Visual data processing method and device based on large language model, equipment and medium

Abstract

The invention relates to the field of artificial intelligence, in particular to a visual data processing method and device based on a large language model, equipment and a medium, and the method comprises the steps: obtaining a question text of a user; classifying the question text through a large language model to obtain question categories; when the question category is a diagnosis category, data information and chart information are determined according to the question text, chart rendering is carried out according to the data information and the chart information, a question and answer result is generated, the question and answer result comprises a rendered chart, and role information in an intelligent investment adviser scene is preset in the large language model; therefore, the user can interact with the role corresponding to the role information in the large language model; and sending the question and answer result to a client device corresponding to the user so as to visually display the question and answer result on the client device. According to the method, the question and answer result including the visual chart can be provided for the question of the user, and the question and answer result is more accurate and better meets the requirements of the user.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence, and in particular to a method, apparatus, device and medium for visual data processing based on a large language model. Background Art

[0002] Smart investment advisor, also known as financial management robot, is an innovative model based on artificial intelligence technology and big data algorithms to provide investors with personalized investment advice and asset management services. In related technologies, the process of smart investment advisor is: the key field slots are extracted after the question is parsed, and the data interface is called to complete the data acquisition. Users are unlikely to enter complete query statements into the system, which leads to a high probability that the data obtained based on the user's question parsing will not match the data required for chart drawing, resulting in poor accuracy of the answer result. Summary of the invention

[0003] The purpose of this application is to provide a visual data processing method, device, equipment and medium based on a large language model, which can provide question and answer results including visual charts for user questions, and the question and answer results are more accurate and better meet the needs of users.

[0004] In a first aspect, a visualization data processing method based on a large language model is provided, comprising:

[0005] Get the user's question text;

[0006] Obtaining a large language model, wherein the large language model is pre-set with role information in a smart investment advisory scenario, so that the user can interact with the role corresponding to the role information in the large language model and the user;

[0007] Inputting the question text into the large language model, classifying the question text through the large language model to obtain a question category; when the question category is a diagnosis category, determining data information and chart information according to the question text, and rendering a chart according to the data information and the chart information to generate a question-and-answer result, wherein the question-and-answer result includes a rendered chart;

[0008] The question-and-answer result is sent to the client device corresponding to the user, so that the question-and-answer result is visually displayed on the client device.

[0009] In a preferred example, the present application can be further configured to: determine data information and chart information according to the question text, including:

[0010] Determine the data information according to the question text, the data information including: target basic information, time period information and indicator information corresponding to at least one target;

[0011] According to the data information, chart information is determined.

[0012] In a preferred example, the present application may be further configured as follows: determining the chart information according to the data information includes:

[0013] Determine a target chart type according to the data information;

[0014] When the target chart type is a general chart, determining chart information according to chart requirements and data information corresponding to the target chart type;

[0015] When the target chart type is a high-definition chart, the target chart is determined from a plurality of preset charts according to the question text, and parameter information of the target chart is used as chart information.

[0016] In a preferred example, the present application may be further configured to: render a chart according to the data information and the chart information to generate a question-and-answer result, including:

[0017] According to the data information, reading first target data corresponding to the data information from a database;

[0018] When the target chart type is a general chart, performing feature analysis on the first target data to obtain basic data features and derived data features;

[0019] Fill in the code block of the general chart corresponding to the target chart type according to the basic data features and the derived data features, realize chart rendering, and generate question and answer results; the code block of the general chart corresponding to the target chart type is determined based on the chart information;

[0020] When the target chart type is a high-definition chart, performing data analysis on the first target data according to parameter information of the target chart to obtain a plurality of chart data corresponding to the parameter information;

[0021] The target chart is located according to the chart information, and the multiple chart data are filled in according to the target chart to obtain the question and answer result.

[0022] In a preferred example, the application can be further configured to: obtain the user's question text, including:

[0023] Get the user's initial question text;

[0024] Analyzing keywords of the initial question text;

[0025] Reasoning is performed based on the keywords to split the initial question text into question texts with at least two tasks.

[0026] In a preferred example, the present application may be further configured as follows: before sending the question-and-answer result to the client device corresponding to the user, the following may also be included:

[0027] Extracting key features of the question and answer results;

[0028] Determining the degree of match between the key feature and the data information;

[0029] If the matching degree is greater than a preset degree threshold, executing the step of sending the question and answer result to the client device corresponding to the user;

[0030] If the matching degree is not greater than the preset degree threshold, the question text is re-input into the large language model until the matching degree is greater than the preset degree threshold, or the number of repetitions reaches the preset number threshold.

[0031] In a preferred example, the present application can be further configured as follows:

[0032] When the question category is not a diagnosis category, second target data corresponding to the question text is determined from a database according to the question text as a question-answering result.

[0033] In a second aspect, a visualization data processing device based on a large language model is provided, comprising:

[0034] The first acquisition module is used to acquire the user's question text;

[0035] A second acquisition module is used to acquire a large language model, in which role information in a smart investment advisory scenario is pre-set, so that the user can interact with the role corresponding to the role information in the large language model and the user;

[0036] A question-answering module, for inputting the question text into the large language model, classifying the question text through the large language model to obtain a question category; when the question category is a diagnosis category, determining data information and chart information according to the question text, and rendering a chart according to the data information and the chart information to generate a question-answering result, wherein the question-answering result includes a rendered chart;

[0037] The sending module is used to send the question and answer results to the client device corresponding to the user, so as to visually display the question and answer results on the client device.

[0038] In a third aspect, an electronic device is provided, including:

[0039] one or more processors;

[0040] Memory;

[0041] One or more applications, wherein one or more applications are stored in a memory and configured to be executed by one or more processors, and one or more programs are configured to: perform operations corresponding to the method shown in any possible implementation of the first aspect.

[0042] In a fourth aspect, a computer-readable storage medium is provided, wherein the storage medium stores at least one instruction, at least one program, a code set or an instruction set, and the at least one instruction, at least one program, a code set or an instruction set is loaded by a processor and executes the steps of the method shown in any possible implementation of the first aspect.

[0043] In a fifth aspect, a computer program product is provided, comprising a computer program, wherein when the computer program is executed by a processor, operations corresponding to the method shown in any possible implementation manner in the first aspect are implemented.

[0044] In summary, the method provided by this application includes the following beneficial technical effects:

[0045] The large language model used in this solution is pre-set with role information in the smart investment advisory scenario, so that the user can interact with the role corresponding to the role information in the large language model; the question text is input into the large language model, and the question text is classified by the large language model to obtain a question category; when the question category is a diagnostic category, data information and chart information are determined according to the question text, and chart rendering is performed according to the data information and the chart information to generate a question and answer result, which includes a rendered chart, and the question type can be determined according to the user's question text through the large language model. When the question type is a diagnostic question, the chart information and the data information presented in the chart can be inferred based on the question text through the large language model, and then the chart rendering is completed based on the chart information and the data information, and the user's question can be provided with a question and answer result including a visualized chart. The question and answer result is more accurate and better meets the needs of the user.

[0046] In addition, the present application also provides a device, equipment and medium, all of which have the above-mentioned beneficial technical effects. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] In order to more clearly illustrate the embodiments of the present application or the technical solutions of the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0048] Figure 1A schematic diagram of a scenario for applying a visual data processing method based on a large language model provided in an embodiment of the present application;

[0049] Figure 2 It is a flowchart of a visual data processing method based on a large language model provided in an embodiment of the present application;

[0050] Figure 3 It is a flowchart of a specific large-model visual analysis and logical reasoning method provided in an embodiment of the present application;

[0051] Figure 4 A schematic diagram of a characterization analysis logic provided in an embodiment of the present application;

[0052] Figure 5 A schematic diagram of a data feature summary process provided in an embodiment of the present application;

[0053] Figure 6 It is a structural diagram of a visual data processing device based on a large language model provided in an embodiment of the present application.

[0054] Figure 7 It is a structural schematic diagram of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0055] This specific embodiment is merely an explanation of the present application and is not a limitation of the present application. After reading this specification, those skilled in the art may make non-creative modifications to the present embodiment as needed, but as long as they are within the scope of the present application, they are protected by the patent law.

[0056] It should be noted that in the optional embodiments of the present application, the object information and other related data involved, when the embodiments in the present application are applied to specific products or technologies, need to obtain the permission or consent of the object, and the collection, use and processing of the relevant data need to comply with the relevant laws, regulations and standards of the relevant countries and regions. In other words, if the embodiments of the present application involve data related to the object, it needs to be obtained with the authorization and consent of the object, the authorization and consent of the relevant departments, and in compliance with the relevant laws, regulations and standards of the country and region. If personal information is involved in the embodiments, the acquisition of all personal information needs to obtain the consent of the individual. If sensitive information is involved, the separate consent of the information subject needs to be obtained. The embodiments also need to be implemented with the authorization and consent of the object.

[0057] In order to make the purpose, technical solution and advantages of the embodiments of the present application clearer, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0058] In addition, the term "and / or" in this article is only a description of the association relationship of associated objects, indicating that there can be three relationships. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. In addition, the character " / " in this article, unless otherwise specified, generally means that the associated objects before and after are in an "or" relationship.

[0059] The current process of visualization technology in smart investment advisors is as follows: the question is parsed to extract the key field slots, the data interface is called to obtain the chart data, and then the chart interface is called to render the chart.

[0060] From the question to the data link, no chart factor is considered. In the scenario of smart investment advisors, the user's intention through asking questions is to obtain chart display, content analysis, and provide insightful and meaningful analysis. It is understandable that there are requirements for data in chart drawing. For example, to draw a company's price-to-earnings ratio with a line chart, only data from multiple time points can be drawn. In addition, in actual scenarios, the query statements entered by users into the system are incomplete and cannot show the true intention, which leads to a high probability that the data obtained based on the user's question analysis and the data required for chart drawing will not match. In the scenario of smart investment advisors, there are two types of problems: failure to draw a chart and poor drawing due to the mismatch between the obtained data and the chart requirements: failure to draw a price-to-earnings ratio curve and lack of year-on-year analysis. Among them, although the query statement does not involve year-on-year analysis, for users as analysts, they not only want to obtain the price-to-earnings ratio of a company, but also need to obtain year-on-year analysis.

[0061] At the same time, for the same content, such as operating income, different charts correspond to different analysis logics. For example, using a pie chart to analyze operating income means that the user wants to see the detailed composition of operating income of different business types or different countries; while using a line chart to analyze operating income means that the user wants to see the changing trend of total operating income over a longer period of time. Therefore, in the link from question to data, in addition to the user's question information, it is also necessary to be able to use chart analysis as a link for final data decision-making to present users with more meaningful investment analysis.

[0062] Therefore, the present application provides a visual data processing method based on a large language model, which involves the fields of artificial intelligence, smart investment consulting, etc.

[0063] In order to better understand the solution provided by the embodiment of the present application, the solution is described below in conjunction with a specific application scenario.

[0064] In one embodiment, please refer to Figure 1 , Figure 1 A schematic diagram of a scenario for the application of a visualization data processing method based on a large language model provided in an embodiment of the present application. The visualization data processing method based on a large language model can be applied to a visualization data processing system based on a large language model.

[0065] The visual data processing system based on the large language model includes a server and a terminal device, and the terminal device can be used by users to input question text. Among them, the terminal device includes but is not limited to: mobile terminals such as mobile phones, laptops, digital broadcast receivers, PDAs (personal digital assistants), PADs (tablet computers), PMPs (portable multimedia players), etc., and fixed terminals such as digital TVs, desktop computers, etc. The electronic device is a server, which can be an independent physical server, a server cluster or a distributed system composed of multiple physical servers, or a cloud server that provides cloud computing services. The terminal device and the electronic device can be directly or indirectly connected via wired or wireless communication, and the embodiments of the present application are not limited here.

[0066] In one feasible manner, the terminal device can implement the visual data processing method based on the large language model offline; in another feasible manner, the user inputs a question text through the terminal device, the terminal device sends the question text to the server, and the server implements the visual data processing method based on the large language model based on the question text to obtain the question and answer results, and feeds back to the terminal device.

[0067] It is understandable that the above is only an example and is not limited to this embodiment.

[0068] The present application embodiment provides a method for processing visual data based on a large language model. Figure 2 As shown, the method provided in the embodiment of the present application can be executed by an electronic device, and the electronic device is a server, or a terminal device. The method includes:

[0069] S101, obtaining a user's question text;

[0070] In the embodiment of the present application, the user may ask questions through the question-and-answer platform or the question-and-answer application to obtain the answer information in the smart investment advisory scenario. For example, the user may enter "compare the net cash flow of company a and company b" or "the net profit of company c" or "compare the net profit and R&D expenses of multiple companies" or "the pharmaceutical rating of company d".

[0071] Correspondingly, after receiving the question input by the user, the electronic device will execute the question-answering service corresponding to the question, obtain the question-answering result, and return it to the user. In the embodiment of the present application, the user can send the question text to the electronic device by keyboard typing, voice conversion, image recognition, etc., which is not limited here.

[0072] For example, a user asked about the support and resistance levels of company A, MACD (Moving Average Convergence Divergence), KDJ, and BOLL (Bollinger Bands) at 04:40:19 on Monday, 2024-08-12, before the market opened. At this time, the large language model can infer the data of the previous trading day (last Friday).

[0073] In one possible scenario, in order to improve accuracy, the user's question words can also include background knowledge to facilitate the large language model to understand the problem more deeply. For example, the question words are "Take market chart reasoning as an example: it can be based on whether the current time is a trading day and the opening time. When the user analyzes the market indicators and technical indicators of a single trading day, the corresponding market chart can be presented to the user. Market chart usage: used to analyze market indicators (such as closing price, trading volume, etc.) and technical indicators (such as MACD, KDJ, moving average, support level, resistance level, etc.) of a single trading day. The start date and the end date should be the same, representing a specific trading day. This chart is very suitable for intraday analysis and short-term technical insights. Market chart trading day data determination rules: For the analysis data of daily trading frequency, when inferring the specific end date, the opening status of the market should be considered according to the current time. The market is usually open for trading from 09:30 am to 16:00 pm on the trading day. If the market has not opened or is a non-trading day, the end date should be adjusted to the previous trading day. Among them, analyze the support level, pressure level, MACD, KDJ, and BOLL of company a".

[0074] In another possible scenario, the user's initial question text is obtained; the keywords of the initial question text are analyzed through the large language model; and then reasoning is performed based on the keywords to split the initial question text into question texts with at least two tasks. Specifically, after obtaining the user's initial question text, the large language model is used to conduct an in-depth analysis of this text, because keywords are the key to understanding user intent and splitting tasks. Once the keywords are identified, the large language model will use reasoning capabilities to split the initial question text into question texts with at least two (or even more) independent tasks based on the keywords and the relationship between them. The large language model can handle these tasks separately to provide users with more accurate and efficient services.

[0075] S102, obtaining a large language model; the large language model is pre-set with role information in the smart investment advisory scenario, so that the user can interact with the role corresponding to the role information in the large language model and the user;

[0076] The embodiment of the present application does not limit the large language model, and it can be trained using corresponding training samples according to financial analysis tasks.

[0077] For the large language model, you can create your own personalized role dialogue robot by providing relevant background information. In the embodiment of the present application, the role information in the smart investment advisory scenario is used to create a financial analyst role, so that the large language model can interact with the user in the role of a financial analyst and provide users with insightful and meaningful analysis; the large language model is used to infer the visualization plan based on the user's question and context, and the visualization plan mainly includes the selected chart and the data information collected by the chart usage instructions.

[0078] Role information includes but is not limited to: roles and tasks.

[0079] S103, inputting the question text into the large language model, classifying the question text through the large language model, and obtaining a question category; when the question category is a diagnosis category, determining data information and chart information according to the question text, and rendering a chart according to the data information and the chart information, generating a question and answer result, and the question and answer result includes a rendered chart;

[0080] In an embodiment of the present application, the data information is the data information required to analyze the chart, and the chart information includes the chart type, chart title, chart name, and information of each parameter in the chart.

[0081] The large language model will determine the data and chart information based on the question text, the model's internal language knowledge, and the generation strategy when the question category is a diagnostic category and the generation strategy when the question category is a non-diagnostic category, and then obtain the rendered chart and reply information to form the question and answer result. It will select the most appropriate chart and reply language to construct the question and answer result and generate an accurate reply.

[0082] Among them, the analysis logic corresponding to different types of questions is quite different. For example, for diagnostic questions, such as stock diagnosis questions, the required charts include bar charts, line charts, pie charts, etc., and the data information is the information related to the target in the question, such as stocks, indicators and time; while, for non-diagnostic questions, such as stock selection questions, the chart is a list, and the data information is the screening conditions. Therefore, in an embodiment of the present application, after obtaining the question text, the input question text can be classified and labeled using a large language model. Specifically, different categories of questions have different keywords, and the large language model is used to extract the key features of the question text; then, based on the key features, the questions are classified to determine the question category.

[0083] When the question category is not a diagnostic category, the large language model can be used to directly determine the second target data corresponding to the question text from the database based on the question text as the question and answer result; wherein the second target data may be simply data excluding charts, or may include charts, which is no longer limited in the embodiments of the present application.

[0084] When the question category is a diagnostic category, the data information and chart information are determined according to the question text through a large language model, and chart rendering is performed based on the data information and chart information to generate a question and answer result, which includes a rendered chart.

[0085] In the large model processing process, data information and chart information are obtained from the question to data stage, and then, in the data to chart rendering stage, the first target data can be obtained based on the data information, and the chart rendering is performed based on the first target data and the chart information to generate the question and answer results. At this time, the question and answer results include not only charts but also simple text analysis. It should be noted that the target data can be read from the knowledge graph pre-stored in the electronic device. The knowledge graph can be a knowledge graph constructed based on information in a database in the financial field. Of course, it can also be directly obtained from other credit websites, and the embodiments of this application are not limited.

[0086] S104: Send the question and answer result to the client device corresponding to the user, so as to visually display the question and answer result on the client device.

[0087] It can be seen that the big language model used in the embodiment of the present application is pre-set with role information in the smart investment advisory scenario, so that the user can interact with the role corresponding to the role information in the big language model and the user; the question text is input into the big language model, and the question text is classified by the big language model to obtain the question category; when the question category is a diagnostic category, the data information and chart information are determined according to the question text, and the chart is rendered according to the data information and the chart information to generate a question and answer result. The question and answer result includes a rendered chart, and the question type can be determined according to the user's question text through the big language model. When the question type is a diagnostic question, the chart information and the data information presented in the chart can be inferred based on the question text through the big language model, and then the chart rendering is completed based on the chart information and the data information, and the user's question can be provided with a question and answer result including a visual chart. The question and answer result is more accurate and more in line with the needs of the user.

[0088] In a possible implementation of the embodiment of the present application, determining data information and chart information according to the question text in S103 includes:

[0089] Determine data information according to the question text, the data information including: basic target information, time period information and indicator information corresponding to at least one target;

[0090] Determine the chart information based on the data information.

[0091] In the embodiment of the present application, the data information can be determined according to the question text through the large language model, and the data information is used as a screening condition to read the corresponding first target data from the database. Among them, the target refers to the enterprise or stock code, the time period refers to the start and end time, and the indicator information includes net profit, net profit year-on-year, price-earnings ratio and other information.

[0092] Through the large language model, chart information can be determined based on data information.

[0093] For example, the data information includes: company A, the time period information is four quarters, and the indicator information is net profit and year-on-year net profit; the basic composition of the chart information is determined according to the data information, such as the chart type, analysis frequency, chart title, chart name, and other parameters in the chart; then, according to the data information, the corresponding data is read from the database, and finally the question and answer results are formed, as shown in Table 1:

[0094] Table 1 Company a's 4 quarterly net profit and year-on-year data

[0095]

[0096] It can be seen that in the embodiment of the present application, the large language model can analyze the user's intention based on the question text, and obtain data information for collecting data from the database, including target basic information, time period information and indicator information corresponding to at least one target. Appropriate chart information can be determined based on the data information to realize reasoning of the visualization plan.

[0097] A possible implementation of the embodiment of the present application is to determine chart information according to data information, including:

[0098] Determine the target chart type based on the data information;

[0099] When the target chart type is a general chart, the chart information is determined according to the chart requirements and data information corresponding to the target chart type;

[0100] When the target chart type is a high-definition chart, the target chart is determined from multiple preset charts based on the question text, and the parameter information of the target chart is used as the chart information.

[0101] The large language model has the ability to select charts and reason about data, and can be used to determine the target chart type based on data information.

[0102] In the embodiment of the present application, the chart types are divided into two categories: general charts and advanced charts. The data in the tables corresponding to the two types of charts are obtained in different ways. The general chart completes the acquisition of all the first target data from the database based on the inferred data information and the chart requirements corresponding to the target chart type, and then the data can be displayed using a bar chart, a line chart or a pie chart.

[0103] A high-definition chart is a scenario in which a large language model determines, based on the question text, that the data in the database cannot be directly used, but the data needs to be processed and analyzed. Then, the language model uses multiple pre-stored preset charts to automatically analyze the optimal target chart, and uses the parameter information of the target chart as the chart information. The parameter information of the target chart is no longer limited in this application, as long as the target chart can be located through the parameter information of the target chart.

[0104] It can be understood that each preset chart corresponds to a type of question and answer. The large language model can automatically locate the target chart through the question text. After determining the target chart, it automatically obtains the title, name and other information of the chart based on the correspondence between the pre-stored charts and chart information.

[0105] It should be noted that in the embodiment of the present application, a chart background that can be understood by the large language model is pre-constructed, including the usage and parameters of each chart. The large language model selects a chart suitable for user question analysis by understanding the chart usage and related rules, and then infers data information that conforms to both financial logic and chart requirements based on the chart parameters and data rules.

[0106] It can be seen that in the embodiments of the present application, the large language model can infer the target chart type that meets the user's needs based on the data information and classify the charts; for the case where the parameters in the chart can be determined directly from the first target data, the chart type is determined as a general chart; for the case where the first target data needs to be analyzed and the analyzed data is used as the parameter value in the chart, the chart type is determined as a high-definition chart; the solution provided by the present application is highly applicable and can meet a variety of situations.

[0107] A possible implementation of the embodiment of the present application is to render a chart according to the data information and the chart information to generate a question-and-answer result, including:

[0108] According to the data information, first target data corresponding to the data information is read from a database;

[0109] When the target chart type is a general chart, feature analysis is performed on the first target data to obtain basic data features and derived data features;

[0110] According to the basic data features and derived data features, fill in the code block of the general chart corresponding to the target chart type to achieve chart rendering and generate question and answer results; the code block of the general chart corresponding to the target chart type is determined based on the chart information;

[0111] When the target chart type is a high-definition chart, data analysis is performed on the first target data according to the parameter information of the target chart to obtain a plurality of chart data corresponding to the parameter information;

[0112] Locate the target chart according to the chart information, and fill in multiple chart data according to the target chart to obtain the question and answer results.

[0113] In the embodiment of the present application, first target data corresponding to the data information is read from a database according to the data information.

[0114] If the question type is an analysis type, and the chart type is a general chart, feature analysis can be performed on the acquired first target data to obtain basic data features and derived data features.

[0115] Basic data features: used to describe the single and multiple features of the subject, time and indicators, which can be represented by N (Nominal), O (Ordinal) and I (Interval). N=0 represents a single subject, N=1 represents multiple subjects; O=0 represents a single time point, O=1 represents multiple time points; I=1 represents one indicator. The basic features of the data will determine the range of available charts. For example, for data with N=0, O=1, and I=1, it can be presented using line charts, bar charts, and pie charts.

[0116] Derived data features: Describes more detailed features of the subject, time and indicators. For example, for Ordinal data, there are data volume and time type (annual, quarterly, monthly, daily, etc.). The derived features of data can determine more appropriate charts. For example, for a particularly large amount of data (for example, more than 100 points), it is more appropriate to use a line chart to express trend changes compared to a bar chart or a pie chart.

[0117] In an embodiment of the present application, the acquired data information is input into a data wizard, and the first target data is read through a database such as Pandas and related query functions, and then input into a large language model to complete the summary of feature information.

[0118] Continuing with the above example, taking the net profit and year-on-year growth data of company A in the past four quarters as an example, Pandas is used to read the data information and output the data information: 4 time points, 1 stock, 2 indicators, 2 units, and the minimum order of magnitude between indicators is 2523. Based on the output data information, feature conversion is performed to obtain basic data features and derived features, among which, basic data features: ; Derived data characteristics: number of indicator units = 2; the order of magnitude between indicators is 2523M.

[0119] In the embodiment of the present application, when the target chart type is a general chart, a code block related to a data configuration item is additionally encapsulated for each chart, and the large language model fills the data fields corresponding to the basic data features and derived data features into the corresponding code area through automatic configuration to complete the automatic rendering of the chart. Taking the column chart as an example, the automatic configuration of the general chart: Basic data feature requirements: ; Derived data feature requirements: number of indicator units = 2 or minimum order of magnitude (multiple) of the indicator > 10; Configuration rules: O data attributes automatically fill in the xAxis Attribute slot of the chart; I data attributes automatically fill in the barAxis Attribute and lineAxis Attribute slots of the chart to achieve data rendering.

[0120] In an embodiment of the present application, when the target chart type is a high-definition chart, after obtaining the first target data, the configuration platform is directly requested according to the parameter information of the target chart, and the first target data is analyzed according to the parameter requirements of the target chart to obtain the analysis result, i.e., the chart data, and the analysis result is entered as a parameter, and the rendering of the material is completed after filling.

[0121] It can be seen that in the embodiment of the present application, according to the data information, the first target data corresponding to the data information is read from the database; when the target chart type is a general chart, the features are directly classified and then filled into the code block of the general chart corresponding to the target chart type to realize the rendering of the chart; when the target chart type is a high-definition chart, it is necessary to analyze it first and then fill the analysis results into the high-definition chart.

[0122] A possible implementation of the embodiment of the present application includes, before sending the question-and-answer result to the client device corresponding to the user, further comprising:

[0123] Extract key features of question-answering results;

[0124] Determine the matching degree between key features and data information;

[0125] If the matching degree is greater than a preset degree threshold, executing the step of sending the question and answer result to the client device corresponding to the user;

[0126] If the matching degree is not greater than the preset degree threshold, the question text is re-input into the large language model until the matching degree is greater than the preset degree threshold, or the number of repetitions reaches the preset number threshold.

[0127] In order to ensure the accuracy of the question and answer results, in an embodiment of the present application, after obtaining the question and answer results, key features that can summarize or represent the main content of the entire answer are extracted from the question and answer results; the extracted key features are matched with the key data; when the degree of match is greater than the preset degree threshold, it indicates that the question and answer result is an answer that is highly relevant to the demand; if the degree of match is not greater than the preset degree threshold, the question text will be re-input into the large language model for processing; through multiple iterations and repeated processing, the accuracy and reliability of the question and answer system can be gradually improved to ensure that users obtain high-quality answers.

[0128] Based on any of the above embodiments, the present application embodiment provides a specific large model visualization analysis logical reasoning method, see Figure 3, involving the question-to-data stage and the data-to-chart rendering stage, both of which are implemented based on the large language model. Among them, the question-to-data stage: based on the large language model, infer the logic suitable for the question analysis based on the user's question: what graph to use to present what data, and output it in a JSON format that is easy to parse; data-to-chart rendering stage: parse the output JSON based on the large language model, and call the data interface and the chart interface to complete the final rendering of the chart. In an embodiment of the present application, a data reasoning method that combines a large model with visualization technology is a technology that processes chart information and combines large model COT reasoning (thinking chain reasoning) to supplement the data information parsed in the user's question, thereby realizing the characterization of the analysis logic of the user's question as what graph to use to present what data and the current question type.

[0129] Specifically, in the question-to-data stage:

[0130] (1) Role setting

[0131] Assume that a financial analyst can provide insightful and meaningful analysis to investors. LLM uses the user's question and context to infer the visualization plan for analysis. The visualization plan mainly includes the selected chart and the data information collected by the chart's instructions.

[0132] Exemplarily, the role information may be:

[0133] “Role setting and mission description:

[0134] You are a financial analyst who can provide deep and meaningful analysis to investors. Therefore, you should infer a visualization scheme through a large language model (LLM) based on the query and context to provide analysis. The visualization scheme mainly includes the chart you choose and the data that should be collected based on the chart usage. Therefore, this task mainly includes three steps. First, you should classify the query type into screeners and analyzers. The following are the main characteristics of these two types of queries:

[0135] 1. For Analyzer queries, investors seek detailed insights into a stock or company's financial performance or trends, often focusing on specific financial metrics or time patterns.

[0136] 2. For filter queries, investors filter stocks based on predefined criteria, such as industry or financial metrics, to identify stocks that match their investment preferences.

[0137] Secondly, you should choose the appropriate chart to provide visualization for investors based on the query type and the purpose presented in the chart information. Thirdly, based on the chart you choose, you should infer the appropriate indicators and corresponding key attributes to collect data from the database. It is worth noting that generated charts and functional charts have different properties. Throughout the process, you do not need to directly provide specific charts and their data, but only need to outline your visualization scheme and the data required for each chart.

[0138] In order to complete the task better, you must strictly follow the following rules:

[0139] 1. Chart requirements:

[0140] 1. The inferred visualization solution should be based on the query and its type.

[0141] 2. The inferred visualization solution must make financial sense and provide some insights.

[0142] 3. The inferred visualization scheme should consider comparative relationships, such as industry comparison, year-on-year (YoY) comparison, etc., to enrich the visualization scheme.

[0143] 4. The number of charts in the inferred visualization solution shall not exceed 5.

[0144] 5. For each chart, you should specify its type as a generated chart or a functional chart based on the chart information.

[0145] 6. Generative charts and functional charts should complement each other.

[0146] 7. Each graph derived must be the one presented in the graph information.

[0147] 8. For analyzer queries, each chart should specify the frequency of the data being analyzed, such as daily transactions, quarterly, or annual.

[0148] 9. For analyzer queries, each chart should have a short, meaningful, and descriptive title.

[0149] 10. The title length should be within 10 words.

[0150] 11. The chart type and chart name should be displayed.

[0151] 12. Query types should be displayed to distinguish different solutions.

[0152] 2. Data requirements:

[0153] 1. Based on the inferred visualization scheme and query type, you should infer the required metrics for each generated chart queried by the analyzer.

[0154] 2. For indicators, the chart information for generating charts should be based on inference.

[0155] 3. For each inferred metric, you must infer appropriate start and end dates based on the purpose of the chart as presented in the chart information, the frequency of the data being analyzed, the most recent reporting period for each stock, and the current time.

[0156] 4. For analytical data of different frequencies, the time periods covered by the start and end dates should be different.

[0157] 5. For the data analysis of daily trading frequency, when inferring a specific end date, you should consider the market's opening status based on the current time. Usually the market's trading hours are from 9:30 am to 4:00 pm on the trading day. If the market has not opened on that day or it is a non-trading day, the end date should be adjusted to the previous trading day.

[0158] 6. For indicators about growth rate, you must provide the name of the original indicator and indicate the corresponding operator, such as YoY, QoQ. For example, if the data you want to collect is the year-on-year growth rate of net income, the original indicator is net income and the operator is YoY.

[0159] 7. When inferring the stock code, you should refer to the content in the code information. This helps to infer the target's code as well as the competitor's code.

[0160] 8. For functional charts, you only need to infer the start date, end date, and stock symbol, and the indicator name defaults to the default value.

[0161] 9. For best visualization, you should adjust the start and end date ranges based on the device type.

[0162] 10. For filter queries, you only need to display the queries raised by investors.

[0163] 3. General requirements:

[0164] 1. The output should be presented in JSON format.

[0165] 2. For visualization schemes, the following keys are mainly included: query type and chart.

[0166] 3. For analyzer queries, the chart mainly includes the following keys: Chart Name, Chart Type, Title, Analysis Frequency, and Data.

[0167] 4. For filter queries, the chart mainly includes the following keys: Chart Name, Chart Type, and Data.

[0168] 5. For analyzer queries, the data is a list where each object includes five keys: metric name, operator, start date, end date, and stock symbol.

[0169] 6. Operator keys in the data should be included only when growth rate data is needed.

[0170] 7. Stock symbols is a list where each object includes two keys: symbol and symbol type.

[0171] 8. The code type includes different types, such as Stock, ETF, Index and Crypto.

[0172] 9. For filter queries, the data contains only one key, the query.

[0173] 10. Queries are inquiries raised by investors.

[0174] 4. Example of analysis logic output format:

[0175] json

[0176] Analyzer query:

[0177] {"Query Type":"Analyzer","Chart":[{"Chart Name":"<Chart Name>","Chart Type":"Generate Chart","Title":"<Descriptive Title>","Analysis Frequency":"<Data Analysis Frequency>","Data":[{"Indicator Name":"<Financial Indicator>","Start Date":"<Start Date of Data Collection>","End Date":"<End Date of Data Collection>","Stock Symbol":[{"Symbol":"<Stock Symbol>","Symbol Type":"<Symbol Type>"}]},{"Indicator Name":"<Another Financial Indicator>","Operator":"<Required Operator>","Start Date":"<Start Date of Data Collection>","End Date":"<End Date of Data Collection>","Stock Symbol":[{"Symbol":"<Stock Symbol>","Symbol Type":"<Symbol Type (e.g., Stock, ETF, Index)>"},{"Symbol":"<Another Stock Symbol>","Symbol Type":"<Symbol Type>"}]}]}]}

[0178] Filter query:

[0179] {"Query Type":"Filter","Chart":[{"Chart Name":"<Chart Name>","Chart Type":"Generate Chart","Data":[{"Query":"<Query raised by investors>"}]}]}".

[0180] (2) Tasks.

[0181] Question classification, chart selection, and data reasoning.

[0182] Among them, the question is input into the large language model, and through the large language model for category analysis, the question category is obtained.

[0183] Moreover, a chart background that can be understood by the large language model is constructed, including the usage and parameters of each chart. The large language model selects a chart suitable for user question analysis by understanding the chart usage and related Rules, and then infers data information that conforms to both financial logic and chart requirements based on the chart parameters and data Rules.

[0184] The company background that can be understood by the large language model includes company information, fiscal year and quarter, and the current background includes the current time and Rules for the large model to learn. These two parts of content are used as supplementary inputs to assist the large model in understanding the company and the chart, so that the large language model can obtain chart information and data information based on the question.

[0185] (3)Model output data definition.

[0186] The analysis logic of the user question is characterized as what data to present with what chart and the current question type.

[0187] Specifically, see Figure 4 , Figure 4 , which is a schematic diagram for characterizing the analysis logic provided by the embodiment of the present application.

[0188] User side: User question, what to view;

[0189] Electronic device side: The analysis logic includes:

[0190] Determine whether the question type is a diagnostic question (diagnosis) or a non-diagnostic question (selection);

[0191] Determine data information: code type, target code, start time, end time, indicator name;

[0192] Determine chart information: chart type, analysis frequency, chart title, chart name.

[0193] Exemplarily, the analysis logic corresponding to the user question "Microsoft's price-earnings ratio in the past 10 years" can be characterized as showing Microsoft's price-earnings ratio in the past 10 years with a line chart and generating corresponding JSON data. The engineering parses the JSON content of the model data and requests the data and chart interfaces respectively, and through automatic configuration, completes the rendering of the final chart:

[0194] {

[0195] "Query Type": "Analyzer", question type

[0196] "Charts": [

[0197] {

[0198] "Chart Name": "line",chart name

[0199] "Chart Type": "Generative Chart",

[0200] "Title": "Microsoft P / E Ratio Over the Past 10 Years",

[0201] "Analysis Frequency": "Trade Daily",

[0202] "Data": [

[0203] {

[0204] "Indicator Name": "P / E(TTM)",Indicator name

[0205] "Start Date": "2014-08-10", start time

[0206] "End Date": "2024-08-10", end time

[0207] "Ticker": [

[0208] {

[0209] "Code": "MSFT", target code

[0210] "Code Type": "Stock"

[0211] }]]}]}

[0212] (4) Data characteristics.

[0213] See also Figure 5 , input the acquired data into the data wizard, complete the reading of data information through Pandas and related query functions, input it into the model, and complete the summary of feature information.

[0214] Specifically, if the question type is a diagnostic question and the chart type is a general chart, it is necessary to perform feature analysis on the first target data. The data features are mainly divided into two categories:

[0215] Basic data features: used to describe the single and multiple features of the target, time and indicator, represented by N (Nominal), O (Ordinal) and I (Interval). N=0 represents a single target, N=1 represents multiple targets; O=0 represents a single time point, O=1 represents multiple time points; I=1 represents one indicator. The basic features of the data will determine the range of available charts. For example, for data with N=0, O=1, and I=1, it can be presented using a line chart, a bar chart, or a pie chart.

[0216] Derived data features: Describes more detailed features of the subject, time and indicators. For example, for Ordinal data, there are data volume and time type (annual, quarterly, monthly, daily, etc.). The derived features of data can determine more appropriate charts. For example, for a particularly large amount of data (for example, more than 100 points), it is more appropriate to use a line chart to express trend changes compared to a bar chart or a pie chart.

[0217] Taking NVIDIA's net profit and year-on-year growth data for the past four quarters as an example, the data is read through Pandas and the output data information is: 4 time points, 1 stock, 2 indicators, 2 units, and the minimum order of magnitude between indicators is 2523. The project performs feature conversion based on the output data information to obtain basic data features and derived features: Basic data features: N=0, I=2, O=1; Derived data features: number of indicator units = 2; The order of magnitude between indicators is 2523M.

[0218] (5) Automated configuration.

[0219] General chart automatic configuration: Each chart additionally encapsulates a code block related to the data configuration item, and fills the data field into the corresponding code area through automatic configuration to complete the automatic rendering of the chart. Taking the column chart as an example, the automatic configuration of the general chart: basic data feature requirements: N=0, O=1, I>=2; derived data feature requirements: number of indicator units = 2 or minimum order of magnitude (multiple) of the indicator >10; configuration rules: O data attributes automatically fill in the chart xAxis Attribute slot; I data attributes automatically fill in the chart barAxis Attribute, lineAxis Attribute slot.

[0220] Automatic configuration of high-definition charts: If the obtained chart type is a high-definition chart type, directly request the configuration platform after obtaining the data. The chart name corresponds to the material ID of the configuration platform, and the data part corresponds to the input parameters of the material. After filling, the material rendering is completed.

[0221] It can be seen that the method provided in the embodiment of the present application has the ability of time reasoning, indicator relationship reasoning, comparison object reasoning and high-definition result page reasoning, which greatly improves the visualization experience. After testing, the output rate has reached 85%.

[0222] The following is an introduction to a visualization data processing device based on a large language model provided in an embodiment of the present application. The visualization data processing device based on a large language model described below and the visualization data processing method based on a large language model described above can refer to each other. The visualization data processing device based on a large language model in this embodiment is set in an electronic device, and the visualization data processing method based on a large language model described above can refer to each other. Figure 6 , Figure 6 : is a structural block diagram of a visual data processing device 200 based on a large language model according to one embodiment of the present application, comprising:

[0223] The first acquisition module 210 is used to acquire the user's question text;

[0224] A second acquisition module 220 is used to acquire a large language model, in which role information in a smart investment advisory scenario is pre-set, so that the user can interact with the role corresponding to the role information in the large language model;

[0225] The question-answering module 230 is used to input the question text into the large language model, classify the question text through the large language model, and obtain the question category; when the question category is a diagnosis category, determine the data information and chart information according to the question text, and render the chart according to the data information and the chart information to generate the question-answering result, which includes the rendered chart;

[0226] The sending module 240 is used to send the question and answer results to the client device corresponding to the user, so as to visually display the question and answer results on the client device.

[0227] In one practicable manner, the question-answering module 230 is used to:

[0228] Determine data information according to the question text, the data information including: basic target information, time period information and indicator information corresponding to at least one target;

[0229] Determine the chart information based on the data information.

[0230] In one practicable manner, the question-answering module 230 is used to:

[0231] Determine the target chart type based on the data information;

[0232] When the target chart type is a general chart, the chart information is determined according to the chart requirements and data information corresponding to the target chart type;

[0233] When the target chart type is a high-definition chart, the target chart is determined from multiple preset charts based on the question text, and the parameter information of the target chart is used as the chart information.

[0234] In one practicable manner, the question-answering module 230 is used to:

[0235] According to the data information, first target data corresponding to the data information is read from a database;

[0236] When the target chart type is a general chart, feature analysis is performed on the first target data to obtain basic data features and derived data features;

[0237] According to the basic data features and derived data features, fill in the code block of the general chart corresponding to the target chart type to achieve chart rendering and generate question and answer results; the code block of the general chart corresponding to the target chart type is determined based on the chart information;

[0238] When the target chart type is a high-definition chart, data analysis is performed on the first target data according to the parameter information of the target chart to obtain a plurality of chart data corresponding to the parameter information;

[0239] Locate the target chart according to the chart information, and fill in multiple chart data according to the target chart to obtain the question and answer results.

[0240] In an implementable manner, the first acquisition module 210 is used to:

[0241] Get the user's initial question text;

[0242] Analyze the keywords of the initial question text;

[0243] Reasoning is performed based on keywords to split the initial question text into question texts with at least two tasks.

[0244] In one achievable manner, the present invention further includes:

[0245] Extraction module, used to extract key features of question-answering results;

[0246] A matching module, used to determine the matching degree between key features and data information;

[0247] A first execution module, configured to execute the step of sending the question-answer result to a client device corresponding to the user if the matching degree is greater than a preset degree threshold;

[0248] The second execution module is used to re-input the question text into the large language model if the matching degree is not greater than the preset degree threshold, until the matching degree is greater than the preset degree threshold, or the number of repetitions reaches the preset number threshold.

[0249] In one achievable manner, the question-answering module is further configured to determine, when the question category is not a diagnosis category, second target data corresponding to the question text from a database as a question-answering result based on the question text.

[0250] An electronic device is provided in an embodiment of the present application, such as Figure 7 As shown, Figure 7 The electronic device 300 shown includes: a processor 301 and a memory 303. The processor 301 and the memory 303 are connected, such as through a bus 302. Optionally, the electronic device 300 may also include a transceiver 304. It should be noted that in actual applications, the transceiver 304 is not limited to one, and the structure of the electronic device 300 does not constitute a limitation on the embodiments of the present application.

[0251] The processor 301 may be a CPU (Central Processing Unit), a general-purpose processor, a DSP (Digital Signal Processor), an ASIC (Application Specific Integrated Circuit), an FPGA (Field Programmable Gate Array) or other programmable logic devices, transistor logic devices, hardware components or any combination thereof. It may implement or execute various exemplary logic blocks, modules and circuits described in conjunction with the disclosure of this application. The processor 301 may also be a combination that implements computing functions, such as a combination of one or more microprocessors, a combination of a DSP and a microprocessor, etc.

[0252] The bus 302 may include a path to transmit information between the above components. The bus 302 may be a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus. The bus 302 may be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 7 Only one thick line is used in the diagram, but this does not mean that there is only one bus or only one type of bus.

[0253] The memory 303 may be a ROM (Read Only Memory) or other types of static storage devices that can store static information and instructions, a RAM (Random Access Memory) or other types of dynamic storage devices that can store information and instructions, or an EEPROM (Electrically Erasable Programmable Read Only Memory), a CD-ROM (Compact Disc Read Only Memory) or other optical disk storage, optical disk storage (including compressed optical disk, laser disk, optical disk, digital versatile disk, Blu-ray disk, etc.), a magnetic disk storage medium or other magnetic storage device, or any other medium that can be used to carry or store the desired program code in the form of instructions or data structures and can be accessed by a computer, but is not limited thereto.

[0254] The memory 303 is used to store the application code for executing the solution of the present application, and the execution is controlled by the processor 301. The processor 301 is used to execute the application code stored in the memory 303 to implement the contents shown in the above method embodiment.

[0255] Figure 7 The electronic device shown is merely an example and should not bring any limitation to the functions and scope of use of the embodiments of the present application.

[0256] An embodiment of the present application provides a computer-readable storage medium, on which a computer program is stored. When the computer-readable storage medium is run on a computer, the computer can execute the corresponding content in the aforementioned method embodiment.

[0257] An embodiment of the present application provides a computer program product, including a computer program, which implements the corresponding contents of the aforementioned method embodiment when the computer program is executed by a processor.

[0258] It should be understood that, although the steps in the flowchart of the accompanying drawings are displayed in sequence as indicated by the arrows, these steps are not necessarily executed in sequence in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least a part of the steps in the flowchart of the accompanying drawings may include multiple sub-steps or multiple stages, and these sub-steps or stages are not necessarily executed at the same time, but can be executed at different times, and their execution order is not necessarily sequential, but can be executed in turn or alternately with other steps or at least a part of the sub-steps or stages of other steps.

[0259] The above are only some implementation methods of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.

Claims

1. A visual data processing method based on a large language model, characterized in that: include: Get the user's question text; Obtaining a large language model, wherein the large language model is pre-set with role information in a smart investment advisory scenario, so that the user can interact with the role corresponding to the role information in the large language model and the user; Inputting the question text into the large language model, and classifying the question text by using the large language model to obtain a question category; When the question category is a diagnosis category, determining data information and chart information according to the question text, and rendering a chart according to the data information and the chart information to generate a question-and-answer result, wherein the question-and-answer result includes a rendered chart; The question-and-answer result is sent to the client device corresponding to the user, so that the question-and-answer result is visually displayed on the client device.

2. The visual data processing method based on a large language model according to claim 1, characterized in that: Determine data information and chart information according to the question text, including: Determine the data information according to the question text, the data information including: target basic information, time period information and indicator information corresponding to at least one target; According to the data information, chart information is determined.

3. The visual data processing method based on a large language model according to claim 2 is characterized in that: Determining chart information according to the data information includes: Determine a target chart type according to the data information; When the target chart type is a general chart, determining chart information according to chart requirements and data information corresponding to the target chart type; When the target chart type is a high-definition chart, the target chart is determined from a plurality of preset charts according to the question text, and parameter information of the target chart is used as chart information.

4. The visual data processing method based on a large language model according to claim 3 is characterized in that: Rendering a chart according to the data information and the chart information to generate a question-and-answer result includes: According to the data information, reading first target data corresponding to the data information from a database; When the target chart type is a general chart, performing feature analysis on the first target data to obtain basic data features and derived data features; Fill in the code block of the general chart corresponding to the target chart type according to the basic data features and the derived data features, realize chart rendering, and generate question and answer results, wherein the code block of the general chart corresponding to the target chart type is determined based on the chart information; When the target chart type is a high-definition chart, performing data analysis on the first target data according to parameter information of the target chart to obtain a plurality of chart data corresponding to the parameter information; The target chart is located according to the chart information, and the multiple chart data are filled in according to the target chart to obtain the question and answer result.

5. The visual data processing method based on a large language model according to claim 2 is characterized in that: Get the user's question text, including: Get the user's initial question text; Analyzing keywords of the initial question text; Reasoning is performed based on the keywords to split the initial question text into question texts with at least two tasks.

6. The visual data processing method based on a large language model according to claim 2 is characterized in that: Before sending the question and answer result to the user's corresponding client device, it also includes: Extracting key features of the question and answer results; Determining the degree of match between the key feature and the data information; If the matching degree is greater than a preset degree threshold, executing the step of sending the question and answer result to the client device corresponding to the user; If the matching degree is not greater than the preset degree threshold, the question text is re-input into the large language model until the matching degree is greater than the preset degree threshold, or the number of repetitions reaches the preset number threshold.

7. The visual data processing method based on a large language model according to claim 1 is characterized in that: Also includes: When the question category is not a diagnosis category, second target data corresponding to the question text is determined from a database according to the question text as a question-answering result.

8. A visual data processing device based on a large language model, characterized in that: include: The first acquisition module is used to acquire the user's question text; A second acquisition module is used to acquire a large language model, in which role information in a smart investment advisory scenario is pre-set, so that the user can interact with the role corresponding to the role information in the large language model and the user; A question-answering module, used for inputting the question text into the large language model, and classifying the question text through the large language model to obtain a question category; When the question category is a diagnosis category, determining data information and chart information according to the question text, and rendering a chart according to the data information and the chart information to generate a question-and-answer result, wherein the question-and-answer result includes a rendered chart; The sending module is used to send the question and answer results to the client device corresponding to the user, so as to visually display the question and answer results on the client device.

9. An electronic device, characterized in that: include: one or more processors; Memory; One or more applications, wherein the one or more applications are stored in the memory and are configured to be executed by the one or more processors, and the one or more applications are configured to: execute the steps of the visual data processing method based on a large language model according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The storage medium stores at least one instruction, at least one program, a code set or an instruction set, and the at least one instruction, at least one program, a code set or an instruction set is loaded by a processor and executes the steps of the visual data processing method based on a large language model according to any one of claims 1 to 7.

Images