A method for structured data interaction extraction and transformation based on large language model driving

Abstract

The application relates to a structured data interaction extraction and transformation system and method based on a large language model, the system comprising a multi-table generation module, a table iteration derivation module and a TableQuilts interaction module, the method comprising the following steps: 1. a user inputs unstructured text, and the system pre-processes the input text; 2. the original text is divided into subtexts with different themes according to the content characteristics, and the system extracts tables with multiple themes according to the content of the divided text; the application has the following superior technical effects: structured table information with multiple themes can be efficiently and accurately extracted from unstructured text, the extraction process is continuously optimized through the interaction between the user and the large language model to meet the personalized needs of the user, the system improves the information processing efficiency and accuracy, enhances the interactivity and flexibility in the data analysis process, is suitable for various application scenarios, and has wide application prospects and commercial value.

Description

Technical Field

[0001] This invention relates to the field of human-computer interaction, and in particular to a method for extracting and transforming structured data interaction based on a large language model. Background Technology

[0002] The field of artificial intelligence (AI) has witnessed the rapid development and widespread application of Large Language Models (LLMs) in recent years. These models have significantly pushed the boundaries of AI technology by processing and understanding complex language data. From Google's Transformer networks to OpenAI's GPT series, each technological advancement has significantly improved the processing power and application breadth of language models. Large language models have now become central to modern computer science and AI research, not only changing how people process language but also expanding the scope of machine learning applications across numerous industries.

[0003] Large language models are trained on massive datasets, typically including extensive textual data collected from the internet. In this way, the models learn the deep structure and complex semantic information of language, enabling them to generate coherent, logically sound text and perform effective reasoning and question answering. For example, through pre-training and fine-tuning, these models can be customized for specific tasks to meet a wide range of needs, from simple classification tasks to complex text generation and dialogue systems. The training process of large language models typically involves multi-layered, large-scale neural network architectures capable of capturing rich linguistic features and extracting useful information from vast text corpora.

[0004] In data analysis, compared to structured information, unstructured text does not conform to common formats or structures, and much of it contains information on multiple topics. In many scenarios, it's necessary to explore for interesting structured information within this unstructured text. For example, when a text writer receives an information-rich news report, they sometimes need to extract one or more data tables that respect the original facts and have clear themes, so that readers can understand the text more quickly and grasp the different themes the journalist wants to emphasize without having to read large sections of text.

[0005] With the explosive growth of unstructured data, it now constitutes an increasingly large proportion of information content on the internet. Much of the massive amounts of data from various sources are in the form of unstructured text, such as social media comments and emails. Because unstructured text lacks a fixed structure and is flexible in its representation, it contains a wealth of information. As the quantity and complexity of unstructured data continue to increase, traditional data analysis and processing methods are struggling to meet the growing demands. However, this data typically lacks a fixed structure, making it difficult to process and analyze using simple rules or models. Therefore, an efficient method is needed to extract structured information from this unstructured text so that users can quickly and accurately obtain the information they need.

[0006] In practical tasks, it is often necessary to explore the characteristics of information within the text and then extract the required structured information (usually in tabular form). Indeed, extracting structured tables from unstructured text is a challenging task. Existing methods utilize the capabilities of natural language processing models or large language models to automate the text-to-table conversion. However, these methods do not consider actual human needs during table extraction. When table extraction needs to be tailored to specific requirements, manual extraction or conversational AI capabilities are necessary. For unfamiliar text, manually extracting tables is time-consuming, and even with conversational AI, prompts need to be designed based on the text and table content. Therefore, existing methods face two challenges when handling demand-driven text-to-table information extraction tasks: the complexity of the text content and the uncertainty of user needs.

[0007] With the gradual development of large language models, they have not only demonstrated outstanding performance in traditional information extraction tasks but also shown great application potential in general information extraction tasks driven by natural language. The performance of large language models in these tasks provides a solution for structured information extraction from unstructured text: Regarding the complexity of text content, due to the strong generalization ability of large language models, information can be extracted in both zero-shot and few-shot scenarios, thus possessing general information extraction capabilities across different text types. Regarding the uncertainty of user needs, although large language models have shown application potential for general information extraction, it is difficult to rely solely on their capabilities to meet the ever-changing needs of humans. Therefore, information extraction can be achieved through human-large language model interaction, focusing on the role of humans in the information extraction process. During the interaction with the large language model, attention should be paid to human feedback and the construction of appropriate cue word strategies for model inference or fine-tuning, thereby achieving demand-oriented information extraction.

[0008] Existing text-to-table conversion methods have the following technical problems:

[0009] The complexity of the text content: The source text is complex and diverse, with significant differences in content and underlying structure between different texts. Complex text may contain information on multiple topics, and a single structured table cannot fully represent this multi-topic information. Analyzing and processing this text is a time-consuming and complex process.

[0010] Uncertainty of User Needs: The characteristics of unstructured text data are not easily apparent, and user needs are diverse and constantly changing. Automated methods struggle to accurately extract tabular information that meets specific user needs. During data exploration, users may continuously adjust and discover new information needs, thus requiring flexible and interactive methods to meet these evolving requirements.

[0011] The present invention aims to solve the above-mentioned technical problems, specifically including:

[0012] Efficient Text Information Extraction: This paper presents a system based on a large language model that can efficiently extract structured tabular information from complex unstructured text, covering multiple topics. Leveraging the powerful natural language processing capabilities of the large language model, the system achieves general information extraction for different types of text.

[0013] Interactivity that meets user needs: Design a human-computer collaborative table extraction and exploration system. Through the interaction between humans and a large language model, combined with user feedback and needs, continuously optimize and adjust the extraction strategy to generate table information that meets the personalized needs of users.

[0014] Multi-turn conversational prompting strategy: Develop a multi-turn chained prompting strategy to guide the large language model to gradually extract important information from the text through multi-turn conversations and generate multi-topic structured tables, thereby ensuring the accuracy and comprehensiveness of the table information.

[0015] Visual interactive interface: Provides an intuitive visual interactive interface through which users can input or upload text and interact with the generated tables to understand the text content more quickly, discover new information needs, and generate new tables. Summary of the Invention

[0016] To address the shortcomings of existing technologies, this invention provides a structured data interaction extraction and transformation system and method based on a large language model.

[0017] This invention provides a structured data interaction extraction and transformation system based on a large language model, comprising:

[0018] The system includes a multi-table generation module, a table iteration and derivation module, and a TableQuilts interaction module. The multi-table generation module first splits the original text and then extracts an initial multi-topic table based on the split text. The table iteration and derivation module allows users to interactively derive new tables by combining information from the initial tables with their own needs. The TableQuilts interaction module implements interactive functions during table derivation, reflects the user's historical operations, and helps users intuitively understand the derivation relationships between tables.

[0019] The present invention further provides the aforementioned method for extracting and transforming structured data interactions based on a large language model, comprising:

[0020] Step 1. The user inputs unstructured text, and the system preprocesses the input text;

[0021] Step 2. Based on the content characteristics of the original text, it is divided into sub-texts with different thematic meanings. The system will then extract multi-thematic tables based on the content of the divided texts.

[0022] Step 3. The system visualizes the initial tables and performs content checks on them. Tables with content quality issues are marked in the TableQuilts area. At the same time, the system interactively prompts the user with the relationships between the initial tables.

[0023] Step 4. The user selects the derivation mode according to their own needs, and the system derives a new generation of tables based on the selected table and the derivation mode.

[0024] Step 5. Users perform interactive operations in the TableQuilts area during the table derivation stage. The system's process waiting function manages the orderly progress of the table extraction process for unstructured text.

[0025] Step 6. The system is designed with a comprehensive error handling mechanism. When a network error or system abnormality occurs, the system records the current operation status and prompts the user to restore the operation.

[0026] Furthermore, in step 1, the user inputs unstructured text, and the system preprocesses the input text, specifically including:

[0027] Users can input or upload raw unstructured text through the system interface. The system supports users to manually input unstructured text, including: text, pasted text, and uploaded local text files. The input text, pasted text, and uploaded local text files will be automatically parsed and displayed in the text input area. The purpose of this process is to enable the system to receive text from various formats and sources, ensuring the system's applicability and flexibility. The system preprocesses the input text and uploaded local text files to ensure the content is concise and standardized.

[0028] Furthermore, in step 2, the original text is divided into sub-texts with different thematic meanings based on its content characteristics. The system will then extract multi-thematic tables based on the content of the divided texts, specifically including:

[0029] After the user displays the original text information in the system's input area, the prompts from the multi-table generation module are invoked. The prompts are divided into two stages: the first stage groups texts with different themes into subtexts based on the content characteristics of the original text, and separates these subtexts with different themes using delimiters; the second stage generates a multi-theme table based on the text segmented in the previous stage, where:

[0030] The first phase of the prompts uses the prompts from the divide part of the divide-and-generate prompts, and divides the original text into sub-texts with different thematic meanings based on the content characteristics of the original text. The original long text is divided into several parts with different thematic content using delimiters.

[0031] In the second stage, the system will extract multi-topic tables based on the content of the divided text. The method of generating multiple tables using the divide-and-generate prompt completely splits each sub-text and then uses the generate prompt to extract table information containing the title meaning from each sub-text. However, this method of generating tables completely splits different sub-texts, making it difficult to consider the relationship between tables when generating tables. This results in significant differences in the format of tables with similar topics. Furthermore, the tables have high requirements for the structure of the data, and some information that is difficult to represent in tables will be discarded, resulting in significant information loss.

[0032] To account for the relationships between subtexts and reduce information loss, the system incorporates prompts for generating multiple tables. These prompts include: task description, given segmented text, explicit output format, and notes. The task description prompt clarifies the task requirements first. Then, the entire segmented text is used as input to generate multi-topic tables based on the segmented text, rather than generating tables for each subtext independently. This helps maintain a consistent table generation style when generating table structures and extracting table content, and allows for the generation of tables with similar structures and expressions for similar subtexts.

[0033] To ensure that the output format of the large language model aligns with the target, the prompt word template explicitly defines the output format and provides an example of the target format. Each table's information includes three parts: table topic, table header information, and table body information. In addition, the "generate" prompt word guides the large language model to fully consider the information in the text that is difficult to structure when generating tables, adding supplementary information columns to record notes on this information, thereby reducing information loss. By using the above two-stage prompt words, users can extract initial multi-topic tables from the original text. Based on these initial tables, users can gain a preliminary understanding of the original text and clarify which topics the original text data roughly contains. This helps users combine the content characteristics of the original text and the derived initial tables to conduct subsequent data exploration.

[0034] Furthermore, in step 3, the system visualizes the initial tables and performs content checks, marking tables with content quality issues in the TableQuilts area. Simultaneously, it interactively prompts the user with the relationships between the initial tables, specifically including:

[0035] The system extracts the initial multi-topic table from the text in the text input area. Users use this system to iteratively derive new tables from the original multi-topic table. As the number of derived tables increases, the aesthetics and convenience of table viewing in the table list area need to be guaranteed. To this end, the system's table list area has been designed to support users to scroll up and down to view all tables, as well as scroll left and right to view the complete information details of the tables.

[0036] like Figure 1 As shown, the TableQuilts diagram in the TableQuilts interactive module is the system's main visualization format. The TableQuilts diagram is composed of different colored rectangular blocks. This visualization format is displayed in the TableQuilts area. The construction process of the TableQuilts diagram is as follows: Figure 2As shown, information containing all table attributes is stored in JSON format. By parsing the derivation relationship between tables in the JSON data and considering the position and layout relationship of the target graphic elements, the JSON data is converted into a two-dimensional matrix form that is easy to render TableQuilts graphs. Each element of the two-dimensional matrix corresponds to a rectangular element of the TableQuilts graph. Finally, the two-dimensional matrix is ​​rendered into a graphical form. Adding interactive functionality to this graphical form yields the TableQuilts graph visualization.

[0037] When passing Figure 1 After the multi-table generation module obtained the initial representative cells, the TableQuilts area would display rectangular elements representing the initial representative cells. In order to provide users with operation suggestions during interaction, elements describing the characteristics and relationships of the tables were mapped on the TableQuilts. The system was designed with a table content inspection function. When rendering the TableQuilts diagram, all tables would be inspected one by one. The system could identify tables with complex cell structures and tables with missing values ​​and mark them on the TableQuilts area.

[0038] To help users quickly understand the relationships between initial grids, the system calculates the similarity between them and uses right-angled arrows to establish associations between similar tables. When the mouse hovers over the gray rectangle corresponding to an initial grid, a right-angled arrow will point from the current rectangle to other rectangles with similar relationships. This mapping between initial grid characteristics and table relationships helps users quickly clarify their next steps in the initial stage. For example, users can derive a new generation of tables based on tables they are interested in and those related to them, or derive tables from a topic table that they are interested in but has quality issues.

[0039] Furthermore, in step 4, the user selects a derivation pattern according to their needs, and the system derives a new generation of tables based on the selected table and the derivation pattern, specifically including:

[0040] Users can choose from three derivation modes: random exploration, restricted exploration, and deterministic modification, to meet table derivation requirements with varying degrees of explicitness.

[0041] Random exploration is suitable for the initial exploration stage, when users do not have a deep understanding of the initial table information. Users can read the table's topic description information and select a single table with a topic of interest or derive a new generation of tables from multiple related tables.

[0042] Restricted exploration focuses on users' subjective needs. Users express their information points of interest or personalized derived needs in the form of natural language. At this time, when deriving single or multiple tables, the user's needs will be fully considered, and a series of multi-topic tables that meet the user's needs will be derived. The user's needs do not impose strong constraints on the form of the target table, that is, the theme of the derived new table is not unique. In this mode, the large model will take into account the balance between exploration and demand constraints.

[0043] Deterministic modification refers to a situation where the user has a clear requirement to modify a certain form. The user only needs to provide a clear modification suggestion, and the system will modify the form based on the content of the modification suggestion and the corresponding prompt words, and display the modified form in the form derivation list. The system will adjust the content and structure of the form based on the user's specific modification requirements to generate a form that meets the user's expectations.

[0044] The system supports flexible switching between multiple derivation modes. Users can choose the most suitable derivation method according to their actual needs and continuously optimize the table content. The system's flexibility ensures that users can adjust their strategies in a timely manner according to changes in needs during the exploration and derivation of tables. The derived tables are displayed on the interface, and users can view and manage the derivation results through the interface. The system provides filtering and screening functions to help users efficiently manage and compare different table information. Users can filter out the most suitable tables according to specific needs. The system supports table version management, allowing users to view and trace different versions of tables to ensure data integrity and consistency. The system also provides table export and sharing functions, allowing users to export table results in various formats for easy use in other applications or sharing with others.

[0045] Furthermore, in step 5, the user performs interactive operations in the TableQuilts area during the table derivation stage. The system's process waiting function manages the orderly progress of the unstructured text table extraction process, specifically including:

[0046] User interactions during the table derivation phase are achieved through... Figure 1The TableQuilts interactive module is implemented in the TableQuilts area. The numbered rectangles in the TableQuilts area correspond one-to-one with the tables in the table overview list. The numbers in the rectangles represent the table numbers that the rectangles represent. The table numbers correspond to the topic numbers in the table overview list. Each column represents a derivation relationship, that is, which parent representative cells derived which child representative cells. For each row, it can be understood that each table in the table overview list participated in which derivation process. The gray rectangles arranged vertically in a column represent tables of the same generation. Multiple tables obtained through the multi-table generation module are the initial representative cells. The tables derived from the initial representative cells are the second representative cells, and so on. For each generation of tables, the left side describes how the table was obtained, and the right side describes how these tables derive new tables.

[0047] Given that the system's workflow utilizes chained prompts, the correct sequence of the entire process must be maintained during operation. To prevent user inconsistencies or network factors from affecting the workflow order, a workflow waiting function is implemented. Specifically, the system has a stop-and-wait function for the multi-topic table generation module and the table iterative derivation module within the unstructured text table extraction workflow. For the iterative table exploration module, when table derivation begins, the derivation process will be marked as active. While active, repeated clicks of the derivation button will trigger a wait prompt to prevent repeated or erroneous operations. The active state of the table iterative derivation module will be deactivated when a network error occurs or after obtaining the derived sub-cells. The stop-and-wait function for the multi-table generation module will then be activated. The principle is similar. When the system generates the initial multi-topic table, the multi-table generation module will be set to working status. Both the multi-topic table generation module and the table iteration and derivation module have multiple exit mechanisms to prevent user errors and network problems while maximizing system efficiency. The system provides operation prompts, including prompts for repeated clicks, no table selection, and no input of requirements. When the user does not select a derived table, does not submit a natural language request, or repeatedly clicks the function button, the system will prompt the user to perform the correct operation through pop-ups or graphical language, avoiding the trouble and time loss caused by misoperation. The system provides detailed user guides and help documents, which users can consult and learn at any time, improving the ease of use and user experience of the system.

[0048] Furthermore, in step 6, the system is designed with a robust error handling mechanism. When a network error or system anomaly occurs, the system automatically records the current operation status and prompts the user to restore the operation. Specifically, this includes:

[0049] Users can continue unfinished operations through the system's recovery function, ensuring data security and operational continuity. The system supports multi-level error monitoring and logging, allowing administrators to quickly locate and resolve problems by viewing logs, thus improving system stability and reliability. Through this mechanism, the system can respond quickly and restore normal operation in the event of an anomaly, reducing the risk of user operation interruption and data loss.

[0050] The superior technical effects of this invention are as follows:

[0051] 1. The large language model-driven structured data interactive extraction and transformation system and method described in this invention realizes efficient and accurate extraction of multi-topic structured tabular information from unstructured text. Through user interaction with the large language model, the extraction process is continuously optimized to meet users' personalized needs. While improving information processing efficiency and accuracy, the system enhances the interactivity and flexibility of the data analysis process. It is suitable for various application scenarios, such as news reports, financial statements, and market analysis, and has broad application prospects and commercial value.

[0052] 2. The structured data interactive extraction and transformation system and method driven by the large language model described in this invention can quickly identify and extract multi-topic information in text through a multi-round chain prompt word strategy. Compared with traditional table extraction methods for text data that rely on manual rules or machine learning models, it significantly improves the speed and efficiency of data extraction when processing complex unstructured text. The system can iteratively generate multi-topic structured tables in a short time, so that users do not need to spend a lot of time manually processing and analyzing text.

[0053] 3. The structured data interaction extraction system and method driven by a large language model described in this invention utilizes the powerful natural language processing capabilities of the large language model to accurately understand the semantic and contextual information in text, thereby extracting high-quality structured tables. Through an improved divide-and-generate cue word strategy, the text is divided into multiple sub-texts according to themes, and corresponding tables are generated, ensuring that the content of the generated tables is consistent with the original text, minimizing information loss and misunderstanding. Furthermore, the system considers the relationships between tables when generating them, ensuring the consistency of table format and content.

[0054] 4. The large language model-driven structured data interactive extraction and transformation system and method of the present invention provide multiple table derivation modes, including random exploration, restricted exploration and deterministic modification, so that users can freely choose and generate structured tables that meet their personalized needs according to their own needs and preferences; the system dynamically adjusts and optimizes the table generation strategy through interaction with users and in combination with user feedback and needs, so as to ensure that the generated tables can accurately reflect the user's needs.

[0055] 5. The large language model-driven structured data interactive extraction and transformation system and method described in this invention, with its designed TableQuilts interactive module, provides an intuitive visual interactive interface. Users can intuitively view and manage the relationship and content characteristics between the initial table and derived tables through the TableQuilts diagram. The system establishes a mapping relationship between the row data of the table and the original text. Users can click on the row data of the table to view its corresponding part in the original text. This interactive method helps users quickly understand the relationship between the table content and the original text, improving data analysis efficiency.

[0056] 6. The large language model-driven structured data interaction extraction and transformation system and method of this invention features a comprehensive system prompt and control module to ensure the sequential operation of the system and the smoothness of the user experience. The system includes a process waiting function to prevent user errors or network problems from affecting normal system operation. The system provides rich operation prompts, including prompts for repeated clicks, unselected tables, and no input requirements, ensuring users can use the system correctly and avoid operational errors. Furthermore, the system is designed with an error handling and recovery mechanism. When network errors or operational anomalies occur, the system can identify the current operation status and prompt the user to recover the operation, ensuring the continuity of operation. Attached Figure Description

[0057] Figure 1 This is a flowchart illustrating the structured data interaction extraction and transformation method driven by the large language model described in this invention.

[0058] Figure 2 This is a schematic diagram illustrating the construction process of TableQuilts, the main visualization form of the large language model-driven structured data interaction extraction and transformation system described in this invention. Implementation

[0059] The following is in conjunction with the instruction manual appendix. Figure 1-2 The present invention provides a detailed description of the technical solution of the large language model-driven structured data interaction extraction and transformation system and method described herein.

[0060] like Figure 1 As shown, the present invention provides a structured data interaction extraction and transformation system based on a large language model, comprising:

[0061] The system comprises three modules: a multi-table generation module, a table iteration and derivation module, and a TableQuilts interactive module. The multi-table generation module first splits the original text and then extracts an initial multi-topic table based on the split text. The table iteration and derivation module allows users to interactively derive new tables by combining information from the initial tables with their own needs. The TableQuilts interactive module implements the main interactive functions during table derivation. TableQuilts also reflects the user's historical operations and helps users intuitively understand the derivation relationships between tables. Throughout the data transformation process, it not only utilizes the capabilities of a large language model but also allows users to participate in decision-making in steps such as choosing the derivation mode and interacting with visualizations.

[0062] The present invention further provides the aforementioned method for extracting and transforming structured data interactions based on a large language model, comprising:

[0063] Step 1. The user inputs unstructured text, and the system preprocesses the input text, specifically including:

[0064] Users can input or upload raw unstructured text through the system interface. The system supports users to manually input unstructured text, including: text, pasted text, and uploaded local text files. The input text, pasted text, and uploaded local text files will be automatically parsed and displayed in the text input area, so that the system can receive text of various formats and sources, ensuring the applicability and flexibility of the system. The system preprocesses the input text and uploaded local text files to ensure the conciseness and standardization of the content.

[0065] Step 2. Based on the content characteristics of the original text, it is segmented into sub-texts with different thematic meanings. The system will then extract multi-thematic tables based on the content of the segmented text, specifically including:

[0066] After the user displays the original text information in the system's input area, the prompts from the multi-table generation module are invoked. The prompts are divided into two stages: the first stage groups texts with different themes into subtexts based on the content characteristics of the original text, and separates these subtexts with different themes using delimiters; the second stage generates a multi-theme table based on the text segmented in the previous stage. Specifically:

[0067] The first phase of the prompts uses the prompts from the divide part of the divide-and-generate prompts, and divides the original text into sub-texts with different thematic meanings based on the content characteristics of the original text. The original long text is divided into multiple parts with different thematic content using delimiters.

[0068] In the second stage, the system will extract multi-topic tables based on the content of the divided text. The method of generating multiple tables using the divide-and-generate prompt completely splits each subtext and then uses the generate prompt to extract table information containing the title meaning from each subtext. However, this method of generating tables completely splits different subtexts, making it difficult to consider the relationship between tables when generating tables. This results in significant differences in the format of tables with similar topics. In addition, since tables have high requirements for the structure of data, some information that is difficult to represent in tables will be discarded, which will lead to significant information loss.

[0069] To account for the relationships between subtexts and reduce information loss, the system incorporates prompts for generating multiple tables. Specifically, these prompts include: task description, given segmented text, clear output format, and precautions. The task description prompt clarifies the task requirements first. Then, the entire segmented text is used as input, and multi-topic tables are generated based on the segmented text, rather than generating tables independently for each subtext. This helps to maintain a consistent table generation style when generating table structures and extracting table content, meaning that similar subtexts will generate tables with similar structures and expressions.

[0070] To ensure that the output format of the large language model aligns with the target, the prompt word template explicitly defines the output format and provides an example of the target format. The information for each table includes the table topic, header information, and body information. In addition, the "generate" prompt word guides the large language model to consider the information in the text that is difficult to structure when generating tables, and adds supplementary information columns to record this information in the tables, thereby reducing information loss. By using the above two-stage prompt words, users can extract the initial multi-topic tables from the original text. Based on the multi-topic tables, users can gain a preliminary understanding of the original text and clarify which topics the original text data roughly contains. This helps users combine the content characteristics of the original text and the derived initial tables to conduct subsequent data exploration.

[0071] Step 3. The system visualizes the initial tables and performs content checks, marking tables with content quality issues in the TableQuilts area. Simultaneously, it interactively prompts the user with the relationships between the initial tables, specifically including:

[0072] The system extracts the initial multi-topic table from the text in the text input area. Users use this system to iteratively derive new tables from the original multi-topic table. As the number of derived tables increases, the aesthetics and convenience of table viewing in the table list area need to be guaranteed. To this end, the system's table list area has been designed to support users to scroll up and down to view all tables, as well as scroll left and right to view the complete information details of the tables.

[0073] like Figure 1 As shown, the TableQuilts graph in the TableQuilts interactive module is the system's main visualization format. The TableQuilts graph is composed of different colored rectangular blocks. This visualization format is displayed in the TableQuilts area. The construction process of the TableQuilts graph is as follows: Figure 2 As shown, information containing all table attributes is stored in JSON format. By parsing the derivation relationship between tables in the JSON data and considering the position and layout relationship of the target graphic elements, the JSON data is converted into a two-dimensional matrix form that is easier to render TableQuilts graphs. Each element of the two-dimensional matrix corresponds to a rectangular element of the TableQuilts graph. The two-dimensional matrix is ​​rendered into a graphical form. Adding interactive functionality to this graphical form yields the TableQuilts graph visualization.

[0074] When passing Figure 1 After the multi-table generation module obtained the initial representative cells, the TableQuilts area would display rectangular elements representing the initial representative cells. In order to provide users with operation suggestions during interaction, elements describing the characteristics and relationships of the tables were mapped on the TableQuilts. The system was designed with a table content inspection function. When rendering the TableQuilts diagram, all tables would be inspected one by one. This function could identify tables with complex cell structures and tables with missing values ​​and mark them on the TableQuilts area.

[0075] To help users quickly understand the relationships between initial grids, the system calculates the similarity between initial grids and establishes a connection between similar tables using right-angled arrows. When the mouse hovers over the gray rectangle corresponding to an initial grid, a right-angled arrow will point from the current rectangle to other rectangles with similar relationships. The mapping between initial grid characteristics and table relationships helps users quickly clarify their next steps in the initial stage. For example, users can derive a new generation of tables based on tables they are interested in and those related to them, or derive tables from a topic table that they are interested in but has quality issues.

[0076] Step 4. The user selects a derivation pattern according to their needs. The system then derives a new generation of tables based on the selected table and the derivation pattern, specifically including:

[0077] Users can choose from three derivation modes: random exploration, restricted exploration, and deterministic modification, to meet table derivation requirements with varying degrees of explicitness.

[0078] Random exploration is suitable for the initial exploration stage, when users do not have a deep understanding of the initial table information. Users can read the table's topic description information and select a single table with a topic of interest or derive a new generation of tables from multiple related tables.

[0079] Restricted exploration focuses on users' subjective needs. Users express their information points of interest or personalized derived needs in the form of natural language. At this time, when deriving single or multiple tables, the user's needs will be fully considered, and a series of multi-topic tables that meet the user's needs will be derived. The user's needs do not impose strong constraints on the form of the target table, that is, the theme of the derived new table is not unique. In this mode, the large model will take into account the balance between exploration and demand constraints.

[0080] Deterministic modification refers to a situation where the user has a clear requirement to modify a certain form. The user only needs to provide a clear modification suggestion, and the system will modify the form based on the content of the modification suggestion and the corresponding prompt words, and display the modified form in the form derivation list. The system will adjust the content and structure of the form based on the user's specific modification requirements to generate a form that meets the user's expectations.

[0081] The system supports flexible switching between multiple derivation modes. Users can choose the most suitable derivation method according to their actual needs and continuously optimize the table content. The system's flexibility ensures that users can adjust their strategies in a timely manner according to changes in needs during the exploration and derivation of tables. The derived tables are displayed on the interface, and users can view and manage the derivation results through the interface. The system provides filtering and screening functions to help users efficiently manage and compare different table information. Users can filter out the most suitable tables according to specific needs. The system supports table version management, allowing users to view and trace back different versions of tables to ensure data integrity and consistency. The system also provides table export and sharing functions, allowing users to export table results in various formats for easy use in other applications or sharing with others.

[0082] Step 5. Users interact with the table derivation stage in the TableQuilts area. The system's workflow waiting function manages the orderly progress of the unstructured text table extraction process, specifically including:

[0083] User interactions during the table derivation phase are achieved through... Figure 1The TableQuilts interactive module is implemented in the TableQuilts area. The numbered rectangles in the TableQuilts area correspond one-to-one with the tables in the table overview list. The numbers in the rectangles represent the table numbers that the rectangles represent. The table numbers correspond to the topic numbers in the table overview list. Each column represents a derivation relationship, that is, which parent representative cells derived which child representative cells. For each row, it can be understood that each table in the table overview list participated in which derivation process. The gray rectangles arranged vertically in a column represent tables of the same generation. Multiple tables obtained through the multi-table generation module are the initial representative cells. The tables derived from the initial representative cells are the second representative cells, and so on. For each generation of tables, the left side describes how the table was obtained, and the right side describes how these tables derive new tables.

[0084] Given that the system's workflow utilizes chained prompts, the correct sequence of the entire process must be maintained during operation. To prevent user inconsistencies or network factors from affecting the workflow sequence, the system incorporates a workflow waiting function. Specifically, the system includes a stop-and-wait function for the multi-topic table generation module and the table iterative derivation module within the unstructured text table extraction workflow. For the iterative table exploration module, when table derivation begins, the derivation process is marked as active. While active, repeated clicks of the derivation button by the user will trigger a wait prompt to prevent repetitive or erroneous operations. The active state of the table iterative derivation module will be deactivated when a network error occurs or the derived sub-cells are obtained. The stop-and-wait principle for the multi-table generation module is explained below. Similarly, when the system generates the initial multi-topic table, the multi-table generation module will be set to working status. Multiple exit mechanisms are implemented for both the multi-topic table generation module and the table iteration and derivation module to prevent user errors and ensure system efficiency in case of network errors. The system provides rich operation prompts, including prompts for repeated clicks, no table selection, and no input of requirements. When a user does not select a derived table, does not submit a natural language request, or repeatedly clicks a function button, the system will prompt the user to perform the correct operation through pop-ups or graphical language, avoiding the inconvenience and time loss caused by misoperation. The system provides detailed user guides and help documents, which users can consult and learn at any time, improving the ease of use and user experience.

[0085] Step 6. The system is designed with a robust error handling mechanism. When a network error or system anomaly occurs, the system records the current operation status and prompts the user to perform operation recovery. Specifically, this includes:

[0086] Users can continue unfinished operations through the system's recovery function, ensuring data security and operational continuity. The system supports multi-level error monitoring and logging, allowing administrators to quickly locate and resolve problems by viewing logs, thus improving system stability and reliability. Through this mechanism, the system can respond quickly and restore normal operation in the event of an anomaly, reducing the risk of user operation interruption and data loss.

[0087] This invention is not limited to the above-described embodiments. Any modifications, improvements, or substitutions that can be conceived by those skilled in the art without departing from the essential content of this invention fall within the protection scope of this invention.

Claims

1. A method for structured data interaction extraction and transformation based on a large language model, comprising a multi-table generation module, a table iteration and derivation module, and a TableQuilts interaction module; wherein, The multi-table generation module first splits the original text, and then extracts the initial multi-topic tables based on the split text; the table iteration and derivation module allows users to interactively derive new tables by combining the information of the initial tables with their own needs. The TableQuilts interactive module implements interactive functions during table derivation. It also reflects the user's historical operations and helps the user intuitively understand the derivation relationships between tables. The specific steps are as follows: Step 1. The user inputs unstructured text, and the system preprocesses the input text; Step 2. Based on the content characteristics of the original text, it is divided into sub-texts with different thematic meanings. The system will then extract multi-thematic tables based on the content of the divided texts. Step 3. The system visualizes the initial tables and performs content checks on them. Tables with content quality issues are marked in the TableQuilts area. At the same time, the system interactively prompts the user with the relationships between the initial tables. Step 4. The user selects the derivation mode according to their own needs, and the system derives a new generation of tables based on the selected table and the derivation mode. Step 5. Users perform interactive operations in the TableQuilts area during the table derivation stage. The system's process waiting function manages the orderly progress of the table extraction process for unstructured text. Step 6. The system is designed with a comprehensive error handling mechanism. When a network error or system abnormality occurs, the system records the current operation status and prompts the user to restore the operation.

2. According to the structured data interaction extraction and transformation method based on a large language model driven by claim 1, in step 1, the user inputs unstructured text, and the system preprocesses the input text, specifically including: Users can input or upload raw unstructured text through the system interface. The system supports users to manually input unstructured text, including: text, pasted text, and uploaded local text files. The input text, pasted text, and uploaded local text files will be automatically parsed and displayed in the text input area. The purpose of this process is to enable the system to receive text from various formats and sources, ensuring the system's applicability and flexibility. The system preprocesses the input text and uploaded local text files to ensure the content is concise and standardized.

3. According to the structured data interaction extraction and transformation method driven by a large language model as described in claim 1, in step 2, the original text is divided into sub-texts with different thematic meanings based on its content characteristics. The system will then extract multi-thematic tables based on the content of the divided texts, specifically including: After the user displays the original text information in the system's input area, the multi-table generation module is invoked. The prompting word process is divided into two stages: the first stage gathers texts with different themes into different subtexts based on the content characteristics of the original text, and separates these subtexts with different themes using delimiters; the second stage generates a multi-theme table based on the text divided in the previous stage.

4. According to the structured data interaction extraction and transformation method based on a large language model driven by claim 1, in step 3, the system visualizes the initial data tables and performs content checks on the tables, marking tables with content quality problems in the TableQuilts area, and simultaneously using an interactive format to prompt the user about the relationships between the initial data tables, specifically including: The system extracts the initial multi-topic table from the text in the text input area. Users use the system to iteratively derive new tables from the original multi-topic table. As the number of derived tables increases, the aesthetics and convenience of table viewing in the table list area need to be guaranteed. To this end, the system's table list area has been designed to support users to scroll up and down to view all tables and scroll left and right to view the complete information details of the tables.

5. In the structured data interaction extraction and transformation method based on a large language model driven by claim 1, in step 4, the user selects a derivation mode according to their own needs, and the system derives a new generation table based on the selected table and the derivation mode, specifically including: Users can choose from three derivation modes: random exploration, restricted exploration, and deterministic modification, to meet table derivation requirements with varying degrees of explicitness. Random exploration is suitable for the initial exploration stage, when users do not have a deep understanding of the initial table information. Users can read the table's topic description information and select a single table with a topic of interest or derive a new generation of tables from multiple related tables. Restricted exploration focuses on users' subjective needs. Users express their information points of interest or personalized derived needs in the form of natural language. At this time, when deriving single or multiple tables, the user's needs will be fully considered, and a series of multi-topic tables that meet the user's needs will be derived. The user's needs do not impose strong constraints on the form of the target table, that is, the theme of the derived new table is not unique. In this mode, the large model will take into account the balance between exploration and demand constraints. Deterministic modification refers to a situation where the user has a clear requirement to modify a certain form. The user only needs to provide a clear modification suggestion, and the system will modify the form based on the content of the modification suggestion and the corresponding prompt words, and display the modified form in the form derivation list. The system will adjust the content and structure of the form based on the user's specific modification requirements to generate a form that meets the user's expectations. The system supports flexible switching between multiple derivation modes. Users can choose the most suitable derivation method according to their actual needs and continuously optimize the table content. The system's flexibility ensures that users can adjust their strategies in a timely manner according to changes in needs during the exploration and derivation of tables. The derived tables are displayed on the interface, and users can view and manage the derivation results through the interface. The system provides filtering and screening functions to help users efficiently manage and compare different table information. Users can filter out the most suitable tables according to specific needs. The system supports table version management, allowing users to view and trace different versions of tables to ensure data integrity and consistency. The system also provides table export and sharing functions, allowing users to export table results in various formats for easy use in other applications or sharing with others.

6. In the structured data interactive extraction and transformation method based on a large language model driven by claim 1, in step 5, the user performs interactive operations in the TableQuilts area for the table derivation stage. The system's process waiting function manages the orderly progress of the table extraction process for unstructured text, specifically including: User interaction during the table derivation phase is achieved through the TableQuilts interaction module. The numbered rectangles in the TableQuilts area correspond one-to-one with the tables in the table overview list. The numbers in the rectangles represent the table numbers they represent, and the table numbers correspond to the topic numbers in the table overview list. Each column represents a derivation relationship, that is, which parent cells derived which child cells. For each row, it can be understood that each table in the table overview list participated in which derivation steps. The gray rectangles arranged vertically in a column represent tables of the same generation. Multiple tables obtained through the multi-table generation module are the initial representative cells, and the tables derived from the initial representative cells are the second representative cells, and so on. For each generation of tables, the left side describes how the table was obtained, and the right side describes how these tables derive new tables. The system's workflow utilizes chained prompts, ensuring the correct sequence of operations. To prevent user errors or network issues from affecting the workflow order, a workflow waiting function is implemented. Specifically, the multi-topic table generation module and the table iterative derivation module within the unstructured text table extraction process have a stop-wait function. For the iterative table exploration module, when table derivation begins, the derivation process is marked as active. While active, repeated clicks of the derivation button will trigger a wait prompt to prevent repetitive or erroneous operations. If a network error occurs or the derived sub-cells are obtained, the iterative derivation module will be paused. The working state will be terminated. The stopping and waiting principle of the multi-table generation module is similar. When the system performs the initial generation of multi-topic tables, the multi-table generation module will be set to the working state. The working states of both the multi-topic table generation module and the table iteration derivation module are equipped with multiple exit mechanisms to ensure system efficiency while preventing user misoperation and network errors. The system provides operation prompts, including prompts for repeated clicks, no table selection, and no input of requirements. When the user does not select a derived table, does not submit a natural language requirement, or repeatedly clicks the function button, the system will prompt the user to perform the correct operation through pop-ups or graphical language to avoid the trouble and time loss caused by misoperation.

7. In the structured data interaction extraction and transformation method based on a large language model driven by claim 1, in step 6, the system is designed with a robust error handling mechanism. When a network error or system anomaly occurs, the system automatically records the current operation status and prompts the user to restore the operation, including: Users can continue unfinished operations through the system's recovery function, ensuring data security and operational continuity. The system supports multi-level error monitoring and logging, allowing administrators to quickly locate and resolve problems by viewing logs, thus improving system stability and reliability. In the event of anomalies, the system can respond quickly and restore normal operation, reducing the risk of user operation interruption and data loss.

Images