Data processing method and electronic equipment

Abstract

The embodiment of the invention provides a data processing method and electronic equipment. The data processing method comprises the steps that first input data and a response result for the first input data are obtained; obtaining evaluation information of the response result; and performing corresponding processing on the response result by using a target processing strategy based on the evaluation information so as to output a target file, wherein at least one characteristic parameter of the target file is superior to a corresponding parameter of the response result, and the target file better conforms to the intention of a target user relative to the response result.

Description

Technical Field

[0001] The present application relates to but is not limited to the field of information technology, and in particular to a data processing method and an electronic device. Background Art

[0002] With the continuous advancement of information technology, users' demands for data processing are becoming increasingly diversified and personalized. They are no longer satisfied with simple data query and statistics, but hope to conduct in-depth analysis and mining of data to obtain more valuable information. In related technologies, data processing relies too much on manual experience and rules, lacks intelligence and adaptive capabilities, and makes the accuracy and compliance of the processed results low, which cannot meet user needs. Summary of the invention

[0003] In view of this, an embodiment of the present application provides a data processing method and an electronic device.

[0004] The technical solution of the embodiment of the present application is implemented as follows:

[0005] In a first aspect, an embodiment of the present application provides a data processing method, comprising: obtaining first input data and a response result to the first input data; obtaining evaluation information of the response result; performing corresponding processing on the response result with a target processing strategy based on the evaluation information to output a target file; wherein at least one characteristic parameter of the target file is better than a corresponding parameter of the response result, and the target file is more in line with the target user's intention than the response result.

[0006] In a second aspect, an embodiment of the present application provides an electronic device, comprising a processor and at least one processing model capable of running on the processor, wherein the processing model can be called by a target application to perform at least one of the following: obtaining first input data and a response result to the first input data; obtaining evaluation information for the response result; performing corresponding processing on the response result with a target processing strategy based on the evaluation information to output a target file; wherein at least one characteristic parameter of the target file is better than a corresponding parameter of the response result, and the target file is more in line with the target user's intention than the response result.

[0007] In a third aspect, an embodiment of the present application provides a data processing device, including:

[0008] A first obtaining module, used to obtain first input data and a response result to the first input data;

[0009] A second obtaining module is used to obtain evaluation information of the response result;

[0010] A processing module, used for processing the response result accordingly with a target processing strategy based on the evaluation information to output a target file;

[0011] Among them, at least one characteristic parameter of the target file is better than the corresponding parameter of the response result, and the target file is more in line with the target user's intention than the response result.

[0012] In a fourth aspect, an embodiment of the present application provides a storage medium having a computer program stored thereon, wherein the computer program implements some or all of the steps in the above method when executed by a processor.

[0013] It should be understood that the above general description and the following detailed description are merely exemplary and explanatory, and are not intended to limit the technical solutions of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] The drawings herein are incorporated into the specification and constitute a part of the specification. These drawings illustrate embodiments consistent with the present application and are used together with the specification to illustrate the technical solution of the present application.

[0015] Figure 1 A schematic diagram of an implementation flow of a data processing method provided in an embodiment of the present application;

[0016] Figure 2 A schematic diagram of an implementation flow of an image processing method provided in an embodiment of the present application;

[0017] Figure 3 A schematic diagram of the implementation process of a multi-dimensional image evaluation method for generating images based on text provided in an embodiment of the present application;

[0018] Figure 4 A schematic diagram of an image processing system architecture provided in an embodiment of the present application;

[0019] Figure 5 A schematic diagram of a hardware entity of an electronic device provided in an embodiment of the present application;

[0020] Figure 6 A schematic diagram of the composition of a data processing device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0021] In order to make the purpose, technical solutions and advantages of the present application clearer, the technical solutions of the present application are further elaborated in detail below in conjunction with the drawings and embodiments. The described embodiments should not be regarded as limiting the present application. All other embodiments obtained by ordinary technicians in the field without making creative work are within the scope of protection of the present application.

[0022] In the following description, reference is made to “some embodiments”, which describe a subset of all possible embodiments, but it will be understood that “some embodiments” may be the same subset or different subsets of all possible embodiments and may be combined with each other without conflict.

[0023] It should be pointed out that the terms "first\second\third" involved in the embodiments of the present application are merely to distinguish similar objects and do not represent a specific ordering of the objects. It can be understood that "first\second\third" can be interchanged in a specific order or sequence where permitted, so that the embodiments of the present application described here can be implemented in an order other than that illustrated or described here.

[0024] It will be understood by those skilled in the art that, unless otherwise defined, all terms (including technical terms and scientific terms) used herein have the same meaning as the general understanding of those skilled in the art in the field to which the embodiments of the present application belong. It should also be understood that terms such as those defined in general dictionaries should be understood to have meanings consistent with the meanings in the context of the prior art, and will not be interpreted with idealized or overly formal meanings unless specifically defined as here.

[0025] The method provided in the embodiment of the present application can be executed by an electronic device, wherein the electronic device can be a laptop, a tablet computer, a desktop computer, a set-top box, a mobile device (for example, a mobile phone, a portable music player, a personal digital assistant, a dedicated messaging device, a portable gaming device) and other types of terminals, and can also be implemented as a server. The server can be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content delivery networks (CDNs), and big data and artificial intelligence platforms.

[0026] The present application embodiment provides a data processing method, such as Figure 1 As shown, the data processing method includes the following steps S101 to S103, wherein:

[0027] Step S101, obtaining first input data and a response result to the first input data;

[0028] Here, the first input data may be multimedia data such as text data, voice data, image data, etc., or may be data related to generating a user interface (UI) required by the user.

[0029] The response result for the first input data may be a generated result obtained by generating and processing the first input data using the target processing model, such as a generated image, audio, text, video or UI prototype, etc. The target processing model may be an artificial intelligence large model deployed on a local terminal device (user terminal device), or a cloud large model that can be called by a local terminal device. The large model may be a generative processing model, such as a text-to-image model, an image-to-image model, a text-to-text model, a video generation model, a UI generation model, etc.

[0030] In some embodiments, the response result of the first input data is different according to the first input data. The task content triggered by different input data is different, and the type of generated response result may also be different. For example, if the first input data is text data, the response result of the text data can be an image; if the first input data is voice data, the response result of the voice data can also be an image; if the first input data is image data, the response result of the image data can be a paragraph of text; if the first input data is text data, the response result of the text data can be an image; if the first input data is data related to the UI required by the user, the response result of the data related to the UI required by the user can be a generated UI control or the entire UI interface.

[0031] Step S102, obtaining evaluation information of the response result;

[0032] In some embodiments, the evaluation information of the response result is different depending on the response result. For example, if the response result is an image, the evaluation information of the image may be an image quality score, such as the score of image parameters such as image resolution and image element layout; if the response result is a text, the evaluation information of the text may be a text quality score, such as the score of text fluency and accuracy; if the response result is a video, the corresponding evaluation information may be video evaluation information, such as video resolution, frame rate, transition effect, etc.; if the response result is a UI, the evaluation information of the UI may be a UI quality score.

[0033] In some embodiments, the evaluation information of the response result may also be the user's evaluation of the response result. For example, if the response result is an image, the user's evaluation of the response result may be a score given to the image by the user based on his or her subjective feelings, or a description of the user's feelings about the image in words (for example, the image is colorful but the details are not clear enough); or the user may make suggestions for modifying the image (for example, the details are not clear enough, the target elements are not prominent enough, the light or texture does not meet the user's expectations, etc.).

[0034] In some embodiments, the evaluation information of the response result may also include historical evaluation data, for example, evaluation data of the model used in the process of generating the response result for the first input data (for example, a raw image model, a raw text model, a raw video model, a raw audio model, etc.), and evaluation data of the model performing the corresponding generation task (for example, evaluation data of the raw image model performing the image generation task, evaluation data of the raw text model performing the text generation task, etc.).

[0035] Step S103, processing the response result accordingly with a target processing strategy based on the evaluation information to output a target file;

[0036] Here, the target processing strategy can be a strategy formulated based on the evaluation information to guide how to process the response results. The target processing strategy includes but is not limited to: repair, regenerate, replace the generation model, switch to cloud generation, update file parameters, etc.

[0037] In some embodiments, repair may include: repairing errors or defects in the response result, such as removing noise in an image, correcting grammatical errors in a text, etc. Regeneration may include: if the quality of the response result does not meet the requirements, you may choose to regenerate the response result to obtain a higher quality target file. Replacing the generation model may include: if the generation model currently used does not work well, you may try to replace it with another model to improve the quality of the output target file. Switching to cloud generation may include: if the performance of the local device is limited, you may choose to switch the generation task to the cloud and use the more powerful computing power of the cloud for processing. Updating file parameters may include: adjusting the parameters of the generated file based on the feedback of the evaluation information, such as image resolution, text style, etc., to optimize the output target file.

[0038] Here, the target file is the final file that is output by processing the response result accordingly using the target processing strategy. The target file can be an image, text, video, UI, etc.

[0039] In some embodiments, for raw image scenarios, an image processing strategy can be used to beautify the generated image (for example, enhancing the contrast of the image, sharpening the edges, etc.), or a cloud-based raw image model or other raw image model can be used to compare the quality of the generated image, and a generated image with better quality can be selected as the target image, or the generated image can be modified and re-processed to obtain an image that better meets the requirements.

[0040] In some embodiments, for raw video scenes, an image processing model or a video editing model can be used to beautify the video (for example, adding filters, adjusting video speed, etc.), or a raw image model can be used to replace the target object or video frame of the generated video, or the prompt words corresponding to the generated video can be modified and the raw video can be re-processed to obtain video content that better meets the requirements.

[0041] In some implementations, for the text model scenario, a document decoration model or a beautified document model may be used to decorate and beautify the text (eg, adjust the font, paragraph format, etc.) to obtain a higher quality text.

[0042] It should be noted that at least one characteristic parameter of the target file is better than the corresponding parameter of the response result, and the target file is more in line with the target user's intention than the response result.

[0043] In some implementations, the characteristic parameters of the target file are different depending on the target file. For example, if the target file is an image, the characteristic parameters of the image may be image brightness, clarity, style, etc.; if the target file is audio, the characteristic parameters of the audio may be audio quality, bit rate, etc.; if the target file is a video, the characteristic parameters of the video may be video bit rate, resolution, color, etc.; if the target file is text, the characteristic parameters of the text may be repetition rate, error rate, etc.

[0044] Here, the target user intent may be the actual need or expectation of the user that triggers the generation of the task.

[0045] In some embodiments, if the user wants to generate a bright and clear picture, the target file should be better than the response result in these two characteristic parameters to better meet the user's expectations. If the user wants to generate a piece of audio with good sound quality, the target file should be better than the response result in audio quality and bit rate to provide a better auditory experience. If the user wants to generate an accurate text, the target file should be better than the response result in text repetition rate and error rate to ensure the accuracy and readability of the text.

[0046] In the embodiments of the present application, on the one hand, by obtaining evaluation information on the response result, it is possible to identify deficiencies in the response result, and based on the evaluation information, adopt a target processing strategy to perform corresponding processing on the response result, thereby outputting a target file that is superior to the response result in one or more characteristic parameters; on the other hand, since at least one characteristic parameter of the target file is superior to the corresponding parameter of the response result, the target file is more in line with the actual needs or expectations of the target user relative to the response result, thereby improving user satisfaction and experience.

[0047] In some embodiments, the implementation of step S102 "obtaining evaluation information of the response result" may include at least one of the following steps S111 to S114, wherein:

[0048] Step S111, performing quality assessment on the response result to obtain the evaluation information;

[0049] Here, the purpose of quality assessment is to obtain evaluation information of the response results. This evaluation information can reflect the performance of the response results in specific dimensions, such as accuracy, completeness, consistency, etc. Through quality assessment, deficiencies or potential problems in the response results can be identified, providing a basis for subsequent processing or improvement.

[0050] In some embodiments, when the response result is an image, a text similarity evaluation method can be used to determine the similarity between the text obtained after the generated image is converted into text and the reference text. The text similarity can be measured by calculating the vocabulary overlap, semantic similarity and other indicators between the two texts.

[0051] In some implementations, when the response result is converted to text and is determined to contain multiple fields, each field may be scored for accuracy to evaluate the performance of the response result in each field. The field accuracy score may be formulated based on specific requirements of the field (e.g., data type, format, value range, etc.).

[0052] In some embodiments, when the response result is an image, a multi-dimensional image feature scoring method can also be used to evaluate the quality of the image. The multi-dimensional image feature scoring can include multiple aspects such as image clarity, contrast, color saturation, brightness, etc.

[0053] Step S112, obtaining feedback information of the target user on the response result, and obtaining the evaluation information based on the feedback information;

[0054] Here, the target user may be the owner or current user of the current user-end device (e.g., a smart phone, a smart computer, a smart home, etc.), or the user of other devices that have a communication connection with the current user-end device, for example, the other party in a social interaction (e.g., a chat, a video call, etc.), or other participants in a meeting. It should be noted that the current user-end device has artificial intelligence (AI) processing capabilities or an AI large model or an AI generation model.

[0055] Here, the feedback information can be the user's evaluation of the response result, for example, a score for the response result (image / text / video / audio), or a descriptive evaluation of the response result (for example, this image is clear and colorful, this text is fluent but some parts are not accurate enough, etc.), or suggestions for modifying the response result, etc.

[0056] In some implementations, when the user provides a clear quantitative rating, the rating can be used as evaluation information.

[0057] In some embodiments, when the user provides descriptive evaluation content, it may be necessary to use natural language processing technology (e.g., sentiment analysis, keyword extraction, etc.) to extract the user's evaluation tendency and key information, and then generate a quantitative score based on the evaluation tendency and key information, and use the quantitative score as the evaluation information or directly use the descriptive evaluation content as the evaluation information.

[0058] In some embodiments, when a user provides a modification suggestion, it may be necessary to evaluate the feasibility and effectiveness of the modification suggestion through a machine learning model or manual review, and then generate a comprehensive evaluation information based on the evaluation results.

[0059] Step S113, obtaining first historical evaluation data for a first processing device and / or a first processing model, and obtaining the evaluation information based on the first historical evaluation data, wherein the first processing model is a model for generating the response result in response to the first input data, and the first processing device is a device for obtaining the response result by using the first processing model;

[0060] Here, the first processing device may be a device that processes input data using a specific model (e.g., the first processing model) and generates a response result. For example, the first processing device may be a server, a computer, a smart phone, a tablet computer, or other devices with data processing capabilities.

[0061] Here, the first processing model may be a model for responding to the first input data and generating a response result. The first processing model may be a raw image model (for generating images), a raw text model (for generating text), a raw video model (for generating videos), a raw audio model (for generating audio), etc.

[0062] Here, the first historical evaluation data may be the evaluation data of the model used in the process of generating a response result for the first input data (for example, a raw image model, a raw text model, a raw video model, a raw audio model, etc.), as well as the evaluation data of the model performing the corresponding generation task (for example, the evaluation data of the raw image model performing the image generation task, the evaluation data of the raw text model performing the text generation task, etc.).

[0063] In some implementations, the first historical evaluation data may be used as evaluation information. For example, if the first historical evaluation data gives a clear quantitative rating to the user, the rating may be used as evaluation information.

[0064] In some implementations, the first historical evaluation data may need to be converted to generate evaluation information. For example, if the first historical data is a descriptive evaluation content given by the user, it may be necessary to extract the user's evaluation tendency and key information through natural language processing technology (e.g., sentiment analysis, keyword extraction, etc.), and then generate a quantitative score based on the evaluation tendency and key information, and use the quantitative score as the evaluation information.

[0065] In some implementations, if there are multiple pieces of first historical evaluation data, different weights may need to be given to the pieces according to their importance or reliability. Then, the evaluation data may be weighted averaged according to the weights to obtain comprehensive evaluation information.

[0066] Step S114, obtaining second historical evaluation data for a target application, and obtaining the evaluation information based on the second historical evaluation data, wherein the target application is an application that responds to the first input data to output the response result.

[0067] Here, the target application may be an application that can respond to the first input data and output a response result. The target application may be an image generation application (for generating images), a text generation application (for generating text), a video generation application (for generating videos), an audio generation application (for generating audio), etc.

[0068] Here, the second historical evaluation data may be evaluation data of response results generated by the target application when processing tasks in the past, or may be historical evaluation data of the target application by a specific user.

[0069] In some embodiments, when the second historical evaluation data is historical evaluation data of a specific user on the target application, personalized evaluation information for the user can be generated based on the historical evaluation data. For example, the user's past evaluation data on different versions or different functions of the application (e.g., good review rate, bad review rate, etc.) can be analyzed to generate evaluation information to understand the user's preferences and needs for the application.

[0070] In an embodiment of the present application, on the one hand, evaluation information is obtained based on the feedback information of the target user on the response result, which can more accurately reflect the user's satisfaction and needs; on the other hand, evaluation information is obtained based on the first historical evaluation data of the first processing device and / or the first processing model, which provides a performance record of past processing tasks, and helps to understand the performance of the first processing device and / or the first processing model; on the other hand, evaluation information is obtained based on the second historical evaluation data of the target application, which can predict the future market performance and user satisfaction of the target application.

[0071] In some embodiments, the implementation of step S111 “performing a quality assessment on the response result” may include at least one of the following steps S121 to S122, wherein:

[0072] Step S121, performing recognition processing on the response result to obtain first recognition data, and obtaining the evaluation information based at least on the similarity between the first input data and the first recognition data;

[0073] Here, the recognition processing may be image-to-text conversion processing (converting an image into text), image recognition processing (recognizing objects, scenes, etc. in an image), feature extraction processing (extracting features of an image), video-to-text processing (converting video content into text), text summarization processing, abstract processing, etc. It should be noted that the recognition processing may be performed by a corresponding processing model (e.g., a text-to-image model, a video-to-text model, etc.) or a processing program, or a processing program logic embodied in a non-model form of program code.

[0074] In some embodiments, the response result is processed accordingly according to the type of the first input data or the user intent represented, or the processing task information corresponding to the first input data. For example, if the first input data is text and an image is required to be generated, the corresponding processing is text-to-image processing; if the first input data is text and another text is required to be generated, the corresponding processing may be semantic recognition and intent understanding processing, or a summary or abstract processing of word expansion processing; if the first input data is text and a video is required to be generated, the corresponding processing may be text-to-video processing.

[0075] It should be noted that the type of the first recognition data is the same as or different from the type of the first input data. For example, if the first input data is text data and the recognition process is a text summary process, then the first recognition data is also text data, that is, the type of the first recognition data is the same as the type of the first input data. If the first input data is text data and the recognition process is a text-to-image process, then the first recognition data is image data, that is, the type of the first recognition data is different from the type of the first input data.

[0076] In some embodiments, image quality can be determined based on text similarity, and can be further determined in combination with the field accuracy of the first recognition data to determine text quality, or the image feature accuracy of the first recognition data to determine image quality, or the audio feature accuracy of the first recognition data to determine audio quality.

[0077] Step S122, identifying the user intention represented by the first input data, and performing quality evaluation on the response result based on the user intention with a corresponding evaluation strategy, where different user intentions correspond to different evaluation strategies.

[0078] Here, the user intention may be the purpose of the task or instruction that the user wishes to complete through the first input data. For example, the user may wish to generate a specific type of image, text report, video, etc.

[0079] Here, the corresponding evaluation strategy is selected according to the user's intention. Different user intentions (e.g., generating images, generating text, generating audio, generating video, etc.) will have different evaluation criteria and methods.

[0080] In some embodiments, the generated response results are evaluated for quality according to the selected evaluation strategy. For example, for generated images, the accuracy may be evaluated by extracting multi-dimensional image features. For generated text, the accuracy may be evaluated by comparing text similarity and text field accuracy. For generated audio, the quality may be evaluated by comparing audio features. For generated video, the quality may be evaluated by checking video parameters.

[0081] In an embodiment of the present application, on the one hand, by using the similarity between the first input data and the first identification data, it is possible to evaluate whether the response result accurately reflects the content of the input data; on the other hand, by selecting a corresponding evaluation strategy based on the user's intention, it is possible to ensure that the quality evaluation of the response result is highly relevant to the user's actual needs, thereby improving the pertinence and effectiveness of the quality evaluation.

[0082] In some embodiments, the implementation of step S111 “performing a quality assessment on the response result” may include at least one of the following steps S131 to S133, wherein:

[0083] Step S131, when the first input data is text data, using a first processing model to calculate a similarity score between the first input data and the first recognition data, and using the similarity score as the evaluation information;

[0084] Here, the first processing model can be a self-attention mechanism model (Bidirectional EncoderRepresentations from Transformers, Transformer BERT).

[0085] In some embodiments, for the first input data being text data and the first recognition data being image-to-text data, a self-attention mechanism model is used to calculate a similarity score between the text data and the image-to-text data, and the similarity score is used as evaluation information to evaluate the quality of the image generated by the text-generated graph model. The higher the similarity score, the more the generated image matches the original text data, and thus the higher the quality of the generated image can be considered.

[0086] Step S132, using a second processing model to calculate a similarity score between the first input data and the first recognition data, performing accuracy calculation on key fields included in the first recognition data, and weighting the similarity score and the accuracy calculation result of the key field based on corresponding weights to obtain the evaluation information;

[0087] Here, the second processing model can also be a self-attention mechanism model. The key field can be an explicit field and / or an implicit field.

[0088] In some embodiments, for the first input data being text data, first, the text data is parsed to obtain image feature fields corresponding to the text data (e.g., art style, lighting, material, art medium, clarity, etc.), or implicit fields in the text data can be automatically extracted by natural language processing technology; secondly, a clustering algorithm (K-means) is used to perform cluster analysis on each field and divide it into 9 categories, each category representing the main features under the field; then, the generated image is classified according to the category corresponding to each field, and the classification accuracy is calculated, so as to use the classification accuracy as the score of the field. Among them, the number of categories of the classification task is 10, including 9 cluster categories and 1 user-specific description category, and the user-specific description category may be used to accommodate images that do not conform to the cluster category but have a specific user description. The classification accuracy can be the ratio of the number of correctly classified images to the total number of images. Finally, the scores of all fields are weighted and summed (the weight can be adjusted according to the importance of the field) to obtain the field comprehensive score, and the text similarity score and the field comprehensive score are weighted and summed to obtain the final score of the image. This score is used to quantify the overall quality of the generated image. For example, the first input data input by the user may be classified by fields and then matched with corresponding features of the image, such as whether the text description of light is reflected in the generated image, to score the generated image.

[0089] Step S133, when the response result is image data, use the third processing model to extract image features of multiple dimensions of the image data, calculate the accuracy score of each dimension based on the image features, and weight the accuracy scores of the various dimensions to obtain the evaluation information.

[0090] Here, the third processing model may be a deep learning model. The image features of multiple dimensions of the image data may be the subject, lighting, color, artistic style, clarity, etc. of the image. Next, the multiple dimensional image features are taken as an example of subject and artistic style.

[0091] In some embodiments, first, a deep learning model is used to extract image features of multiple dimensions of image data to obtain the theme and artistic style of the image; then, the accuracy scores of the theme and artistic style are calculated respectively; finally, the accuracy score of the theme and the accuracy score of the artistic style are weighted to obtain the quality score of the overall image.

[0092] In the embodiment of the present application, for text data, both the overall similarity and the accuracy of key fields are considered; for image data, image features are extracted from multiple dimensions and the accuracy is scored. This multi-dimensional evaluation can more comprehensively reflect the quality of the response results.

[0093] In some embodiments, the implementation of step S103 "processing the response result accordingly with the target processing strategy based on the evaluation information" may include at least one of the following steps S141 to S144, wherein:

[0094] Step S141, when the evaluation information indicates that the response result meets the user intention indicated by the first input data, directly output the response result as the target file, or generate the response result to obtain the target file;

[0095] Here, the response result being in accordance with the user intent represented by the first input data may be that the quality of the generated image / the quality of the generated text is good, meeting the user's expectations or not meeting the user's goals. For example, the number and content of keywords in the generated text are the same as the number and content of keywords in the first input data, or the number and content of features in the generated image are the same as the number and content of features in the first input data, etc.

[0096] In some embodiments, when the number and content of keywords in the generated text are the same as the number and content of keywords in the first input data, the generated text may not be subjected to any additional processing, but may be directly output as a target file, or the generated text may be subjected to a generation process again. This generation process may be a model that is the same as the first generation process, or may be a different model. Finally, the text subjected to the regeneration process may be output as a target file.

[0097] In some embodiments, when the number and content of features in the generated image are the same as the number and content of features in the first input data, the generated image may not be subjected to any additional processing, but may be directly output as a target file, or the generated image may be subjected to a regeneration process. This generation process may be a model that is the same as the first generation process, or may be a different model. Finally, the image subjected to the regeneration process may be output as a target file.

[0098] Step S142, when the evaluation information indicates that the response result does not conform to the user intention indicated by the first input data, repairing the response result to obtain the target file;

[0099] Here, the response result not meeting the user intent represented by the first input data may be that the quality of the generated image / the quality of the generated text is poor, not meeting the user's expectations or not meeting the user's goals. For example, the keyword content in the generated text is not completely the same as the keyword content in the first input data, or the feature content in the generated image is not completely the same as the feature content in the first input data, etc.

[0100] In some embodiments, when the feature content in the generated image is not completely the same as the feature content in the first input data, the non-completely identical features (e.g., light intensity, color warmth) are determined; then, the light intensity and color warmth in the generated image are repaired respectively using the repair standards defined by the user for the non-completely identical features, and the repaired image is used as the target file.

[0101] Step S143, when the evaluation information indicates that the response result does not conform to the user intention indicated by the first input data, the first input data is generated again and then the response result is replaced as the target file, or the response result is compared with the generated result of the first input data to output the target file;

[0102] In some embodiments, when the feature content in the generated image is not completely the same as the feature content in the first input data, the first input data is generated and processed again through a local model or a cloud model to obtain a regenerated image, and the regenerated image is used to replace the generated image of the first input data as the target image.

[0103] In some embodiments, when the feature content in the generated image is not exactly the same as the feature content in the first input data, the image of the first input data is compared with the image generated again for the first input data, and the image with better quality is selected as the target image for output, or both images are output as target images.

[0104] Step S144, when the evaluation information is negative feedback information provided by the target user, the first input data and the negative feedback information are generated and processed again to replace the response result as the target file, or the characteristic parameters of the response result are adjusted based on the negative feedback information to output the target file.

[0105] Here, the negative feedback information provided by the target user may be a negative evaluation of the generated image by the target user, for example, the generated image does not conform to the style, color matching, or image clarity.

[0106] In some embodiments, when the evaluation information does not match the style of the generated image provided by the target user, the feedback information of the mismatch between the first input data and the generated image style is input as a prompt word into the first generation model or a different processing model for generation processing to obtain a regenerated image, and the regenerated image is used to replace the generated image as the target file.

[0107] Here, if the response result is an image, then the characteristic parameters of the image may be style, color, etc.; if the response result is a video, then the characteristic parameters of the video may be frame rate, resolution, etc.; if the response result is audio, then the characteristic parameters of the audio may be bit rate, timbre, etc.; if the response result is text, then the characteristic parameters of the text may be display layout, font, font size, etc.

[0108] In some embodiments, for different types of response results, targeted adjustments are made according to their specific characteristic parameters. For example, if the negative feedback information provided by the user is that the image style and color matching are inconsistent, then the image style and color matching are adjusted to make the output target file more in line with the user's aesthetic requirements; if the negative feedback information provided by the user is that the frame rate and resolution of the video are inconsistent, then the frame rate and resolution of the video are adjusted to improve the playback effect of the output target file; if the negative feedback information provided by the user is that the audio bit rate and timbre are inconsistent, then the audio bit rate and timbre are adjusted to optimize the sound quality of the output target file; if the negative feedback information provided by the user is that the text display layout, font and font size are inconsistent, then the text display layout, font and font size are adjusted to improve the readability and visual effect of the output target file.

[0109] In an embodiment of the present application, on the one hand, when the evaluation information indicates that the response result meets the user's intention, the response result can be directly output or a simple generation process can be performed to improve efficiency; when the evaluation information indicates that the response result does not meet the user's intention, a repair process or a regeneration process can be adopted to ensure the accuracy of the output target file; on the other hand, by adjusting the response result according to the negative feedback information provided by the user, the target file can better meet the user's needs, thereby improving user satisfaction.

[0110] In some embodiments, the implementation of step S142 "repairing the response result" may include the following steps S151 to S153, wherein:

[0111] Step S151, obtaining multi-dimensional features of the response result;

[0112] Here, if the response result is an image, the multi-dimensional features of the image may be the subject, lighting, color, artistic style, clarity, etc. of the image.

[0113] In some embodiments, a target detection model is used to identify the main objects and scenes in an image; an image illumination estimation network is used to analyze the illumination distribution and brightness in the image; a color extraction algorithm is used to extract the main colors and tonal distribution in the image; a style extraction model is used to extract the artistic style in the image; and a clarity detection algorithm is used to evaluate the clarity of the image.

[0114] Step S152, obtaining target dimension features and their weight information based on the first input data;

[0115] Here, if the first input data is text data, then the target dimension feature may be the theme, style, lighting, color scheme, etc. in the text data.

[0116] In some implementations, natural language processing technology is used to parse text data to obtain themes, styles, lighting, colors, etc. in the text data. The weight information of each target dimension feature can be user-defined, for example, the weight information corresponding to the theme is 0.5, the weight information corresponding to the style is 0.3, the weight information corresponding to the lighting is 0.1, and the weight information corresponding to the color is 0.1.

[0117] Step S153, adjusting the multi-dimensional features of the response result based on the target dimensional features and their weight information to obtain the target file.

[0118] In some embodiments, the features extracted from the text data and the weights corresponding to each feature are used as a standard, and the corresponding features of the generated image are adjusted using this standard so that the image is more visually consistent with the content described in the text or the user's intention. For example, the theme of the image is adjusted using the theme and the weight information corresponding to the theme; the style of the image is adjusted using the style and the weight information corresponding to the style; the lighting of the image is adjusted using the lighting and the weight information corresponding to the lighting; and the color matching of the image is adjusted using the color matching and the weight information corresponding to the color matching, to obtain the adjusted image, and the adjusted image is used as the target file.

[0119] In an embodiment of the present application, by obtaining the target dimensional features and their weight information from the first input data, the multi-dimensional features of the response results can be accurately adjusted according to the user's needs, thereby significantly improving the quality of the response results and ensuring that the target file ultimately generated is more in line with the user's expectations and preferences.

[0120] In some embodiments, the implementation of step S142 “repairing the response result” may include at least one of the following steps S161 to S164, wherein:

[0121] Step S161, obtaining the target user's operation behavior data for the response result, and updating the characteristic parameters of the response result based on the operation behavior data to obtain the target file;

[0122] Here, for example, if the response result is an image, the operation behavior data can be the user's click, drag, zoom, rotate and other operations on the image, as well as the user's setting of the adjustment range of image properties (for example, light intensity, color warmth, etc.) through controls on the interface (for example, sliders, buttons, etc.).

[0123] In some embodiments, the light intensity in the image is updated according to the adjustment range of the light intensity, and the color warmth in the image is updated according to the adjustment range of the color warmth, to obtain an updated image, and the updated image is used as a target file, so that the target file better meets the user's expectations and needs, thereby improving the user's satisfaction and experience.

[0124] Step S162, generating the target file based on the obtained selection operation of the recommended data for the target data in the response result and / or the editing operation for the target data;

[0125] Here, the recommendation data of the target data may be recommended options for specific words (such as words in the generated text) or image elements (such as elements in the generated image) in the response result.

[0126] In some implementations, the user can select one of these options to replace the generated word or image element, or modify the generated word or image element to generate a target file that meets the user's expectations.

[0127] In some implementations, the user may also directly edit and modify the target data in the response result, for example, modify an element in the generated image, adjust certain words or sentences in the generated text, etc., and ultimately generate a target file that meets the user's expectations.

[0128] Step S163, generating the target file based on the configuration operation obtained for the configuration options in the configuration window of the response result, wherein the configuration options are used to update the feature data in the response result;

[0129] Here, the configuration operation may be selecting new feature data, adjusting parameters or attributes of feature data, etc. Feature data is the part of the response result that needs to be updated or replaced, and the feature data may be a target keyword, a target image frame, a target object in an image, or a target segment in a voice, etc.

[0130] In some embodiments, the user can update or replace the feature data in the response result by selecting, adjusting or setting the configuration options in the configuration window. For example, the user selects a target object to replace the object in the image, generates the replaced image, and uses the replaced image as the target file.

[0131] Step S164, obtaining context data associated with the response result, and using the context data to update the response result to obtain the target file.

[0132] Here, the context data may be the user's historical behavior data, environmental information, time information, geographic location information, user preference settings, etc. The update process may be to adjust the format, content, layout, presentation method, etc. of the response result.

[0133] In some embodiments, the response result is updated using the obtained context data. For example, if the context data indicates that the user is viewing information within a specific time period, then adjusting the response result to only display relevant data within the time period and using the relevant data within the time period as the target file can ensure that the response result is always consistent with the current context data, thereby providing a better user experience.

[0134] In an embodiment of the present application, on the one hand, by obtaining the target user's operational behavior data for the response result and updating the characteristic parameters of the response result based on the operational behavior data, the target file is made closer to the user's actual needs and preferences; on the other hand, the characteristic data in the response result is updated through the configuration options in the configuration window, so that the user can adjust the configuration options according to their own needs, thereby generating a target file that meets specific requirements; on the other hand, the response result is updated using the context data associated with the response result, so that the target file is consistent with the current context, which helps to improve the relevance and accuracy of the target file.

[0135] In some embodiments, the data processing method may include at least one of the following steps S171 to S172, wherein:

[0136] Step S171, obtaining an intermediate file obtained after corresponding processing of the response result, re-evaluating the intermediate file, and performing corresponding processing on the intermediate file based on the obtained re-evaluation result to output the target file;

[0137] Here, the intermediate file may be a file obtained after repairing the response result.

[0138] In some embodiments, for example, if the intermediate file is a repaired image, the evaluation result of the repaired image can be obtained by calculating the structural similarity index or peak signal-to-noise ratio of the repaired image; based on the evaluation result, an adaptive adjustment algorithm can be used to further improve the image quality, for example, adjusting the image brightness, contrast, sharpness and other parameters to ensure that the image quality meets the requirements.

[0139] Step S172: Record the target user's evaluation information on the response result, so as to optimize the processing model for processing the first input data based on the evaluation information.

[0140] Here, the evaluation information may be feedback information of the user on the response result, and the feedback information may include the repair standard and the characteristics of the first input data.

[0141] In some implementations, a processing model for processing the first input data is optimized based on a repair criterion and features of the first input data, so that the optimized processing model generates high-quality response results in the future.

[0142] It should be noted that the feedback information is stored in the database as a reference for future response result generation and repair processes. When similar input data is received again, the historical records in the database can be consulted to understand the user's expectations and feedback on such input data, thereby generating an image that better meets the user's expectations.

[0143] In the embodiments of the present application, on the one hand, by re-evaluating the intermediate files, problems in the processing process can be discovered and corrected in a timely manner, which helps to gradually approach the user's expectations, thereby improving the quality of the target file that is finally output; on the other hand, recording the user's evaluation information on the response results and optimizing the processing model based on this information can make the model more accurate and efficient in future processing tasks.

[0144] In some embodiments, the implementation of step S101 “obtaining first input data and a response result to the first input data” may include at least one of the following steps S181 to S183, wherein:

[0145] Step S181, obtaining a first processing instruction for the first input data, and performing generation processing on the first input data using a first processing model matching the first processing instruction to obtain the response result;

[0146] Here, the first processing instruction may be triggered by a user in some way (e.g., clicking a button, selecting a menu item, etc.). The first processing instruction may be of various types, such as a picture generation instruction, a text generation instruction, a video generation instruction, a UI generation instruction, an audio generation instruction, etc. These instructions correspond to different generation tasks, such as generating an image, generating text, generating a video, generating a user interface, generating audio, etc.

[0147] Here, for each type of instruction, there is a processing model that matches it. These processing models are trained and optimized to perform specific generation tasks. For example, for the image generation instruction, the corresponding processing model is the image generation model; for the text generation instruction, the corresponding processing model is the text generation model; and so on.

[0148] In some embodiments, first, when a first processing instruction is received, a corresponding processing model is called according to the type of the first processing instruction; then, the first input data is passed to the called processing model, and the processing model processes the first input data and generates a response result, and the response result meets the requirements of the generation task specified by the user through the first processing instruction. For example, in an image generation application, a user can trigger a generation instruction by clicking on a generation control, and then call the generation model to process the input data to generate an image. In a text generation application, a user can trigger a text generation task by selecting a generation instruction, and then call the generation model to process the input data to generate a piece of text. Similarly, for generation tasks such as video, audio, and user interface, users can also trigger the system's generation processing flow through corresponding instructions.

[0149] Step S182, generating and processing the first target data in the first input data based on the type of the first input data and the represented user intention to obtain the response result;

[0150] Here, the type of the first input data can be of various types, such as text data, image data, video data, etc. The first input data not only contains specific content, but also implies the user's intention or demand. For example, text data may express the content that the user wants to generate; image data may reflect the visual information that the user wants to convert or enhance.

[0151] Here, the first target data may be all or part of the first input data. For example, in text data, specific words are generated, such as nouns, adjectives, etc. In image data, specific objects are generated, such as a dog, grass, etc. in the original input image.

[0152] In some embodiments, the first target data in the first input data is generated based on the type of the first input data and the user intent represented. The specific method of the generation process depends on the type of the input data and the user intent. For example, for text data, semantic analysis or intent understanding may be performed to extract keywords, and text, images, or videos are generated based on these keywords. For image data, image-text conversion may be performed to convert the information in the image into text, and then a new image or video is generated based on the text.

[0153] Step S183, integrating the first input data and the context data of its source application to obtain second input data, and using a generative model to generate the second input data to obtain the response result.

[0154] Here, the context data of the source application may be additional information associated with the first input data. The additional information may come from the application that generates the first input data, and generally includes historical information, user behavior, environmental status, etc. related to the first input data.

[0155] In some embodiments, first, the first input data and the contextual data of its source application are integrated to form a more complete and richer data set, namely, the second input data; then, the second input data is passed to the generative model for processing to generate a response result.

[0156] In the embodiments of the present application, on the one hand, different processing models can be flexibly called through different processing instructions to meet diverse processing needs; on the other hand, by integrating the first input data and context data, the user's intentions and needs can be more comprehensively understood, and a more complete and accurate response result can be generated.

[0157] In the field of image generation, especially in the task of text-to-image generation, although existing technologies have been able to generate high-quality images, these images usually cannot fully meet the specific needs of users and the description of the prompt words. Although the image style transfer technology in the related art can adjust the style of the image, it still has certain limitations when dealing with complex prompt word descriptions and multi-dimensional restoration.

[0158] The image processing in the related art has the following shortcomings: insufficient comprehensive evaluation and adjustment capabilities in multiple dimensions, such as subject matter, lighting, color matching, artistic style, clarity, etc.;

[0159] Style transfer algorithms usually require a global style application to the image, which makes it difficult to fine-tune it to meet specific prompt word requirements; they lack the function of allowing users to customize restoration standards, making it difficult to meet personalized needs.

[0160] The embodiment of the present application provides an image processing method, which performs image restoration by combining the evaluation of multiple dimensions of the image (e.g., subject, lighting, color, artistic style, clarity, etc.) through a multi-dimensional restoration model. Accurate style transfer and image restoration are achieved by effectively utilizing style prompt words and their weights. Users are allowed to customize restoration standards and remember prompt word features for future use.

[0161] The present application embodiment provides an image processing method, such as Figure 2 As shown, the following steps may be included:

[0162] Step S201, generating an initial image according to the prompt word;

[0163] Here, the prompt word can include a description of the image theme, style, lighting, color scheme, etc.

[0164] In some implementations, a user submits a prompt word (eg, a mountain cabin at sunset, soft lighting, warm tones) through an interface, and the system generates an initial image based on the acquired prompt word.

[0165] Step S202, using a deep learning model to perform multi-dimensional feature extraction on the initial image;

[0166] In some embodiments, a target detection model is used to identify the main objects and scenes in the initial image; an image illumination estimation network is used to analyze the illumination distribution and brightness in the initial image; a color extraction algorithm is used to extract the main colors and tonal distribution in the initial image; a style is used to extract the artistic style in the initial image; and a clarity detection algorithm is used to evaluate the clarity of the initial image.

[0167] Step S203, analyzing the prompt words to extract relevant style features and corresponding weight information;

[0168] In some embodiments, natural language processing techniques are used to analyze the prompt words to identify key features related to style (e.g., soft lighting, warm tones, etc.) A weight is assigned to each extracted style feature to indicate its importance in the final image.

[0169] Step S204, adjusting the style of the initial image using a style conversion algorithm according to the style features and corresponding weight information;

[0170] In some embodiments, a style transfer algorithm is used to perform style adjustments on corresponding features (eg, lighting, color) of an initial image based on style features and corresponding weight information to obtain an intermediate image for further optimization.

[0171] Step S205, fine-tuning the intermediate image according to the restoration standard defined by the user to ensure that the final image meets the user's requirements;

[0172] In some embodiments, an interface is provided for users to customize restoration standards (eg, a slider to adjust light intensity, color warmth, etc.) to set adjustment parameters; the intermediate image is further adjusted and optimized according to the adjustment parameters to achieve the desired effect.

[0173] Step S206, evaluating the quality of the restored image and performing necessary optimization;

[0174] In some embodiments, the image quality is evaluated using evaluation indicators (eg, structural similarity index, peak signal-to-noise ratio, etc.), and the image quality is further optimized through an adaptive adjustment algorithm to ensure that the image meets the description of the prompt word.

[0175] Step S207, recording the user's feedback and prompt word features to provide a reference for future image generation and restoration processes.

[0176] In some embodiments, user feedback on the final image is collected, and the restoration criteria and prompt word features are recorded. The feedback and feature data are stored in a database for reference and optimization of future image generation processes.

[0177] In the embodiment of the present application, the accuracy of image generation can be significantly improved through multi-dimensional evaluation and restoration methods to meet more complex user needs. The multi-style transfer algorithm can achieve more flexible and refined style adjustment. Allowing users to customize restoration standards according to their personal preferences improves user participation and satisfaction.

[0178] With the development of artificial intelligence technology, the technology of generating images based on text has made significant progress. In related technologies, the corresponding images are generated through natural language descriptions, which shows great potential in applications such as artistic creation, virtual reality, and advertising design. However, how to effectively evaluate the quality of these generated images, especially the comprehensive evaluation in multiple dimensions, is still a technical problem that has not been fully solved.

[0179] There are mainly the following methods for evaluating the image generation quality in related technologies:

[0180] (1) Manual evaluation: Manual evaluation is the most direct and traditional method for image quality evaluation. It is usually performed by experts or ordinary users to score the generated images. This method is highly subjective, inefficient, and difficult to form a unified evaluation standard. Different evaluators often give different evaluation results due to differences in aesthetics, professional knowledge, and experience background, resulting in inconsistent evaluation results.

[0181] (2) Single-dimensional automated evaluation: Some existing automated evaluation methods are mainly based on the similarity calculation between images and text descriptions. Such methods usually use natural language processing techniques, such as cosine similarity and Jaccard similarity, to match the text description generated from the image with the input text. However, such methods only focus on the correlation between the image content and the text description, while ignoring the quality performance of the image in other dimensions, such as color matching, composition beauty, detail restoration, etc.

[0182] (2) Evaluation based on Generative Adversarial Networks (GAN): In recent years, GAN has been widely used in image generation tasks, and some GAN-based image quality evaluation methods have gradually been developed. These methods train a discriminant network to distinguish between generated images and real images, and then give an evaluation score. However, when evaluating the authenticity and quality of an image, these methods often ignore the match between the image and the text description, and cannot fully reflect the overall quality of image generation.

[0183] The present application provides a multi-dimensional image evaluation method based on text-generated images. The generated image is converted into a text description through an image-text conversion model (img2text), and the similarity is calculated with the original text input by the user to obtain a text similarity score; the prompt word input by the user is parsed to extract multiple attribute fields that need to be evaluated (for example, color, composition, subject, etc.); the description under each attribute field is clustered to generate 9 categories, and on this basis, the corresponding fields of the generated image are classified, and the number of categories is 10; the score of each field is calculated by the accuracy of the classification, and the weighted sum of all field scores is obtained to obtain the final field comprehensive score; the final image score is the sum of the text similarity score and the field comprehensive score. This method is used to perform comprehensive, automated, and multi-dimensional quality evaluation of images generated based on text, and is applicable to multiple fields such as image generation, artistic creation, and advertising design.

[0184] This application provides a multi-dimensional image evaluation method based on text-generated images, such as Figure 3 As shown, the following steps may be included:

[0185] Step S301, generating image text description;

[0186] In some embodiments, a pre-trained img2text model is used to convert the generated image into a natural language text description. The text description captures the main elements and features in the image and provides a basis for subsequent similarity calculations.

[0187] Step S302, text similarity calculation;

[0188] In some embodiments, the text description generated by img2text is compared with the original text input by the user, and a text similarity score is calculated using a natural language processing model. For example, a Transformer BERT model can be used to calculate cosine similarity through the encoded vector.

[0189] Step S303, field analysis and clustering;

[0190] In some embodiments, the text prompt words input by the user are parsed to extract multiple important fields. These fields can be explicitly specified (e.g., art style, lighting, material, art medium, clarity, etc.), or implicit fields automatically extracted by natural language processing technology. All values ​​of each field are clustered and divided into 9 categories, representing the main features under the field. The clustering algorithm uses K-means, and the number of categories is specified as 9.

[0191] Step S304, field score calculation;

[0192] In some embodiments, the generated images are classified according to the categories corresponding to each field, and the classification accuracy is calculated. The number of categories for this classification task is 10, including 9 clustering categories and 1 user-specific description category. The classification accuracy is the score of the field.

[0193] Step S305, calculating the field comprehensive score;

[0194] In some implementations, a weighted sum is performed on the scores of all fields, and the weights can be adjusted according to the importance of the fields.

[0195] Step S306: Calculate the final score of the image.

[0196] In some implementations, the text similarity score is added to the field comprehensive score to obtain the final score of the image. This score can be used to quantify the overall quality of the generated image.

[0197] In the embodiment of the present application, a natural language processing model based on deep learning (such as BERT, GPT, etc.) is used to calculate the text similarity. Through a deep understanding of the text semantics, the algorithm can capture the similarity between the generated image and the input text at the semantic level. The field description is clustered using the K-means clustering algorithm, and the various possibilities under the field are divided into 9 categories. These categories can effectively represent the main features of the field and contribute to subsequent classification tasks. For each attribute field, the bootstrap language image pre-training (Bootstrapping Language-Image Pre-training, BLIP) algorithm is used to classify the generated image. By classifying in 10 categories, the higher the accuracy, the more the performance of the field is in line with expectations. Not only the similarity between the image and the text description is considered, but also the multi-dimensional image feature evaluation is combined to make the evaluation results more comprehensive. The entire evaluation process is fully automated, without manual intervention, reducing the influence of subjective factors. Through the field clustering and classification method, the evaluation process is highly standardized and repeatable.

[0198] The present application embodiment provides a system architecture for image processing, such as Figure 4 As shown, the system architecture includes an interface layer 41, a business logic layer 42, a model layer 43 and a data layer 44, wherein the interface layer 41 is used to input prompt words and images, provide custom restoration standards, display images before and after restoration, and obtain user feedback on images. The business logic layer 42 includes an image quality assessment module 421 and an image restoration module 422; the image quality assessment module 421 is used for image text description generation, text similarity calculation, image feature extraction, and score calculation and evaluation; the image restoration module 422 is used for style prompt word analysis, style conversion algorithm application, and custom standard application; the image quality assessment module and the image restoration module work together to ensure that the system can generate high-quality images according to the prompt words input by the user, and perform restoration and optimization. The model layer 43 includes model training, model fine-tuning, model update, model deployment, img2text model, text similarity model, image feature extraction model, and style migration model. The data layer 44 includes prompt words and images, image quality assessment results, image restoration history, and user feedback records.

[0199] In the embodiments of the present application, (1) automatic image quality assessment based on multi-dimensional features is specifically: combining multi-dimensional features, such as color consistency, composition rationality, and the fit between scene and text, to conduct a comprehensive assessment. By extracting low-level and high-level features of the image through a deep learning model, the system can evaluate all aspects of the image, not only limited to technical indicators, but also including its artistic expression and semantic rationality. The results of such an assessment can better reflect the actual quality of the image, especially in the generation scenario of image and text matching. This automatic assessment based on multi-dimensional features allows the system to have a more comprehensive and accurate understanding of the image. Multi-dimensional similarity analysis of images and texts: It utilizes the combination of natural language processing and computer vision technology to generate images through text descriptions and conduct multi-dimensional similarity assessments on them. In traditional image quality assessments, only a single dimension of the image (such as clarity and resolution) is often considered. However, the present application introduces a text-image comparison mechanism to evaluate the matching degree between the generated image and the original text description. The specific evaluation dimensions include color matching, composition, and text in the image. Figure 1Consistency, clarity of the main object, semantic fit of the scene, etc. This multi-dimensional analysis can comprehensively evaluate whether the image accurately reflects the content of the text description, avoiding the limitations of a single-dimensional evaluation. Multi-level feature extraction and quantitative evaluation: In order to achieve a more detailed image quality assessment, this application automatically extracts multi-level features of the image based on deep learning technology. Low-level features include the color, texture, edge clarity, etc. of the image, while high-level features cover the scene semantics of the image, the semantic information of the main object, and the complex structure. By quantifying these features, this application constructs a scoring mechanism that can score each dimension and combine the scores of each dimension according to preset weights to form an overall quality score. This method not only makes the evaluation more comprehensive, but also improves the accuracy of the evaluation.

[0200] (2) Automated image restoration based on multi-dimensional features, specifically: not only traditional image enhancement (such as noise reduction and clarity optimization) is performed, but also restoration is performed in dimensions such as lighting adjustment, color optimization, and style consistency based on techniques such as style transfer. Unlike traditional restoration that only corrects technical defects, the system can intelligently identify the artistic style of the image and automatically adjust the image style according to preset restoration standards or text descriptions. For example, in advertising design, the restored image must not only be clear, but also conform to the overall brand style or theme. This multi-dimensional restoration greatly enhances the visual effect and artistic expression of the image. Multi-dimensional image restoration: Using techniques such as style transfer and image enhancement, the image is automatically restored in multiple dimensions such as lighting, color matching, and artistic style. Specifically, the restoration process is not limited to basic enhancement of the image (such as denoising, clarity enhancement, etc.), but more importantly, the image style, color tone, light and shadow effects, etc. are finely adjusted through deep learning models and style transfer algorithms. For example, if the image is too dark or has color deviations, the system can automatically optimize the brightness and tone to ensure that the repaired image not only has better visual effects, but is also more consistent with the artistic style of the original text. It is suitable for scenes where high-quality images need to be generated in artistic creation and advertising design. User-defined restoration standards and weight distribution: During the image restoration process, this application allows users to customize restoration standards according to personal preferences, such as adjusting the brightness, color saturation, and clarity of the image. More importantly, users can flexibly control the priority of each element in the restoration process by setting weights in different dimensions. For example, some users may pay more attention to the accuracy of the image composition, while other users may be more concerned about the color expression of the image. Through this weight distribution mechanism, users can customize the restoration effect to ensure that the generated image is more in line with personal aesthetic needs and enhance the user's sense of participation and control.

[0201] (3) Automated and standardized evaluation and repair process, specifically: By using deep learning technology, the system can automatically perform the entire process from image generation to quality assessment and then to style repair. Without human intervention, the system automatically evaluates and repairs images according to preset standards to ensure quality consistency and efficiency when processing images in batches. This not only reduces human operation errors, but also improves processing speed, which is particularly suitable for scenarios that require rapid processing of a large number of images, such as the generation and repair of e-commerce product images. Fully automated evaluation and repair process: The entire process from image generation to quality assessment and then to style repair is fully automated. This application is based on deep learning technology and natural language processing technology to achieve highly automated process management. After entering text, the system can automatically generate images, and then perform quality assessment and style repair in turn. The entire process does not require human intervention, reducing human subjective influence. This can not only maintain consistency and standardization of evaluation and repair in large-scale image generation tasks, but also significantly improve operational efficiency. It is suitable for application scenarios that require batch processing of images, such as advertising generation, creative design, etc.) Efficiency and repeatability: Due to the use of a fully automated processing process, this method can maintain efficiency and repeatability in different image generation and repair tasks. Whether it is generating a series of images with the same style or applying the same evaluation and restoration standards in different tasks, the automated evaluation and restoration process of this application can be effectively executed. By introducing a standardized quality evaluation and restoration mechanism, it not only greatly reduces the errors caused by human operation, but also ensures high efficiency and accuracy when applied in different scenarios. This makes the solution particularly suitable for industries that need to process massive amounts of images, such as e-commerce, advertising design, game development and other fields.

[0202] (4) User-defined optimization and enhanced interactive experience: users can adjust the restoration criteria according to their personal preferences, such as brightness, color saturation or style expression, and the system can remember these preferences for future restoration tasks. Through this customization and preference recording, users can control the restoration effect more finely and ensure that the generated image is more in line with personalized needs. In addition, the system also integrates a user feedback mechanism to continuously optimize the restoration algorithm through user feedback, and ultimately provide a more accurate and personalized image restoration experience. This user-involved design not only enhances the adaptability of the system, but also makes image processing more flexible and intelligent. User-defined restoration preference recording: During the restoration process, the system can record the user's personalized preferences and automatically apply these preferences in subsequent image processing. For example, when a user often adjusts specific lighting effects or improves specific color expressions when restoring images, the system can automatically learn these preferences and adaptively adjust parameters in subsequent restorations, thereby reducing the user's repeated adjustment operations. This mechanism greatly improves the user experience and makes the restoration process more personalized and intelligent, especially for designers and creators who frequently use the system. Sustainable optimization user feedback mechanism: The system not only records the user's personalized preferences, but also continuously optimizes the algorithm through user feedback. After each restoration is completed, the user can rate or give comments on the restoration effect of the image. The system collects these feedbacks to further train and optimize its restoration model so that it can generate and restore images that better meet user expectations in the future. This sustainable optimization mechanism enables the system to continuously improve itself, ultimately providing users with more efficient and accurate services, and improving the overall experience and quality of image restoration.

[0203] It should be noted that in the embodiment of the present application, if the above method is implemented in the form of a software function module and sold or used as an independent product, it can also be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the embodiment of the present application is essentially or the part that contributes to the relevant technology can be embodied in the form of a software product, which is stored in a storage medium, including a number of instructions to enable an electronic device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the methods described in each embodiment of the present application. The aforementioned storage medium includes: various media that can store program codes, such as a U disk, a mobile hard disk, a read-only memory (ROM), a disk or an optical disk. In this way, the embodiment of the present application is not limited to any specific combination of hardware and software.

[0204] An embodiment of the present application provides an electronic device, comprising a processor and at least one processing model capable of running on the processor, wherein the processing model can be called by a target application to execute at least one of the following: obtaining first input data and a response result to the first input data; obtaining evaluation information for the response result; performing corresponding processing on the response result with a target processing strategy based on the evaluation information to output a target file; wherein at least one characteristic parameter of the target file is better than a corresponding parameter of the response result, and the target file is more in line with the target user's intention relative to the response result.

[0205] The embodiment of the present application provides a computer-readable storage medium on which a computer program is stored, and the computer program implements the above method when executed by a processor. The computer-readable storage medium may be transient or non-transient.

[0206] The embodiment of the present application provides a computer program product, which includes a non-transitory computer-readable storage medium storing a computer program, and when the computer program is read and executed by a computer, some or all of the steps in the above method are implemented. The computer program product can be implemented specifically by hardware, software or a combination thereof. In an optional embodiment, the computer program product is embodied as a computer storage medium, and in another optional embodiment, the computer program product is embodied as a software product, such as a software development kit (Software Development Kit, SDK) and the like.

[0207] It should be noted that the embodiment of the present application provides a hardware entity of an electronic device, such as Figure 5 As shown, the hardware entity of the electronic device 400 includes: a processor 410 and at least one processing model 420 capable of running on the processor 410, and the processing model 420 can be called by the target application to perform at least one of the following: obtaining first input data and a response result to the first input data; obtaining evaluation information for the response result; performing corresponding processing on the response result with a target processing strategy based on the evaluation information to output a target file; wherein at least one characteristic parameter of the target file is better than the corresponding parameter of the response result, and the target file is more in line with the target user's intention relative to the response result.

[0208] The present application embodiment provides a data processing device, such as Figure 6 As shown, including:

[0209] A first obtaining module 510 is used to obtain first input data and a response result to the first input data;

[0210] A second obtaining module 520 is used to obtain evaluation information of the response result;

[0211] A processing module 530 is used to process the response result according to the evaluation information using a target processing strategy to output a target file;

[0212] Among them, at least one characteristic parameter of the target file is better than the corresponding parameter of the response result, and the target file is more in line with the target user's intention than the response result.

[0213] It should be noted here that the description of the above storage medium and device embodiments is similar to the description of the above method embodiments, and has similar beneficial effects as the method embodiments. For technical details not disclosed in the storage medium and device embodiments of this application, please refer to the description of the method embodiments of this application for understanding.

[0214] It should be understood that "one embodiment" or "an embodiment" mentioned throughout the specification means that specific features, structures or characteristics related to the embodiment are included in at least one embodiment of the present application. Therefore, "in one embodiment" or "in an embodiment" appearing throughout the specification does not necessarily refer to the same embodiment. In addition, these specific features, structures or characteristics can be combined in one or more embodiments in any suitable manner. It should be understood that in various embodiments of the present application, the size of the serial number of each step / process mentioned above does not mean the order of execution, and the execution order of each step / process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiment of the present application. The serial numbers of the embodiments of the present application mentioned above are for description only and do not represent the advantages and disadvantages of the embodiments.

[0215] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the existence of other identical elements in the process, method, article or device including the element.

[0216] In the several embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as: multiple units or components can be combined, or can be integrated into another system, or some features can be ignored or not executed. In addition, the coupling, direct coupling, or communication connection between the components shown or discussed can be through some interfaces, and the indirect coupling or communication connection of the devices or units can be electrical, mechanical or other forms.

[0217] The units described above as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units; they may be located in one place or distributed on multiple network units; some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0218] In addition, all functional units in the embodiments of the present application may be integrated into one processing unit, or each unit may be a separate unit, or two or more units may be integrated into one unit; the above-mentioned integrated units may be implemented in the form of hardware or in the form of hardware plus software functional units.

[0219] A person skilled in the art can understand that all or part of the steps of implementing the above method embodiment can be completed by hardware related to program instructions, and the aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it executes the steps of the above method embodiment; and the aforementioned storage medium includes: mobile storage devices, read-only memories (ROM), magnetic disks or optical disks, etc., various media that can store program codes.

[0220] Alternatively, if the above-mentioned integrated unit of the present application is implemented in the form of a software function module and sold or used as an independent product, it can also be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application can essentially or in other words, the part that contributes to the relevant technology can be embodied in the form of a software product, which is stored in a storage medium and includes a number of instructions for an electronic device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the methods described in each embodiment of the present application. The aforementioned storage medium includes: various media that can store program codes, such as mobile storage devices, ROMs, magnetic disks, or optical disks.

[0221] The above is only an implementation method of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application.

Claims

1. A data processing method, comprising: Obtaining first input data and a response result to the first input data; Obtaining evaluation information of the response result; Based on the evaluation information, the response result is processed accordingly with a target processing strategy to output a target file; Among them, at least one characteristic parameter of the target file is better than the corresponding parameter of the response result, and the target file is more in line with the target user's intention than the response result.

2. The method according to claim 1, wherein: Obtaining evaluation information of the response result, including at least one of the following: Performing quality assessment on the response result to obtain the evaluation information; Obtaining feedback information of the target user on the response result, and obtaining the evaluation information based on the feedback information; Obtaining first historical evaluation data for a first processing device and / or a first processing model, and obtaining the evaluation information based on the first historical evaluation data, wherein the first processing model is a model that generates the response result in response to the first input data, and the first processing device is a device that obtains the response result using the first processing model; Second historical evaluation data for a target application is obtained, and the evaluation information is obtained based on the second historical evaluation data, wherein the target application is an application that responds to the first input data to output the response result.

3. The method according to claim 2, wherein: Performing a quality assessment on the response result includes at least one of the following: Performing recognition processing on the response result to obtain first recognition data, and obtaining the evaluation information based at least on a similarity between the first input data and the first recognition data; The user intention represented by the first input data is identified, and the quality of the response result is evaluated with a corresponding evaluation strategy based on the user intention, where different user intentions correspond to different evaluation strategies.

4. The method according to claim 3, wherein: Performing a quality assessment on the response result includes at least one of the following: In the case where the first input data is text data, a similarity score between the first input data and the first recognition data is calculated using a first processing model, and the similarity score is used as the evaluation information; Calculating a similarity score between the first input data and the first recognition data using a second processing model, performing accuracy calculation on key fields included in the first recognition data, and weighting the similarity score and the accuracy calculation results of the key fields based on corresponding weights to obtain the evaluation information; In the case where the response result is image data, the third processing model is used to extract image features of multiple dimensions of the image data, the accuracy score of each dimension is calculated based on the image features, and the accuracy scores of the various dimensions are weighted to obtain the evaluation information.

5. The method according to claim 1, wherein: Based on the evaluation information, the response result is processed accordingly with a target processing strategy, including at least one of the following: When the evaluation information indicates that the response result meets the user intention indicated by the first input data, the response result is directly output as the target file, or the response result is generated to obtain the target file; When the evaluation information indicates that the response result does not conform to the user intention indicated by the first input data, repairing the response result to obtain the target file; If the evaluation information indicates that the response result does not conform to the user intention indicated by the first input data, the first input data is regenerated and then replaced with the response result as the target file, or the response result is compared with the generated result of the first input data to output the target file; In the case where the evaluation information is negative feedback information provided by the target user, the first input data and the negative feedback information are generated and processed again to replace the response result as the target file, or the characteristic parameters of the response result are adjusted based on the negative feedback information to output the target file.

6. The method according to claim 5, wherein: Performing repair processing on the response result includes: Acquire multi-dimensional features of the response result; Acquire target dimension features and weight information thereof based on the first input data; The multi-dimensional features of the response result are adjusted based on the target dimensional features and their weight information to obtain the target file.

7. The method according to claim 5, wherein: The step of repairing the response result further includes at least one of the following: Obtaining operation behavior data of the target user for the response result, and updating characteristic parameters of the response result based on the operation behavior data to obtain the target file; generating the target file based on the obtained selection operation of the recommended data for the target data in the response result and / or the editing operation for the target data; generating the target file based on the configuration operation obtained for the configuration options in the configuration window of the response result, wherein the configuration options are used to update the feature data in the response result; Context data associated with the response result is obtained, and the response result is updated using the context data to obtain the target file.

8. The method according to claim 1, further comprising at least one of the following: Obtaining an intermediate file obtained after corresponding processing of the response result, re-evaluating the intermediate file, and performing corresponding processing on the intermediate file based on the obtained re-evaluation result to output the target file; The target user's evaluation information on the response result is recorded to optimize the processing model for processing the first input data based on the evaluation information.

9. The method according to claim 1, wherein: Obtaining first input data and a response result to the first input data includes at least one of the following: Obtaining a first processing instruction for the first input data, and performing generation processing on the first input data using a first processing model matching the first processing instruction to obtain the response result; generating and processing the first target data in the first input data based on the type of the first input data and the represented user intention to obtain the response result; The first input data and context data of its source application are integrated to obtain second input data, and the second input data is generatively processed using a generative model to obtain the response result.

10. An electronic device comprising a processor and at least one processing model capable of running on the processor, wherein the processing model can be called by a target application to perform at least one of the following: Obtaining first input data and a response result to the first input data; Obtaining evaluation information of the response result; Based on the evaluation information, the response result is processed accordingly with a target processing strategy to output a target file; in, At least one characteristic parameter of the target file is better than a corresponding parameter of the response result, and the target file is more consistent with the target user's intention than the response result.

Images