Data analysis methods, systems and storage media

Abstract

This specification provides a data analysis method, system, and storage medium. The data analysis method includes: the data analysis system first acquires a first image set and a second image set. The first image set includes multiple first images labeled with a first tag, and the second image set includes multiple second images labeled with a second tag. The first and second tags are image evaluation metrics for the same target dimension, and the image evaluation metric for the first tag is superior to that for the second tag. Then, the data analysis system uses a pre-defined multimodal large language model to analyze the first and second image sets in the target dimension, obtaining data analysis results. The data analysis results include at least the quantitative features corresponding to the first tags.

Description

Technical Field

[0001] This specification relates to the field of artificial intelligence technology, and in particular to a data analysis method, system and storage medium. Background Technology

[0002] With the rapid development of artificial intelligence technology, multimodal large language models have been widely used to process image-related tasks, such as image scoring, image content review, and image generation.

[0003] Currently, to improve the performance of large language models on target tasks, corresponding evaluation criteria are typically introduced into the prompt words, or these criteria are provided to the large language model as reference content from an external knowledge base. When performing the target task, the large language model needs to self-constrain or adjust itself based on the given evaluation criteria. These evaluation criteria are defined and described by humans using natural language according to the task requirements.

[0004] However, such manually described evaluation criteria are vague qualitative descriptions. When large language models use these vague evaluation criteria, they may misinterpret or misunderstand them, leading to deviations in the task processing and affecting user experience and the credibility of task processing.

[0005] The information in the background section is merely information known only to the inventor and does not imply that such information had entered the public domain before the date of this application, nor does it imply that it can be considered prior art in this disclosure. Summary of the Invention

[0006] This specification provides a data analysis method, system, and storage medium that can analyze and generate a superior image in the target dimension by using two image sets with different labels in the same target dimension.

[0007] To achieve the above objectives, the embodiments in this specification adopt the following technical solutions: In a first aspect, this specification provides a data analysis method, the method comprising: acquiring a first image set and a second image set, the first image set including multiple first images labeled with a first tag, and the second image set including multiple second images labeled with a second tag, wherein the first tag and the second tag are image evaluation metrics of the same target dimension, and the image evaluation metrics of the first tag are superior to the image evaluation metrics of the second tag; and performing target dimension analysis on the first image set and the second image set using a preset multimodal large language model to obtain data analysis results, wherein the data analysis results include at least the quantitative features corresponding to the first tags.

[0008] In some embodiments, the step of using a preset multimodal large language model to perform target dimension analysis on the first image set and the second image set to obtain data analysis results includes: generating a first prompt instruction based on the first image set and the second image set, wherein the first prompt instruction is used to guide the preset multimodal large language model to perform target dimension inference based on the first image set and the second image set, to obtain quantized common features among multiple first images and quantized difference features among multiple first images and multiple second images, and to determine the quantized features corresponding to the first label based on the quantized common features and the quantized difference features, thereby generating the data analysis results; and inputting the first prompt instruction into the preset multimodal large language model to obtain the data analysis results output by the preset multimodal large language model.

[0009] In some embodiments, the data analysis result further includes a second prompt instruction; correspondingly, the first prompt instruction is also used to guide the preset multimodal large language model to generate a second prompt instruction based on the quantization features corresponding to the first label, wherein the second prompt instruction includes instructions to guide the preset multimodal large language model to perform a target task using the quantization features corresponding to the first label.

[0010] In some embodiments, the method further includes: inputting a third prompt instruction to the preset multimodal large language model, the third prompt instruction including: a target task and quantized features corresponding to the first label, the third prompt instruction being used to guide the preset multimodal large language model to generate a fourth prompt instruction corresponding to the target task based on the quantized features corresponding to the first label, the fourth prompt instruction being used to guide the preset multimodal large language model to execute the target task.

[0011] In some embodiments, after obtaining the data analysis results output by the preset multimodal large language model, the method further includes: verifying the accuracy dimension of the data analysis results to obtain an accuracy verification result, wherein the accuracy verification result is used to characterize whether the accuracy of the quantization feature corresponding to the first label meets a first preset requirement; and when the accuracy verification result indicates that the quantization feature corresponding to the first label does not meet the first preset requirement, updating the first prompt instruction, and obtaining the data analysis results using the updated first prompt instruction.

[0012] In some embodiments, obtaining the first image set and the second image set includes: displaying a first interactive interface and a second interactive interface using a target page; receiving images uploaded by users based on the first interactive interface and the second interactive interface; and determining that the image received based on the first interactive interface is the first image, determining that the image received based on the second interactive interface is the second image, thereby obtaining the first image set and the second image set.

[0013] In some embodiments, the target page also displays a third interactive interface, and the method further includes: receiving reference data based on the third interactive interface, the reference data including at least one of the following: reference information of the first image, reference information of the second image, domain information, and feature dimension information.

[0014] In some embodiments, the step of using a preset multimodal large language model to analyze the first image set and the second image set in the target dimension to obtain data analysis results includes: generating a fifth prompt instruction based on the first image set, the second image set, and the reference data; the fifth prompt instruction is used to guide the preset multimodal large language model to perform image quality dimension inference based on the first image set and the second image set, referencing the reference data, to obtain quantized common features among multiple first images and quantized difference features among multiple first images and multiple second images, and to determine the quantized features corresponding to the first label based on the quantized common features and the quantized difference features, thereby generating the data analysis results; and inputting the fifth prompt instruction into the preset multimodal large language model to obtain the data analysis results output by the preset multimodal large language model.

[0015] In some embodiments, before performing target dimension analysis on the first image set and the second image set using a preset multimodal large language model, the method further includes: testing the feature recognition accuracy dimension of the general multimodal large language model based on a test sample set to obtain an evaluation index of the general multimodal large language model in the feature recognition accuracy dimension; when the evaluation index meets a second preset requirement, determining the general multimodal large language model as the preset multimodal large language model; and when the evaluation index does not meet the second preset requirement, fine-tuning the general multimodal large language model so that the evaluation index of the fine-tuned general multimodal large language model in the feature recognition accuracy dimension meets the second preset requirement, and determining the fine-tuned general multimodal large language model as the preset multimodal large language model.

[0016] Secondly, this specification also provides a data analysis system, comprising: at least one storage medium storing at least one instruction set for identifying quantization features corresponding to a first tag; and at least one processor communicatively connected to the at least one storage medium, wherein, when the data analysis system is running, the at least one processor reads the at least one instruction set and implements the method provided in the first aspect according to the instructions of the at least one instruction set.

[0017] Thirdly, this specification also provides a computer-readable non-volatile storage medium, wherein the computer-readable non-volatile storage medium stores at least one instruction set, which, when executed by at least one processor, implements the method provided in the first aspect.

[0018] Other functionalities of the data analysis methods, systems, and storage media provided in this specification are partially listed in the following description. The inventive aspects of the data analysis methods, systems, and storage media provided in this specification can be fully understood through practice or use of the methods, apparatus, and combinations described in the detailed examples below. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of this specification, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this specification. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 A schematic diagram illustrating an application scenario of the data analysis method provided according to embodiments of this specification is shown. Figure 2 A hardware structure diagram of a computing system provided according to an embodiment of this specification is shown; Figure 3 A flowchart of a data analysis method provided according to an embodiment of this specification is shown; Figure 4 A schematic diagram of a target page provided according to an embodiment of this specification is shown; Figure 5 A schematic diagram illustrating a method for obtaining data analysis results based on a first image set and a second image set, according to an embodiment of this specification, is shown. Figure 6 A schematic diagram illustrating a method for obtaining data analysis results based on a first image set, a second image set, and reference data, according to an embodiment of this specification, is shown. Figure 7 A schematic diagram of another target page provided according to an embodiment of this specification is shown; and Figure 8 A schematic diagram is shown illustrating an embodiment of this specification that generates prompt instructions based on data analysis results. Detailed Implementation

[0021] The following description provides specific application scenarios and requirements for this specification, intended to enable those skilled in the art to make and use the contents of this specification. Various partial modifications to the disclosed embodiments will be apparent to those skilled in the art, and the general principles defined herein can be applied to other embodiments and applications without departing from the spirit and scope of this specification. Therefore, this specification is not limited to the embodiments shown, but rather to the widest scope consistent with the claims.

[0022] The terminology used herein is for the purpose of describing particular exemplary embodiments only and is not restrictive. For example, unless the context clearly indicates otherwise, the singular forms “a,” “an,” and “the” used herein may also include the plural forms. When used in this specification, the terms “comprising,” “including,” and / or “containing” mean that the associated integers, steps, operations, elements, and / or components are present, but do not exclude the presence of one or more other features, integers, steps, operations, elements, components, and / or groups, or that other features, integers, steps, operations, elements, components, and / or groups may be added to the system / method.

[0023] Considering the following description, these and other features of this specification, as well as the operation and function of the related components of the structure, and the economy of assembly and manufacture of the parts, can be significantly improved. All of these form part of this specification with reference to the accompanying drawings. However, it should be clearly understood that the drawings are for illustrative and descriptive purposes only and are not intended to limit the scope of this specification. It should also be understood that the drawings are not drawn to scale.

[0024] The flowcharts used in this specification illustrate operations implemented according to some embodiments of this specification. It should be clearly understood that the operations in the flowcharts may not be implemented in a sequential order. Instead, the operations may be implemented in reverse order or simultaneously. Furthermore, one or more additional operations may be added to the flowcharts. One or more operations may be removed from the flowcharts.

[0025] In some embodiments, a Large Language Model (LLM) can also be simply referred to as a large model. A LLM is a natural language processing model based on deep learning techniques, typically with billions to hundreds of billions or even more parameters, possessing powerful language understanding and generation capabilities. LLMs can employ the Transformer architecture or its variants (such as GPT, BERT, etc.), which utilizes an attention mechanism to globally model sequential data, efficiently handling long-distance dependencies and thus performing exceptionally well in natural language tasks. LLMs learn the statistical features and semantic relationships of language through pre-training on large-scale corpora, giving them excellent generalization capabilities. The core capabilities of LLMs include, but are not limited to: understanding contextual semantics, generating coherent and grammatically correct text, performing logical reasoning, and handling multi-task scenarios. Their usage typically includes two modes: direct inference and fine-tuning. In direct inference mode, the user guides the LLM to generate specific outputs by designing prompts. Prompts can be task descriptions or instructions in text form, used to stimulate the LLM's semantic understanding and generation capabilities. In fine-tuning mode, large language models are further trained on small-scale datasets within a specific domain to optimize their performance on specific tasks. The powerful generalization capabilities and flexibility of large language models make them an important tool in the field of artificial intelligence, providing efficient and accurate solutions for automated text generation and understanding.

[0026] In this specification, large language models can possess the ability to understand and generate data from other modalities (such as visual and audio data). In this case, large language models can also be called multimodal large language models (MLLMs). MLLMs provide a richer and more natural interactive experience by integrating multiple types of input and output, such as text, images, and sound. The core advantage of MLLMs lies in their ability to process and understand information from different modalities and fuse this information to accomplish complex tasks. For example, the Vision Language Model (VLM) discussed in this specification is a branch of MLLMs; a VLM can analyze an image and generate descriptive text. In other examples, MLLMs can also generate corresponding images or videos based on text descriptions. This cross-modal understanding and generation capability makes MLLMs widely applicable across multiple fields.

[0027] It should be noted that the key technologies of large language models can be found in the detailed description in the paper "A Survey of Large Language Models" (paper number: arXiv:2303.18223v16, published on March 11, 2025, public link: https: / / doi.org / 10.48550 / arXiv.2303.18223), and will not be repeated here.

[0028] The following section introduces the application scenarios of this manual.

[0029] The technical solutions provided in this specification are applicable to scenarios where multimodal large language models are used to process image-related tasks. These tasks include parsing, analyzing, judging, evaluating, or generating image data to output results that meet task requirements. Task-related evaluation criteria can be introduced into this process to improve processing effectiveness. Using evaluation criteria as constraints allows multimodal large language models to more accurately perform various image-related tasks such as image scoring, image content review, and image generation.

[0030] However, when multimodal large language models process image-related tasks, the evaluation criteria introduced are often defined and described manually using natural language based on task requirements. Because manually described evaluation criteria involve significant subjective judgment and cannot be accurately quantified, using these vague criteria to constrain the image-related task processing of multimodal large language models can lead to misinterpretations or biases in the model's understanding of the evaluation criteria. Alternatively, due to insufficient human experience, the defined evaluation criteria may be inaccurate. These factors can all cause deviations in the task processing results of multimodal large language models in image-related tasks, resulting in low reliability of the processing results and poor user experience.

[0031] For example, in the field of image evaluation, the quality of an image is judged based on the evaluator's aesthetic ability. However, when evaluators use multimodal large language models to perform image evaluation tasks, the aesthetic standards described by natural language are difficult to quantify, and cannot provide accurate and clear constraints or guidance for the multimodal large language models.

[0032] Furthermore, when performing image evaluation tasks, multimodal large language models may misinterpret or misunderstand the aesthetic standards of natural language descriptions, leading to discrepancies between the evaluation results output by the multimodal large language model and the evaluator's judgment. This severely impacts the user experience and makes it difficult for users to trust the task processing results output by the multimodal large language model.

[0033] To this end, this specification provides a data analysis method that can be executed by a data analysis system. The method includes: the data analysis system first acquiring a first image set and a second image set. The first image set includes multiple first images labeled with a first tag, and the second image set includes multiple second images labeled with a second tag. The first tag and the second tag are image evaluation metrics for the same target dimension, and the image evaluation metric for the first tag is superior to the image evaluation metric for the second tag. Then, the data analysis system uses a preset multimodal large language model to perform target dimension analysis on the first image set and the second image set to obtain data analysis results. The data analysis results include at least the quantitative features corresponding to the first tags.

[0034] In the solution provided in this manual, the data analysis system first obtains a first image set corresponding to a first label with a better evaluation metric under the same target dimension, and a second image set corresponding to a second label with a relatively worse evaluation metric. The data analysis system then uses a pre-set multimodal large language model to perform target dimension reasoning and analysis on the two image sets, mining and obtaining the quantitative features of the first label. In this way, the solution provided in this manual achieves automatic analysis to obtain the quantitative features corresponding to the first label based on image sets with different evaluation metrics under the same target dimension using a pre-set multimodal large language model. Using the obtained quantitative features as evaluation criteria enables the multimodal large language model to have clear, accurate, and quantifiable evaluation criteria when processing image-related tasks. This effectively improves the problem of biased task processing results caused by ambiguous or low-quality evaluation criteria, enhances the accuracy and reliability of the multimodal large language model in processing image-related tasks, and improves the user experience.

[0035] Figure 1 A schematic diagram illustrating an application scenario of the data analysis method provided according to embodiments of this specification is shown. For example... Figure 1 As shown, the application scenario 100 may include a data analysis system 11, a terminal device 12, and a multimodal large language model 13.

[0036] refer to Figure 1The data analysis system 11 can be deployed on a device or device cluster with data processing capabilities. For example, the data analysis system 11 can be deployed on a server, server cluster, cloud server, or other such device. In this case, the physical device corresponding to the data analysis system 11 can store data or instructions for executing the data analysis methods described in this specification, and can execute or be used to execute the data or instructions. In some embodiments, the data analysis system 11 may include a hardware device with data information processing capabilities and the necessary programs required to drive the hardware device to work. The multimodal large language model 13 can be deployed on the same physical device as the data analysis system 11, or on a different device or device cluster than the data analysis system 11; this specification does not impose any limitations on this.

[0037] As an example, terminal device 12 may be an electronic device capable of interacting with data analysis system 11. For example, terminal device 12 may include mobile devices, tablets, laptops, built-in devices in motor vehicles, or similar content, or any combination thereof. In some embodiments, the mobile device may include wearable devices, camera devices, smart mobile devices, virtual reality devices, augmented reality devices, or similar devices, or any combination thereof. In some embodiments, wearable devices include smartwatches, smart bracelets, smart glasses, etc. In some embodiments, the smart mobile device may include smartphones, personal digital assistants, gaming devices, navigation devices, etc., or any combination thereof. In some embodiments, the virtual reality device or augmented reality device may include head-mounted displays, virtual reality headsets, virtual reality glasses, virtual reality patches, augmented reality headsets, augmented reality glasses, augmented reality patches, or similar content, or any combination thereof.

[0038] In some embodiments, the terminal device 12 may have one or more applications (APPs) installed. The APPs provide the ability or interface to access the data analysis system 11. The APPs include, but are not limited to: web browser APPs, search APPs, chat APPs, shopping APPs, video APPs, financial management APPs, instant messaging tools, email clients, social media platform software, etc. In some embodiments, the terminal device 12 may have a target APP installed. The target APP is the application corresponding to the data analysis system 11.

[0039] In some embodiments, the target app can respond to user operations by receiving the tags of image data from an image dataset uploaded by the user. Then, it sends the image data and corresponding tags to a data analysis system 11. Based on the received image data and corresponding tags, the data analysis system 11 calls a multimodal large language model 13 to analyze the image data and obtain the quantized features corresponding to the first tag. Furthermore, the target app can also respond to user operations by obtaining a file or prompt containing the quantized features, and instruct the multimodal large language model 13 to perform a target task based on the file or prompt containing the quantized features. The multimodal large language model 13 performing the target task and the multimodal large language model 13 analyzing the data can be the same model or different models.

[0040] It should be understood that Figure 1 The number of data analysis systems 11, terminal devices 12, and multimodal large language models 13 shown in the diagram is merely illustrative. Depending on implementation needs, any number of data analysis systems 11, terminal devices 12, and multimodal large language models 13 can be included.

[0041] Figure 2 A hardware structure diagram of a computing system provided according to an embodiment of this specification is shown. The computing system 200 can serve as... Figure 1 The data analysis system 11 in the manual executes the data analysis methods described herein.

[0042] like Figure 2 As shown, the computing system 200 may include at least one storage medium 230 and at least one processor 220. In some embodiments, the computing system 200 may also include a communication port 250 and an internal communication bus 210. The computing system 200 may also include I / O components 260.

[0043] The internal communication bus 210 can connect to different system components. For example, the internal communication bus 210 can connect to storage medium 230, processor 220, communication port 250, and I / O component 260, etc.

[0044] I / O component 260 supports input / output between computing system 200 and other components.

[0045] Communication port 250 is used for data communication between computing system 200 and the outside world. For example, communication port 250 can be used for data communication between computing system 200 and a network. Communication port 250 can be a wired communication port or a wireless communication port.

[0046] Storage medium 230 may include a data storage device. The data storage device may be a non-transitory storage medium or a temporary storage medium. For example, the data storage device may include one or more of a disk 232, a read-only storage medium (ROM) 234, or a random access storage medium (RAM) 235. Storage medium 230 also includes at least one instruction set stored in the data storage device. The instruction set may include computer program code, which may include programs, routines, objects, components, data structures, procedures, modules, etc.

[0047] At least one processor 220 may be communicatively connected to at least one storage medium 230. When the computing system 200 is running, at least one processor 220 reads the at least one instruction set and executes the data analysis method provided in this specification according to the instructions of the at least one instruction set. The processor 220 may perform the steps included in the data analysis method. The processor 220 may be in the form of one or more processors. In some embodiments, the processor 220 may include one or more hardware processors, such as a microcontroller, microprocessor, reduced instruction set computer (RISC), application-specific integrated circuit (ASIC), application-specific instruction set processor (ASIP), central processing unit (CPU), graphics processing unit (GPU), physical processing unit (PPU), microcontroller unit, digital signal processor (DSP), field-programmable gate array (FPGA), advanced RISC machine (ARM), programmable logic device (PLD), any circuit or processor capable of performing one or more functions, or any combination thereof.

[0048] For illustrative purposes only, the accompanying drawings show only one processor 220 for the computing system 200. However, it should be noted that the computing system 200 may also include multiple processors; therefore, the operations and / or method steps disclosed herein may be executed by one processor or by multiple processors in combination. For example, if the processor 220 of the computing system 200 described in this specification executes steps A and B, it should be understood that steps A and B may also be executed jointly or separately by two different processors 220 (e.g., a first processor executes step A, a second processor executes step B, or the first and second processors jointly execute steps A and B).

[0049] Figure 3 A flowchart of a data analysis method according to an embodiment of this specification is shown. As previously described, the computing system 200 can perform the data analysis method of this specification as a data analysis system.

[0050] like Figure 3 As shown, data analysis methods may include: S310: Obtain a first image set and a second image set. The first image set includes multiple first images labeled with a first label, and the second image set includes multiple second images labeled with a second label. The first label and the second label are image evaluation metrics of the same target dimension, and the image evaluation metrics of the first label are better than the image evaluation metrics of the second label.

[0051] In some embodiments, the target dimension can be an evaluation dimension of image quality in different scenarios. For example, in e-commerce scenarios, the target dimension can be the conversion rate of the product image; in marketing design scenarios, the target dimension can be the attractiveness of the advertising image to users; in the field of user interface design, the target dimension can be the ease of use of the user interface; and in the field of image evaluation, the target dimension can be the quality of the image's aesthetics. Here, the conversion rate refers to the proportion of users who completed a specified target action (such as purchasing, adding to favorites, or recommending) after viewing the product image, out of all users who viewed that product image.

[0052] Correspondingly, in e-commerce scenarios, the first label can be high conversion rate, and the second label can be low conversion rate. Therefore, multiple first images in the first image set are all images with high conversion rates, and multiple second images in the second image set are all images with low conversion rates. In marketing design scenarios, the first label can be attractive to users, and the second label can be unattractive to users. Therefore, multiple first images in the first image set are all images with high user appeal, and multiple second images in the second image set are all images with low user appeal. In the field of user interface design, the first label can be easy to use, and the second label can be difficult to use. Therefore, multiple first images in the first image set are all images of easy-to-use user interfaces, and multiple second images in the second image set are all images of difficult-to-use user interfaces. Furthermore, in the field of image evaluation, the first label can be high aesthetic quality, and the second label can be low aesthetic quality. Therefore, multiple first images in the first image set are all images of relatively high aesthetic quality, and multiple second images in the second image set are all images of relatively low aesthetic quality.

[0053] In some embodiments, the first image set and the second image set may be acquired in different ways. For example, the data analysis system may display multiple images on a terminal device and, in response to user operations, determine whether each image in the multiple images is labeled with a first label or a second label. The image labeled with the first label is taken as the first image, and the image labeled with the second label is taken as the second image, thus obtaining the first image set and the second image set.

[0054] Alternatively, in some embodiments, the data analysis system may display a target page on a terminal device and use the target page to receive a first image and a second image uploaded by the user.

[0055] Figure 4 A schematic diagram of a target page provided according to an embodiment of this specification is shown.

[0056] In some embodiments, the data analysis system can be displayed on a terminal device. Figure 4 The target page shown includes a first interactive interface, a second interactive interface, and a third interactive interface. It also includes a first image display area, a second image display area, and an interactive area for starting the generation of evaluation criteria.

[0057] As an example, after the data analysis system displays the first and second interactive interfaces on the target page on the terminal device, it can respond to user operations using the terminal device, receive images uploaded by the user based on the first and second interactive interfaces, determine the image received based on the first interactive interface as the first image, determine the image received based on the second interactive interface as the second image, and obtain the first image set and the second image set.

[0058] For example, when a terminal device detects a click operation on the first interactive interface, it can display an image upload interface. In response to detecting a selection and confirmation operation on the image upload interface, the terminal device uploads the image indicated by the selection operation to the data analysis system. These images uploaded through the first interactive interface are called the first images and have a first tag. The data analysis system can also process the received first images (such as...) Figure 4 The one shown Figure 1 , Figure 2 and Figure 3 It will be displayed in the first image display area on the target page.

[0059] When the terminal device detects a click operation on the second interactive interface, it can display an image upload interface. Upon detecting a selection and confirmation operation on the image upload interface, the terminal device uploads the image indicated by the selection operation to the data analysis system. These images uploaded through the second interactive interface are called second images and have second tags. The data analysis system can also process the received second images (such as...) Figure 4 The one shown Figure 4 , Figure 5 and Figure 6 It will be displayed in the second image display area on the target page.

[0060] When the terminal device detects a click operation on the interactive area where evaluation criteria are initially generated, it can instruct the data analysis system to perform relevant analysis steps based on the first image set and the second image set.

[0061] In this embodiment, the data analysis system displays a target page on the terminal device and receives a first image and a second image using the target page. By using a graphical interface to upload images, users only need to upload high-quality images through the first interactive interface and low-quality images through the second interactive interface. The data analysis system can then obtain a first image set and a second image set for analysis. This process eliminates the need for users to describe the evaluation criteria, simplifying the user's operation and effectively avoiding interference caused by differences in user subjective descriptions, making image data acquisition more intuitive, accurate, and efficient.

[0062] In some embodiments, reference Figure 4 The data analysis system can also receive reference data based on a third interactive interface. The reference data includes at least one of the following: reference information of the first image, reference information of the second image, domain information, and feature dimension information.

[0063] As an example, the third interactive interface may include drop-down menus, checkboxes, and text input boxes.

[0064] The dropdown menu can provide multiple domain options, and the checkbox can include multiple feature dimension options. When the terminal device detects an operation on the dropdown menu and / or checkbox in the third interactive interface, it can respond to the operation to determine the domain and feature dimensions selected by the user, and obtain domain information and feature dimension information. The terminal device can also respond to input operations on the text input box by receiving reference information for the first image and reference information for the second image input by the user.

[0065] When the terminal device detects a click operation on the interactive area where the evaluation criteria are generated, it can upload reference information to the data analysis system and instruct the data analysis system to perform relevant analysis steps based on the first image set, the second image set, and the reference data.

[0066] In this embodiment, the data analysis system displays a target page on the terminal device and uses that target page to receive reference data. This simplifies the user's workflow while enabling the data analysis process to perform targeted reasoning based on the reference data, resulting in more targeted data analysis results that better meet the user's personalized needs.

[0067] S320: Analyze the target dimension of the first image set and the second image set using a preset multimodal large language model to obtain data analysis results. The data analysis results shall include at least the quantitative features corresponding to the first label.

[0068] In some embodiments, the preset multimodal large language model can be a general multimodal large language model or a multimodal large language model that has been specifically fine-tuned.

[0069] As an example, a data analysis system can first test the feature recognition accuracy of a general multimodal large language model based on a test sample set, and then obtain an evaluation index of the general multimodal large language model in the feature recognition accuracy dimension.

[0070] For example, a test sample set may include multiple test samples and the actual recognition results of each test sample. The data analysis system can use a general multimodal large language model to process multiple test samples, obtain the predicted recognition result of each test sample, and determine the feature recognition accuracy (i.e., the evaluation index of feature recognition accuracy) of the general multimodal large language model based on the difference between the predicted recognition result and the actual recognition result of each test sample.

[0071] When the feature recognition accuracy meets the second preset requirement, the data analysis system can determine the general multimodal language model as the preset multimodal language model. When the evaluation metric does not meet the second preset requirement, the data analysis system can fine-tune the general multimodal language model so that the fine-tuned general multimodal language model meets the second preset requirement in terms of feature recognition accuracy, and determine the fine-tuned general multimodal language model as the preset multimodal language model.

[0072] For example, suppose the second preset requirement is a feature recognition accuracy rate greater than 95%. When the feature recognition accuracy rate is 98%, the data analysis system can determine that the feature recognition accuracy rate meets the second preset requirement, that is, the performance of the general multimodal large language model in the feature recognition dimension meets the requirements, and the general multimodal large language model can be used for data analysis.

[0073] When the feature recognition accuracy is 80%, the data analysis system can determine that the feature recognition accuracy does not meet the second preset requirement, meaning that the performance of the general multimodal language model in the feature recognition dimension cannot meet the requirements. In this case, the data analysis system can fine-tune the general multimodal language model through reinforcement learning, supervised fine-tuning, etc., to enhance its performance in the feature recognition dimension.

[0074] Then, the data analysis system can test the feature recognition accuracy dimension of the fine-tuned general multimodal language model, and after determining that the evaluation index of the fine-tuned general multimodal language model in the feature recognition accuracy dimension meets the second preset requirement, the fine-tuned general multimodal language model will be used for data analysis.

[0075] In this embodiment, the data analysis system can perform performance testing on a general multimodal language model in the feature extraction dimension. When the performance of the general multimodal language model in the feature recognition dimension meets the requirements, it is used as the preset multimodal language model to perform data analysis tasks. When the performance of the general multimodal language model in the feature recognition dimension does not meet the requirements, it is fine-tuned so that the performance of the fine-tuned multimodal language model in the feature recognition dimension meets the requirements, and then used as the preset multimodal language model to perform data analysis tasks. In this way, the multimodal language model performing the task can have better performance in the feature extraction dimension, thereby improving the overall quality and reliability of the data analysis results.

[0076] Figure 5 A schematic diagram is shown illustrating a data analysis result obtained from a first image set and a second image set according to an embodiment of this specification.

[0077] In some embodiments, the data analysis system may obtain data analysis results based solely on a first image set and a second image set.

[0078] As an example, see reference Figure 5 The data analysis system can first generate a first prompt instruction based on the first image set and the second image set. The first prompt instruction is used to guide the preset multimodal large language model to perform target dimension reasoning based on the first image set and the second image set, obtain the quantized common features between multiple first images, as well as the quantized difference features between multiple first images and multiple second images, and determine the quantized features corresponding to the first label based on the quantized common features and quantized difference features, and generate data analysis results.

[0079] Here, an example template for a first prompt instruction is provided. In this example, the target dimension is the degree of aesthetic quality of the image, the first label is high aesthetic quality, the second label is low aesthetic quality, multiple first images in the first image set are all images with high aesthetic quality, and multiple second images in the second image set are all images with low aesthetic quality.

[0080] An example of a template for the first prompt instruction is as follows: "You are a professional [image aesthetics] analysis expert. Your task is to perform [aesthetic quality] inference based on data from two image sets, extract quantitative features of [high aesthetic quality], and generate a data analysis report."

[0081] Input data: First image set: This set contains multiple images labeled as [high aesthetic quality].

[0082] The second image set contains several images labeled as [low aesthetic quality].

[0083] Task objective: Identify common quantization features in the first image set: find out which dimensions all [high aesthetic quality] images have consistent numerical features.

[0084] Identify quantitative differences between the first and second image sets: find out in which dimensions there are significant quantitative differences between [high aesthetic quality] images and [low aesthetic quality] images.

[0085] Determine the quantitative characteristics corresponding to the first label: Based on the above analysis, summarize the typical quantitative descriptions of [high aesthetic quality] in key aesthetic dimensions.

[0086] Output requirements: Data analysis results include: First image set common features: List the commonalities in each dimension.

[0087] Differences between the first and second image sets: List the dimensions of significant differences and provide a quantitative comparison.

[0088] First-level label quantification features: Describes the typical characteristics of a [high aesthetic quality] image using specific numerical values. The “[ ]” is a placeholder. The data analysis system can fill in the corresponding content in the placeholder according to the specific situation of the application (such as the specific type of the target dimension) to get the first prompt instruction.

[0089] Then, the data analysis system can input the obtained first prompt command into a preset multimodal large language model and obtain the data analysis results output by the preset multimodal large language model. The data analysis results include at least the quantitative features corresponding to the first label.

[0090] In some embodiments, the data analysis results may include, in addition to the quantitative features corresponding to the first label, a second prompt instruction.

[0091] Correspondingly, the first prompt instruction is also used to guide the preset multimodal large language model to generate a second prompt instruction based on the quantization features corresponding to the first label. The second prompt instruction includes instructions to guide the preset multimodal large language model to perform the target task using the quantization features corresponding to the first label.

[0092] As an example, when the data analysis results also include a second prompt, the template for the first prompt could be: "You are a professional [image aesthetics] analysis expert. Your task is to perform [aesthetic quality] inference based on data from two image sets, extract quantitative features of [high aesthetic quality], and generate a data analysis report."

[0093] Input data: First image set: This set contains multiple images labeled as [high aesthetic quality].

[0094] The second image set contains several images labeled as [low aesthetic quality].

[0095] Task objective: Identify common quantization features in the first image set: find out which dimensions all [high aesthetic quality] images have consistent numerical features.

[0096] Identify quantitative differences between the first and second image sets: find out in which dimensions there are significant quantitative differences between [high aesthetic quality] images and [low aesthetic quality] images.

[0097] Determine the quantitative characteristics corresponding to the first label: Based on the above analysis, summarize the typical quantitative descriptions of [high aesthetic quality] in key aesthetic dimensions.

[0098] Based on the quantization features corresponding to the first label, a second prompt instruction is generated. The second prompt instruction is a prompt instruction to guide the multimodal large language model to execute the [target task] using the quantization features corresponding to the first label.

[0099] Output requirements: Data analysis results include: First image set common features: List the commonalities in each dimension.

[0100] Differences between the first and second image sets: List the dimensions of significant differences and provide a quantitative comparison.

[0101] First label quantification feature: Describes the typical performance of a [high aesthetic quality] image using specific numerical values.

[0102] Second prompt command: Output the second prompt command as a string. The data analysis system can input the obtained first prompt instruction into a preset multimodal large language model and obtain the data analysis results output by the preset multimodal large language model. The data analysis results include at least the quantitative features corresponding to the first label and the second prompt instruction.

[0103] In this embodiment, the data analysis system uses a preset multimodal large language model to analyze the first image set and the second image set to obtain the quantitative features of the first tag. During this process, the data analysis system only analyzes the commonalities and differences between the images, without analyzing user descriptions, thus avoiding interference from differences in subjective user descriptions and making the obtained quantitative features of the first tag more accurate.

[0104] Figure 6 A schematic diagram is shown illustrating a data analysis result obtained from a first image set, a second image set, and reference data, according to an embodiment of this specification.

[0105] In some embodiments, the data analysis system can obtain data analysis results based on a first image set, a second image set, and reference data.

[0106] As an example, see reference Figure 6 The data analysis system can first use a preset multimodal large language model to analyze the first and second image sets in the target dimension to obtain data analysis results. The target dimension can be the image quality dimension. Specifically, one possible implementation is as follows: A fifth prompt instruction is generated based on the first image set, the second image set, and reference data. This fifth prompt instruction guides the preset multimodal large language model to infer the image quality dimension based on the reference data, the first image set, and the second image set. It then obtains the quantized common features among multiple first images and the quantized difference features between multiple first images and multiple second images. Based on the quantized common features and quantized difference features, it determines the quantized features corresponding to the first label and generates data analysis results. Finally, the fifth prompt instruction is input into the preset multimodal large language model to obtain the data analysis results output by the preset multimodal large language model.

[0107] Here, an example template for a fifth prompt instruction is provided. In this example, the image quality dimension can be the degree of good or bad in terms of image aesthetic quality, with the first label being high aesthetic quality and the second label being low aesthetic quality. Multiple first images in the first image set are all images with high aesthetic quality, and multiple second images in the second image set are all images with low aesthetic quality.

[0108] An example of the template for the fifth prompt instruction is as follows: "You are a professional [image aesthetics] analysis expert. Your task is to perform [aesthetic quality] inference based on data from two image sets, extract quantitative features of [high aesthetic quality], and generate a data analysis report."

[0109] Input data: The first image set contains multiple images labeled as [high aesthetic quality].

[0110] The second image set contains multiple images labeled as [low aesthetic quality].

[0111] Reference Data: Information about the first image: Description of the first image as a [high aesthetic quality] image. Information about the second image: Description of the second image as a [low aesthetic quality] image. Domain Information: The application domain of the image. Feature Dimension Information: The key feature dimensions.

[0112] Task objective: Identify common quantization features in the first image set: find out which dimensions all [high aesthetic quality] images have consistent numerical features.

[0113] Identify quantitative differences between the first and second image sets: find out in which dimensions there are significant quantitative differences between [high aesthetic quality] images and [low aesthetic quality] images.

[0114] Determine the quantitative characteristics corresponding to the first label: Based on the above analysis, summarize the typical quantitative descriptions of [high aesthetic quality] in key aesthetic dimensions.

[0115] During task execution, focus on the dimensions involved in the feature dimension information and consider the domain information provided in the domain information.

[0116] Output requirements: Data analysis results include: First image set common features: List the commonalities in each dimension.

[0117] Differences between the first and second image sets: List the dimensions of significant differences and provide a quantitative comparison.

[0118] First-level label quantification features: Describes the typical appearance of a [high aesthetic quality] image using specific numerical values ​​(especially the values ​​in the dimensions involved in the feature dimension information). Then, the data analysis system can input the obtained fifth prompt instruction into a preset multimodal large language model and obtain the data analysis results output by the preset multimodal large language model. The data analysis results include at least the quantitative features corresponding to the first label.

[0119] Alternatively, the data analysis results may include at least the quantitative features corresponding to the first label and the second prompt instruction. When the data analysis results include at least the quantitative features corresponding to the first label and the second prompt instruction, the fifth prompt instruction template can be modified in accordance with the above-described method for modifying the first prompt instruction template, which will not be elaborated upon here.

[0120] In this embodiment, the data analysis system uses a preset multimodal large language model to analyze the first image set, the second image set, and reference data to obtain the quantitative features of the first label. During this process, the data analysis system can analyze the commonalities and differences between images based on the guidance of the reference data in areas such as domain and dimension. This allows the data analysis process to combine the reference data for targeted reasoning, making the obtained data analysis results more targeted and better suited to the user's personalized needs.

[0121] Figure 7A schematic diagram of another target page provided according to an embodiment of this specification is shown.

[0122] In some embodiments, after the data analysis system obtains the data analysis results output by the preset multimodal large language model, it can display the data analysis results on the terminal device through the target page. Figure 7 The target page in the text is the page that displays the data analysis results.

[0123] refer to Figure 7 The target page includes a display area for the quantitative features of the first tag, a display area for the second prompt instruction, a download interaction area for the quantitative features, and a download interaction area for the prompt instruction. When the data analysis results only include the quantitative features of the first tag, the display area for the second prompt instruction and the download interaction area for the prompt instruction can be left empty. Alternatively, the data analysis system can merge the display area for the second prompt instruction into the display area for the quantitative features of the first tag, and merge the download interaction area for the prompt instruction into the download interaction area for the quantitative features.

[0124] Specifically, the display area for the quantized features of the first tag can display the obtained quantized features of the first tag in the form of a string. Furthermore, the terminal device can download the quantized features of the first tag as a file to its local storage space in response to a click operation on the quantized feature download interaction area.

[0125] Furthermore, when the data analysis results include a second prompt instruction, the display area for the second prompt instruction can show the obtained second prompt instruction as a string. Additionally, the terminal device can download the second prompt instruction as a file to its local storage space in response to a click operation applied to the prompt instruction download interaction area.

[0126] Figure 8 A schematic diagram is shown illustrating an embodiment of this specification that generates prompt instructions based on data analysis results.

[0127] In some embodiments, reference Figure 8 When the data analysis results only include the quantitative features corresponding to the first label, the data analysis system can also input a third prompt instruction to the preset multimodal large language model. The third prompt instruction includes: the target task and the quantitative features corresponding to the first label.

[0128] The third prompt instruction is used to guide the preset multimodal large language model to generate the fourth prompt instruction corresponding to the target task based on the quantized features corresponding to the first label. The fourth prompt instruction is used to guide the preset multimodal large language model to execute the target task.

[0129] Here, an example template for a third prompt instruction is provided. In this example, the first label is "high aesthetic quality," and the quantization feature corresponding to the first label is obtained according to the steps described above.

[0130] An example of a template for the third prompt instruction is as follows: "You are a prompting engineer, specializing in designing instructions for multimodal large language models. Your task is to generate a fourth prompting instruction for a target task based on user-provided quantized features. This fourth prompting instruction will be directly used to guide the pre-defined multimodal large language model to perform the target task and ensure that the model fully understands and utilizes these quantized features during inference."

[0131] Input information Quantization features: [Quantization features corresponding to the first label].

[0132] Target Task: [Description of the target task] Mission Objectives Based on the above input, you need to generate a fourth prompt command. This fourth prompt command should: Directly targeting the objective task, it guides a pre-defined multimodal large language model to perform the task. Quantized features corresponding to the first label are embedded as key reference information in the instructions, ensuring the model can infer or generate based on these numerical standards. The language is clear, specific, and adaptable to the output requirements of multimodal inputs (such as images + text).

[0133] Output requirements The final output is simply the text of the fourth prompt instruction, requiring no further explanation. That is, the output fourth prompt instruction should be suitable for direct copying and pasting into the input box of the multimodal large language model. The "[ ]" is a placeholder. The data analysis system can fill in the corresponding content in the placeholder according to the specific situation of the application (such as the specific target task) to obtain the third prompt instruction.

[0134] Then, the data analysis system can input the obtained third prompt instruction into the preset multimodal large language model and obtain the fourth prompt instruction output by the preset multimodal large language model.

[0135] It should be noted that since the third prompt instruction only contains text, it can guide a preset multimodal or non-multimodal large language model to perform reasoning and generate a fourth prompt instruction. The process of inputting the third prompt instruction into a preset multimodal large language model to generate a fourth prompt instruction, as described in this specification, is merely an example. In practical applications, the third prompt instruction can be input into any large language model to guide it to perform reasoning and generate a fourth prompt instruction.

[0136] The fourth prompt instruction and the second prompt instruction mentioned above are both used to guide the preset multimodal large language model to perform the target task based on the quantized features corresponding to the first label. The multimodal large language model performing the target task can be the preset multimodal large language model, or it can be other multimodal large language models. This specification does not impose any restrictions on the multimodal large language model performing the target task.

[0137] In this embodiment, the data analysis system obtains the quantified features corresponding to the first label and the fourth prompt instruction through different prompt instructions. During the acquisition of the fourth prompt instruction, the third prompt instruction can more effectively guide the large model in reasoning, effectively improving the generation quality of the fourth prompt instruction, thereby enhancing the quality and accuracy of the subsequent execution of the target task.

[0138] In some embodiments, the data analysis system can also verify the accuracy dimension of the data analysis results to obtain an accuracy verification result. The accuracy verification result is used to characterize whether the accuracy of the quantitative feature corresponding to the first label meets the first preset requirement. Furthermore, when the accuracy verification result indicates that the quantitative feature corresponding to the first label does not meet the first preset requirement, the first prompt instruction is updated, and the updated first prompt instruction is used to obtain the data analysis result.

[0139] As an example, the data analysis system can execute multiple test tasks using a multimodal large language model based on the quantitative features corresponding to the first label, obtaining multiple output results for each test task. The data analysis system can then display these multiple output results to the user and receive multiple quality evaluation metrics from the user based on these results. Finally, the data analysis system can determine whether the accuracy of the quantitative features corresponding to the first label meets a first preset requirement based on the distribution of these multiple quality evaluation metrics across the quality dimensions.

[0140] For example, assuming the quality evaluation index is scored from 0 to 10, the first preset requirement is that the proportion of evaluation indicators scoring 8 or above is no less than 85%. When the evaluation indicators scoring 8 or above account for 90% of all evaluation indicators, the data analysis system can determine that the accuracy verification result meets the first preset requirement. When the evaluation indicators scoring 8 or above account for 80% of all evaluation indicators, the data analysis system can determine that the accuracy verification result does not meet the first preset requirement.

[0141] In this embodiment, the data analysis system can verify the accuracy of the quantization features corresponding to the first label. When the accuracy verification result does not meet the first preset requirement, it indicates that the quantization features corresponding to the first label are not accurate enough and need to be regenerated. In this way, the accuracy of the quantization features corresponding to the first label can be verified, so that the final quantization features corresponding to the first label can more accurately represent the characteristics of the first label, thereby improving the quality and reliability of subsequent target task execution.

[0142] In some implementations, the data analysis system can also guide a pre-defined multimodal language model to output the quantized features of the second label. In this case, the guidance instructions used to guide the pre-defined multimodal language model to perform the target task need to guide the pre-defined multimodal language model to avoid using the quantized features of the second label when performing the target task. The method for obtaining the quantized features of the second label is the same as the method for obtaining the quantized features of the first label, and will not be described in detail here.

[0143] In summary, the data analysis method and system provided in this specification allow the data analysis system to first obtain a first image set corresponding to a first label with a better evaluation metric under the same target dimension, and a second image set corresponding to a second label with a relatively worse evaluation metric. The data analysis system then utilizes a pre-set multimodal large language model to perform target dimension reasoning and analysis on the two image sets, mining and obtaining the quantitative features of the first label. In this way, the solution provided in this specification achieves automatic analysis to obtain the quantitative features corresponding to the first label based on image sets with different evaluation metrics under the same target dimension using a pre-set multimodal large language model. Using the obtained quantitative features as evaluation criteria enables the multimodal large language model to have clear, accurate, and quantifiable evaluation criteria when processing image-related tasks. This effectively improves the problem of biased task processing results caused by ambiguous or low-quality evaluation criteria, enhances the accuracy and reliability of the multimodal large language model in processing image-related tasks, and improves the user experience.

[0144] Furthermore, the quantitative features corresponding to the first label can also serve as a knowledge base, providing reference for other tasks. In this way, the experience of senior data analysts can be extended to more scenarios, reducing the experience requirements for other data analysts and thus improving the task execution efficiency and quality of the entire data analysis team. Moreover, for the data analysis team, using unified quantitative features can maintain consistent standards within the team, reducing communication costs and improving work efficiency.

[0145] This specification, in another aspect, provides a computer-readable non-transitory storage medium storing at least one instruction set that identifies quantization features corresponding to a first tag. When the at least one instruction set is executed by a processor, it instructs the processor to perform the steps of the data analysis method described in this specification. In some possible implementations, various aspects of this specification can also be implemented as a program product comprising program code. When the program product is run on a computing system 200, the program code causes the computing system 200 to perform the steps of the data analysis method described in this specification. The program product for implementing the above method may employ a portable compact disc read-only memory (CD-ROM) containing program code and may run on the computing system 200. However, the program product of this specification is not limited thereto. In this specification, the readable storage medium may be any tangible medium containing or storing a program that may be used by or in conjunction with an instruction execution system. The program product may employ any combination of one or more readable media. The readable medium may be a readable signal medium or a readable storage medium. The readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of readable storage media include: electrical connections having one or more wires, portable disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. The computer-readable storage medium may include data signals propagated in baseband or as part of a carrier wave, carrying readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A readable storage medium may also be any readable medium other than a readable storage medium that can send, propagate, or transmit programs for use by or in connection with an instruction execution system, apparatus, or device. Program code contained on a readable storage medium may be transmitted using any suitable medium, including but not limited to wireless, wired, optical fiber, RF, etc., or any suitable combination thereof. Program code for performing the operations described herein can be written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Java and C++, and conventional procedural programming languages ​​such as C or similar languages. The program code can be executed entirely on computing system 200, partially on computing system 200, as a standalone software package, partially on computing system 200 and partially on a remote computing device, or entirely on a remote computing device.

[0146] The foregoing has described specific embodiments of this specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims may be performed in a different order than that shown in the embodiments and may still achieve the desired result. Furthermore, the processes depicted in the drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are possible or may be advantageous.

[0147] In summary, after reading this detailed disclosure, those skilled in the art will understand that the foregoing detailed disclosure is presented by way of example only and is not restrictive. Although not explicitly stated herein, those skilled in the art will understand that this specification requires various reasonable changes, improvements, and modifications to the embodiments. These changes, improvements, and modifications are intended to be made by this specification and are within the spirit and scope of the exemplary embodiments described herein.

[0148] Furthermore, certain terms in this specification have been used to describe embodiments of this specification. For example, "an embodiment," "an embodiment," and / or "some embodiments" mean that a particular feature, structure, or characteristic described in connection with that embodiment may be included in at least one embodiment of this specification. Therefore, it is to be emphasized and understood that two or more references to "an embodiment" or "an embodiment" or "alternative embodiment" in various parts of this specification do not necessarily refer to the same embodiment. Moreover, specific features, structures, or characteristics may be suitably combined in one or more embodiments of this specification.

[0149] It should be understood that in the foregoing description of the embodiments in this specification, various features are combined in a single embodiment, drawing, or description for the purpose of simplifying the description and to aid in understanding a feature. However, this does not mean that the combination of these features is necessary, and those skilled in the art, upon reading this specification, may readily identify some of the devices as separate embodiments. That is, the embodiments in this specification can also be understood as an integration of multiple secondary embodiments. And the content of each secondary embodiment is valid even if it contains fewer than all the features of a single foregoing disclosed embodiment.

[0150] Every patent, patent application, publication of a patent application, and other material such as articles, books, specifications, publications, documents, articles, etc., cited herein, except for those inconsistent with or conflicting with this document, or those having a restrictive effect on the widest scope of the claims, may be incorporated herein by reference for all purposes now or hereafter associated with this document. Furthermore, in the event of any inconsistency or conflict between the description, definition, and / or use of relevant terms in any material and the description, definition, and / or use of relevant terms in this document, the terms in this document shall prevail.

[0151] Finally, it should be understood that the embodiments disclosed herein are illustrative of the principles of the embodiments described in this specification. Other modified embodiments are also within the scope of this specification. Therefore, the embodiments disclosed in this specification are merely examples and not limitations. Those skilled in the art can implement the applications described in this specification using alternative configurations based on the embodiments in this specification. Therefore, the embodiments in this specification are not limited to the embodiments precisely described in the applications.

Claims

1. A data analysis method, the method comprising: Obtain a first image set and a second image set. The first image set includes multiple first images labeled with a first tag, and the second image set includes multiple second images labeled with a second tag. The first tag and the second tag are image evaluation metrics of the same target dimension, and the image evaluation metric for the first tag is superior to the image evaluation metric for the second tag. The first image set and the second image set are analyzed in terms of target dimension using a preset multimodal large language model to obtain data analysis results, which include at least the quantized features corresponding to the first label.

2. The method according to claim 1, wherein, The step involves using a pre-defined multimodal large language model to analyze the target dimension of the first image set and the second image set, obtaining data analysis results, including: A first prompt instruction is generated based on the first image set and the second image set. This first prompt instruction guides the preset multimodal large language model to perform target-dimensional inference based on the first and second image sets, acquiring quantized common features among multiple first images and quantized difference features between multiple first images and multiple second images. Based on the quantized common features and the quantized difference features, the quantized features corresponding to the first label are determined, generating the data analysis result. The first prompt instruction is input into the preset multimodal large language model to obtain the data analysis results output by the preset multimodal large language model.

3. The method according to claim 2, wherein, The data analysis results also include: a second prompt instruction; Accordingly, the first prompt instruction is also used to guide the preset multimodal large language model to generate a second prompt instruction based on the quantization features corresponding to the first label. The second prompt instruction includes instructions to guide the preset multimodal large language model to perform the target task using the quantization features corresponding to the first label.

4. The method according to claim 1, wherein, The method further includes: A third prompt instruction is input into the preset multimodal large language model. The third prompt instruction includes: the target task and the quantized features corresponding to the first label. The third prompt instruction is used to guide the preset multimodal large language model to generate a fourth prompt instruction corresponding to the target task based on the quantized features corresponding to the first label. The fourth prompt instruction is used to guide the preset multimodal large language model to execute the target task.

5. The method according to claim 2, wherein, After obtaining the data analysis results output by the preset multimodal large language model, the method further includes: The data analysis results are verified for accuracy, resulting in an accuracy verification result. This accuracy verification result characterizes whether the accuracy of the quantified feature corresponding to the first label meets a first preset requirement. When the accuracy verification result indicates that the quantization feature corresponding to the first label does not meet the first preset requirement, the first prompt instruction is updated, and the updated first prompt instruction is used to obtain the data analysis result.

6. The method according to claim 1, wherein, The acquisition of the first image set and the second image set includes: The target page is used to display the first and second interactive interfaces; Receive images uploaded by the user based on the first interactive interface and the second interactive interface; and The image received based on the first interactive interface is determined to be the first image, and the image received based on the second interactive interface is determined to be the second image, thus obtaining the first image set and the second image set.

7. The method according to claim 6, wherein, The target page also displays a third interactive interface, and the method further includes: Reference data is received based on the third interactive interface, and the reference data includes at least one of the following: The reference information of the first image, the reference information of the second image, the domain information, and the feature dimension information.

8. The method according to claim 7, wherein, The step involves using a pre-defined multimodal large language model to analyze the target dimension of the first image set and the second image set, obtaining data analysis results, including: A fifth prompt instruction is generated based on the first image set, the second image set, and the reference data. This fifth prompt instruction guides the preset multimodal large language model to perform image quality dimension inference based on the reference data, the first image set, and the second image set. It then obtains quantized common features among multiple first images and quantized difference features among multiple first images and multiple second images, and determines the quantized features corresponding to the first label based on the quantized common features and the quantized difference features, generating the data analysis result. The fifth prompt instruction is input into the preset multimodal large language model to obtain the data analysis results output by the preset multimodal large language model.

9. The method according to claim 1, wherein, Before performing target dimension analysis on the first image set and the second image set using a pre-defined multimodal large language model, the method further includes: Based on the test sample set, the feature recognition accuracy dimension of the general multimodal large language model is tested, and the evaluation index of the general multimodal large language model in the feature recognition accuracy dimension is obtained. When the evaluation index meets the second preset requirement, the general multimodal large language model is determined to be the preset multimodal large language model; and When the evaluation index does not meet the second preset requirement, the general multimodal large language model is fine-tuned so that the evaluation index of the fine-tuned general multimodal large language model in the dimension of feature recognition accuracy meets the second preset requirement, and the fine-tuned general multimodal large language model is determined to be the preset multimodal large language model.

10. A data analysis system, comprising: At least one storage medium stores at least one instruction set for identifying the quantization features corresponding to the first tag; as well as At least one processor is communicatively connected to the at least one storage medium, wherein when the data analysis system is running, the at least one processor reads the at least one instruction set and implements the method as described in any one of claims 1-9 according to the instructions of the at least one instruction set.

11. A computer-readable non-volatile storage medium, wherein, The computer-readable non-volatile storage medium stores at least one instruction set, which, when executed by at least one processor, implements the method as described in any one of claims 1-9.

Images