Financial announcement data processing system, methods, equipment and storage media

Abstract

This application proposes a financial announcement data processing system, method, device, and storage medium, relating to the field of data processing technology. It includes: a semantic extraction module for extracting key metadata and specific text paragraphs from an input document based on preset extraction rules, constructing a structured data framework based on the extracted data, and outputting an initial report file based on the structured data framework and a predefined template; a report parsing module for extracting quantitative financial indicators from the input document, standardizing the extracted quantitative financial indicators, filling the financial data columns of the initial report file with structured financial indicator data based on the standardized quantitative financial indicators, and generating a financial report file; and a specialized refinement module for performing semantic reconstruction and format standardization operations on target indicators in the financial report file, generating a target report file. The aforementioned financial announcement data processing system can automatically generate structured reports based on financial announcement documents.

Description

Technical Field

[0001] This application relates to the field of data processing technology, and in particular to a financial announcement data processing system, method, device and storage medium. Background Technology

[0002] Financial analysis is based on financial data and uses a series of specialized analytical methods to analyze a user's past and present profitability, solvency, and growth potential related to financing and investment activities.

[0003] Currently, financial institutions and research institutions conduct financial analysis on listed companies, requiring them to analyze financial announcements such as preliminary earnings reports. However, existing methods for analyzing financial announcements primarily involve manually reading PDF documents and manually entering data into Excel spreadsheets. Manually processing unstructured data is typically inefficient and costly. Summary of the Invention

[0004] In view of this, this application proposes a financial announcement data processing system, method, apparatus and storage medium.

[0005] In a first aspect, this application provides a financial announcement data processing system, which includes: a semantic extraction module, a report parsing module, and a special refinement module; The semantic extraction module is used to extract key metadata and specific text paragraphs from the input document based on preset extraction rules, construct a structured data framework based on the extracted data, and output an initial report file based on the structured data framework and a predefined template. The report parsing module is used to extract quantitative financial indicators from the input document, standardize the extracted quantitative financial indicators, fill the structured financial indicator data into the financial data column of the initial report file according to the standardized quantitative financial indicators, and generate a financial report file. The specialized refining module is used to perform semantic reconstruction and format standardization operations on the target indicators in the financial statement file to generate the target report file; wherein, the semantic reconstruction operation includes extracting periodic data from the target indicators to determine the periodic data; and calculating periodic variation indicators based on the periodic data.

[0006] In one embodiment, the semantic extraction module includes: a preprocessing submodule and a data extraction submodule; The preprocessing submodule is used to extract text content from the input document and integrate it into a text format file stream, and to perform semantic normalization preprocessing on the text format file stream. The data extraction submodule is used to locate feature semantic segments in the text format file stream using regular expressions, perform denoising and standardization processing based on the feature semantic segments, extract structured metadata and structured text paragraphs, construct a structured data framework based on the structured metadata and structured text paragraphs, and output an initial report file based on the structured data framework and a predefined template.

[0007] In one embodiment, the preprocessing submodule includes: an identification unit and a semantic parsing unit; The recognition unit is used to convert the input document into an image when the input document is a scanned PDF document or an image PDF document, perform OCR recognition on the converted image, and generate a recognized document; when the input document is a text PDF document, the input document is directly used as the recognized document. The semantic parsing unit is used to perform page-by-page semantic parsing on the identified document, extract the text content and integrate it into a text format file stream, and perform semantic normalization preprocessing on the text format file stream.

[0008] In one embodiment, the financial announcement data processing system further includes: an announcement date parsing module; The announcement date parsing module is used to perform semantic conversion on the extracted date format when Arabic numeral or Chinese numeral date is extracted from the text format file stream to generate a standardized date format; when Arabic numeral or Chinese numeral date is not extracted, the date is extracted from the last page of the text format file stream and semantic conversion is performed to generate a standardized date format.

[0009] In one embodiment, the report parsing module includes: a quantitative value extraction submodule, a unit of measurement processing submodule, and a financial indicator quantitative parsing submodule; The quantitative value extraction submodule is used to convert the input document into a temporary docx document, extract table row-level semantic data from the temporary docx document, and extract the quantitative values ​​of financial indicators based on the table row-level semantic data; when it is determined that the table semantics are missing, the input document is parsed using the full-text semantic extraction mode, and the quantitative values ​​of financial indicators are extracted based on keywords. The unit of measurement processing submodule is used to detect units of measurement from the temporary docx document using regular expressions, and to perform numerical scaling processing on the extracted financial indicator quantification values ​​according to the units of measurement of the predefined template and the detected units of measurement, so as to generate corresponding quantitative financial indicators. The financial indicator quantification and analysis submodule is used to standardize the quantitative financial indicators, fill the structured financial indicator data into the financial data column of the initial report file according to the standardized quantitative financial indicators, and generate a financial report file.

[0010] In one embodiment, the specialized refining module includes: a semantic reconstruction submodule and a standardization submodule; The semantic reconstruction submodule is used to extract the previous period value and the current period value of the corresponding target indicator, and to determine the rate of change based on the previous period value and the current period value of the corresponding target indicator; The standardization submodule is used to convert the quantized value of the target indicator into a percentage format when the numerical semantics of the target indicator are detected to be a percentage.

[0011] In one embodiment, the structured data framework includes a key-value mapping structure for multiple fields, with the initial value of each field set as a default placeholder and each field corresponding to a column index of the table.

[0012] Secondly, this application also provides a method for processing financial announcement data, the method comprising: Extract key metadata and specific text paragraphs from the input document based on preset extraction rules; A structured data framework is constructed based on the extracted data, and an initial report file is output based on the structured data framework and a predefined template. Quantitative financial indicators are extracted from the input document, and the extracted quantitative financial indicators are standardized. The structured financial indicator data is filled into the financial data columns of the initial report file based on the standardized quantitative financial indicators to generate the financial report file; The target indicators in the financial statement file are subjected to semantic reconstruction and format standardization operations to generate the target report file; wherein, the semantic reconstruction operation includes extracting periodic data from the target indicators to determine the periodic data; and calculating periodic variation indicators based on the periodic data.

[0013] Thirdly, this application also provides an electronic device, including a processor and a memory; the memory has a computer program stored thereon, wherein the computer program, when executed by the processor, implements the financial announcement data processing method as described in the first aspect.

[0014] Fourthly, this application also provides a computer storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the financial announcement data processing method as described in the first aspect.

[0015] The financial announcement data processing system proposed in this application has the following advantages over related technologies: 1. The semantic extraction module extracts key metadata and specific text paragraphs from the input document based on preset extraction rules. A structured data framework is constructed based on the extracted data, and an initial report file is output based on this framework and a predefined template. Then, the report parsing module extracts quantitative financial indicators from the input document, standardizes these indicators, and fills the financial data columns of the initial report file with the standardized quantitative financial indicators, generating the financial report file. This distinguishes between semantic and quantitative data, avoiding the problem of mixing semantic and quantitative data, and improving processing efficiency and result reliability. Furthermore, the specialized refinement module performs semantic reconstruction and format standardization operations on the target indicators in the financial report file, generating the target report file automatically. This process highlights and analyzes key financial indicators for easy user review and analysis.

[0016] 2. The financial announcement data processing system of this application integrates regular expression matching and OCR technology to construct a multimodal semantic extraction framework, which can handle heterogeneous data sources such as text-based PDFs, scanned PDFs and image-based PDFs, thereby enhancing the robustness of semantic extraction.

[0017] 3. The financial announcement data processing system of this application can automatically extract results from input documents to specified cells in a table template, and intelligently format and output data of types such as numerical values, percentages, and text, forming an integrated closed loop from document upload to structured report generation, meeting the high-efficiency production needs of large-scale batch processing of financial data. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments or related technologies of this application, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a schematic diagram of the structure of a financial announcement data processing system according to one embodiment of this application; Figure 2 This is a schematic diagram of the structure of a financial announcement data processing system according to another embodiment of this application; Figure 3 This is a schematic diagram of the structure of the financial announcement data processing system in another embodiment of this application; Figure 4 This is a schematic diagram of the structure of a financial announcement data processing system in another embodiment of this application; Figure 5 This is a schematic diagram of the structure of a financial announcement data processing system in another embodiment of this application; Figure 6 This is a schematic diagram of a portion of the target report file in one embodiment of this application; Figure 7 This is a flowchart illustrating a financial announcement data processing method in one embodiment of this application.

[0020] Explanation of reference numerals in the attached figures: 10-Financial Announcement Data Processing System, 11-Semantic Extraction Module, 111-Preprocessing Submodule, 1111-Identification Unit, 1112-Semantic Parsing Unit, 112-Data Extraction Submodule, 12-Report Parsing Module, 121-Quantitative Value Extraction Submodule, 122-Unit of Measurement Processing Submodule, 123-Financial Indicator Quantitative Parsing Submodule, 13-Specialized Refinement Module, 131-Semantic Reconstruction Submodule, 132-Standardization Submodule, 14-Input Layer, 15-Output Layer, 16-Announcement Date Parsing Module. Detailed Implementation

[0021] The technical solutions in the embodiments of this application will be clearly and completely described below with reference to the embodiments of this application. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0022] As described in the background section, financial analysis is based on financial data and employs a series of specialized analytical methods to analyze a user's past and present profitability, solvency, and growth capabilities related to financing and investment activities.

[0023] Currently, financial institutions and research institutions conduct financial analysis on listed companies, requiring them to analyze financial announcements such as preliminary earnings reports. However, existing methods for analyzing financial announcements primarily involve manually reading PDF documents and manually entering data into Excel spreadsheets. Manually processing unstructured data is typically inefficient and costly.

[0024] Based on this, in some embodiments, such as Figure 1 As shown, this application provides a financial announcement data processing system 10, which includes: a semantic extraction module 11, a report parsing module 12, and a special refinement module 13.

[0025] The semantic extraction module 11 is used to extract key metadata and specific text paragraphs from the input document based on preset extraction rules, construct a structured data framework based on the extracted data, and output an initial report file based on the structured data framework and a predefined template. The input document (e.g., company announcements, financial reports, etc.) is generally a PDF document. Key metadata may include company identifiers, securities codes, announcement identifiers, etc.

[0026] It is understandable that the semantic extraction module 11 mainly extracts text content. Based on pre-set extraction rules (such as regular expression matching), it accurately filters and extracts key metadata and specific text paragraphs with business value. Then, it integrates these extracted scattered data according to preset logical relationships to build a well-organized and formatted structured data framework. Finally, using this structured data framework as the data source, combined with a predefined report template (including fixed column settings, format requirements, etc.), it automatically fills in the data and generates an initial report file (such as an Excel spreadsheet), thereby completing the transformation from the original input document to a standardized initial report.

[0027] The report parsing module 12 is used to extract quantitative financial indicators from the input document, standardize the extracted quantitative financial indicators, fill the structured financial indicator data into the financial data column of the initial report file according to the standardized quantitative financial indicators, and generate a financial report file.

[0028] The report parsing module 12 primarily processes financial data. Its function is to automatically identify and extract key quantitative financial indicators from input documents (such as financial tables). Then, it standardizes these extracted raw indicators (e.g., standardizing currency units, correcting data precision, etc., to ensure comparability of data from different sources). Finally, it maps the standardized financial indicators to the corresponding financial data columns in the initial report file according to preset rules, ultimately generating a complete and formatted financial report file. This process automates the accurate extraction of financial data from unstructured text and its integration into standard reports, effectively improving the efficiency and accuracy of financial data processing compared to manual input.

[0029] The specialized refinement module 13 is used to perform semantic reconstruction and format standardization operations on target indicators in financial statement files to generate target report files; among them, the semantic reconstruction operation includes extracting periodic data from target indicators to determine periodic data; and calculating periodic variation indicators based on periodic data.

[0030] The Specialized Refinement Module 13 primarily conducts in-depth analysis of key indicators in financial statement documents. It extracts the values ​​of target indicators across different periods from the financial statements, identifies the corresponding periodic data for each period, and then calculates periodic variation indicators based on this periodic data. Next, it standardizes the format of the target indicators (including periodic data and periodic variation indicators), unifying their presentation and ensuring consistent, readable, and understandable data. Through these two operations, the Specialized Refinement Module 13 transforms scattered target indicators in financial statements into target report documents that combine trend analysis value with a standardized format, providing a more accurate, focused, and standardized data foundation for subsequent financial analysis and decision support.

[0031] It should be noted that in applications, such as Figure 2 As shown, the financial announcement data processing system 10 may also include an input layer 14 and an output layer 15. The input layer 14 is used to import input documents in batches and can also import predefined templates. The output layer 15 is used to output target report files.

[0032] The aforementioned financial announcement data processing system 10 extracts key metadata and specific text paragraphs from the input document using a semantic extraction module 11 based on preset extraction rules. It then constructs a structured data framework based on the extracted data and outputs an initial report file using the structured data framework and a predefined template. Next, the report parsing module 12 extracts quantitative financial indicators from the input document and standardizes these indicators. The standardized quantitative financial indicators are then used to fill the financial data columns of the initial report file with structured financial indicator data, generating the financial report file. This distinguishes between semantic and quantitative data, avoiding the problem of mixed processing and improving processing efficiency and result reliability. Furthermore, the specialized refinement module 13 performs semantic reconstruction and format standardization operations on the target indicators in the financial report file, generating the target report file automatically. This system highlights and analyzes key financial indicators for easy user review and analysis.

[0033] In some embodiments, such as Figure 3 As shown, the semantic extraction module 11 includes a preprocessing submodule 111 and a data extraction submodule 112.

[0034] The preprocessing submodule 111 is used to extract text content from the input document and integrate it into a text format file stream, and to perform semantic normalization preprocessing on the text format file stream. The semantic normalization preprocessing may include removing redundant punctuation (such as commas), unifying irregular characters (such as Chinese dashes to standard format), and standardizing measurement symbols (such as percentages) to ensure semantic consistency in subsequent pattern matching.

[0035] The data extraction submodule 112 is used to locate feature semantic fragments in a text format file stream using regular expressions, perform denoising and standardization processing based on the feature semantic fragments, extract structured metadata and structured text paragraphs, construct a structured data framework based on the structured metadata and structured text paragraphs, and output an initial report file based on the structured data framework and a predefined template.

[0036] It is understandable that the preprocessing submodule 111 extracts text content from the input document and integrates it into a text format file stream, which can effectively avoid the impact of tabular data on the extraction of key metadata and specific text paragraphs. The data extraction submodule 112 performs in-depth processing on the text format file stream after semantic normalization preprocessing. First, based on the regular expression pattern preset according to business requirements (multiple regular expressions with different expressions can be set), it accurately scans the normalized text stream, locates and locks the feature semantic segments that conform to business rules; then, it performs noise reduction processing on these segments (such as removing non-core information such as announcement numbers, irrelevant remarks, and redundant modifiers) and standardization processing (such as converting dates to "YYYY-MM-DD" format and aligning the metadata expression to preset standards), extracting structured metadata and structured text paragraphs; then, according to preset field mapping rules (such as the key-value relationship corresponding to Excel column indexes), it integrates these structured data into a clear structured data framework; finally, using this framework as the data source, combined with predefined report templates, it automatically fills in the data and outputs the initial report file, completing the automated conversion from unstructured text to standardized initial report.

[0037] In one embodiment, a structured data framework includes a key-value mapping structure with multiple fields, where the initial value of each field is set to a default placeholder, and each field corresponds to a column index of the table.

[0038] For example, a key-value mapping structure info containing 59 fields can be constructed, with the default placeholder " "Initialization involves mapping each field to an Excel column index (e.g., index 0 maps to column A, and index 48 maps to column AS). This structure provides a formalized storage system for semantic extraction results, ensuring the standardization of data organization."

[0039] Based on this, corresponding to the aforementioned embodiments, the metadata and its processing can be as follows: Company Identifier (Column A, Index 0): Uses regular expressions to capture semantic fragments, removes redundant whitespace and noise characters, and generates the company name; Securities Abbreviation (Column B, Index 1): Extracts the semantic text after "Securities Abbreviation" or "Stock Abbreviation", removes irrelevant word order (such as announcement identifiers), performs deduplication, and separates with slashes; Securities Code (Column C, Index 2): Matches six-digit codes, removes duplicates, and connects with slashes to ensure code uniqueness; Bond Financial Instruments (Columns D, E, Indexes 3, 4): Extracts bond codes and their abbreviations, handling semantic missing scenarios; Announcement Title (Column F, Index 5): Captures the semantic context between "Earnings Reports" or extracts semantic fragments containing "Earnings Reports"; Announcement Identifier (Column G, Index 6): Matches announcement numbers using multi-pattern regular expressions (such as "Temporary XXXX-XXXX") and performs format normalization.

[0040] The process of locating and processing specific text paragraphs can be as follows: AS column (index 48): Captures high-priority semantic tags such as "the main reasons for exceeding 30%" through the find-content-by-keywords submodule and extracts them to the semantic sentence "II."; AX column (index 49): Extracts the semantic content between "Explanation of the difference from the previous performance forecast" and "IV." to record the deviation analysis of the performance forecast; AZ column (index 51): Calls extract-auditor-info to extract the semantics of the auditing institution between the two punctuation marks after "audited by the accounting firm"; BA column (index 52): Extracts the semantics of risk disclosure between "This announcement" and "Please pay attention to investment risks" through extract-risk-warning; AR column (index 43): Extracts the semantics of performance summary between "Explanation of operating performance and financial condition" and "III."

[0041] In some embodiments, such as Figure 4 As shown, the preprocessing submodule 111 includes: an identification unit 1111 and a semantic parsing unit 1112.

[0042] The recognition unit 1111 is used to convert the input document into an image when the input document is a scanned PDF document or an image PDF document, perform OCR recognition on the converted image, and generate a recognized document; when the input document is a text PDF document, the input document is directly used as the recognized document.

[0043] The semantic parsing unit 1112 is used to perform page-by-page semantic parsing of the identified document, extract the text content and integrate it into a text format file stream, and perform semantic normalization preprocessing on the text format file stream.

[0044] It is understandable that the input document is not necessarily a text-based PDF document. Therefore, to ensure the accuracy of the target report file, non-text-based PDF documents can be pre-processed for recognition. Specifically, the recognition unit 1111 judges the format of the input document. If it detects a scanned PDF document (i.e., the document consists of images, and the text cannot be directly extracted) or an image-based PDF document (embedding a large number of images as text content), a PDF parsing tool is called to convert the document into an image sequence (such as JPEG / PNG format). Then, OCR (Optical Character Recognition) technology is used to recognize the text in each image, converting the text in the images into editable text to generate a recognized document. If the input document is a text-based PDF document (the text is stored in text format and can be directly extracted), it is directly used as the recognized document, skipping the image conversion and OCR steps to improve efficiency.

[0045] After receiving the document, the semantic parsing unit 1112 performs page-by-page semantic parsing. First, it traverses the document page by page, extracting the text content (including body text, titles, and text in tables) from each page. Then, it reorganizes and integrates the extracted text according to preset text structure rules (such as heading levels and paragraph separators), forming a continuous and complete text format file stream. Next, it performs semantic normalization preprocessing on this file stream, ensuring that the text stream meets the standards of semantic coherence and uniform format, providing high-quality input for the accurate processing of the subsequent data extraction submodule 112. This process, by implementing differentiated processing strategies based on document type, ensures the text recognition capability of the scanned document, optimizes the processing efficiency of the original text document, and improves text quality through semantic normalization, laying a solid foundation for subsequent metadata extraction and report generation.

[0046] It should be noted that when the regular expression extraction hit rate is lower than the threshold, it can automatically switch to the backup mode, i.e., the OCR recognition mode, and regenerate the recognition document.

[0047] In some embodiments, such as Figure 5 As shown, the financial announcement data processing system 10 also includes an announcement date parsing module 16.

[0048] The announcement date parsing module 16 is used to perform semantic conversion on the extracted date format when Arabic numeral or Chinese numeral date is extracted from the text format file stream to generate a standardized date format; when Arabic numeral or Chinese numeral date is not extracted, the date is extracted from the last page of the text format file stream and semantic conversion is performed to generate a standardized date format.

[0049] It can be understood that when the announcement date parsing module 16 detects an explicit date identifier in the text format file stream (such as "December 31, 2023" or "December 31st, 2023"), the module identifies the date format type, then converts the date identifier into a standard format (for example, "二〇二三" → "2023"), and uniformly standardizes it in the YYYY-MM-DD format (such as "2023-12-31"). If the document does not directly contain date information in the compliant format, the module will automatically switch to the last page extraction strategy, locate potential date text by analyzing fixed positions (such as the footer, signature) or feature fields (such as "release date", "signature time") on the last page of the document, then obtain the date content through OCR recognition (if the last page is an image) or text matching (if it is text), and finally perform the same operations of converting Chinese numerals to Arabic numerals and format standardization. This dual-strategy design effectively covers the common forms of date information in the document, and ensures the generation of a unified standard date format through intelligent judgment and format conversion, providing an accurate data basis for time series analysis in subsequent report generation.

[0050] Exemplarily, the announcement date can correspond to columns BE and BF, indexes 55 and 56. The cn2an library can be used to implement the semantic conversion of Chinese numerals to generate the standardized date format "YYYY-MM-DD".

[0051] In some embodiments, as Figure 3 shown, the report parsing module 12 includes: a quantitative value extraction sub-module 121, a measurement unit processing sub-module 122, and a financial indicator quantitative analysis sub-module 123.

[0052] The quantitative value extraction sub-module 121 is used to convert the input document into a temporary docx document, extract table row-level semantic data from the temporary docx document, and extract financial indicator quantitative values based on the table row-level semantic data; when it is determined that the table semantics are missing, a full-text semantic extraction mode is adopted to parse the input document, and financial indicator quantitative values are extracted based on keywords.

[0053] The measurement unit processing sub-module 122 is used to detect measurement units from the temporary docx document through regular expressions, and perform numerical scaling processing on the extracted financial indicator quantitative values according to the dimension of the predefined template and the detected measurement units to generate corresponding quantitative financial indicators.

[0054] The financial indicator quantitative analysis sub-module 123 is used to perform standardization processing on the quantitative financial indicators, and fill the structured financial indicator data into the financial data column of the initial report file according to the standardized quantitative financial indicators to generate a financial report file.

[0055] It is understandable that the quantitative value extraction submodule 121 converts the input document into a temporary docx format, utilizes its structured characteristics to extract table row-level semantic data (such as the correspondence between table headers and cells), and identifies financial indicators (such as "operating revenue") and their corresponding values ​​through predefined rules. If the table structure is incomplete or semantics are missing, it automatically switches to the full-text semantic extraction mode, and locates the quantitative values ​​in the scattered data based on keywords (such as "operating revenue") combined with contextual analysis. The unit of measurement processing submodule 122 scans the document using regular expressions to identify the unit of measurement (e.g., "ten thousand yuan") after each quantified value, and performs numerical scaling according to the standard units of measurement in a predefined template (e.g., unifying them to "yuan"), ensuring that all indicators have consistent units of measurement. The financial indicator quantification and analysis submodule 123 performs standardization processing on the converted values, including format unification (e.g., retaining four decimal places), outlier correction (e.g., negative value verification), and data type conversion (e.g., percentage to decimal). Then, the standardized indicators are automatically filled into the financial data columns of the initial report file according to predefined mapping relationships (e.g., "operating revenue" corresponds to the "main business revenue" column in the report), ultimately generating a structured financial report file that can be directly used for analysis. This process, through multimodal extraction, intelligent unit conversion, and a standardized pipeline, effectively solves the problems of scattered data, inconsistent units, and diverse formats in financial documents, providing a high-quality structured data foundation for financial analysis.

[0056] For example, pdf2docx is used to convert the input PDF document into a temporary DOCX format, and table-level semantic data is extracted. Regular expressions are used to detect units of measurement (such as "yuan", "ten thousand yuan", "hundred million yuan") from the semantics of the DOCX paragraphs, with "yuan" as the default. If table semantics are missing, the process degenerates into full-text semantic extraction mode, calling pdfplumber to parse the PDF document, extracting quantified values ​​based on keywords (such as "operating revenue"), and calculating columns AC, AD, and AE (indexes 28, 29, and 30). After extracting the table-level semantic data, non-numeric characters in the semantic string can be cleaned (replacing Chinese commas, periods, etc.), converted to floating-point numbers, percentage semantics processed, and four decimal places retained for quantified value extraction. Unit standardization includes scaling the values ​​according to the detected units of measurement (such as "ten thousand yuan" corresponding to 10000) to ensure consistency with the units of measurement in the Excel template.

[0057] The process of quantitative analysis of financial indicators can be as follows: Locate the income statement and balance sheet using the keyword "project" to divide the semantic structure of the tables. Match quantitative indicator keywords (such as "total operating revenue" and "net profit attributable to shareholders of listed companies") and map the results to the financial statement file, for example, columns G to AQ of the financial statement file, and mark percentage semantic columns (such as rate of change).

[0058] It should be noted that after extracting all the financial indicators required for calculating derived indicators, the financial indicator quantification and analysis submodule 123 can also calculate derived indicators based on preset formulas, and then fill the derived indicators into the corresponding positions in the financial statement file. For example, column AC (index 28) can be: current period non-recurring net profit / owners' equity, column AD (index 29) can be: previous period non-recurring net profit / owners' equity, and column AE (index 30) can be: year-on-year change rate. Values ​​can be retained to 4 decimal places.

[0059] In some embodiments, such as Figure 3 As shown, the specialized refining module 13 includes: semantic reconstruction submodule 131 and standardization submodule 132.

[0060] The semantic reconstruction submodule 131 is used to extract the previous period value and the current period value of the corresponding target indicator, and to determine the rate of change based on the previous period value and the current period value of the corresponding target indicator.

[0061] The standardization submodule 132 is used to convert the quantized value of the target indicator into a percentage format when the numerical semantics of the target indicator are detected to be a percentage.

[0062] Understandably, the semantic reconstruction submodule 131 first extracts the previous period value (e.g., 2022 data) and the current period value (e.g., 2023 data) of the target indicator (e.g., weighted average return on net assets) from the financial statements based on predefined rules. It then automatically calculates the rate of change ((current period value - previous period value) / previous period value) by comparing these two values. The standardization submodule 132 simultaneously performs semantic detection on the target indicator. If it identifies a numerical semantic as a percentage (e.g., "year-on-year growth of 0.25" actually represents 25%), it automatically converts it to a standardized percentage format (e.g., 0.25 → 25%), ensuring that all percentage data are presented in a consistent format. For example, percentage semantic columns (e.g., rate of change) are formatted as "0.0000%", and numerical columns are formatted as "0.0000", ignoring invalid semantic values ​​(e.g., None or "N / A"). These two submodules, through data reconstruction and format standardization, improve the readability and analytical value of financial statement documents, providing a more intuitive and accurate data foundation for subsequent decision support. Based on the solutions described above, a partial schematic diagram of the final target report file is shown below. Figure 6 As shown.

[0063] Based on the same inventive concept, this application also provides a method for processing financial announcement data, such as... Figure 7 As shown, the financial announcement data processing method includes the following steps S701 to S705.

[0064] S701: Extracts key metadata and specific text paragraphs from the input document based on preset extraction rules.

[0065] S702: Construct a structured data framework based on the extracted data, and output the initial report file based on the structured data framework and predefined templates.

[0066] S703: Extract quantitative financial indicators from the input document and standardize the extracted quantitative financial indicators.

[0067] S704: Fill the structured financial indicator data into the financial data columns of the initial report file based on the standardized quantitative financial indicators to generate the financial report file.

[0068] S705: Perform semantic reconstruction and format standardization operations on the target indicators in the financial statement files to generate the target report files. The semantic reconstruction operation includes extracting periodic data from the target indicators to determine the periodic data; and calculating periodic variation indicators based on the periodic data.

[0069] In some embodiments, step S701, which involves extracting key metadata and specific text paragraphs from the input document based on preset extraction rules, includes: extracting text content from the input document and integrating it into a text format file stream; performing semantic normalization preprocessing on the text format file stream; locating feature semantic fragments using regular expressions; performing denoising and normalization processing based on the feature semantic fragments; and extracting structured metadata and structured text paragraphs.

[0070] In some embodiments, extracting text content from an input document and integrating it into a text format file stream includes: if the input document is identified as a scanned PDF document or an image-based PDF document, converting the input document into an image, performing OCR recognition on the converted image, and generating a recognized document; if the input document is a text-based PDF document, directly using the input document as the recognized document; performing page-by-page semantic parsing on the recognized document, extracting text content and integrating it into a text format file stream, and performing semantic normalization preprocessing on the text format file stream.

[0071] In some embodiments, the financial announcement data processing method further includes: when Arabic numeral format dates or Chinese numeral format dates are extracted from a text format file stream, performing semantic conversion on the extracted format dates to generate a standardized date format; when Arabic numeral format dates or Chinese numeral format dates are not extracted, extracting dates from the last page of the text format file stream and performing semantic conversion to generate a standardized date format.

[0072] In some embodiments, step S703, extracting quantitative financial indicators from the input document, includes: converting the input document into a temporary docx document; extracting table row-level semantic data from the temporary docx document; extracting quantitative values ​​of financial indicators based on the table row-level semantic data; when it is determined that table semantics are missing, parsing the input document using a full-text semantic extraction mode; extracting quantitative values ​​of financial indicators based on keywords; detecting units of measurement from the temporary docx document using regular expressions; performing numerical scaling processing on the extracted quantitative values ​​of financial indicators according to the units of measurement of a predefined template and the detected units of measurement; and generating corresponding quantitative financial indicators.

[0073] In some embodiments, step S705 involves performing semantic reconstruction and format standardization operations on the target indicators in the financial statement files, including: extracting the previous period value and current period value of the corresponding target indicator, and determining the rate of change based on the previous period value and current period value of the corresponding target indicator; and converting the quantitative value of the target indicator into a percentage format when the numerical semantics of the target indicator are detected to be a percentage.

[0074] It should be noted that the financial announcement data processing method provided in this application embodiment and the financial announcement data processing system provided in this application embodiment are based on the same application concept. Therefore, the specific implementation of this embodiment can refer to the implementation of the aforementioned financial announcement data processing system, and the repeated parts will not be described again.

[0075] In some embodiments, an electronic device provided in this application includes a processor and a memory; the memory stores a computer program, wherein the computer program, when executed by the processor, implements the above-described financial announcement data processing method.

[0076] Specifically, the processor may include, for example, a general-purpose microprocessor, an instruction set processor and / or an associated chipset and / or a special-purpose microprocessor (e.g., an application-specific integrated circuit (ASIC)), etc. The processor may also include onboard memory for caching purposes. The processor may be a single processing unit or multiple processing units for performing different actions of the method flow according to embodiments of this application.

[0077] Memory can be any medium capable of containing, storing, transmitting, propagating, or transmitting instructions. For example, memory can include, but is not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices, instruments, or propagation media. Specific examples of memory include: magnetic storage devices such as magnetic tape or hard disk drives (HDDs); optical storage devices such as optical discs (CD-ROMs); and also random access memory (RAM) or flash memory; and / or wired / wireless communication links.

[0078] This application also provides a computer-readable medium storing a computer program thereon, which, when executed by a processor, implements the above-described financial announcement data processing method. This computer-readable medium may be included in the device / apparatus / system described in the above embodiments; or it may exist independently and not assembled into that device / apparatus / system. The aforementioned computer-readable medium carries one or more programs, which, when executed, implement the method as described in the embodiments of this application.

[0079] According to embodiments of this application, a computer-readable medium may be a computer-readable signal medium or a computer-readable storage medium, or any combination thereof. A computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this application, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In this application, a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. Computer-readable signal media can also be any computer-readable medium other than computer-readable storage media, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to: wireless, wired, optical fiber, radio frequency signals, etc., or any suitable combination thereof.

[0080] Those skilled in the art will understand that the features described in the various embodiments of this application can be combined and / or combined in various ways, even if such combinations or combinations are not explicitly described in this application. In particular, the features described in the various embodiments of this application can be combined and / or combined in various ways without departing from the spirit and teachings of this application. All such combinations and / or combinations fall within the scope of this application. Therefore, the scope of this application should not be limited to the above embodiments. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A financial announcement data processing system, characterized in that, The financial announcement data processing system includes: a semantic extraction module, a report parsing module, and a specialized refinement module; The semantic extraction module includes a preprocessing submodule and a data extraction submodule. The preprocessing submodule is used to extract text content from the input document and integrate it into a text format file stream, and perform semantic normalization preprocessing on the text format file stream. The data extraction submodule is used to locate feature semantic segments in the text format file stream using regular expressions, perform denoising and normalization processing based on the feature semantic segments, extract structured metadata and structured text paragraphs, construct a structured data framework based on the structured metadata and structured text paragraphs, and output an initial report file based on the structured data framework and a predefined template. The report parsing module includes a quantitative value extraction submodule, a unit of measurement processing submodule, and a financial indicator quantitative parsing submodule. The quantitative value extraction submodule converts the input document into a temporary docx document, extracts table row-level semantic data from the temporary docx document, and extracts quantitative values ​​of financial indicators based on the table row-level semantic data. When table semantics are missing, the input document is parsed using a full-text semantic extraction mode, and quantitative values ​​of financial indicators are extracted based on keywords. The unit of measurement processing submodule detects units of measurement from the temporary docx document using regular expressions, performs numerical scaling on the extracted quantitative values ​​of financial indicators according to the units of measurement in the predefined template and the detected units of measurement, and generates corresponding quantitative financial indicators. The financial indicator quantitative parsing submodule standardizes the quantitative financial indicators, fills the structured financial indicator data into the financial data columns of the initial report file based on the standardized quantitative financial indicators, and generates a financial report file. The specialized refining module is used to perform semantic reconstruction and format standardization operations on the target indicators in the financial statement file to generate the target report file; wherein, the semantic reconstruction operation includes extracting periodic data from the target indicators to determine the periodic data; and calculating periodic variation indicators based on the periodic data.

2. The financial announcement data processing system as described in claim 1, characterized in that, The preprocessing submodule includes: a recognition unit and a semantic parsing unit; The recognition unit is used to convert the input document into an image when the input document is a scanned PDF document or an image PDF document, perform OCR recognition on the converted image, and generate a recognized document; when the input document is a text PDF document, the input document is directly used as the recognized document. The semantic parsing unit is used to perform page-by-page semantic parsing on the identified document, extract the text content and integrate it into a text format file stream, and perform semantic normalization preprocessing on the text format file stream.

3. The financial announcement data processing system as described in claim 1, characterized in that, The financial announcement data processing system also includes: an announcement date parsing module; The announcement date parsing module is used to perform semantic conversion on the extracted date format when Arabic numeral or Chinese numeral date is extracted from the text format file stream to generate a standardized date format; when Arabic numeral or Chinese numeral date is not extracted, the date is extracted from the last page of the text format file stream and semantic conversion is performed to generate a standardized date format.

4. The financial announcement data processing system as described in claim 1, characterized in that, The specialized refinement module includes: a semantic reconstruction submodule and a standardization submodule; The semantic reconstruction submodule is used to extract the previous period value and the current period value of the corresponding target indicator, and to determine the rate of change based on the previous period value and the current period value of the corresponding target indicator; The standardization submodule is used to convert the quantized value of the target indicator into a percentage format when the numerical semantics of the target indicator are detected to be a percentage.

5. The financial announcement data processing system as described in claim 1, characterized in that, The structured data framework contains a key-value mapping structure for multiple fields, with the initial value of each field set as a default placeholder, and each field corresponding to a column index of the table.

6. A method for processing financial announcement data, characterized in that, The financial announcement data processing method includes: Extract text content from the input document and integrate it into a text format file stream, then perform semantic normalization preprocessing on the text format file stream; Regular expressions are used to locate feature semantic fragments, and denoising and standardization are performed on the feature semantic fragments to extract structured metadata and structured text paragraphs. A structured data framework is constructed based on the structured metadata and the structured text paragraphs, and an initial report file is output based on the structured data framework and a predefined template. The input document is converted into a temporary docx document. Table row-level semantic data is extracted from the temporary docx document, and quantitative values ​​of financial indicators are extracted based on the table row-level semantic data. When the table semantics are found to be missing, the input document is parsed using a full-text semantic extraction mode, and quantitative values ​​of financial indicators are extracted based on keywords. The unit of measurement is detected from the temporary docx document using regular expressions. The extracted quantitative values ​​of financial indicators are numerically scaled according to the units of measurement of the predefined template and the detected units of measurement to generate corresponding quantitative financial indicators. The extracted quantitative financial indicators are then standardized. The structured financial indicator data is filled into the financial data columns of the initial report file based on the standardized quantitative financial indicators to generate the financial report file; The target indicators in the financial statement file are subjected to semantic reconstruction and format standardization operations to generate the target report file; wherein, the semantic reconstruction operation includes extracting periodic data from the target indicators to determine the periodic data; and calculating periodic variation indicators based on the periodic data.

7. An electronic device, characterized in that, It includes a processor and a memory; the memory stores a computer program, wherein the computer program, when executed by the processor, implements the financial announcement data processing method as described in claim 6.

8. A computer storage medium, characterized in that, It stores a computer program, wherein the computer program, when executed by a processor, implements the financial announcement data processing method as described in claim 6.

Images