includes a script, the audio corresponding to the repaired script is obtained, and the audio of the target media content is replaced based on this audio to obtain the repaired media content. The audio corresponding to the repaired script is played when the repaired media content is displayed.

[0137] If the target media content contains images, optical character recognition technology can be used to recognize the image text to obtain a list of image texts of the target media content. Identify the image texts in the list of recognized image texts that do not meet the media publication conditions as the image texts to be repaired. When displaying the repaired media content, mask the repaired image texts. For example Figure 7 As shown, assume that the image texts of the target media content include the first text and the second text. Assume that it is recognized that the second text does not meet the media publication conditions. Then, when displaying, mask the second text. For example, use mosaic masking, etc. It should be noted that the element masking the second text needs to be close to the image background or does not affect the display of other texts.

[0138] If a prompt needs to be displayed in the target media content, the list of image texts can be recognized to determine the missing prompt, and the missing prompt is added to the target media content so that the missing prompt is displayed when the repaired media content is displayed. For example Figure 8 As shown, assume that the target prompt needs to be displayed synchronously when the target media content is displayed. However, the target prompt is missing in the target media content. Then, the target prompt can be added to the target media content and displayed.

[0139] If the target media content contains storyboards, the storyboard information can be recognized to determine the storyboards that do not meet the media publication conditions, delete the storyboards that do not meet the media publication conditions in the target media content, and do not display the storyboards that do not meet the media publication conditions when displaying the repaired media content.

[0140] For the media content repair method provided in this embodiment, the audio played when displaying the repaired media content corresponds to the repaired script. When displaying the repaired media content, the target image text is masked. The repaired media content has a target prompt missing in the target media content added, and the repaired media content has the storyboards that do not meet the media publication conditions removed. Therefore, it can ensure that the displayed repaired media content meets the media publication conditions and improve the repair effect of the media content.

[0141] As a specific application example, a client of a media platform is installed on a mobile phone. The media platform is used to manage media content, such as videos. A target media video can be created in the client, and the server of the media platform evaluates whether the target media video meets the media publishing conditions to obtain the reason for the unsuccessful publication. When the client obtains the target media content that has not been successfully published and the corresponding reason for the unsuccessful publication, it can use the media content repair method of the present disclosure to obtain the repaired media content and display it locally. At the same time, when other clients obtain the target media content that has not been successfully published and the corresponding reason for the unsuccessful publication, they can also use the media content repair method of the present disclosure to obtain the repaired media content and display it locally. Among them, the repaired media content can be obtained by the server using the generation method of the repaired media content provided by the present disclosure to repair the target media video. The repaired media content can also be obtained by the client using the generation method of the repaired media content provided by the present disclosure to repair the target media video.

[0142] It should be noted that the overall concept of the media content repair method of the present disclosure is as follows: decompose the basic repair operations of the media content into repair tasks that the target language model can execute, and aggregate the repaired media content generated after the target language model outputs the rendering protocol. For example, if the media content is a video, the repair tasks include script repair tasks, image text repair, caption repair, storyboard repair, etc. Therefore, the target language model can be used to accurately find the content that has not been successfully published in the media content, and then perform targeted repair on this content, so as to realize the automatic repair of the media content that has not been successfully published, reduce the repair time, and improve the repair efficiency.

[0143] It should be noted that when training the target language model for the media content repair method of the present disclosure, media content that has been successfully published and adopted by users can be collected online, regarded as external reward information, and these reward information can be used as input to continuously align the target language model with the media content adopted by users. In addition, the target language model itself is used to repeatedly sample the media content that has not been successfully published or the repair samples, and multiple versions of repair samples are generated offline. Then, an evaluation model or manual evaluation is used to evaluate whether the generated repair samples meet the media publishing conditions to obtain the publishing results of the repair samples. Then, the successfully published repair samples are used as new training data to train the target language model, and the output distribution of the target language model is continuously adjusted in the direction of a higher publishing success probability to improve the accuracy of the target language model in repairing media content.

[0144] In this embodiment, a media content repair device is further provided. This device is used to implement the above-mentioned embodiments and preferred implementation manners, and those that have been described will not be repeated. As used hereinafter, the term "module" can be a combination of software and / or hardware that can achieve a predetermined function. Although the devices described in the following embodiments are preferably implemented in software, implementation in hardware, or a combination of software and hardware is also possible and contemplated.

[0145] This embodiment provides a media content repair device, as Figure 9 shown, including:

[0146] A data acquisition module 401, configured to acquire target media content that has not been successfully published and the corresponding reason for the unsuccessful publication;

[0147] A data processing module 402, configured to obtain and display the repaired media content according to the first media information and the reason of the target media content. The repaired media content is obtained by processing the target media content for corresponding repair tasks based on the target language model and the task prompt information. The task prompt information corresponds to the repair task and is obtained based on the first media information and the reason.

[0148] In some optional implementation manners, the media content repair device of the present disclosure further includes a data repair module, and the data repair module is configured to generate the repaired media content. Among them, the data repair module includes:

[0149] An information extraction unit, configured to extract information from the target media content to obtain the first media information corresponding to at least one repair task;

[0150] A requirement acquisition unit, configured to acquire the repair requirements corresponding to the repair task;

[0151] An information generation unit, configured to obtain the task prompt information corresponding to the repair task based on the reason, the first media information, and the repair requirements;

[0152] A data repair unit, configured to process the target media content for corresponding repair tasks based on the task prompt information and the target language model to obtain the second media information of at least one repair task, so as to obtain the repaired media content.

[0153] In some optional implementation manners, the information generation unit includes:

[0154] A template acquisition subunit, configured to acquire a prompt information template corresponding to the repair task;

[0155] A data fusion subunit, configured to fuse the first media information, the reason, and the repair requirements based on the prompt information template to obtain the task prompt information corresponding to the repair task.

[0156] In some alternative embodiments, the data fusion subunit is specifically configured to: extract the attribute information of the target media content; fuse the attribute information, reasons, first media information, and repair requirements based on the prompt information template to obtain task prompt information corresponding to the repair task.

[0157] In some alternative embodiments, the media content repair device of the present disclosure further includes a model training module, and the model training module is used to determine a target language model. Among them, the model training module includes:

[0158] A training data acquisition unit, configured to acquire media release conditions and media content samples that have not been successfully released, and the media release conditions include release conditions corresponding to the repair task;

[0159] A first model training unit, configured to adjust the parameters of the initial language model based on the media release conditions to obtain a first language model;

[0160] A first sample generation unit, configured to acquire successfully released repair samples corresponding to the media content samples, and generate a first sample pair based on the media content samples and the successfully released repair samples;

[0161] A second model training unit, configured to adjust the parameters of the first language model based on the first sample pair to obtain a second language model;

[0162] A second sample generation unit, configured to acquire unsuccessfully released repair samples corresponding to the media content samples, and generate a second sample pair based on the media content samples, the successfully released repair samples, and the unsuccessfully released repair samples;

[0163] A third model training unit, configured to adjust the parameters of the second language model based on the second sample pair to obtain the target language model.

[0164] In some alternative embodiments, the first sample generation unit includes:

[0165] A first repair subunit, configured to repair the media content samples to obtain first repair samples;

[0166] A first result acquisition subunit, configured to acquire the release results of the first repair samples;

[0167] A second repair subunit, configured to, if the first repair samples are not successfully released, repair the media content samples and / or the first repair samples multiple times to obtain multiple second repair samples;

[0168] A second result acquisition subunit, configured to acquire the release results of each of the second repair samples;

[0169] The first sample determination subunit is configured to determine the second repair sample as the repaired sample with successful release if any second repair sample is successfully released.

[0170] In some alternative embodiments, the first sample generation unit further includes:

[0171] The second sample determination subunit is configured to determine the first repair sample as the repaired sample with successful release if the first repair sample is successfully released.

[0172] In some alternative embodiments, the second sample generation unit includes:

[0173] The third sample determination subunit is configured to obtain the unreleased repair samples from the unreleased first repair samples and / or the unreleased second repair samples.

[0174] In some alternative embodiments, if the target media content includes a script and at least one repair task includes a first repair task for repairing the script, the data repair unit includes:

[0175] The script repair subunit is configured to process the script of the target media content for the first repair task based on the target language model and the task prompt information of the first repair task, to obtain the second media information of the first repair task.

[0176] In some alternative embodiments, if the target media content includes an image and at least one repair task includes a second repair task for repairing the image text, the data repair unit further includes:

[0177] The text repair subunit is configured to process the image text of the target media content for the second repair task based on the target language model and the task prompt information of the second repair task, to obtain the second media information of the second repair task.

[0178] In some alternative embodiments, if at least one repair task includes a third repair task for repairing the missing prompt, the data repair unit further includes:

[0179] The prompt repair subunit is configured to process the target media content for the third repair task based on the target language model and the task prompt information of the third repair task, to obtain the second media information of the third repair task, where the second media information includes the missing prompt.

[0180] In some alternative embodiments, if the target media content includes storyboards and at least one repair task includes a fourth repair task for repairing the storyboards, the data repair unit further includes:

[0181] The storyboard repair subunit is configured to perform the fourth repair task on the storyboard of the target media content based on the target language model and the task prompt information of the fourth repair task, so as to obtain the second media information of the fourth repair task.

[0182] In some alternative embodiments, the repaired media content includes a repaired script, and the audio played when the repaired media content is displayed corresponds to the repaired script; and / or, the repaired media content includes target image text that does not meet the media release conditions, and the target image text is obscured when the repaired media content is displayed; and / or, the repaired media content newly adds prompt words missing from the target media content; and / or, the repaired media content removes storyboards that do not meet the media release conditions.

[0183] The further functional descriptions of the above-mentioned various modules and units are the same as those in the corresponding embodiments above, and will not be elaborated here.

[0184] The media content repair device in this embodiment is presented in the form of functional units. Here, the unit refers to an ASIC (Application Specific Integrated Circuit) circuit, a processor and a memory that execute one or more software or fixed programs, and / or other devices that can provide the above functions.

[0185] The embodiments of the present disclosure also provide an electronic device having the above-mentioned Figure 9 shown media content repair device.

[0186] Please refer to Figure 10 , Figure 10 which is a structural block diagram of an electronic device provided by an alternative embodiment of the present disclosure. As shown in Figure 10 , the electronic device includes: one or more processors 501, a memory 502, and interfaces for connecting various components, including a high-speed interface and a low-speed interface. Each component communicates with each other using different buses and can be installed on a common motherboard or installed in other ways as needed. The processor can process instructions executed within the electronic device, including instructions stored in the memory or on the memory to display graphical information of the GUI on an external input / output device (such as a display device coupled to the interface). In some alternative embodiments, if necessary, multiple processors and / or multiple buses can be used together with multiple memories and multiple memories. Similarly, multiple electronic devices can be connected, and each device provides some necessary operations (such as a server array, a set of blade servers, or a multi-processor system). Figure 10 In

[0187] The processor 501 can be a central processing unit, a network processor, or a combination thereof. Among them, the processor 501 can further include a hardware chip. The above-mentioned hardware chip can be an application-specific integrated circuit, a programmable logic device, or a combination thereof. The above-mentioned programmable logic device can be a complex programmable logic device, a field programmable gate array, a generic array logic, or any combination thereof.

[0188] Among them, the memory 502 stores instructions that can be executed by at least one processor 501, so that at least one processor 501 executes the method shown in the above embodiments.

[0189] The memory 502 can include a program storage area and a data storage area. Among them, the program storage area can store an operating system and application programs required for at least one function; the data storage area can store data created according to the use of the electronic device, etc. In addition, the memory 502 can include a high-speed random access memory, and can also include a non-transitory memory, such as at least one disk storage device, a flash memory device, or other non-transitory solid-state storage devices. In some alternative embodiments, the memory 502 can optionally include a memory remotely provided with respect to the processor 501, and these remote memories can be connected to the electronic device through a network. Examples of the above-mentioned network include but are not limited to the Internet, an enterprise intranet, a local area network, a mobile communication network, and combinations thereof.

[0190] The memory 502 can include a volatile memory, for example, a random access memory; the memory can also include a non-volatile memory, for example, a flash memory, a hard disk, or a solid-state drive; the memory 502 can also include a combination of the above types of memories.

[0191] The electronic device further includes an input device 503 and an output device 504. The processor 501, the memory 502, the input device 503, and the output device 504 can be connected through a bus or other means, Figure 10 Taking the connection through the bus as an example.

[0192] The input device 503 can receive input digital or character information, and generate key signal inputs related to the user settings and function controls of the electronic device, such as a touch screen, a keypad, a mouse, a trackpad, a touchpad, a pointing stick, one or more mouse buttons, a trackball, a joystick, etc. The output device 504 can include a display device, an auxiliary lighting device (for example, an LED), and a tactile feedback device (for example, a vibration motor), etc. The above-mentioned display device includes but is not limited to a liquid crystal display, a light-emitting diode, a display, and a plasma display. In some alternative embodiments, the display device can be a touch screen.

[0193] Embodiments of the present disclosure also provide a computer-readable storage medium. The methods according to the embodiments of the present disclosure can be implemented in hardware, firmware, or be implemented as computer code that can be recorded on a storage medium, or be implemented as computer code that is originally stored in a remote storage medium or a non-transitory machine-readable storage medium and downloaded through a network and will be stored in a local storage medium, so that the methods described herein can be processed by such software stored on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. Among them, the storage medium can be a magnetic disk, an optical disk, a read-only memory, a random access memory, a flash memory, a hard disk, or a solid-state drive, etc.; further, the storage medium can also include a combination of the above-mentioned types of memories. It can be understood that a computer, a processor, a microprocessor controller, or programmable hardware includes a storage component that can store or receive software or computer code. When the software or computer code is accessed and executed by the computer, the processor, or the hardware, the methods shown in the above embodiments are implemented.

[0194] A part of the present disclosure can be applied as a computer program product, for example, computer program instructions. When executed by a computer, through the operation of the computer, the methods and / or technical solutions according to the present disclosure can be called or provided. Those skilled in the art should be able to understand that the forms in which computer program instructions exist in a computer-readable medium include, but are not limited to, source files, executable files, installation package files, etc. Correspondingly, the ways in which computer program instructions are executed by a computer include, but are not limited to: the computer directly executes the instruction, or the computer compiles the instruction and then executes the corresponding compiled program, or the computer reads and executes the instruction, or the computer reads and installs the instruction and then executes the corresponding installed program. Here, the computer-readable medium can be any available computer-readable storage medium or communication medium accessible by the computer.

[0195] Although the embodiments of the present disclosure have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of the present disclosure, and such modifications and variations all fall within the scope defined by the appended claims.

Claims

1. A method for repairing media content, characterized in that: The method comprises: Obtain the target media content that was not successfully published and the corresponding reason for the unsuccessful publication; Based on the first media information of the target media content and the reason, repaired media content is obtained and displayed, wherein the repaired media content is obtained after performing a corresponding repair task on the target media content based on a target language model and task prompt information, the task prompt information corresponds to the repair task, and the task prompt information is obtained based on the first media information and the reason.

2. The method for repairing media content according to claim 1, characterized in that: The method for generating the repaired media content includes: Extracting information from the target media content to obtain first media information corresponding to at least one repair task; Obtaining a repair requirement corresponding to the repair task; Based on the reason, the first media information and the repair requirement, obtaining task prompt information corresponding to the repair task; Based on the task prompt information and the target language model, the target media content is processed with a corresponding repair task to obtain second media information of the at least one repair task, so as to obtain repaired media content.

3. The method for repairing media content according to claim 2, characterized in that: The obtaining, based on the reason, the first media information and the repair requirement, task prompt information corresponding to the repair task includes: Obtaining a prompt information template corresponding to the repair task; The first media information, the reason, and the repair requirement are integrated based on the prompt information template to obtain task prompt information corresponding to the repair task.

4. The method for repairing media content according to claim 3, characterized in that: The fusing the first media information, the reason and the repair requirement based on the prompt information template to obtain task prompt information corresponding to the repair task further includes: Extracting attribute information of the target media content; The attribute information, the reason, the first media information and the repair requirement are integrated based on the prompt information template to obtain task prompt information corresponding to the repair task.

5. The method for repairing media content according to any one of claims 1 to 4, characterized in that: The target language model is determined by: Acquire media publishing conditions and media content samples that have not been successfully published, wherein the media publishing conditions include publishing conditions corresponding to the repair task; Adjusting parameters of the initial language model based on the media publishing condition to obtain a first language model; Acquire a successfully published repair sample corresponding to the media content sample, and generate a first sample pair based on the media content sample and the successfully published repair sample; Based on the first sample pair, adjusting parameters of the first language model to obtain a second language model; Acquire an unsuccessfully published repair sample corresponding to the media content sample, and generate a second sample pair based on the media content sample, the successfully published repair sample, and the unsuccessfully published repair sample; Based on the second sample pair, the parameters of the second language model are adjusted to obtain the target language model.

6. The method for repairing media content according to claim 5, characterized in that: The step of obtaining a successfully published repair sample corresponding to the media content sample includes: Repairing the media content sample to obtain a first repaired sample; Obtaining the publishing result of the first repair sample; If the first repair sample is not published successfully, repairing the media content sample and / or the first repair sample multiple times to obtain multiple second repair samples; Obtaining the publishing results of each of the second repair samples; If any of the second repair samples is successfully released, the second repair sample is determined as the successfully released repair sample.

7. The method for repairing media content according to claim 6, characterized in that: The step of obtaining a successfully published repair sample corresponding to the media content sample further includes: If the first repair sample is published successfully, the first repair sample is determined as the successfully published repair sample.

8. The method for repairing media content according to claim 6, characterized in that: The obtaining of an unpublished repair sample corresponding to the media content sample includes: The unsuccessfully published repair sample is obtained from the unsuccessfully published first repair sample and / or the unsuccessfully published second repair sample.

9. The method for repairing media content according to claim 2, characterized in that: If the target media content includes a script, and the at least one repair task includes a first repair task for repairing the script, then based on the task prompt information and the target language model, processing the target media content with the corresponding repair task to obtain second media information of the at least one repair task includes: Based on the target language model and the task prompt information of the first repair task, the script of the target media content is processed by the first repair task to obtain second media information of the first repair task.

10. The method for repairing media content according to claim 2, characterized in that: If the target media content includes an image, and the at least one repair task includes a second repair task for repairing image text, then based on the task prompt information and the target language model, processing the target media content with the corresponding repair task to obtain second media information of the at least one repair task includes: Based on the target language model and the task prompt information of the second repair task, the image text of the target media content is processed for the second repair task to obtain second media information of the second repair task.

11. The method for repairing media content according to claim 2, characterized in that: The at least one repair task includes a third repair task for repairing the missing prompt language, and the processing of the target media content with the corresponding repair task based on the task prompt information and the target language model to obtain the second media information of the at least one repair task includes: Based on the target language model and the task prompt information of the third repair task, the target media content is processed for the third repair task to obtain second media information of the third repair task, where the second media information includes the missing prompt.

12. The method for repairing media content according to claim 2, characterized in that: If the target media content includes storyboards, and the at least one repair task includes a fourth repair task for repairing the storyboards, then based on the task prompt information and the target language model, processing the target media content with the corresponding repair task to obtain second media information of the at least one repair task includes: Based on the target language model and the task prompt information of the fourth repair task, the fourth repair task is processed on the storyboards of the target media content to obtain second media information of the fourth repair task.

13. The method for repairing media content according to claim 1, characterized in that: The repaired media content includes a repaired script, and the audio played when the repaired media content is displayed corresponds to the repaired script; And / or, the repaired media content contains target image text that does not meet the media publishing conditions, and the target image text is covered when the repaired media content is displayed; And / or, the repaired media content is newly added with the prompt words missing from the target media content; And / or, the repaired media content removes the storyboards that do not meet the media publishing conditions.

14. A media content repair device, characterized in that: The device comprises: A data acquisition module, used to acquire the target media content that has not been successfully published and the corresponding reasons for the unsuccessful publication; A data processing module is used to obtain and display repaired media content based on the first media information of the target media content and the reason, wherein the repaired media content is obtained after processing the target media content with a corresponding repair task based on a target language model and task prompt information, wherein the task prompt information corresponds to the repair task, and the task prompt information is obtained based on the first media information and the reason.

15. An electronic device, characterized in that: include: A memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the method for repairing media content according to any one of claims 1 to 13 by executing the computer instructions.

16. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a computer to execute the media content repair method according to any one of claims 1 to 13.

17. A computer program product, characterized in that The method comprises computer instructions, wherein the computer instructions are used to cause a computer to execute the media content repairing method according to any one of claims 1 to 13.

Images