Semiconductor data analysis and visualization method and device based on large model

Abstract

The invention provides a semiconductor data analysis and visualization method and device based on a large model, and relates to the technical field of data processing. The method comprises the steps that demand information input by a user is analyzed through a large language model, and key information corresponding to the demand information is obtained; the key information comprises time information, industry information, index information and geographic information; determining a target data set most relevant to the key information from a preset database by utilizing a data set automatic selection algorithm; determining a visualization strategy of the demand information according to the demand information and the target data set; and according to the visualization strategy and the large language model, generating a visualization code corresponding to the visualization strategy, and executing the visualization code to automatically generate a visualization chart. According to the scheme, the problems of automatic analysis and visualization of complex data in the semiconductor feasibility research report are solved, dynamic data display is supported, and the data visualization effect is improved.

Description

Technical Field

[0001] The present application relates to the field of data processing technology, and in particular to a semiconductor data analysis and visualization method and device based on a large model. Background Art

[0002] In the semiconductor industry, data comes from a variety of sources, including market dynamics, manufacturing capacity, capital expenditures, etc. Existing technologies usually rely on manual data analysis and static report generation, which is not only time-consuming but also prone to errors. Traditional solutions mainly include manual data analysis using spreadsheets and static chart generation using basic charting tools. These methods lack automation and intelligence and cannot effectively respond to rapidly changing market demands.

[0003] The existing technology uses spreadsheet software (such as Excel), which cannot process large amounts of data, has cumbersome manual operations, and lacks dynamic analysis capabilities; the existing technology uses business intelligence tools (such as Tableau, Power BI), which have a steep learning curve, complex initial settings, and limited real-time data processing capabilities; the existing technology uses programming languages ​​and data analysis libraries, which require programming skills, take a long time to develop, and require manual editing of charts; the existing technology uses dedicated semiconductor data analysis software, which is expensive, has a limited scope of application, and lacks versatility.

[0004] It can be seen that in the semiconductor industry, the existing methods of generating charts have a low degree of automation, and most solutions rely on manual operations, which are inefficient and prone to errors. They are not intelligent enough and lack the ability to use artificial intelligence for deep data mining and analysis. Summary of the invention

[0005] The purpose of this application is to provide a semiconductor data analysis and visualization method and device based on a large model to solve the problems of low automation and insufficient intelligence in the feasibility study report for semiconductors in the prior art.

[0006] To achieve the above objectives, an embodiment of the present application provides a semiconductor data analysis and visualization method based on a large model, comprising: Utilize the large language model to parse the demand information input by the user and obtain key information corresponding to the demand information; the key information includes time information, industry information, indicator information and geographic information; Using an automatic data set selection algorithm, determine a target data set that is most relevant to the key information from a preset database; Determining a visualization strategy for the demand information according to the demand information and the target data set; According to the visualization strategy and the large language model, a visualization code corresponding to the visualization strategy is generated, and the visualization code is executed to automatically generate a visualization chart.

[0007] Optionally, using an automatic data set selection algorithm to determine a target data set most relevant to the key information from a preset database includes: Initializing each data set in the preset database; each data set includes multiple feature information, and the feature information includes time characteristics, industry characteristics, indicator type, geographical scope and data reliability; Analyzing the key information using an automatic data set selection algorithm, and determining a matching score of the feature information corresponding to the key information in each data set and a reliability score of each data set; According to the matching score of each piece of feature information and the reliability score, weighted sum is performed to calculate the total matching score of each data set; The data set with the highest total matching score is selected from all the data sets to be determined as the target data set.

[0008] Optionally, before determining a visualization strategy for the demand information according to the demand information and the target data set, the method further includes: extracting target data from the target data set; Using the large language model, verifying whether the target data corresponds to the demand information; If the correspondence is inconsistent, reacquire the target data set or supplement the missing corresponding features in the target data set, and re-execute the step of extracting the target data in the target data set; If the correspondence is consistent, a step of determining a visualization strategy for the demand information according to the demand information and the target data set is performed.

[0009] Optionally, determining a visualization strategy for the demand information according to the demand information and the target data set includes: Utilizing the large language model, parsing the demand information, and determining the user intent corresponding to the demand information; Using the large language model, extracting visualization parameters in the target data set; Selecting a visualization chart type and a color scheme based on the user intent and the visualization parameters using the large language model; A visualization strategy for the demand information is determined based on the visualization chart type and color scheme.

[0010] Optionally, generating a visualization code corresponding to the visualization strategy according to the visualization strategy and the large language model, and executing the visualization code to automatically generate a visualization chart includes: Understanding the semantic intent of the visualization strategy to obtain multiple chart attributes and multiple indicators of the visualization chart to be generated; Performing code generation processing according to the plurality of chart attributes, the plurality of indicators and the large language model to dynamically generate visualization code corresponding to the visualization strategy; The visualization code is executed to draw a visualization chart corresponding to the visualization strategy.

[0011] Optionally, after generating a visualization code corresponding to the visualization strategy, the method further includes: Testing the visualization code, and when the test result fails, displaying the reason for the test failure to the user; If the reason for the test failure is a code reason, regenerate a target code for the failed code by using the large language model in combination with the context of the failed code; The visualization code of the entire text is retested. If the test result still fails, a new visualization code is generated and re-verified in response to a code request input by the user.

[0012] Optionally, the method further comprises: Displaying the visualization chart to the user; Acquire interaction data between the user and the visualization chart according to a preset period; Adjusting the visualization strategy according to the interaction data; According to the adjusted visualization strategy, an optimized visualization chart is obtained, and the optimized visualization chart is displayed to the user.

[0013] To achieve the above-mentioned purpose, the embodiment of the present application further provides a semiconductor data analysis and visualization device based on a large model, comprising: The first processing module is used to parse the demand information input by the user using the large language model to obtain key information corresponding to the demand information; the key information includes time information, industry information, indicator information and geographic information; A first determination module is used to determine the target data set most relevant to the key information from a preset database by using an automatic data set selection algorithm; A second determination module, configured to determine a visualization strategy for the demand information according to the demand information and the target data set; The second processing module is used to generate a visualization code corresponding to the visualization strategy according to the visualization strategy and the large language model, and execute the visualization code to automatically generate a visualization chart.

[0014] To achieve the above-mentioned purpose, an embodiment of the present application further provides a readable storage medium on which a program or instruction is stored, and when the program or instruction is executed by a processor, the steps in the method described above are implemented.

[0015] To achieve the above objectives, an embodiment of the present application further provides a computer program product, including computer instructions, which implement the steps of the above method when executed by a processor.

[0016] The beneficial effects of the above technical solution of the present application are as follows: In the embodiment of the present application, a large language model is used to parse the demand information input by the user to obtain the key information corresponding to the demand information; the key information includes time information, industry information, indicator information and geographic information; an automatic data set selection algorithm is used to determine the target data set most relevant to the key information from a preset database; a visualization strategy for the demand information is determined based on the demand information and the target data set; a visualization code corresponding to the visualization strategy is generated based on the visualization strategy and the large language model, and the visualization code is executed to automatically generate a visualization chart. The solution of the present application solves the problem of automated analysis and visualization of complex data in semiconductor feasibility reports, supports dynamic data display, and improves data visualization effects. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 A flowchart of a semiconductor data analysis and visualization method based on a large model provided in an embodiment of the present application; Figure 2 A structural diagram of a semiconductor data analysis and visualization device based on a large model provided in an embodiment of the present application. DETAILED DESCRIPTION

[0018] In order to make the technical problems, technical solutions and advantages to be solved by the present application clearer, a detailed description will be given below with reference to the accompanying drawings and specific embodiments.

[0019] It should be understood that the references to "one embodiment" or "an embodiment" throughout the specification mean that the specific features, structures, or characteristics associated with the embodiment are included in at least one embodiment of the present application. Therefore, the references to "in one embodiment" or "in an embodiment" appearing throughout the specification do not necessarily refer to the same embodiment. In addition, these specific features, structures, or characteristics may be combined in any suitable manner in one or more embodiments.

[0020] In the embodiments of the present application, the term "and / or" describes the association relationship of the associated objects, indicating that three relationships may exist. For example, A and / or B may represent: A exists alone, A and B exist at the same time, and B exists alone. The character " / " generally indicates that the associated objects before and after are in an "or" relationship.

[0021] In the embodiments of the present application, the term "plurality" refers to two or more than two, and other quantifiers are similar.

[0022] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only 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.

[0023] The embodiment of the present application provides a semiconductor data analysis and visualization method and device based on a large model. The method and device are based on the same application concept. Since the method and device solve the problem in a similar way, the implementation of the device and the method can refer to each other, and the repeated parts will not be repeated.

[0024] like Figure 1 As shown, a semiconductor data analysis and visualization method based on a large model is provided in an embodiment of the present application, comprising: Step 11, using a large language model to parse the demand information input by the user, and obtain key information corresponding to the demand information; the key information includes time information, industry information, indicator information and geographic information.

[0025] In this application, the user can input demand information in natural language once and obtain the corresponding visualization chart. This application uses natural language technology to parse the demand information and determine the key information corresponding to the demand information. The key information includes time information, industry information, indicator information and geographic information. For example, the user inputs "need total revenue and investment data of global manufacturer A between 2019 and 2021", and the key information extracted by the natural language technology of the large language model is: time range: 2019 to 2021; industry segmentation: A; indicator type: total revenue, investment (capital expenditure); geographic scope: global.

[0026] For another example, the demand information can also be a description of the chart generation. That is, the demand information is the description text when the object needs to generate a chart. The demand information can be, for example, "help to count the changes in sales and production of XX and XX companies, and give a chart, the specific data is XX".

[0027] To give a specific example, at least one human-computer interaction method can be provided to the user to allow the user to input corresponding requirement information, such as providing the user with an input text box to allow the user to input text-type requirement information, or multiple human-computer interaction methods such as voice input, gesture input, etc.

[0028] Step 12: using an automatic data set selection algorithm to determine a target data set that is most relevant to the key information from a preset database.

[0029] Step 13, determining a visualization strategy for the demand information according to the demand information and the target data set; Step 14: Generate visualization code corresponding to the visualization strategy according to the visualization strategy and the large language model, and execute the visualization code to automatically generate a visualization chart.

[0030] In the embodiment of the present application, the large language model includes a semiconductor data analysis and visualization algorithm (SDAVA), which combines natural language processing, data selection and automatic visualization technology to realize intelligent analysis of semiconductor data: (1) Natural language query processing and data processing: Use the language model to parse user queries, determine the target data set most relevant to the key information from the preset database, and automatically identify the multidimensional characteristics and time series characteristics of the data, handle the diversity of the data, and ensure consistency. (2) Data extraction and preparation: Generate target code to automatically extract and prepare data, perform data cleaning and format conversion, and ensure the integrity and accuracy of the data. (3) Visualization code generation: According to the requirement information and the target data set, determine the visualization strategy of the requirement information, generate visualization code suitable for the visualization strategy, and select the best chart type and style to dynamically display data trends. (4) Execution and image generation: Execute the generated code in a controlled environment, generate and save visualization charts.

[0031] Furthermore, the chart types of the visualization chart include: at least one of a bar chart, a line chart, a scatter chart, a pie chart, a radar chart, a dashboard, a chord chart, a flow chart, a time series chart, a mulberry chart, and a flow map.

[0032] It should be noted that the present application only requires the input of natural language once, and the corresponding target code can be generated using a large language model. The execution of the target code can generate a visual chart, and there is no need for multiple interactions to generate a visual chart.

[0033] Optionally, step 12 of the present application includes: Initializing each data set in the preset database; each data set includes multiple feature information, and the feature information includes time characteristics, industry characteristics, indicator type, geographical scope and data reliability; Analyzing the key information using an automatic data set selection algorithm, and determining a matching score of the feature information corresponding to the key information in each data set and a reliability score of each data set; According to the matching score of each piece of feature information and the reliability score, weighted sum is performed to calculate the total matching score of each data set; The data set with the highest total matching score is selected from all the data sets to be determined as the target data set.

[0034] In the embodiment of the present application, the function of natural language query processing is used to parse the user's natural language query (Q) to select the most relevant target data set. A large language model (M) (such as ChatOpenAI, etc.) can be used to process the query. Initialize each data set in the preset database; each data set includes multiple feature information, and the feature information includes time features, industry features, indicator types, geographical scope and data reliability. Specifically, initialize the data set features: each data set is described by the following features: time range (specific year and month (xx year and xx month), year time period (2020-2024), etc.); industry segmentation (for example, DRAM, NAND, IC, LCD); indicator type (for example, revenue, investment, production capacity, market share); geographical scope (for example, global, m country, n country, etc.); data reliability, ranging from 0 to 1; using the data set automatic selection algorithm to parse the key information, the query string can be parsed: extract the time requirements, industry keywords, indicator keywords and geographical keywords in the query. According to the matching score of each feature information and the reliability score, weighted summation is performed, and the matching score can be calculated by the large model.

[0035] Specifically, the time match score is determined by calculating the overlap rate between the query time range and the dataset time range. Industry match score: uses a predefined industry keyword list to calculate the proportion of matching keywords that appear in the query. Indicator match score: uses a predefined indicator keyword list to calculate the proportion of matching keywords that appear in the query. Geographic match score: uses a predefined geographic keyword list to calculate the proportion of matching keywords that appear in the query. Data reliability score: directly uses the reliability score in the dataset.

[0036] Furthermore, the total matching score is calculated by weighting: for example, the scores of each dimension can be combined using the following weights: time dimension: 30%; industry dimension: 25%; indicator dimension: 25%; geographic dimension: 15%; reliability dimension: 5%.

[0037] Select the target dataset: Calculate the total matching score of each dataset and select the dataset with the highest score as the best match. The specific formula is as follows: Total Score = wt*St+ wi*Si+wm*Sm+ wg*Sg+wr*R; where Total Score represents the total matching score of each data set; wt represents the weight of the time dimension in the data set; wi represents the weight of the industry dimension in the data set; wm represents the weight of the indicator dimension in the data set; wg represents the weight of the geographic dimension in the data set; wt represents the weight of the reliability dimension in the data set; St represents the time matching score, which is determined by the overlap rate between the query time and the data set time; Si represents the industry matching score, which is determined by the matching ratio of the query industry management detection; Sm represents the indicator matching score, which is determined by the matching ratio of the query indicator keywords; Sg represents the geographic matching score, which is determined by the matching ratio of the query geographic keywords; R is used to represent the reliability score, which is determined by the credibility score of the query data set.

[0038] Optionally, before determining a visualization strategy for the demand information according to the demand information and the target data set, the method further includes: Extracting target data from the target data set; Using the large language model, verifying whether the target data corresponds to the demand information; If the correspondence is inconsistent, reacquire the target data set or supplement the missing corresponding features in the target data set, and re-execute the step of extracting the target data in the target data set; If the correspondence is consistent, a step of determining a visualization strategy for the demand information according to the demand information and the target data set is performed.

[0039] In the embodiment of the present application, before step 13, it is necessary to verify whether the target data in the target data set is data corresponding to the required information. If the target data set is selected incorrectly, it is necessary to reselect the target data set. Among them, the function of the large language model can extract and prepare relevant data from the selected data set, and the specific implementation can be to generate and execute code for data extraction.

[0040] Optionally, the above step 13 includes: Utilizing the large language model, parsing the demand information, and determining the user intent corresponding to the demand information; Using the large language model, extracting visualization parameters in the target data set; Selecting a visualization chart type and a color scheme based on the user intent and the visualization parameters using the large language model; A visualization strategy for the demand information is determined based on the visualization chart type and color scheme.

[0041] In the embodiment of the present application, input processing: receiving the file name and demand information of the user query and the selected target data set, and using the large language model to parse the user intent corresponding to the demand information and the key requirements corresponding to the target data set. Dataset feature identification: matching the predefined data set features (time range, industry segmentation, indicator type, geographic range, data reliability) according to the file name, using the large language model to analyze the data content, verify and supplement the predefined features. Visualization strategy formulation: based on the data set features and user intent, use the large language model to select the visualization chart type and color scheme, that is, the large language model can determine the color scheme, axis type and scale, so as to generate a visualization strategy, and determine the visualization strategy of the demand information according to the visualization chart type and color scheme. All the solutions of the present application are automatically generated by the large language model, including the generation of visualization strategies, without the need for human intervention.

[0042] Optionally, generating a visualization code corresponding to the visualization strategy according to the visualization strategy and the large language model, and executing the visualization code to automatically generate a visualization chart includes: Understand the semantic intent of the visualization strategy to obtain multiple chart attributes and multiple indicators of the visualization chart to be generated; Performing code generation processing according to the plurality of chart attributes, the plurality of indicators and the large language model to dynamically generate visualization code corresponding to the visualization strategy; The visualization code is executed to draw a visualization chart corresponding to the visualization strategy.

[0043] In the embodiment of the present application, code generation: using a large model to dynamically generate visualization code corresponding to the visualization strategy, including: a. data loading and preprocessing; b. chart creation (using matplotlib or seaborn); c. style setting and optimization; d. saving the chart as a static image. 5. Self-debugging and optimization. After generating the visualization code, the present application also needs to execute the generated visualization code and capture potential errors, use the large model to analyze the error information and generate repair suggestions, and iteratively modify the visualization code until success. Finally, execute the visualization code, such as executing the generated code in a controlled environment, save the image, return the result of the visualization chart to the image path that the user wants to save, and display the visualization chart to the user.

[0044] Optionally, after generating a visualization code corresponding to the visualization strategy, the method further includes: Testing the visualization code, and when the test result fails, displaying the reason for the test failure to the user; If the reason for the test failure is a code reason, regenerate a target code for the failed code by using the large language model in combination with the context of the failed code; The visualization code of the entire text is retested. If the test result still fails, a new visualization code is generated and re-verified in response to a code request input by the user.

[0045] In an embodiment of the present application, if the execution of the visualization code fails, the code reason for the failure is fed back. If the code reason is a code problem, the large language model obtains the code problem, and re-obtains the context code associated with the failed code, and regenerates part of the code in combination with the code reason and requirement information, and re-verifies the failed code. The verification process is repeated a maximum of three times. If the code cannot be successfully generated after three failures, manual access is required to modify the visualization code, and a new visualization code is generated and re-verified in response to the code request input by the user.

[0046] In a specific implementation, the large language model can obtain the full logical processing flow of automatically generating a chart based on multiple chart attributes, multiple indicators, and chart types of the visualization chart. This application does not need to set a preset template in advance to generate visualization code, and can use the automatic code generation capability of the large language model to dynamically generate visualization code.

[0047] In a specific implementation, the large language model can determine the chart type that is most suitable for displaying a visualization chart. For example, the chart type can be selected from at least one of a bar chart, a line chart, a scatter chart, a pie chart, a radar chart, and a dashboard. After the large language model selects the chart type, it can find the most suitable chart generation template from a web crawler or a preset training library, and use the chart generation template to set whether to display basic attributes such as titles, legends, and prompt boxes. When displaying title attributes, the title position, title name, and subtitle name can be set; when displaying legends, the shape of the legend can be set.

[0048] In this specification, taking the three types of parameters, chart type, chart color and chart name, as an example, the recognition result can be expressed as "chart type: x, chart color: y, chart name: z.", where x is the type of chart, y is the color of the chart, and z is the name of the chart. The parameters of the chart type can be column charts, line charts, pie charts, bar charts, area charts, scatter charts, radar charts, bubble charts, surface charts, combination charts, etc. The parameters of the chart color can be any color, such as blue, red, green, etc. The large language model uses visual codes to draw the parameters corresponding to each part.

[0049] For another example, based on the target data set, a scatter plot and line chart generation algorithm can be used to convert it into a visual chart, and the chart can be adjusted in combination with a layout optimization algorithm. The specific steps for generating a visual chart are: based on the target data set, a scatter plot and line chart generation algorithm is used to generate a basic chart set; based on the basic chart set, an automatic layout optimization algorithm is applied to generate an optimized layout chart; based on the optimized layout chart, dynamic data visualization technology is used to generate a dynamic visual chart; based on the dynamic visual chart, a data-driven design strategy is used to adjust the chart style and color scheme to generate a visual chart.

[0050] Based on the query response data set, scatter plot and line chart generation algorithms are used to generate basic charts. Scatter plots are suitable for showing the relationship between two variables, while line charts are suitable for showing the trend of a variable over time or other factors. By selecting a suitable chart generation algorithm, the data set can be converted into a chart form that is easy to understand and analyze. Based on the basic chart, the automatic layout optimization algorithm is applied to generate an optimized layout chart. The automatic layout optimization algorithm can automatically adjust the size, position and spacing of the chart according to factors such as the type of chart, data volume and visualization requirements to obtain better visual effects and information communication effects. Through layout optimization, the chart can be clearer, easier to read and more beautiful.

[0051] Based on the optimized layout chart, dynamic visual charts are generated using data dynamic visualization technology. Data dynamic visualization technology can change the content, style and interaction mode of the chart in real time according to the user's operation or data update. Through dynamic visualization, it can provide richer data analysis and exploration functions, and enhance the user's interactive experience and decision-making ability. Based on dynamic visual charts, data-driven design strategies are adopted to adjust the chart style and color matching to generate visual charts. Data-driven design strategies can select appropriate chart styles and color schemes according to the characteristics of the data and visualization goals. Through the adjustment of chart style and color matching, the chart can be made more consistent, coordinated and professional.

[0052] Optionally, the method of the present application further includes: Displaying the visualization chart to the user; Acquire interaction data between the user and the visualization chart according to a preset period; Adjusting the visualization strategy according to the interaction data; According to the adjusted visualization strategy, an optimized visualization chart is obtained, and the optimized visualization chart is displayed to the user.

[0053] In an embodiment of the present application, the chart is optimized based on the interaction data collected after a preset period, so that the chart is updated periodically to ensure that the chart can meet the needs of the user. Based on this, the present application displays a visualized chart to the user, and obtains the interaction data between the user and the chart according to a preset period, and the preset period can be within 5 minutes. According to the interaction data, the visualization strategy is adjusted. Then, according to the adjusted visualization strategy, the visualization data is visualized to obtain an optimized chart. The optimized chart is then displayed to the user.

[0054] In a specific embodiment, for example, the demand information obtained is "need total revenue and investment data of global DRAM manufacturers between 2019 and 2021". (1) Natural language query processing: User query (Q): "need total revenue and investment data of global DRAM manufacturers between 2019 and 2021"; use the large language model (M) to parse the query and extract key information: time range: 2019 to 2021; industry segment: DRAM; indicator type: total revenue, investment (capital expenditure); geographical scope: global.

[0055] (2) Automatic data set selection algorithm: Based on the extracted key information, calculate the matching score of each data set: a) "Global Main DRAM Manufacturer's Total Revenue.csv": time matching score: 1.0 (complete match); industry matching score: 1.0 (DRAM); indicator matching score: 0.5 (revenue only); geographic matching score: 1.0 (global); data reliability: 0.85; total matching score = 1.0 * 0.3 + 1.0 * 0.25 + 0.5 * 0.25 + 1.0* 0.15 + 0.85 * 0.05 = 0.8925. b) "DRAM Maker's Total Capex.csv": Time match score: 1.0 (perfect match); Industry match score: 1.0 (DRAM); Indicator match score: 0.5 (investment / capex only); Geographic match score: 1.0 (global); Data reliability: 0.85; Total match score = 1.0 * 0.3 + 1.0 *0.25 + 0.5 * 0.25 + 1.0 * 0.15 + 0.85 * 0.05 = 0.8925.

[0056] From the above scores, we can see that both datasets achieved the highest matching scores, so both datasets will be used.

[0057] (3) Data extraction and preparation: Generate “Python code” to extract and prepare data and verify the target data in the target dataset.

[0058] (4) Visualization code generation: Based on the previous user query: "Need total revenue and investment data for global DRAM manufacturers between 2019 and 2021", the selected dataset file names are: "GlobalMainDRAMManufacturer's Total Revenue.csv" and "DRAM Maker's TotalCapex.csv". Dataset feature identification: Analyze the dataset through a large model and extract visualization parameters: Time range: 2015 to 2023; Industry segmentation: DRAM; Indicator type: Revenue, Capital expenditure; Geographic scope: Global; Data reliability: 0.85; Drawing type: Line chart or bar chart.

[0059] Read the data set content, use the big model to analyze the data content, and develop a visualization strategy: Based on the data set characteristics and user intent, use the big model to select the best chart type: Considering the need to simultaneously display the changes of two indicators (revenue and investment) in a specific time period, it is recommended to use a double Y-axis bar chart. This chart type can clearly show the absolute values ​​and change trends of the two indicators.

[0060] Determine the color scheme, axis type, and scale: Use blue to represent revenue and green to represent capital expenditure. The X-axis is the year, and a double Y-axis is used, with the left Y-axis representing revenue and the right Y-axis representing capital expenditure. Use a linear scale to visually display numerical changes. Further, use the large model to generate Python visualization code.

[0061] Execute the visualization code to generate a visualization chart: Execute the visualization code to generate a visualization chart (image) and return the image path 'DRAM_revenue_capex_chart.png'.

[0062] This practical example shows how to use the large language model of this application to process user queries, select relevant data sets, extract and prepare data, generate visualization code, and finally create a chart containing the total revenue and investment data of DRAM manufacturers from 2019 to 2021. The visualization chart can use a dual Y-axis bar chart to display total revenue and total capital expenditure data at the same time.

[0063] The overall process of the present application may also include: obtaining user demand information, using natural language query processing to parse the user demand information to select the corresponding target data set, obtaining a verification code, using the verification code to verify the consistency of the target data in the target data set, if the data verification is successful, executing visualization code generation, and using the visualization code to generate a visualization image, and determining whether the visualization image is successfully generated, if the visualization image is successfully generated, feedback the visualization image to the user, if the visualization image generation fails, return error information: the reason for the image generation failure. If the data verification step is a data verification failure, a new target data set is re-acquired.

[0064] The large language model of this application has the SDAVA algorithm, which provides the semiconductor industry with a flexible and powerful automated data analysis and visualization method that can efficiently process a variety of queries and data sets and significantly reduce human intervention. By combining natural language processing and dynamic code generation, the algorithm improves the automation and accuracy of data analysis.

[0065] This application can achieve: (a) Intelligent semiconductor data analysis and visualization process: Intelligent semiconductor data analysis and visualization process, covering the complete process from natural language query processing, data set selection, dynamic code generation, data extraction and cleaning, adaptive visualization to result output. It integrates natural language processing, dynamic code generation and data visualization technology to form a specially optimized process that can efficiently process and analyze the complex data needs of the semiconductor industry. (b) Dynamic generation of analysis code suitable for semiconductor data: It can automatically generate Python code based on the characteristics of semiconductor data (such as time series, periodic fluctuations) to achieve efficient data extraction and analysis. (c) Adaptive visualization technology designed specifically for semiconductor data: It can automatically select and configure chart types based on the unique characteristics of semiconductor data (such as multi-dimensional, cross-time), and support complex chart combinations (such as multi-axis charts, trend forecast charts). (d) Natural language processing and semiconductor data matching: It can parse user queries on semiconductor markets and technologies and match the most relevant data sets in multi-source data. This technology optimizes the understanding of semiconductor industry terminology and data characteristics, can identify key requirements in complex queries, and perform efficient data matching and analysis. (e) User-oriented customized report generation process: Users can submit specific writing requirements (such as report topic, content scope, etc.) through a friendly interface. The system will automatically collect and analyze data and generate customized reports that meet user needs.

[0066] The method of this application improves efficiency by automating data processing and visualization generation. Intelligent data selection and analysis are achieved through natural language processing. Dynamic data display is supported to improve data visualization. The algorithm design of this application supports expansion and can easily adapt to new data sets and user needs.

[0067] Compared with the prior art, the present invention has the following effects: (1) Dynamic code generation: Generate custom Python code based on specific queries and datasets instead of using predefined templates.

[0068] (2) Natural language understanding: Utilize advanced language models to parse complex queries and match appropriate data sets and visualization techniques.

[0069] (3) Adaptive visualization: Dynamically adjust visualization type and parameters based on data properties and queries, and support advanced charting technology.

[0070] (4) Error handling and validation: Each step contains a validation mechanism to ensure output quality and automatically retry or adjust the method when failure occurs.

[0071] (5) Scalability: The design supports extension, and new datasets, visualization types, or language model functions can be easily added.

[0072] Reference Figure 2 As shown, the embodiment of the present application also provides a semiconductor data analysis and visualization device based on a large model, including: The first processing module 21 is used to parse the demand information input by the user using a large language model to obtain key information corresponding to the demand information; the key information includes time information, industry information, indicator information and geographic information; A first determination module 22, configured to determine a target data set most relevant to the key information from a preset database using an automatic data set selection algorithm; A second determination module 23, configured to determine a visualization strategy for the demand information according to the demand information and the target data set; The second processing module 24 is used to generate a visualization code corresponding to the visualization strategy according to the visualization strategy and the large language model, and execute the visualization code to automatically generate a visualization chart.

[0073] In the embodiment of the present application, the first determining module 22 includes: A first processing unit, configured to initialize each data set in the preset database; each data set includes a plurality of feature information, wherein the feature information includes time feature, industry feature, indicator type, geographical scope and data reliability; A first determination unit, configured to parse the key information using a data set automatic selection algorithm, and determine a matching score of the feature information corresponding to the key information in each data set and a reliability score of each data set; A second processing unit, configured to perform weighted summation according to the matching score of each feature information and the reliability score, to calculate a total matching score of each data set; The third processing unit is configured to select the data set with the highest total matching score from all the data sets as the target data set.

[0074] In the embodiment of the present application, the semiconductor data analysis and visualization device based on the large model further includes: A third processing module, used for extracting target data from the target data set; A fourth processing module, used to verify whether the target data corresponds to the demand information by using the large language model; A fifth processing module, configured to reacquire the target data set or supplement the missing corresponding features in the target data set if the correspondence is inconsistent, and re-execute the step of extracting the target data in the target data set; The sixth processing module is used to execute the step of determining the visualization strategy of the demand information according to the demand information and the target data set if the correspondence is consistent.

[0075] In the embodiment of the present application, the second determining module 23 includes: A second determining unit, configured to analyze the demand information by using the large language model, and determine the user intent corresponding to the demand information; A fourth processing unit, configured to extract visualization parameters from the target data set using the large language model; a fifth processing unit, configured to select a visualization chart type and a color scheme using the large language model according to the user intention and the visualization parameter; The third determining unit is used to determine the visualization strategy of the demand information according to the visualization chart type and color scheme.

[0076] In the embodiment of the present application, the second processing module 24 includes: A sixth processing unit, configured to understand the semantic intent of the visualization strategy and obtain a plurality of chart attributes and a plurality of indicators of the visualization chart to be generated; a seventh processing unit, configured to perform code generation processing according to the plurality of chart attributes, the plurality of indicators and the large language model, and dynamically generate visualization code corresponding to the visualization strategy; An eighth processing unit is configured to execute the visualization code to draw a visualization chart corresponding to the visualization strategy.

[0077] In the embodiment of the present application, the semiconductor data analysis and visualization device based on the large model includes: A seventh processing module is used to test the visual code and, when the test result fails, to display the reason for the test failure to the user; An eighth processing module, configured to regenerate a target code for the failed code by using the large language model and combining the context of the failed code if the reason for the test failure is a code reason; The ninth processing module is used to retest the visualization code of the full text, and if the test result still fails, generate a new visualization code in response to a code request input by the user and re-verify it.

[0078] In the embodiment of the present application, the semiconductor data analysis and visualization device based on the large model comprises: A tenth processing module, configured to display the visual chart to the user; An acquisition module, used for acquiring the interaction data between the user and the visual chart according to a preset period; an eleventh processing module, configured to adjust the visualization strategy according to the interaction data; The twelfth processing module is used to obtain an optimized visualization chart according to the adjusted visualization strategy, and display the optimized visualization chart to the user.

[0079] Among them, the implementation embodiments of the above-mentioned semiconductor data analysis and visualization method based on a large model are all applicable to the embodiments of the semiconductor data analysis and visualization device based on a large model, and can also achieve the same technical effect.

[0080] A readable storage medium according to an embodiment of the present application stores a program or instruction thereon. When the program or instruction is executed by a processor, the steps in the semiconductor data analysis and visualization method based on a large model as described above are implemented, and the same technical effect can be achieved. To avoid repetition, it will not be described here.

[0081] The processor is the processor in the semiconductor data analysis and visualization method based on a large model described in the above embodiment. The readable storage medium includes a computer readable storage medium, such as a computer read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0082] The present application also provides a computer program product, including computer instructions, which, when executed by a processor, implement the above Figure 1 The various processes of the method embodiment shown can achieve the same technical effect, and will not be described again here to avoid repetition.

[0083] The above exemplary embodiments are described with reference to the accompanying drawings, and many different forms and embodiments are feasible without departing from the spirit and teachings of the present application. Therefore, the present application should not be constructed as a limitation of the exemplary embodiments proposed herein. More specifically, these exemplary embodiments are provided so that the present application will be perfect and complete, and the scope of the present application will be conveyed to those who are familiar with the technology. In these figures, the component sizes and relative sizes may be exaggerated for clarity. The terms used here are only based on the purpose of describing specific exemplary embodiments and are not intended to be limiting. As used herein, unless the text clearly indicates otherwise, the singular forms "one", "an" and "the" are intended to include these multiple forms. It will be further understood that the terms "comprising" and / or "including" when used in this specification indicate the presence of the features, integers, steps, operations, components and / or components, but do not exclude the presence or increase of one or more other features, integers, steps, operations, components, components and / or their groups. Unless otherwise indicated, when stated, a range of values ​​includes the upper and lower limits of that range and any subranges therebetween.

[0084] The above is a preferred embodiment 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 described in the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.

Claims

1. A semiconductor data analysis and visualization method based on a large model, characterized in that: include: Utilize the large language model to analyze the demand information input by the user and obtain key information corresponding to the demand information; The key information includes time information, industry information, indicator information and geographic information; Using an automatic data set selection algorithm, determine a target data set that is most relevant to the key information from a preset database; Determining a visualization strategy for the demand information according to the demand information and the target data set; According to the visualization strategy and the large language model, a visualization code corresponding to the visualization strategy is generated, and the visualization code is executed to automatically generate a visualization chart.

2. The method according to claim 1, characterized in that Using the automatic data set selection algorithm, determine the target data set most relevant to the key information from the preset database, including: Initializing each data set in the preset database; each data set includes multiple feature information, and the feature information includes time characteristics, industry characteristics, indicator type, geographical scope and data reliability; Analyzing the key information using an automatic data set selection algorithm, and determining a matching score of the feature information corresponding to the key information in each data set and a reliability score of each data set; According to the matching score of each piece of feature information and the reliability score, weighted sum is performed to calculate the total matching score of each data set; The data set with the highest total matching score is selected from all the data sets to be determined as the target data set.

3. The method according to claim 1, characterized in that Before determining a visualization strategy for the demand information according to the demand information and the target data set, the method further includes: extracting target data from the target data set; Using the large language model, verifying whether the target data corresponds to the demand information; If the correspondence is inconsistent, reacquire the target data set or supplement the missing corresponding features in the target data set, and re-execute the step of extracting the target data in the target data set; If the correspondence is consistent, a step of determining a visualization strategy for the demand information according to the demand information and the target data set is performed.

4. The method according to claim 1, characterized in that Determining a visualization strategy for the demand information according to the demand information and the target data set includes: Utilizing the large language model, parsing the demand information, and determining the user intent corresponding to the demand information; Using the large language model, extracting visualization parameters in the target data set; Selecting a visualization chart type and a color scheme based on the user intent and the visualization parameters using the large language model; A visualization strategy for the demand information is determined based on the visualization chart type and color scheme.

5. The method according to claim 1, characterized in that Generating a visualization code corresponding to the visualization strategy according to the visualization strategy and the large language model, and executing the visualization code to automatically generate a visualization chart, including: Understand the semantic intent of the visualization strategy to obtain multiple chart attributes and multiple indicators of the visualization chart to be generated; Performing code generation processing according to the plurality of chart attributes, the plurality of indicators and the large language model to dynamically generate visualization code corresponding to the visualization strategy; The visualization code is executed to draw a visualization chart corresponding to the visualization strategy.

6. The method according to claim 5, characterized in that After generating the visualization code corresponding to the visualization strategy, the method further includes: Testing the visualization code, and when the test result fails, displaying the reason for the test failure to the user; If the reason for the test failure is a code reason, regenerate a target code for the failed code by using the large language model in combination with the context of the failed code; The visualization code of the entire text is retested. If the test result still fails, a new visualization code is generated and re-verified in response to a code request input by the user.

7. The method according to claim 1, characterized in that The method further comprises: Displaying the visualization chart to the user; Acquire interaction data between the user and the visualization chart according to a preset period; Adjusting the visualization strategy according to the interaction data; According to the adjusted visualization strategy, an optimized visualization chart is obtained, and the optimized visualization chart is displayed to the user.

8. A semiconductor data analysis and visualization device based on a large model, characterized in that: include: A first processing module is used to parse the demand information input by the user using a large language model to obtain key information corresponding to the demand information; The key information includes time information, industry information, indicator information and geographic information; A first determination module is used to determine the target data set most relevant to the key information from a preset database by using an automatic data set selection algorithm; A second determination module, configured to determine a visualization strategy for the demand information according to the demand information and the target data set; The second processing module is used to generate a visualization code corresponding to the visualization strategy according to the visualization strategy and the large language model, and execute the visualization code to automatically generate a visualization chart.

9. A readable storage medium having a program or instruction stored thereon, characterized in that: When the program or instruction is executed by a processor, the steps in the method according to any one of claims 1 to 7 are implemented.

10. A computer program product, characterized in that The method comprises computer instructions, which, when executed by a processor, implement the steps of the method according to any one of claims 1 to 7.

Images