Industrial research-oriented natural language query and automatic visualization generation method

Abstract

The invention relates to the technical field of industrial research, and discloses an industrial research-oriented natural language query and automatic visualization generation method, which comprises the following steps of: acquiring a natural language query statement of a user, adapting to different accent and speech speed through a speech recognition module, and converting the natural language query statement into a text; a natural language analysis model with sentiment analysis and intention mining capabilities is used for extracting key information such as analysis indexes, time ranges and industry categories, and emphasis is extracted by combining historical records and context error correction clarification and adjustment; on the basis of the key information and a multi-industry data feature library containing an intelligent query optimization algorithm, query scripts adaptive to different industry data structures are automatically generated and executed; and generating a matching chart according to a query result in combination with user preferences and an analysis scene, associating a traceability identifier, and labeling a credibility level. According to the method, professional SQL writing and visual configuration capability are not needed, analysis is directly driven by a natural language, and the operation threshold is reduced; and intelligent chart adaptation avoids subjective deviation.

Description

Technical Field

[0001] This invention relates to the field of industrial research technology, specifically to a natural language query and automated visualization generation method for industrial research. Background Technology

[0002] In the field of industry research, with the explosive growth of data volume, the ability to efficiently acquire and analyze data has become a key requirement. Industry research involves multi-dimensional data such as macroeconomics and specific industries. Researchers need to extract valuable information from massive amounts of data to support decision-making, and traditional analysis methods urgently need to be improved in terms of data processing efficiency and ease of use.

[0003] While common technologies possess natural language query or visualization capabilities, they suffer from several drawbacks: natural language queries require users to manually write complex SQL statements, posing a high barrier to entry for non-technical personnel; visualization configuration relies on manual selection, making it susceptible to subjective biases and resulting in unintuitive information delivery; and cross-industry queries require frequent script adjustments due to differences in field structures, making seamless switching impossible and failing to meet the requirements of industry research. Therefore, this paper proposes a natural language query and automated visualization generation method for industry research. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a natural language query and automated visualization generation method for industry research, which solves the technical problems of high operational threshold, large subjective bias in visualization, and poor cross-industry query adaptability.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a natural language query and automated visualization generation method for industry research, comprising:

[0006] S1 Data Acquisition:

[0007] First, the user inputs a natural language query statement, and then a natural language query recognition module for voice input is set up to convert speech with different accents and speaking speeds into text query statements.

[0008] S2 Key Information Extraction and Semantic Error Correction and Clarification:

[0009] Subsequently, a natural language parsing model with sentiment analysis and intent deep mining capabilities is used to extract key information from the statements obtained in step S1. The key information includes analysis indicators, time range, and industry category information. At the same time, the natural language parsing model performs intelligent semantic correction and clarification of ambiguous and ambiguous query statements based on the user's historical query records and current query context. Furthermore, it adjusts the focus of key information extraction based on the sentiment tendency and potential intent in the user's input statement.

[0010] S3 generates and executes database query scripts:

[0011] Based on the key information extracted in step S2, and combined with a multi-industry data feature library with intelligent query optimization algorithm, a database query script adapted to different industry data structures is automatically generated and executed. At the same time, the field structure and relationship information unique to each industry data are pre-stored in the multi-industry data feature library.

[0012] S4 generates matching visualization charts:

[0013] Based on the query results of the database query script in step S3, according to the structure and type, combined with the user's historical visualization preferences and the current analysis scenario, a matching visualization chart is generated. At the same time, the query results and visualization charts are associated with the data source traceability identifier, and a credibility level is generated and labeled by combining the data source authority rating and data update time.

[0014] Preferably, the natural language query recognition module in step S1, which uses voice input, includes an accent adaptation unit and a speech rate adjustment unit. The accent adaptation unit pre-stores a speech feature database of typical regional accents and uses dynamic comparison and matching to recognize different regional accents. The speech rate adjustment unit automatically determines the speech rate type based on the time interval characteristics of the speech signal, performs rhythm calibration on excessively fast or slow speech, and then converts it into a text query statement. After conversion, a text preview interface is automatically generated for user confirmation. By setting the accent adaptation unit and the speech rate adjustment unit, different regional accents can be accurately identified and the speech rate and rhythm can be automatically calibrated. This effectively solves the problem of speech recognition errors caused by accent differences and speech rate variations, eliminating the need for users to repeatedly adjust their voice input, improving input efficiency and accuracy, and providing a more reliable foundation for subsequent analysis processes.

[0015] Preferably, the natural language parsing model in step S2 integrates an industry-specific knowledge graph. This knowledge graph includes core indicator systems for each industry, upstream and downstream industry relationships, and a professional terminology dictionary. When extracting key information, the model calls upon the knowledge graph to verify and complete the initial information. For ambiguous or unclear queries, in addition to considering the user's historical query records and the current context, 2-3 candidate clarification schemes are simultaneously generated and presented to the user in a pop-up window. After the user selects a scheme, the model performs semantic correction and clarification based on the selected scheme. Integrating the industry-specific knowledge graph allows for verification and completion of the initial extracted information, ensuring the accuracy and completeness of key information. Generating multiple candidate clarification schemes for the user to choose from more accurately resolves ambiguity and avoids potential biases from a single correction scheme, improving the semantic correction and clarification effect and making key information extraction more aligned with user needs.

[0016] Preferably, the intelligent query optimization algorithm in step S3 includes a query path planning submodule and a redundancy filtering submodule. The query path planning submodule first matches the optimal data access path from a multi-industry data feature database based on the industry category involved in the key information, reducing the response time of cross-database queries. The redundancy filtering submodule performs syntax checks and logic optimization on the generated initial query script, removing duplicate query conditions and redundant field calls. The query path planning submodule matches the optimal data access path, reducing the response time of cross-database queries and improving query efficiency. The redundancy filtering submodule performs syntax checks and logic optimization on the initial query script, removing duplicate conditions and redundant field calls, making the query script more concise and efficient, reducing the database execution burden, and speeding up the retrieval of query results.

[0017] Preferably, in step S4, when generating matching visualization charts, a chart type matching rule library is first constructed. This rule library predefines the correspondence between different data structures and chart types: time series data is preferentially matched with line charts, categorical data with bar charts, percentage-based data with pie charts, and industry-related data with industry chain structure diagrams. By constructing this rule library, the optimal chart type is automatically matched based on different data structures, avoiding the subjectivity and blindness of manual selection. This allows the generated charts to more intuitively and accurately display data characteristics, making information transmission more efficient, helping users quickly understand the data's meaning, and improving the efficiency and quality of data analysis and decision-making.

[0018] Preferably, in step S2, after extracting key information, the natural language parsing model simultaneously executes a latent correlation indicator mining process. This involves mining related latent correlation indicators based on the extracted core analytical indicators and combining them with an industry-specific knowledge graph. A prompt box is then displayed asking the user whether to include these indicators in the analysis. Upon user confirmation, key information about the relevant latent indicators is extracted simultaneously. The natural language parsing model's execution of the latent correlation indicator mining process, by mining related latent indicators based on core indicators and prompting the user to include them in the analysis, broadens the analytical dimensions, avoids overlooking important information, and allows users to conduct more comprehensive and in-depth industry research, uncover the potential value of data, and provide richer evidence for decision-making.

[0019] Preferably, in step S3, a data quality verification module is synchronously set up during the execution of the database query script. This module verifies the completeness, consistency, and validity of the raw data obtained from the query. When the proportion of missing data exceeds a preset threshold, an automatic completion mechanism is triggered, which involves interpolating data from similar time periods within the same domain or calling backup data sources. When logical conflicts occur, the conflicting data is simultaneously marked and annotated with a special color in the subsequently generated visualization charts. The conflict situation is also explained in the source tracing identifier. By setting up the data quality verification module to verify the completeness, consistency, and validity of the raw data obtained from the query, the automatic triggering of the completion mechanism resolves the data missing problem, and the marking of conflicting data with special annotations in the charts ensures data quality, avoids deviations in analysis results due to data issues, and improves the reliability and accuracy of the analysis results.

[0020] Preferably, the visualization chart generated in step S4 is equipped with dynamic interactive functions. These functions include dimension switching, data drill-down, and chart linkage. Dimension switching allows users to switch between different analysis dimensions within the same chart. Data drill-down allows users to click on a specific data node in the chart to drill down and view the detailed data of that node. Chart linkage allows users to adjust the filter conditions of one chart when multiple related charts are generated simultaneously, and the results are updated synchronously in other related charts. These interactive functions, such as dimension switching allowing users to flexibly switch analysis perspectives, data drill-down enabling viewing detailed data, and chart linkage ensuring synchronous updates of related charts, enhance user interactivity with the charts, allowing users to explore the data more deeply, discover relationships and patterns between data, and improve the depth and flexibility of data analysis.

[0021] Preferably, the data source traceability identification function in step S4 is further implemented as follows: First, a unique traceability code is generated for each query result field. Blockchain technology is used to record the entire data flow information from its original source to the final visualized display. Simultaneously, a data source blacklist mechanism is established. When a data source is detected to have questionable authority or outdated updates, the credibility level of its associated query results is automatically lowered, and a warning label is displayed in the visualized chart. This allows for clear traceability of data sources. When a data source has problems, its credibility level is promptly lowered and a warning label is displayed, allowing users to clearly understand data reliability and avoiding decision-making errors due to the use of unreliable data, thus ensuring the scientific rigor of industry research.

[0022] Preferably, the credibility rating labeling function further includes: firstly, generating a credibility score by comprehensively considering three dimensions—data source authority rating, data update time, and data completeness—through a multi-dimensional evaluation model; secondly, adjusting the labeling method according to the display scenario of the visualization chart through a dynamic labeling strategy; and thirdly, establishing a user feedback channel to support users in raising objections to the credibility labeling results and submitting supporting materials. After review, the data source evaluation model is updated, and the credibility score is comprehensively generated through the multi-dimensional evaluation model. The dynamic labeling strategy adjusts the labeling method according to the scenario, and the user feedback channel is established. This makes credibility labeling more comprehensive, flexible, and accurate, enabling timely updates to the evaluation model based on user feedback, providing users with more reliable credibility references, helping users to more rationally judge the value of data, and improving the credibility and effectiveness of industry research.

[0023] Compared with existing technologies, this invention provides a natural language query and automated visualization generation method for industrial research, which has the following beneficial effects:

[0024] This natural language query and automated visualization generation method for industry research directly drives the analysis process through natural language. Combined with a voice input recognition module that adapts to different accents and speaking speeds, it significantly lowers the operational threshold. Users do not need professional SQL writing skills or complex data visualization configuration skills; they only need to input their query requirements in natural language. The system also supports voice input and automatically adapts to different accents and speaking speeds to drive the entire analysis process. Simultaneously, the system automatically extracts key information and clarifies ambiguities through a semantic parsing model, enabling industry researchers without technical backgrounds to efficiently complete data exploration. Furthermore, the visualization chart generation process features intelligent adaptation capabilities. The system automatically recommends the optimal chart type based on the data structure characteristics of the query results, while optimizing the display style based on user historical preferences and the current analysis scenario. This avoids subjective biases from manual selection, ensuring the intuitiveness and accuracy of information delivery. Moreover, relying on a pre-built multi-industry data feature library, the system can automatically identify differences in field structures across different industries and generate appropriate query scripts, achieving seamless switching from macroeconomic indicators to subdivided industry data. Attached Figure Description

[0025] Figure 1 This is a schematic diagram of the method flow of the present invention. Detailed Implementation

[0026] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0027] Please see Figure 1 This invention provides a technical solution: a natural language query and automated visualization generation method for industry research, comprising the following steps:

[0028] S1 Data Acquisition:

[0029] First, the user inputs a natural language query statement, and then a natural language query recognition module for voice input is set up to convert speech with different accents and speaking speeds into text query statements.

[0030] The natural language query recognition module in the form of voice input includes an accent adaptation unit and a speech rate adjustment unit. The accent adaptation unit pre-stores a speech feature library of typical regional accents and realizes the recognition of different regional accents through dynamic comparison and matching. The speech rate adjustment unit automatically determines the speech rate type based on the time interval characteristics of the speech signal, performs rhythm calibration on speech that is too fast or too slow, and then converts it into a text query statement. After the conversion is completed, a text preview interface is automatically generated for the user to confirm. The specific implementation process of the above method is as follows:

[0031] First, the system enters a state where it waits for the user to input a natural language query. This input method is versatile, supporting both traditional text input and a natural language query recognition module for voice input, to meet the needs of different users in different scenarios. When the user selects voice input, the voice signal is received by the system. At this point, the natural language query recognition module for voice input begins to function, comprising two key components: an accent adaptation unit and a speech rate adjustment unit.

[0032] The accent matching unit pre-stores a multi-regional typical accent speech feature library. This feature library is like a vast speech "dictionary," containing the unique speech features of people from different regions. When a user's speech signal enters the accent matching unit, the system activates a dynamic comparison and matching mechanism. It compares various features in the user's speech, such as pitch, pronunciation style, and intonation variations, with the typical accent speech features in the feature library one by one. Through this meticulous comparative analysis, the system can accurately identify the user's regional accent type, thus preparing for the subsequent accurate conversion of speech into text.

[0033] Next comes the speech rate adjustment unit. This unit automatically determines the speech rate type based on the time interval characteristics of the speech signal. It analyzes the time interval between each word and phrase in the speech to determine whether the user is speaking too fast, too slow, or at a normal pace. If the speech rate is determined to be too fast, the system will perform rhythm calibration, appropriately slowing down the playback speed to make each word pronounced more clearly and completely; if the speech rate is too slow, the system will speed up the speech rhythm, eliminating unnecessary pauses and dragging, making the speech more concise and fluent. The speech processed by the speech rate adjustment unit will be converted into text query statements at a more suitable pace.

[0034] After successfully converting speech into text query statements, the system doesn't immediately use it for subsequent analysis. Instead, it automatically generates a text preview interface. This interface acts as a "confirmation window," clearly displaying the converted text to the user. Users can carefully review the converted statement to ensure its accuracy and completeness in expressing their query intent. If the user finds problems with the conversion result, such as typos, grammatical errors, or missing information, they can promptly modify and adjust it. Only after the user confirms the text content is correct will the system use this user-confirmed text query statement as the final data, passing it to the next step. This provides an accurate foundation for subsequent key information extraction and semantic correction. Through this rigorous process, the S1 data acquisition stage ensures the acquisition of high-quality, accurate natural language query statements that reflect user needs, laying a solid foundation for the smooth progress of the entire industry research and analysis process.

[0035] S2 Key Information Extraction and Semantic Error Correction and Clarification:

[0036] Subsequently, a natural language parsing model with sentiment analysis and intent deep mining capabilities is used to extract key information from the statements obtained in step S1. The key information includes analysis indicators, time range, and industry category information. At the same time, the natural language parsing model performs intelligent semantic correction and clarification of ambiguous and ambiguous query statements based on the user's historical query records and current query context. Furthermore, it adjusts the focus of key information extraction based on the sentiment tendency and potential intent in the user's input statement.

[0037] The natural language parsing model integrates an industry domain knowledge graph, which includes the core indicator system of each industry, the relationship between upstream and downstream industries, and a professional terminology dictionary. When extracting key information, the model calls the knowledge graph to verify and complete the preliminary information. For ambiguous or ambiguous queries, in addition to combining the user's historical query records and the current context, it also generates 2-3 candidate clarification solutions and presents them to the user in the form of a pop-up window. After the user selects a solution, the model completes semantic correction and clarification based on the selected solution.

[0038] After extracting key information, the natural language parsing model simultaneously executes the implicit correlation indicator mining process. That is, based on the extracted core analysis indicators, it combines industry domain knowledge graphs to mine related implicit correlation indicators and asks the user in the form of a prompt box whether to include them in the analysis scope. After the user confirms, the key information of the relevant implicit indicators can be extracted simultaneously. The specific process of the above method is as follows:

[0039] The natural language processing model first performs a comprehensive scan and analysis of the acquired statement to extract key information. This key information covers aspects such as analytical indicators, time range, and industry category. For example, if the user inputs "view the sales figures of the automotive industry in the past six months," the model will accurately identify "sales figures" as the analytical indicator, "past six months" as the time range, and "automotive industry" as the industry category.

[0040] While extracting key information, the model meticulously examines the query based on the user's historical query history and the current query context. If it finds any ambiguity or vagueness in the query, such as when the user enters "to learn about the sales of products related to the electronics industry," where the term "related products" is ambiguous, the model will activate an intelligent semantic correction and clarification mechanism. It not only considers the user's past query habits but also takes into account the context of the entire query to comprehensively analyze the possible accurate intent.

[0041] Furthermore, the industry-specific knowledge graph integrated within the model plays a crucial role. This knowledge graph contains core indicator systems for each industry, upstream and downstream industry relationships, and a dictionary of specialized terminology. After extracting initial information, the model invokes the knowledge graph for verification and completion. For example, when an analytical indicator for a particular industry is extracted, the knowledge graph will confirm whether the indicator is a core indicator for that industry; if not, it will indicate a potential error or suggest supplementing with a more suitable indicator.

[0042] For ambiguous or unclear queries, the model, after combining user history and contextual analysis, will simultaneously generate 2-3 candidate clarification solutions. These solutions will be clearly presented to the user in a pop-up window, with each solution detailing its rationale and possible outcomes. Users can choose the solution that best suits their needs and intentions. The model then performs semantic correction and clarification based on the user's selected solution, ensuring that subsequent analysis is based on accurate statements.

[0043] In addition, after extracting key information, the natural language processing model simultaneously executes a process to mine implicit correlation indicators. Based on the extracted core analytical indicators, it combines industry-specific knowledge graphs to delve deeper into related implicit correlation indicators. For example, when analyzing sales revenue in the automotive industry, the model might uncover implicit indicators related to sales revenue, such as the sales percentage of different car models and sales region distribution. After uncovering these implicit correlation indicators, the model will prompt the user to include them in the analysis. If the user confirms, the model will simultaneously extract key information from these related implicit indicators, further enriching the content and dimensions of the analysis and providing strong support for subsequent more comprehensive and in-depth industry research. Through this series of rigorous and intelligent processes, the S2 stage can accurately extract key information, effectively resolve semantic ambiguity and ambiguity issues, and mine potential correlation information, providing a high-quality data foundation for the entire industry research and analysis process.

[0044] S3 generates and executes database query scripts:

[0045] Based on the key information extracted in step S2, and combined with a multi-industry data feature library with intelligent query optimization algorithm, a database query script adapted to different industry data structures is automatically generated and executed. At the same time, the field structure and relationship information unique to each industry data are pre-stored in the multi-industry data feature library.

[0046] The intelligent query optimization algorithm includes a query path planning submodule and a redundancy filtering submodule. The query path planning submodule first matches the optimal data access path from a multi-industry data feature library based on the industry category involved in the key information, thereby reducing the response time of cross-database queries. The redundancy filtering submodule performs syntax checks and logic optimization on the generated initial query script to remove duplicate query conditions and redundant field calls.

[0047] During the execution of the database query script, a data quality verification module is set up synchronously. The data quality verification module will perform integrity, consistency and validity verification on the original data obtained by the query. When the proportion of missing data exceeds the preset threshold, the completion mechanism is automatically triggered, that is, by interpolating data from similar time periods in the same field or calling backup data sources to supplement the data. When there is a logical conflict in the data, the conflicting data is marked synchronously and marked with a special color in the subsequently generated visualization chart. At the same time, the conflict situation is explained in the source tracing mark. The specific implementation process of the above method is as follows:

[0048] First, based on the key information extracted from S2 and combined with a multi-industry data feature library, the system automatically generates database query scripts adapted to different industry data structures. The multi-industry data feature library pre-stores the unique field structures and relationships of data from each industry, providing a foundation for generating accurate query scripts. For example, if the key information pertains to manufacturing, the system will refer to the field structures in the manufacturing data feature library, such as product model, production batch, and production time, as well as the relationships between these fields, such as the correspondence between product model and production batch, to construct the framework for the query script.

[0049] The intelligent query optimization algorithm plays a crucial role in this process, comprising a query path planning submodule and a redundancy filtering submodule. The query path planning submodule first performs precise matching across multiple industry data feature databases based on the industry category of the key information to find the optimal data access path. For example, when the key information pertains to the electronics industry, and the query requests sales data for a specific electronic product, this submodule analyzes the storage distribution of electronics industry data in the database and selects to access the relevant data table or partition storing the sales data for that electronic product. This avoids querying across multiple unrelated databases, significantly reducing cross-database query response time and improving query efficiency.

[0050] The redundancy filtering submodule performs meticulous syntax checks and logic optimizations on the generated initial query script. It carefully examines every query condition and field call in the script, removing duplicate conditions. For example, if both "product type = mobile phone" and "product type is mobile phone" appear in the query script, this submodule will remove one. Simultaneously, it removes redundant field calls, retaining only those closely related to key information analysis, making the query script more concise and efficient.

[0051] After the query script is generated and optimized, the system will begin executing the database query script. During execution, a data quality verification module is simultaneously configured. This module performs comprehensive and rigorous integrity, consistency, and validity checks on the raw data retrieved from the query. It checks for missing data, such as missing production date data in a product's sales data. When the proportion of missing data exceeds a preset threshold, the system automatically triggers a data completion mechanism. This mechanism employs two methods: first, interpolation using data from similar time periods within the same domain; for example, if sales data for a certain month is missing, it can refer to sales data from adjacent months and interpolate according to a certain pattern. Second, it uses a backup data source for supplementation; if a portion of the main data source is severely missing, the system will retrieve the corresponding data from a pre-set backup data source for supplementation.

[0052] When logical conflicts arise in the data, the data quality verification module will simultaneously mark the conflicting data. For example, for the same batch of the same product, the production record shows a production quantity of 1000 units, but the sales record shows a sales quantity of 1200 units, which constitutes a logical conflict. The system will mark these conflicting data and annotate them with a special color in the subsequently generated visualization charts. Simultaneously, the traceability label will provide a detailed explanation of the conflict, allowing users to clearly understand the data issues and make reasonable analyses and decisions. Through this series of rigorous processes, the S3 stage can efficiently and accurately generate and execute database query scripts, providing reliable data support for the subsequent generation of matching visualization charts.

[0053] S4 generates matching visualization charts:

[0054] Based on the query results of the database query script in step S3, according to the structure and type, combined with the user's historical visualization preferences and the current analysis scenario, a matching visualization chart is generated. At the same time, the query results and visualization charts are associated with the data source traceability identifier, and a credibility level is generated and labeled by combining the data source authority rating and data update time.

[0055] When generating matching visualization charts, a chart type matching rule library is first constructed. The rule library predefines the correspondence between different data structures and chart types, namely, time series data is preferentially matched with line charts, categorical data is preferentially matched with bar charts, percentage-based data is preferentially matched with pie charts, and industry-related data is preferentially matched with industry chain structure diagrams.

[0056] Furthermore, the generated visualization charts are equipped with dynamic interactive functions, including dimension switching, data drill-down, and chart linkage. The dimension switching allows users to switch between different analysis dimensions in the same chart. The data drill-down allows users to click on a specific data node in the chart and drill down to view the detailed data of that node. The chart linkage allows users to operate the filter conditions of one chart when multiple related charts are generated at the same time, and the results of other related charts are updated synchronously.

[0057] The data source traceability and identification function is further implemented as follows: First, a unique traceability code is generated for each query result field. The entire chain flow information of data from the original source to the final visualization is recorded through blockchain technology. At the same time, a data source blacklist mechanism is established. When a data source is detected to have authority disputes or be outdated, the credibility level of its associated query results is automatically reduced and a warning label is marked in the visualization chart.

[0058] The credibility rating labeling function further includes: firstly, generating a credibility score by comprehensively considering three dimensions—authoritative rating of the data source, data update time, and data completeness—through a multi-dimensional evaluation model; secondly, adjusting the labeling method according to the display scenario of the visualization chart through a dynamic labeling strategy; and thirdly, building a user feedback channel to support users in raising objections to the credibility labeling results and submitting supporting materials. After review, the data source evaluation model is updated. The specific implementation process of the above methods is as follows:

[0059] First, the system builds a rule base for matching chart types. This rule base predefines the correspondence between different data structures and chart types. For example, for time series data, since its focus is on showing the trend of data changes over time, line charts are prioritized. When the query results show the sales volume of an industry over the past few years, the system will automatically select a line chart to present the data, allowing users to clearly see the fluctuations in sales volume. Categorical data is prioritized for bar charts; for example, bar charts can visually compare the sales performance of a product in different regions. Percentage-based data is prioritized for pie charts; for example, pie charts clearly show the market share of different product types in an industry. Industry-related data is prioritized for supply chain diagrams; if the query results involve the relationships between upstream and downstream enterprises in an industry, a supply chain diagram can clearly show the upstream and downstream connections between these enterprises.

[0060] Next, the system will combine the user's historical visualization preferences with the current analysis scenario to select the most suitable chart type from the chart type matching rule library to generate the visualization chart. If the user has frequently used bar charts when analyzing similar industry data before, and the current analysis scenario is also a comparison of different categories of data, the system will prioritize generating a bar chart;

[0061] The generated visualizations also feature rich dynamic interactive functions. The dimension switching function allows users to switch between different analysis dimensions within the same chart. For example, in a chart displaying sales data for an industry, originally shown by region, users can switch the display dimension to product type using the dimension switching function, thus analyzing sales data from different perspectives. The data drill-down function allows users to click on specific data nodes in a chart to drill down and view detailed breakdowns of that node's data. For instance, in a bar chart displaying the overall sales of an industry, clicking on the bar representing a region allows users to drill down and view the sales details of different products within that region. The chart linking function is suitable for situations where multiple related charts are generated simultaneously. When the filter conditions of one chart are adjusted, the other related charts will update their displays synchronously. For example, if two related charts displaying sales and profit margins for an industry are generated simultaneously, when data for a specific time period is filtered in the sales chart, the profit margin chart will also automatically update to reflect that time period.

[0062] Simultaneously, the system associates source identification tags with query results and visualization charts. First, a unique source code is generated for each query result field. Then, blockchain technology records the entire data flow from its original source to the final visualization. This is like giving each piece of data an "ID card," allowing users to clearly understand where the data came from and what processing it underwent. Furthermore, the system establishes a data source blacklist mechanism. When a data source is detected to have questionable authority or outdated updates, the credibility level of its associated query results is automatically lowered, and a warning label is added to the visualization charts to remind users to refer to them with caution.

[0063] Finally, there's the credibility rating function. The system generates a credibility score by comprehensively considering three dimensions: data source authority rating, data update time, and data completeness, using a multi-dimensional evaluation model. For example, a query result with a high data source authority rating, timely data updates, and good data completeness will receive a high credibility score. Then, a dynamic rating strategy adjusts the rating method based on the visualization chart's display scenario. For instance, in important industry analysis reports, prominent colors and detailed text descriptions are used to indicate the credibility rating; in general daily analysis, simple icons are used. Simultaneously, the system establishes a user feedback channel, allowing users to raise objections to the credibility rating results and submit supporting materials. After review, if the user's materials are reasonable, the data source evaluation model will be updated to improve the accuracy of subsequent data credibility assessments. Through this rigorous and meticulous process, the S4 stage generates high-quality, practical, and reliable visualization charts, providing strong support for industry research.

[0064] This solution directly drives the analysis process through natural language, and with the voice input recognition module adapting to different accents and speaking speeds, it significantly lowers the operational threshold. Users do not need professional SQL writing skills or complex data visualization configuration skills; they only need to input their query requirements through natural language. The system also supports voice input and automatically adapts to different accents and speaking speeds to drive the entire analysis process. At the same time, the system automatically extracts key information and clarifies ambiguities through a semantic parsing model, enabling industry researchers without technical backgrounds to efficiently complete data exploration. Secondly, the visualization chart generation process has intelligent adaptation capabilities. The system can automatically recommend the optimal chart type based on the data structure characteristics of the query results, and optimize the display style by combining user historical preferences and the current analysis scenario, avoiding subjective biases from manual selection and ensuring the intuitiveness and accuracy of information delivery. Furthermore, relying on a pre-built multi-industry data feature library, the system can automatically identify the differences in field structure between different industries and generate adapted query scripts, achieving seamless switching from macroeconomic indicators to subdivided industry data.

[0065] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0066] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A natural language query and automated visualization generation method for industry research, characterized in that, Includes the following steps: S1 Data Acquisition: First, the user inputs a natural language query statement, and then a natural language query recognition module for voice input is set up to convert speech with different accents and speaking speeds into text query statements. S2 Key Information Extraction and Semantic Error Correction and Clarification: Subsequently, a natural language parsing model with sentiment analysis and intent deep mining capabilities is used to extract key information from the statements obtained in step S1. The key information includes analysis indicators, time range, and industry category information. At the same time, the natural language parsing model performs intelligent semantic correction and clarification of ambiguous and ambiguous query statements based on the user's historical query records and current query context. Furthermore, it adjusts the focus of key information extraction based on the sentiment tendency and potential intent in the user's input statement. S3 generates and executes database query scripts: Based on the key information extracted in step S2, and combined with a multi-industry data feature library with intelligent query optimization algorithm, a database query script adapted to different industry data structures is automatically generated and executed. At the same time, the field structure and relationship information unique to each industry data are pre-stored in the multi-industry data feature library. S4 generates matching visualization charts: Based on the query results of the database query script in step S3, according to the structure and type, combined with the user's historical visualization preferences and the current analysis scenario, a matching visualization chart is generated. At the same time, the query results and visualization charts are associated with the data source traceability identifier, and a credibility level is generated and labeled by combining the data source authority rating and data update time.

2. The natural language query and automated visualization generation method for industry research as described in claim 1, characterized in that: The natural language query recognition module in step S1, which uses voice input, includes an accent adaptation unit and a speech rate adjustment unit. The accent adaptation unit stores a speech feature library of typical regional accents in advance and recognizes different regional accents through dynamic comparison and matching. The speech rate adjustment unit automatically determines the speech rate type based on the time interval characteristics of the speech signal, performs rhythm calibration on speech that is too fast or too slow, and then converts it into a text query statement. After the conversion is completed, a text preview interface is automatically generated for the user to confirm.

3. The natural language query and automated visualization generation method for industry research as described in claim 1, characterized in that: The natural language parsing model in step S2 integrates an industry domain knowledge graph. This knowledge graph includes the core indicator system of each industry, the upstream and downstream industry relationships, and a professional terminology dictionary. When extracting key information, the model calls the knowledge graph to verify and complete the extracted preliminary information. For ambiguous or ambiguous query statements, in addition to combining the user's historical query records and the current context, 2-3 candidate clarification solutions are generated simultaneously and presented to the user in the form of a pop-up window. After the user selects a solution, the model completes semantic correction and clarification based on the selected solution.

4. The natural language query and automated visualization generation method for industry research as described in claim 1, characterized in that: The intelligent query optimization algorithm in step S3 includes a query path planning submodule and a redundancy filtering submodule. The query path planning submodule first matches the optimal data access path from a multi-industry data feature library based on the industry category involved in the key information, thereby reducing the response time of cross-database queries. The redundancy filtering submodule performs syntax checks and logic optimization on the generated initial query script to remove duplicate query conditions and redundant field calls.

5. The natural language query and automated visualization generation method for industry research according to claim 1, characterized in that: When generating matching visualization charts in step S4, a chart type matching rule library is first constructed. The rule library predefines the correspondence between different data structures and chart types, namely, time series data is preferentially matched with line charts, categorical data is preferentially matched with bar charts, percentage-based data is preferentially matched with pie charts, and industry-related data is preferentially matched with industry chain structure diagrams.

6. The natural language query and automated visualization generation method for industry research according to claim 1, characterized in that: In step S2, after extracting key information, the natural language parsing model simultaneously executes the implicit correlation indicator mining process. That is, based on the extracted core analysis indicators, it mines related implicit correlation indicators by combining industry domain knowledge graphs, and asks the user in the form of a prompt box whether to include them in the analysis scope. After the user confirms, the key information of the relevant implicit indicators can be extracted simultaneously.

7. The natural language query and automated visualization generation method for industry research according to claim 1, characterized in that: In step S3, during the execution of the database query script, a data quality verification module is set up simultaneously. The data quality verification module will verify the integrity, consistency and validity of the original data obtained by the query. When the proportion of missing data exceeds the preset threshold, the completion mechanism is automatically triggered, that is, by interpolating data from similar time periods in the same field or calling backup data sources to supplement the data. When there is a logical conflict in the data, the conflicting data is marked simultaneously and marked with a special color in the subsequently generated visualization chart. At the same time, the conflict situation is explained in the source tracing identifier.

8. The natural language query and automated visualization generation method for industry research according to claim 1, characterized in that: The visualization chart generated in step S4 is equipped with dynamic interactive functions, including dimension switching, data drill-down, and chart linkage. The dimension switching allows users to switch between different analysis dimensions in the same chart. The data drill-down allows users to click on a specific data node in the chart and drill down to view the detailed data of that node. The chart linkage allows users to operate the filter conditions of one chart when multiple related charts are generated at the same time, and the results of other related charts are updated synchronously.

9. The natural language query and automated visualization generation method for industry research according to claim 1, characterized in that: The data source traceability identification function in step S4 is further implemented as follows: First, a unique traceability code is generated for each query result field. The entire chain flow information of data from the original source to the final visualization is recorded through blockchain technology. At the same time, a data source blacklist mechanism is established. When the authority of the data source is disputed or the update is delayed, the credibility level of its associated query results is automatically reduced and a warning label is marked in the visualization chart.

10. The natural language query and automated visualization generation method for industry research according to claim 1, characterized in that: The credibility level labeling function further includes: firstly, generating a credibility score by comprehensively considering three dimensions—data source authority rating, data update time, and data completeness—through a multi-dimensional evaluation model; secondly, adjusting the labeling method according to the display scenario of the visualization chart through a dynamic labeling strategy; and thirdly, building a user feedback channel to support users in raising objections to the credibility labeling results and submitting supporting materials, and updating the data source evaluation model after review.

Images