Method and device for extracting report data based on large model technology

Abstract

The invention provides a method and device for extracting report data based on a large model technology, and relates to the field of batch data processing, and the method comprises the steps: carrying out the intelligent table splitting processing of obtained report image data through a pre-constructed intelligent report prompt word project, and obtaining an effective target report; carrying out metadata dynamic extraction processing on the effective target report by utilizing a pre-constructed data extraction cue word project to obtain a report label; performing multi-dimensional data integration processing on the effective target report by utilizing a pre-constructed data integration prompt word project and the report label to obtain a report data dictionary; and performing subject semantic mapping on the effective target report according to the report data dictionary to obtain a required data report. According to the method, a complete automatic processing flow can be constructed, and intelligent report splitting, metadata dynamic extraction, multi-dimensional data integration, subject semantic mapping and the like are realized by combining an optical symbol recognition technology with a large model technology.

Description

Technical Field

[0001] This application relates to the field of batch data processing, specifically a method and apparatus for extracting report data based on large model technology. Background Technology

[0002] Current technologies for automated extraction and standardization of financial statement data face numerous challenges in practical applications. There are significant gaps in the level of system automation. While they can handle the scanning and recognition of fixed-format reports, key steps such as report splitting, cross-page data integration, and accounting subject mapping often require substantial manual intervention. This fragmented automation process significantly increases labor costs and time consumption.

[0003] From a technical perspective, existing OCR (Optical Character Recognition) systems lack the ability to parse complex report structures, especially when dealing with non-standardized formats such as multi-page linked reports, abnormal empty columns, column layouts, and mixed annotations. Their recognition accuracy and adaptability drop significantly. More importantly, current systems generally lack effective OCR data quality control mechanisms, failing to perform intelligent cross-validation and error correction on recognition results. Furthermore, with the surge in enterprise data volume, existing technical architectures are struggling to handle the demands of processing massive amounts of multi-source, heterogeneous reports, failing to achieve efficient distributed parallel processing capabilities. This directly restricts the overall efficiency and quality of data entry into the database. Specifically:

[0004] 1. Automated processes are fragmented, and manual intervention is costly.

[0005] The existing system can only automate some aspects of data extraction (such as scanning fixed-format reports), while key steps such as report splitting and extraction, cross-page integration, and subject mapping still rely on manual operation.

[0006] 2. Lacks intelligent recognition capabilities for multi-format reports.

[0007] Existing OCR technologies have limited capabilities in processing complex reports, such as insufficient parsing of multi-page / paged structures. They also have poor tolerance for abnormal formats, lacking the ability to adapt to non-standard formats such as abnormal empty columns, column layouts, and mixed annotations.

[0008] 3. The OCR data quality control mechanism is lacking, and the existing system lacks intelligent verification capabilities for OCR recognition results.

[0009] 4. Insufficient scalability in batch processing: When faced with massive amounts of reports from multiple heterogeneous sources, the existing system struggles to achieve distributed parallel processing.

[0010] This section is intended to provide background or context for the embodiments of the invention set forth in the claims. The description herein is not an admission that it is prior art simply because it is included in this section. Summary of the Invention

[0011] To address the problems in existing technologies, this application provides a method and apparatus for extracting report data based on large model technology. This method can construct a complete automated processing flow and achieve intelligent report splitting, dynamic extraction of metadata, multi-dimensional data integration, and subject semantic mapping by combining optical symbol recognition technology with large model technology.

[0012] To solve the above-mentioned technical problems, this application provides the following technical solution:

[0013] Firstly, this application provides a method for extracting report data based on large model technology, including:

[0014] The acquired report image data is intelligently split into tables using a pre-built intelligent report prompt word project to obtain effective target reports;

[0015] The metadata of the effective target report is dynamically extracted and processed using a pre-built data extraction prompt word project to obtain report tags;

[0016] The effective target report is subjected to multi-dimensional data integration processing using a pre-built data integration prompt word project and the report tags to obtain a report data dictionary;

[0017] Based on the report data dictionary, the effective target reports are semantically mapped to subjects to obtain the required data reports.

[0018] Furthermore, the process of intelligently splitting the acquired report image data using a pre-built intelligent report prompt word engineering method to obtain an effective target report includes:

[0019] Optical symbol recognition is performed on the report image data to obtain the report data page;

[0020] Input the report data page and the intelligent report prompts in the intelligent report prompts project into the pre-trained large model to obtain the target balance sheet, income statement and cash flow statement;

[0021] If the target balance sheet, income statement, or cash flow statement is determined to be complete and without page breaks, the target balance sheet, income statement, or cash flow statement is marked as the valid target report; otherwise, the target balance sheet, income statement, or cash flow statement is merged to generate the valid target report.

[0022] Furthermore, the method of dynamically extracting metadata from the effective target report using a pre-built data extraction prompt word project to obtain report tags includes:

[0023] The effective target report and the data extraction prompts in the data extraction prompts project are input into a pre-trained large model to extract report labels from the effective target report and obtain the report name labels.

[0024] Based on the table name labels and the preset mapping relationship between financial subjects and balances, the data format of the effective target report is parsed to obtain the report labels corresponding to each column of data.

[0025] Furthermore, the multi-dimensional data integration processing of the effective target report using the pre-constructed data integration prompt word project and the report tags yields a report data dictionary, including:

[0026] Based on the data integration prompts in the data integration prompts project and the report labels, the effective target reports are merged across pages to obtain a complete report with subject columns;

[0027] Extract the values ​​corresponding to each account column from the complete account column report;

[0028] The subject columns and their corresponding values ​​are converted into the report data dictionary.

[0029] Further, the step of performing subject semantic mapping on the valid target report based on the report data dictionary to obtain the required data report includes:

[0030] Map each subject in the valid target report to a standard subject based on the report data dictionary;

[0031] The balances of each item in the effective target report are aggregated and mapped to obtain the balances of the standard items.

[0032] Furthermore, the method for extracting report data based on large model technology also includes:

[0033] Dynamic prompts are generated based on the data in the effective target report and preset prompt anchor points to support batch queries to the large model; wherein, the dynamic prompts include intelligent report prompts, data extraction prompts and data integration prompts.

[0034] Secondly, this application provides an apparatus for extracting report data based on large model technology, comprising:

[0035] The report splitting unit is used to intelligently split the acquired report image data using a pre-built intelligent report prompt word project to obtain an effective target report;

[0036] The dynamic extraction unit is used to dynamically extract metadata from the effective target report using a pre-built data extraction prompt word project to obtain report tags;

[0037] The data integration unit is used to perform multi-dimensional data integration processing on the effective target report using a pre-built data integration prompt word project and the report tags to obtain a report data dictionary.

[0038] The semantic mapping unit is used to perform subject semantic mapping on the valid target report according to the report data dictionary to obtain the required data report.

[0039] Furthermore, the report splitting unit includes:

[0040] An optical recognition module is used to perform optical symbol recognition on the report image data to obtain the report data page;

[0041] The three-table filtering module is used to input the report data page and the intelligent report prompts in the intelligent report prompts project into a pre-trained large model to obtain the target balance sheet, income statement and cash flow statement.

[0042] The target report generation module is used to mark the target balance sheet, income statement, or cash flow statement as the valid target report if it is determined that the target balance sheet, income statement, or cash flow statement is complete and has no page breaks; otherwise, it generates the valid target report by merging the target balance sheet, income statement, or cash flow statement respectively.

[0043] Furthermore, the dynamic extraction unit includes:

[0044] The table name tag extraction module is used to input the effective target report and the data extraction prompt words in the data extraction prompt word project into a pre-trained large model to extract the report tags of the effective target report and obtain the table name tags;

[0045] The report label extraction module is used to parse the data format of the effective target report according to the report name label and the preset financial subject and balance mapping relationship, so as to obtain the report label corresponding to each column of data.

[0046] Furthermore, the data integration unit includes:

[0047] The subject report generation module is used to merge the effective target reports across pages based on the data integration prompts in the data integration prompts project and the report tags, so as to obtain a complete subject column report.

[0048] The subject value extraction module is used to extract the values ​​corresponding to each subject column in the complete subject column report;

[0049] The data dictionary generation module is used to convert the subject columns and their corresponding values ​​into the report data dictionary.

[0050] Further, the semantic mapping unit includes:

[0051] The standard subject mapping module is used to map each subject in the valid target report to a standard subject according to the report data dictionary;

[0052] The account balance determination module is used to perform balance aggregation mapping on the balances corresponding to each account in the valid target report to obtain the balance corresponding to the standard account.

[0053] Furthermore, the device for extracting report data based on large model technology also includes:

[0054] The prompt generation unit is used to generate dynamic prompt words based on the data in the effective target report and preset prompt word anchor points to support batch queries to the large model; wherein, the dynamic prompt words include intelligent report prompt words, data extraction prompt words and data integration prompt words.

[0055] Thirdly, this application provides an electronic device including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of the method for extracting report data based on large model technology.

[0056] Fourthly, this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method for extracting report data based on large model technology.

[0057] Fifthly, this application provides a computer program product, including a computer program / instructions that, when executed by a processor, implement the steps of the method for extracting report data based on large model technology.

[0058] To address the problems in existing technologies, this application provides a method and apparatus for extracting report data based on large-scale model technology. Through four core steps—intelligent report splitting based on optical symbol recognition and large-scale models, dynamic metadata extraction, multi-dimensional data integration, and subject semantic mapping—it achieves fully automated conversion from unstructured image files to standardized financial data. First, optical symbol recognition and semantic analysis of the large-scale model accurately locate the target "three tables," effectively solving the parsing challenges of complex scenarios such as multi-page cross-tables and non-standard formats. Second, it dynamically extracts metadata such as customer names and reporting periods and parses column types to build a structured data foundation. Then, it integrates data through cross-page integrity verification and programming languages ​​to ensure the integrity of subject balances. Finally, using semantic association analysis of the large-scale model, it automatically maps non-standard subjects to the enterprise's standard system and completes balance aggregation. The fully automated processing significantly improves the parsing efficiency of heterogeneous reports, while standardized mapping ensures data consistency, effectively supporting the needs of financial analysis and decision-making. Attached Figure Description

[0059] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0060] Figure 1 This is a flowchart illustrating the method for extracting report data based on large model technology in this application embodiment;

[0061] Figure 2 This is a flowchart illustrating the process of obtaining a valid target report in the embodiments of this application;

[0062] Figure 3 This is a flowchart illustrating the process of obtaining report labels in an embodiment of this application;

[0063] Figure 4 This is a flowchart illustrating the process of obtaining the report data dictionary in this application embodiment;

[0064] Figure 5 This is a flowchart illustrating how the required data report is obtained in the embodiments of this application;

[0065] Figure 6 This is a structural diagram of the device for extracting report data based on large model technology in the embodiments of this application;

[0066] Figure 7 This is a structural diagram of the report splitting unit in an embodiment of this application;

[0067] Figure 8 This is a structural diagram of the dynamic extraction unit in the embodiments of this application;

[0068] Figure 9 This is a structural diagram of the data integration unit in an embodiment of this application;

[0069] Figure 10 This is a structural diagram of the semantic mapping unit in the embodiments of this application;

[0070] Figure 11 This is a schematic diagram of the structure of the electronic device in the embodiments of this application;

[0071] Figure 12 This is a flowchart illustrating the overall processing in the embodiments of this application. Detailed Implementation

[0072] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the embodiments of the present invention will be further described in detail below with reference to the accompanying drawings. Here, the illustrative embodiments of the present invention and their descriptions are used to explain the present invention, but are not intended to limit the present invention.

[0073] The information collected in the technical solution of this application is information and data authorized by the user or fully authorized by all parties. The collection, storage, use, processing, transmission, provision, disclosure and application of the relevant data all comply with the relevant laws, regulations and standards of the relevant countries and regions, necessary confidentiality measures have been taken, and they do not violate public order and good morals. Corresponding operation portals are provided for users to choose to authorize or refuse.

[0074] Provide users with corresponding operation entry points, allowing them to choose to agree to or reject the automated decision results; if the user chooses to reject, the process will proceed to the expert decision-making process.

[0075] In one embodiment, see Figure 1 To construct a complete automated processing flow, this application provides a method for extracting report data based on large model technology, which combines optical symbol recognition technology with large model technology to achieve intelligent report splitting, dynamic metadata extraction, multi-dimensional data integration, and subject semantic mapping. The method includes:

[0076] S101: Utilize a pre-built intelligent report prompt word project to perform intelligent table splitting on the acquired report image data to obtain an effective target report;

[0077] S102: Dynamically extract metadata from the effective target report using a pre-built data extraction prompt word project to obtain report tags;

[0078] S103: Using the pre-built data integration prompt word project and the report tags, perform multi-dimensional data integration processing on the effective target report to obtain a report data dictionary;

[0079] S104: Perform subject semantic mapping on the valid target report according to the report data dictionary to obtain the required data report.

[0080] Understandably, current financial statement data processing often faces the following challenges: First, the formats of audit reports, annual reports, and other image files submitted by companies are complex (such as multi-page spreadsheets, non-standard formatting, and differences in account descriptions), making it difficult for traditional OCR technology to accurately extract data from the "three statements (target balance sheet, income statement, and cash flow statement)"; second, the massive amount of heterogeneous reports (PDFs, images, etc.) requires manual screening of key information, which is inefficient and prone to errors; and third, the mapping between non-standard accounts and the company's standard system relies on manual comparison, which cannot meet the timeliness requirements of regulatory reporting.

[0081] To solve the above technical problems, see [link to relevant documentation]. Figure 12 This application's embodiment uses "AI-driven automated data governance" as its core framework, deeply integrating OCR, large-scale models, and programming technologies to construct a closed-loop intelligent processing system. First, through OCR recognition and the semantic understanding capabilities of the large-scale model, it accurately locates and integrates scattered content from the three tables (datasheet, financial statements, and accounting tables) within massive image files, solving challenges such as cross-page table association and non-standard format recognition. Second, based on the large-scale model's prompt word engineering, it dynamically extracts metadata (such as customer name and reporting period) and table column types (subject / amount columns), establishing a structured foundation for subsequent processing. Finally, through semantic association analysis, it automatically maps non-standard subjects to the enterprise's standard system and aggregates multi-subject balance data, achieving a fully automated transformation from non-standard images to standardized financial data. This framework uses AI technology to deconstruct complex data logic, significantly improving the processing efficiency and data consistency of heterogeneous reports, aligning with the core needs for automated and intelligent data governance in current financial digital transformation. The method provided in this application integrates general OCR technology with large model capabilities to construct an end-to-end fully automated data import process from scanned images of third-party enterprise financial statements to standard financial data, realizing intelligent processing of non-standard report data extraction and standardization.

[0082] Specifically, the method provided in this application addresses the challenge of processing non-standard financial statements from third-party enterprises by constructing an automated processing workflow. Through the combination of OCR text recognition and large-scale model technology, it achieves the following standardized processing closed loop:

[0083] Intelligent Report Segmentation: Based on general OCR recognition and large model technology, it accurately separates target reports (balance sheet, income statement, and cash flow statement) from original documents such as audit reports and annual reports in PDF / image format. It employs a dual-engine approach of template matching and semantic analysis to identify table boundaries and automatically segment headings based on their features.

[0084] Dynamic metadata extraction: Structured parsing is performed on the split single-table files to automatically extract key metadata tags, including the enterprise entity name, reporting period, consolidation scope identifier, and account distribution characteristics. A combination of named entity recognition and a rule engine ensures high-confidence capture of customer information fields.

[0085] Multidimensional data integration: For complex report structures with spanning multiple pages or displayed in columns, a table association algorithm is used to detect data fragmentation. Through row and column topology analysis and header semantic comparison, intelligent merging of spanning tables is achieved to form a complete set of data sources.

[0086] Subject Semantic Mapping: Utilizing the semantic understanding capabilities of a large model to convert non-standard subjects. A two-way mapping mechanism is employed, supporting both adaptive semantic matching and providing a manual verification interface to supplement mapping rules, ultimately mapping non-standard accounting subjects to the standardized subject system required by regulators or banks.

[0087] This process achieves end-to-end processing through a modular design, making it particularly suitable for enterprise scenarios that require batch processing of multi-source heterogeneous reports, as well as automated metadata management, complex table integration, and accounting subject standardization. The outputs of each stage can be directly integrated into database systems to meet subsequent needs such as regulatory reporting and business analysis.

[0088] As described above, the method for extracting report data based on large-scale model technology provided in this application achieves fully automated conversion from unstructured image files to standardized financial data through four core steps: intelligent report splitting based on optical symbol recognition and large-scale model, dynamic metadata extraction, multi-dimensional data integration, and subject semantic mapping. First, optical symbol recognition and semantic analysis of the large-scale model accurately locate the target "three tables," effectively solving the parsing challenges of complex scenarios such as multi-page cross-tables and non-standard formats. Second, it dynamically extracts metadata such as customer names and reporting periods and parses column types to build a structured data foundation. Then, it integrates data through cross-page integrity verification and programming languages ​​to ensure the integrity of subject balances. Finally, using semantic association analysis of the large-scale model, it automatically maps non-standard subjects to the enterprise's standard system and completes balance aggregation. This fully automated process significantly improves the parsing efficiency of heterogeneous reports, while standardized mapping ensures data consistency, effectively supporting the needs of financial analysis and decision-making.

[0089] In one embodiment, see Figure 2 The process of using a pre-built intelligent report prompt word engineering method to intelligently split the acquired report image data into tables to obtain an effective target report includes:

[0090] S201: Perform optical symbol recognition on the report image data to obtain the report data page;

[0091] S202: Input the report data page and the intelligent report prompts in the intelligent report prompts project into the pre-trained large model to obtain the target balance sheet, income statement and cash flow statement;

[0092] S203: If the target balance sheet, income statement or cash flow statement is determined to be complete and without page breaks, mark the target balance sheet, income statement or cash flow statement as the valid target report;

[0093] S204: Otherwise, the target balance sheet, income statement or cash flow statement are merged separately to generate the effective target report.

[0094] It is understandable that steps S201 to S204 correspond to the intelligent report splitting process, used to accurately extract the target balance sheet, income statement, and cash flow statement (collectively referred to as the "three statements") from the image files of the client's audit report, annual report, and reports for a specified period. The processed "three statements" serve as the core target files for subsequent processing. The specific process is as follows:

[0095] (1) Filtering table content

[0096] Its purpose is to remove invalid content from non-table pages and locate scanned documents containing tables; this can be achieved based on general OCR table recognition technology; the selection criterion is: only scanned documents containing table elements recognized by OCR are retained. After completion, the next step can be taken.

[0097] (2) Table type determination

[0098] Its purpose is to accurately identify and filter out target documents in the "three tables" category. Technically, it combines prompt word engineering with a large-scale model for classification and judgment. Judgment criteria: Based on the large-scale model's determination of the table's subject category, documents clearly belonging to the "three tables" category are selected.

[0099] For example, prompt word template:

[0100] Based on the obtained source data, please determine whether this page is part of a financial statement and which type it is derived from: [Balance Sheet, Income Statement, Cash Flow Statement] or [other than the above statements]. The source data is as follows: {0}.

[0101] Note:

[0102] {0} is taken from the source data content;

[0103] The large model returns the following results:

[0104] Return to this page type: one of the following: Balance Sheet, Income Statement, Cash Flow Statement, or a different statement.

[0105] (3) Verification of table integrity

[0106] Its purpose is to integrate table content that spans multiple pages into a complete object. Technically, it is based on prompt word engineering and integrity verification of the large model. Verification criteria: If the large model determines that the table content is complete and has no page breaks, it is marked as a valid target file; if page breaks exist, it is classified as an object to be merged.

[0107] For example, prompt word template:

[0108] Based on the obtained source data, determine whether this page contains the key account {1} marked as the end of {0} and its related balance. If the relevant information for this account is found, return "Complete"; otherwise, return "Incomplete". The source data is as follows: {2}.

[0109] Note:

[0110] {0} is taken from one of the following: the balance sheet, income statement, or cash flow statement.

[0111] {1} is taken from the standard account of {0}, such as "Total Liabilities and Owner's Equity" on the balance sheet.

[0112] {2} is taken from the source data content.

[0113] The large model returns the following results:

[0114] One of the categories: [Complete / Incomplete].

[0115] As can be seen from the above description, the method for extracting report data based on large model technology provided in this application can use a pre-built intelligent report prompt word project to intelligently split the acquired report image data into tables to obtain effective target reports.

[0116] In one embodiment, see Figure 3 The step of dynamically extracting metadata from the effective target report using a pre-built data extraction prompt word project to obtain report tags includes:

[0117] S301: Input the effective target report and the data extraction prompts in the data extraction prompts project into a pre-trained large model to extract report labels from the effective target report and obtain the report name labels;

[0118] S302: Based on the table name labels and the preset financial account and balance mapping relationship, the data format of the effective target report is parsed to obtain the report labels corresponding to each column of data.

[0119] It is understandable that steps S301 to S302 correspond to the dynamic extraction process of metadata.

[0120] For files that have been split and categorized, extract report tags and format information to support subsequent data categorization and import, as well as financial account mapping requirements. The specific technical process is as follows:

[0121] (1) Extraction of report labels

[0122] Its purpose is to obtain structured metadata such as customer name, reporting period, and consolidation scope. Technically, this can be achieved using OCR text recognition + prompt word engineering + key information extraction from a large model. Judgment criteria: Core fields such as customer name, reporting period, and consolidation scope must be completely extracted as valid tags.

[0123] For example, prompt word template:

[0124] Please determine the relevant information for {0} based on the obtained source data:

[0125] Company Name: Returns the string that most likely represents the name of the organization;

[0126] Reporting Period: Determine the current period date for the financial statements. The returned format is "xxxx year xx month";

[0127] Scope of merger: [Headquarters, Merger] (one of the following).

[0128] The return result keys include "Company Name, Reporting Period, Consolidation Scope". Please return the results in the order given above. {1}

[0129] Note:

[0130] {0} is taken from one of the following: the balance sheet, income statement, or cash flow statement.

[0131] {1} is taken from the source data content.

[0132] The large model returns the following results:

[0133] This includes "company name, reporting period, and scope of consolidation".

[0134] (2) Report format parsing

[0135] Its purpose is to identify table column types to support the mapping of financial accounts to balances. Technically, it is based on prompt word engineering and large-scale model column type recognition technology. Judgment criteria: It is necessary to clearly distinguish the core column type labels in the table, such as account columns, current period amount columns, and previous period amount columns.

[0136] For example, prompt word template:

[0137] Data was extracted from multiple columns of the report using OCR. The data for each column of the balance sheet was extracted. Please determine the data type of each column: report item column, current period balance column, previous period balance column, empty column, or other column. Please strictly adhere to the above criteria when returning the corresponding column type. The return format is "{0}", and the result should only contain the column type and a comma. All column data is as follows: {1}

[0138] Note:

[0139] {0} is a string concatenated by the program, such as "the type corresponding to column 1, the type corresponding to column 2, the type corresponding to column 3, etc.";

[0140] {1} The source data of each column extracted by the program;

[0141] The large model returns the following results:

[0142] Each column corresponds to a different data type.

[0143] As can be seen from the above description, the method for extracting report data based on large model technology provided in this application can dynamically extract metadata from the effective target report using a pre-built data extraction prompt word project to obtain report tags.

[0144] In one embodiment, see Figure 4 The process of using pre-built data integration prompts and report tags to perform multi-dimensional data integration on the effective target report, resulting in a report data dictionary, includes:

[0145] S401: Merge the effective target reports across pages according to the data integration prompts in the data integration prompts project and the report labels to obtain a complete report of account columns;

[0146] S402: Extract the values ​​corresponding to each account column in the complete account column report;

[0147] S403: Convert the subject columns and their corresponding values ​​into the report data dictionary.

[0148] It is understandable that steps S401 to S403 correspond to the multidimensional data integration process.

[0149] (1) Merging tables across pages

[0150] Its purpose is to merge the content of the same table scattered across multiple pages into a complete dataset by identifying cross-page relationships. Technically, it is based on prompt word engineering and cross-page integrity verification technology using a large model. Judgment criteria: All subject column data must be confirmed to be complete and without cross-page breaks.

[0151] The specific implementation process is as follows:

[0152] Step 1: Based on the large model, extract the page types and column labels of each data source page. Further subdivide the case of multiple tables per page. This involves extracting the data from each page sequentially according to its table label, and then incorporating that page into the source data objects of different tables. Refer to the steps above for extraction and labeling methods.

[0153] Step 2: Merge the existing data by label through code engineering. During the merging process, make comprehensive use of the labels obtained by the above methods to organize the table account columns and balance columns in an orderly manner and merge them according to the source table order.

[0154] (2) Financial Data Analysis

[0155] Its purpose is to extract account and balance data based on report format information to support subsequent standardized processing. Technically, structured data extraction is achieved through programming languages.

[0156] The specific implementation steps are as follows:

[0157] 1. Establish a dynamic data extraction system through tags. Further categorize columns into account columns, balance columns, and other columns. Using large-scale modeling techniques, further categorize the balance column into current period columns and previous period columns. Incorporate all types of tables into the aforementioned tagging system.

[0158] 2. Establish a task mode of extracting data by subject and period at a time. Based on the established tagging system, various reports will be dynamically extracted in a task mode that extracts only one column of subject and one period's balance at a time, thus completing the extraction of all report data.

[0159] 3. Establish an appropriate data storage system: Store data in a dictionary structure, with the key being the subject name and the value containing the corresponding previous and current period balance data (example variable names: dicPreviousSourceData, dicCurrentSourceData), as shown in Table 1.

[0160] Table 1

[0161]

[0162] As can be seen from the above description, the method for extracting report data based on large model technology provided in this application can perform multi-dimensional data integration processing on the effective target report using a pre-constructed data integration prompt word project and the report tags to obtain a report data dictionary.

[0163] In one embodiment, see Figure 5 The step of performing subject semantic mapping on the valid target report based on the report data dictionary to obtain the required data report includes:

[0164] S501: Map each subject in the valid target report to a standard subject according to the report data dictionary;

[0165] S502: Perform balance aggregation semantic mapping on the balances corresponding to each item in the effective target report to obtain the balances corresponding to the standard items; then, generate the required data report based on the standard items and the balances corresponding to the standard items.

[0166] It is understandable that steps S501 to S502 correspond to the subject semantic mapping process.

[0167] (1) Subject standardization mapping

[0168] Its purpose is to map non-standard reporting items to the enterprise's standard reporting item system through semantic matching. Technically, it is based on prompt word engineering and semantic association analysis technology of a large model. Mapping standard: It needs to return the mapping relationship between all items and standard items, allowing multiple non-standard items (such as the key-value pairs of dictionaries dicPreviousSourceData and dicCurrentSourceData) to be mapped to the same standard item.

[0169] This process primarily addresses the issues of diversity and personalization in non-standard report items, and it relies on the following prompt word templates:

[0170] For example, prompt word template:

[0171] The customer's financial statement OCR recognition retrieved the account string "{0}". The bank has standard account names for its financial statements, and you need to determine which standard account this string most likely describes. As an experienced financial expert, you should return the name of the standard account most likely associated with this string. The returned account name is limited to one of the following standard account names, and the account name characters must be exactly the same.

[0172] {1} Corresponding standard subject result: [Standard report subject name, no corresponding] one of them.

[0173] The results are returned in JSON format. Apart from the JSON, no other content is required in the results. The key is the string of the customer's report account, and the value is the name of the standard report account associated with it.

[0174] The standard report item name range is as follows: {2}

[0175] Note:

[0176] {0} Customer report subject string;

[0177] {1} Customer report subject string;

[0178] {2} Standard report subject name set.

[0179] The large model returns the following results:

[0180] This subject may not have a corresponding standard subject name.

[0181] (2) Balance aggregation mapping

[0182] The purpose is to aggregate the balances of multiple accounts in non-standard financial statements to the corresponding standard accounts of the enterprise. Technically, this is achieved through prompt word engineering and a large model to perform multi-source balance correlation analysis. Mapping standard: The mapping relationship between standard and non-standard account balances must be successfully returned. (For example, the standard account "Other Receivables" needs to integrate the balance data of the non-standard accounts "Interest Receivable" and "Other Receivables".)

[0183] As can be seen from the above description, the method for extracting report data based on large model technology provided in this application can perform subject semantic mapping on the effective target report according to the report data dictionary to obtain the required data report.

[0184] In one embodiment, the method for extracting report data based on large model technology further includes:

[0185] Dynamic prompts are generated based on the data in the effective target report and preset prompt anchor points to support batch queries to the large model; wherein, the dynamic prompts include intelligent report prompts, data extraction prompts and data integration prompts.

[0186] Understandably, the specific implementation process for the project prompt word project is as follows:

[0187] 1. A dynamic template library of prompt words based on report format and extracted content.

[0188] By pre-setting anchor points in the prompts to create dynamic prompts, dynamic templates can supplement the prompts with more precise question directions based on the project environment.

[0189] Meanwhile, the project construction dynamic prompt word template library effectively isolates the refactoring risks caused by prompt word changes from the code engineering.

[0190] For example, during the report splitting phase, when determining whether two consecutive reports are the same report, queries can be made in the template regarding the type of report already obtained and whether any key accounts appear, to help the large model make a judgment in a more concrete environment and return the expected results.

[0191] 2. Multi-keyword batch queries and preset responses.

[0192] Multi-keyword batch queries can significantly improve query efficiency. The core of multi-keyword batch queries is to ensure that the large model returns the corresponding keywords in the expected order. By pre-setting the return order and the return range, return efficiency and stability can be guaranteed.

[0193] For example, during the report subject mapping stage, it is necessary to query the large model in batches for subjects that fail to match normally and require them to match the enterprise standard subjects.

[0194] 3. Examples based on prompt word template library and preset strategies (see Table 2).

[0195] Table 2

[0196]

[0197] Anchor point annotation:

[0198] {0}: Generated by code. Based on [Customer Report Subject 1 Conversion Result, Customer Report Subject 2 Conversion Result, ...].

[0199] {1}: Generated from code. Requires the number of subjects that the large model needs to return.

[0200] {2}: Generated by code. Customer report items and their order, for example, [Customer report item 1: Accounts Receivable; Customer report item 2: Accounts Payable...].

[0201] {3}: The standard subject string for internal enterprise reports.

[0202] As can be seen from the above description, the method for extracting report data based on large model technology provided in this application can generate dynamic prompt words based on the data in the effective target report and preset prompt word anchor points, so as to support batch queries to the large model.

[0203] Based on the same inventive concept, this application also provides an apparatus for extracting report data based on large model technology, which can be used to implement the method described in the above embodiments, as described in the following embodiments. Since the principle of the apparatus for extracting report data based on large model technology is similar to that of the method for extracting report data based on large model technology, the implementation of the apparatus for extracting report data based on large model technology can refer to the implementation of the method based on software performance benchmarks, and repeated details will not be elaborated further. As used below, the terms "unit" or "module" can refer to a combination of software and / or hardware that performs a predetermined function. Although the system described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0204] In one embodiment, see Figure 6In order to build a complete automated processing flow, this application provides a device for extracting report data based on large model technology, which combines optical symbol recognition technology with large model technology to achieve intelligent report splitting, dynamic extraction of metadata, multi-dimensional data integration and subject semantic mapping. The device includes: a report splitting unit 601, a dynamic extraction unit 602, a data integration unit 603 and a semantic mapping unit 604.

[0205] The report splitting unit 601 is used to perform intelligent table splitting processing on the acquired report image data using a pre-built intelligent report prompt word project to obtain an effective target report;

[0206] Dynamic extraction unit 602 is used to dynamically extract metadata from the effective target report using a pre-built data extraction prompt word project to obtain report tags;

[0207] Data integration unit 603 is used to perform multi-dimensional data integration processing on the effective target report using a pre-built data integration prompt word project and the report tags to obtain a report data dictionary;

[0208] The semantic mapping unit 604 is used to perform subject semantic mapping on the valid target report according to the report data dictionary to obtain the required data report.

[0209] In one embodiment, see Figure 7 The report splitting unit 601 includes: an optical recognition module 701, a three-table filtering module 702, and a target report generation module 703.

[0210] Optical recognition module 701 is used to perform optical symbol recognition on the report image data to obtain the report data page;

[0211] The three-table filtering module 702 is used to input the report data page and the intelligent report prompts in the intelligent report prompts project into a pre-trained large model to obtain the target balance sheet, income statement and cash flow statement.

[0212] The target report generation module 703 is used to mark the target balance sheet, income statement, or cash flow statement as the valid target report if it is determined that the target balance sheet, income statement, or cash flow statement is complete and has no page breaks; otherwise, it generates the valid target report by merging the target balance sheet, income statement, or cash flow statement respectively.

[0213] In one embodiment, see Figure 8 The dynamic extraction unit 602 includes: a table name tag extraction module 801 and a report tag extraction module 802.

[0214] The table name tag extraction module 801 is used to input the effective target report and the data extraction prompt words in the data extraction prompt word project into a pre-trained large model to extract the report tags of the effective target report and obtain the table name tags;

[0215] The report label extraction module 802 is used to parse the data format of the effective target report according to the report name label and the preset financial subject and balance mapping relationship, so as to obtain the report label corresponding to each column of data.

[0216] In one embodiment, see Figure 9 The data integration unit 603 includes: a subject report generation module 901, a subject value extraction module 902, and a data dictionary generation module 903.

[0217] The subject report generation module 901 is used to merge the effective target report across pages according to the data integration prompt words in the data integration prompt word project and the report label to obtain a complete subject column report;

[0218] The subject value extraction module 902 is used to extract the values ​​corresponding to each subject column in the complete subject column report;

[0219] The data dictionary generation module 903 is used to convert the subject columns and their corresponding values ​​into the report data dictionary.

[0220] In one embodiment, see Figure 10 The semantic mapping unit 604 includes: a standard subject mapping module 1001 and a subject balance determination module 1002.

[0221] The standard subject mapping module 1001 is used to map each subject in the valid target report to a standard subject according to the report data dictionary;

[0222] The account balance determination module 1002 is used to perform balance aggregation mapping on the balances corresponding to each account in the valid target report to obtain the balance corresponding to the standard account.

[0223] In one embodiment, the apparatus for extracting report data based on large model technology further includes:

[0224] The prompt generation unit is used to generate dynamic prompt words based on the data in the effective target report and preset prompt word anchor points to support batch queries to the large model; wherein, the dynamic prompt words include intelligent report prompt words, data extraction prompt words and data integration prompt words.

[0225] From a hardware perspective, in order to build a complete automated processing flow, this application provides an embodiment of an electronic device for implementing all or part of the method for extracting report data based on large model technology. This electronic device specifically includes the following components: Optical symbol recognition technology is combined with large model technology to achieve intelligent report splitting, dynamic metadata extraction, multi-dimensional data integration, and subject semantic mapping.

[0226] The system comprises a processor, a memory, a communications interface, and a bus; wherein the processor, memory, and communications interface communicate with each other via the bus; the communications interface is used to realize information transmission between the device for extracting report data based on large model technology and core business systems, user terminals, and related databases and other related devices; the logic controller can be a desktop computer, tablet computer, or mobile terminal, etc., and this embodiment is not limited to these. In this embodiment, the logic controller can be implemented with reference to the embodiments of the method for extracting report data based on large model technology and the embodiments of the device for extracting report data based on large model technology in the embodiments, the contents of which are incorporated herein, and repeated details will not be described again.

[0227] It is understood that the user terminal may include smartphones, tablet computers, network set-top boxes, portable computers, desktop computers, personal digital assistants (PDAs), in-vehicle devices, smart wearable devices, etc. Among these, the smart wearable devices may include smart glasses, smartwatches, smart bracelets, etc.

[0228] In practical applications, the method for extracting report data based on large model technology can be partially executed on the electronic device side as described above, or all operations can be completed on the client device. The choice can be made based on the processing power of the client device and the limitations of the user's usage scenario. This application does not impose any limitations on this. If all operations are completed on the client device, the client device may further include a processor.

[0229] The aforementioned client device may have a communication module (i.e., a communication unit) that can communicate with a remote server to achieve data transmission. The server may include a server on the task scheduling center side; in other implementation scenarios, it may also include a server on an intermediate platform, such as a server on a third-party server platform that has a communication link with the task scheduling center server. The server may include a single computer device, a server cluster consisting of multiple servers, or a distributed server structure.

[0230] Figure 11This is a schematic block diagram illustrating the system configuration of the electronic device 9600 according to an embodiment of this application. Figure 11 As shown, the electronic device 9600 may include a central processing unit 9100 and a memory 9140; the memory 9140 is coupled to the central processing unit 9100. It is worth noting that... Figure 11 This is an example; other types of structures can also be used to supplement or replace this structure to achieve telecommunications functions or other functions.

[0231] In one embodiment, the method for extracting report data based on large model technology can be integrated into the central processing unit 9100. The central processing unit 9100 can be configured to perform the following controls:

[0232] S101: Utilize a pre-built intelligent report prompt word project to perform intelligent table splitting on the acquired report image data to obtain an effective target report;

[0233] S102: Dynamically extract metadata from the effective target report using a pre-built data extraction prompt word project to obtain report tags;

[0234] S103: Using the pre-built data integration prompt word project and the report tags, perform multi-dimensional data integration processing on the effective target report to obtain a report data dictionary;

[0235] S104: Perform subject semantic mapping on the valid target report according to the report data dictionary to obtain the required data report.

[0236] As described above, the method for extracting report data based on large-scale model technology provided in this application achieves fully automated conversion from unstructured image files to standardized financial data through four core steps: intelligent report splitting based on optical symbol recognition and large-scale model, dynamic metadata extraction, multi-dimensional data integration, and subject semantic mapping. First, optical symbol recognition and semantic analysis of the large-scale model accurately locate the target "three tables," effectively solving the parsing challenges of complex scenarios such as multi-page cross-tables and non-standard formats. Second, it dynamically extracts metadata such as customer names and reporting periods and parses column types to build a structured data foundation. Then, it integrates data through cross-page integrity verification and programming languages ​​to ensure the integrity of subject balances. Finally, using semantic association analysis of the large-scale model, it automatically maps non-standard subjects to the enterprise's standard system and completes balance aggregation. This fully automated process significantly improves the parsing efficiency of heterogeneous reports, while standardized mapping ensures data consistency, effectively supporting the needs of financial analysis and decision-making.

[0237] In another embodiment, the device for extracting report data based on large model technology can be configured separately from the central processing unit 9100. For example, the device for extracting report data based on large model technology can be configured as a chip connected to the central processing unit 9100, and the function of the method for extracting report data based on large model technology can be realized through the control of the central processing unit.

[0238] like Figure 11 As shown, the electronic device 9600 may further include: a communication module 9110, an input unit 9120, an audio processor 9130, a display 9160, and a power supply 9170. It is worth noting that the electronic device 9600 does not necessarily need to include these components. Figure 11 All components shown; in addition, the electronic device 9600 may also include Figure 11 For components not shown, please refer to existing technologies.

[0239] like Figure 11 As shown, the central processing unit 9100, sometimes also referred to as a controller or operating control, may include a microprocessor or other processor device and / or logic device, which receives inputs and controls the operation of various components of the electronic device 9600.

[0240] The memory 9140 may be, for example, one or more of a cache, flash memory, hard drive, removable media, volatile memory, non-volatile memory, or other suitable devices. It may store the aforementioned failure-related information, and also store a program for executing that information. The central processing unit 9100 may execute the program stored in the memory 9140 to perform information storage or processing, etc.

[0241] Input unit 9120 provides input to central processing unit 9100. Input unit 9120 may be, for example, a keypad or touch input device. Power supply 9170 provides power to electronic device 9600. Display 9160 displays images and text. Display may be, for example, an LCD display, but is not limited thereto.

[0242] The memory 9140 can be a solid-state memory, such as a read-only memory (ROM), random access memory (RAM), a SIM card, etc. It can also be a memory that retains information even when power is off, can be selectively erased, and contains more data; examples of this type of memory are sometimes referred to as EPROMs. The memory 9140 can also be some other type of device. The memory 9140 includes a buffer memory 9141 (sometimes referred to as a buffer). The memory 9140 may include an application / function storage unit 9142 for storing application programs and function programs or processes for executing the operation of the electronic device 9600 via the central processing unit 9100.

[0243] The memory 9140 may also include a data storage unit 9143 for storing data, such as contacts, digital data, pictures, sounds, and / or any other data used by the electronic device. The driver storage unit 9144 of the memory 9140 may include various drivers for the electronic device's communication functions and / or for performing other functions of the electronic device (such as messaging applications, address book applications, etc.).

[0244] The communication module 9110 is a transmitter / receiver that sends and receives signals via the antenna 9111. The communication module (transmitter / receiver) 9110 is coupled to the central processing unit 9100 to provide input signals and receive output signals, which is the same as in a conventional mobile communication terminal.

[0245] Based on different communication technologies, multiple communication modules 9110 can be configured in the same electronic device, such as cellular network modules, Bluetooth modules, and / or wireless LAN modules. The communication module (transmitter / receiver) 9110 is also coupled to a speaker 9131 and a microphone 9132 via an audio processor 9130 to provide audio output via the speaker 9131 and receive audio input from the microphone 9132, thereby realizing typical telecommunications functions. The audio processor 9130 may include any suitable buffer, decoder, amplifier, etc. Additionally, the audio processor 9130 is also coupled to a central processing unit 9100, enabling on-device recording via the microphone 9132 and on-device playback of stored sound via the speaker 9131.

[0246] Embodiments of this application also provide a computer-readable storage medium capable of implementing all steps of the method for extracting report data based on large model technology, where the execution subject is a server or client, as described in the above embodiments. The computer-readable storage medium stores a computer program that, when executed by a processor, implements all steps of the method for extracting report data based on large model technology, where the execution subject is a server or client, as described in the above embodiments. For example, when the processor executes the computer program, it implements the following steps:

[0247] S101: Utilize a pre-built intelligent report prompt word project to perform intelligent table splitting on the acquired report image data to obtain an effective target report;

[0248] S102: Dynamically extract metadata from the effective target report using a pre-built data extraction prompt word project to obtain report tags;

[0249] S103: Using the pre-built data integration prompt word project and the report tags, perform multi-dimensional data integration processing on the effective target report to obtain a report data dictionary;

[0250] S104: Perform subject semantic mapping on the valid target report according to the report data dictionary to obtain the required data report.

[0251] As described above, the method for extracting report data based on large-scale model technology provided in this application achieves fully automated conversion from unstructured image files to standardized financial data through four core steps: intelligent report splitting based on optical symbol recognition and large-scale model, dynamic metadata extraction, multi-dimensional data integration, and subject semantic mapping. First, optical symbol recognition and semantic analysis of the large-scale model accurately locate the target "three tables," effectively solving the parsing challenges of complex scenarios such as multi-page cross-tables and non-standard formats. Second, it dynamically extracts metadata such as customer names and reporting periods and parses column types to build a structured data foundation. Then, it integrates data through cross-page integrity verification and programming languages ​​to ensure the integrity of subject balances. Finally, using semantic association analysis of the large-scale model, it automatically maps non-standard subjects to the enterprise's standard system and completes balance aggregation. This fully automated process significantly improves the parsing efficiency of heterogeneous reports, while standardized mapping ensures data consistency, effectively supporting the needs of financial analysis and decision-making.

[0252] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, apparatus, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0253] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (devices), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0254] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0255] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0256] Specific embodiments have been used to illustrate the principles and implementation methods of this invention. The descriptions of the embodiments above are only for the purpose of helping to understand the method and core ideas of this invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this invention. Therefore, the content of this specification should not be construed as a limitation of this invention.

Claims

1. A method for extracting report data based on large model technology, characterized in that, The application comprises the following steps: Intelligent table splitting processing is performed on the obtained report image data by using a pre-constructed intelligent report prompt word template to obtain an effective target report; Metadata dynamic extraction processing is performed on the effective target report by using a pre-constructed data extraction prompt word template to obtain report labels; Multi-dimensional data integration processing is performed on the effective target report by using a pre-constructed data integration prompt word template and the report labels to obtain a report data dictionary; Subject semantic mapping is performed on the effective target report according to the report data dictionary to obtain a required data report.

2. The method for extracting report data based on large model technology according to claim 1, characterized in that, The intelligent table splitting processing on the obtained report image data by using the pre-constructed intelligent report prompt word template to obtain the effective target report comprises the following steps: Optical symbol recognition is performed on the report image data to obtain a report data page; The report data page and intelligent report prompt words in the intelligent report prompt word template are input into a pre-trained neural network model to obtain a target balance sheet, a profit sheet and a cash flow sheet; If it is determined that the target balance sheet, the profit sheet or the cash flow sheet is complete and has no cross-page fault, the target balance sheet, the profit sheet or the cash flow sheet is marked as the effective target report; Otherwise, the cross-page fault part in the target balance sheet is merged to obtain a merged target balance sheet; the cross-page fault part in the profit sheet is merged to obtain a merged profit sheet; the cross-page fault part in the cash flow sheet is merged to obtain a merged cash flow sheet; and the effective target report is generated according to the merged target balance sheet, the merged profit sheet and the merged cash flow sheet.

3. The method for extracting report data based on large model technology according to claim 1, characterized in that, The metadata dynamic extraction processing on the effective target report by using the pre-constructed data extraction prompt word template to obtain the report labels comprises the following steps: The effective target report and data extraction prompt words in the data extraction prompt word template are input into a pre-trained neural network model to extract report labels from the effective target report to obtain a table name label; According to the mapping relationship between the table name label, a preset financial subject and a balance, data format analysis is performed on the effective target report to obtain the report labels corresponding to each column of data.

4. The method for extracting report data based on large model technology according to claim 1, characterized in that, The multi-dimensional data integration processing on the effective target report by using the pre-constructed data integration prompt word template and the report labels to obtain a report data dictionary comprises the following steps: Cross-page table merging is performed on the effective target report according to data integration prompt words in the data integration prompt word template and the report labels to obtain a subject column complete report; Values corresponding to each subject column in the subject column complete report are extracted; The each subject column and the corresponding values are converted into the report data dictionary.

5. The method for extracting report data based on large model technology according to claim 1, characterized in that, The subject semantic mapping on the effective target report according to the report data dictionary to obtain a required data report comprises the following steps: Each subject in the effective target report is mapped to a standard subject according to the report data dictionary; Balance aggregation semantic mapping is performed on the balances corresponding to each subject in the effective target report to obtain the balances corresponding to the standard subject; According to the standard subject and the balance corresponding to the standard subject, the required data report is generated.

6. The method for extracting report data based on large model technology according to claim 1, characterized in that, Further comprising: According to the data in the effective target report and the preset prompt word anchor, a dynamic prompt word is generated to support batch inquiry to a neural network model; wherein the dynamic prompt word includes an intelligent report prompt word, a data extraction prompt word and a data integration prompt word.

7. An apparatus for extracting report data based on large model technology, characterized by comprising: Comprising: A report splitting unit is configured to perform intelligent table splitting processing on the obtained report image data by using a pre-constructed intelligent report prompt word template, to obtain an effective target report; A dynamic extraction unit is configured to perform metadata dynamic extraction processing on the effective target report by using a pre-constructed data extraction prompt word template, to obtain a report label; A data integration unit is configured to perform multi-dimensional data integration processing on the effective target report by using a pre-constructed data integration prompt word template and the report label, to obtain a report data dictionary; A semantic mapping unit is configured to perform subject semantic mapping on the effective target report according to the report data dictionary, to obtain a required data report.

8. An electronic device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, The processor executes the program to realize the steps of the method for extracting report data based on large model technology according to any one of claims 1 to 6.

9. A computer readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to realize the steps of the method for extracting report data based on large model technology according to any one of claims 1 to 6.

10. A computer program product comprising computer programs / instructions, characterized in that, The computer program / instruction is executed by the processor to realize the steps of the method for extracting report data based on large model technology according to any one of claims 1 to 6.

Images