Artificial Intelligence-Based Financial Report Generation Methods and Systems

Abstract

This application relates to the field of data processing technology and discloses a method and system for generating financial reports based on artificial intelligence. The method includes: preprocessing multi-source financial data to obtain target data; obtaining a knowledge base through semantic understanding using a capital extraction algorithm; generating a preliminary report by intelligently matching templates; applying causal relationship verification to analyze logical relationships; optimizing content based on verification results; and updating the report through multi-level review to obtain the final report. This application, by introducing a knowledge-based capital extraction algorithm and a causal relationship extraction model, achieves intelligent processing, deep semantic understanding, and logical verification of financial data, thereby generating high-quality financial reports with rigorous logic and professional content.

Description

Technical Field

[0001] This application relates to the field of data processing technology, and in particular to a method and system for generating financial reports based on artificial intelligence. Background Technology

[0002] As corporate financial reporting systems continue to improve, the preparation of financial reports has become increasingly complex. Traditionally, financial reporting relies primarily on manual methods. Financial personnel extract data from the company's financial system, perform calculations and organize the data using spreadsheet software, and manually write the report text. This approach requires a significant investment of time and energy in data collection, verification, calculation, and report writing. With the advent of the big data era, the volume of corporate financial data has exploded. Internal financial data, external market data, and industry regulatory data have created a multi-source, heterogeneous data environment. This makes traditional manual methods inefficient and prone to errors when processing such massive and complex financial data.

[0003] Most existing technologies can only process structured financial data, with limited capabilities for processing unstructured financial text information and an inability to deeply understand the semantic information behind the financial data. Secondly, existing technologies lack the ability to analyze causal relationships between financial events, making it difficult to logically verify the content of financial reports, potentially leading to logical contradictions or inconsistencies in the generated reports. Furthermore, existing technologies often rely on template-based report generation, lacking intelligent content generation and optimization capabilities, and failing to automatically generate personalized and professional report content based on enterprise characteristics and industry background. In addition, existing technologies lack multi-level, multi-dimensional intelligent review mechanisms for report auditing and updates, failing to comprehensively guarantee the accuracy and standardization of reports. Summary of the Invention

[0004] This application provides an artificial intelligence-based financial report generation method and system, which uses knowledge-based capital extraction algorithms and causal relationship extraction models to achieve intelligent processing, deep semantic understanding and logical verification of financial data, thereby generating high-quality financial reports with rigorous logic and professional content.

[0005] Firstly, this application provides an AI-based financial report generation method, comprising: collecting and preprocessing multi-source financial data to obtain target financial data; inputting the target financial data into a knowledge-based capital extraction algorithm for financial semantic understanding to obtain a financial data knowledge base; performing intelligent report template matching and automatic financial information filling based on the financial data knowledge base to obtain a first financial report; performing financial logic verification on the first financial report using a causal relationship extraction model to identify causal relationships between financial events and compare them with expected models to obtain a financial logic verification report; performing deep learning content generation and optimization on the first financial report based on the financial logic verification report to generate a second financial report; and performing multi-level automatic review and intelligent updates on the second financial report to obtain a target financial report.

[0006] Secondly, this application provides an artificial intelligence-based financial report generation system, the artificial intelligence-based financial report generation system comprising:

[0007] The data acquisition module is used to collect and preprocess multi-source financial data to obtain the target financial data;

[0008] The input module is used to input the target financial data into a knowledge-based capital extraction algorithm for financial semantic understanding, thereby obtaining a financial data knowledge base;

[0009] The filling module is used to perform intelligent report template matching and automatic filling of financial information based on the financial data knowledge base to obtain the first financial report;

[0010] The verification module is used to perform financial logic verification on the first financial report through a causal relationship extraction model, identify the causal relationship between financial events and compare it with the expected model to obtain a financial logic verification report.

[0011] The generation module is used to generate and optimize the first financial report based on the financial logic verification report using deep learning content to generate and optimize the second financial report.

[0012] The update module is used to perform multi-level automatic review and intelligent updates on the second financial report to obtain the target financial report.

[0013] Thirdly, an artificial intelligence-based financial report generation device is provided, comprising: a memory and at least one processor, wherein the memory stores instructions; the at least one processor invokes the instructions in the memory to cause the artificial intelligence-based financial report generation device to execute the aforementioned artificial intelligence-based financial report generation method.

[0014] Fourthly, a computer-readable storage medium is provided, wherein instructions are stored therein, which, when executed on a computer, cause the computer to perform the aforementioned artificial intelligence-based financial report generation method.

[0015] The technical solution provided in this application solves the problems of scattered data sources and inconsistent formats in the traditional financial report preparation process by collecting and preprocessing multi-source financial data, significantly improving data processing efficiency. The knowledge-based capital extraction algorithm overcomes the limitations of traditional methods in understanding the surface of financial data, achieving intelligent parsing of the deep semantics of financial data and accurately identifying the intellectual capital elements implicit in the financial data, enabling the financial report to comprehensively reflect the enterprise's value. The intelligent report template matching and automatic filling technology overcomes the rigidity of traditional template use, dynamically selecting and adjusting the optimal template based on enterprise characteristics and financial data features, achieving personalized report customization. The application of the causal relationship extraction model is the core innovation of this invention. This model, by integrating the characteristics of the Chinese language and graph attention network technology, accurately… Identifying the causal relationships between financial events effectively solves the problems of ambiguous causal relationships and logical contradictions in traditional financial reports, making the report content logically rigorous and clearly organized. Based on deep learning-based content generation and optimization technology, it breaks through the limitations of traditional report generation methods, which suffer from monotonous content and mechanical expression. It intelligently generates professional and accurate explanatory text based on the characteristics of financial data and verification results, enhancing the readability and professional depth of the report. Multi-level automatic review and intelligent update technology constructs a comprehensive quality assurance system. Through multi-dimensional detection such as cross-validation, trend analysis, and semantic matching, it ensures the accuracy, completeness, and standardization of the report, significantly reducing the error rate, improving the efficiency of financial report generation, and further guaranteeing the quality and professional standards of the report, providing a fully intelligent solution for the entire process of corporate financial report preparation. Attached Figure Description

[0016] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a schematic diagram of one embodiment of the artificial intelligence-based financial report generation method in this application.

[0018] Figure 2 This is a schematic diagram of one embodiment of the AI-based financial report generation system in this application.

[0019] Figure 3This is a schematic block diagram of the structure of the artificial intelligence-based financial report generation device in an embodiment of the present invention. Detailed Implementation

[0020] This application provides an artificial intelligence-based financial report generation method and system. The terms first, second, third, fourth, etc. (if present) in the specification, claims, and accompanying drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in a sequence other than that illustrated or described herein. Furthermore, the terms include or have, and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0021] For ease of understanding, the specific process of the embodiments of this application is described below. Please refer to [link / reference]. Figure 1 One embodiment of the artificial intelligence-based financial report generation method in this application includes:

[0022] Step S101: Collect and preprocess multi-source financial data to obtain target financial data;

[0023] Step S102: Input the target financial data into a knowledge-based capital extraction algorithm for financial semantic understanding to obtain a financial data knowledge base;

[0024] Step S103: Based on the financial data knowledge base, perform intelligent report template matching and automatic filling of financial information to obtain the first financial report;

[0025] Step S104: Perform financial logic verification on the first financial report using the causal relationship extraction model, identify the causal relationships between financial events and compare them with the expected model to obtain a financial logic verification report;

[0026] Step S105: Based on the financial logic verification report, perform deep learning content generation and optimization on the first financial report to generate the second financial report;

[0027] Step S106: Perform multi-level automatic auditing and intelligent updating of the second financial report to obtain the target financial report.

[0028] It is understood that the executing entity of this application can be an AI-based financial report generation system, a terminal, or a server; no specific limitation is made here. This application's embodiment uses a server as an example for illustration.

[0029] Specifically, data is collected periodically from multiple data sources, including the company's internal core financial system, external audit report databases, historical financial statements, and external data from regulatory agencies. This multi-source data often has inconsistent formats and complex content, necessitating format conversion to transform the data from different sources into a standardized structure, ensuring smooth subsequent processing. Data conversion involves more than just simple format changes; it also requires data integrity checks to identify missing key financial fields and fill them in according to industry standards, ensuring data accuracy and completeness. After data completion, consistency verification is performed, checking the relationships between major financial statements such as the balance sheet, income statement, and cash flow statement to ensure numerical consistency and logical reasonableness. Sensitive data is anonymized, with tiered anonymization of sensitive information such as customer identifiers and employee compensation to ensure data security. By establishing a financial indicator correlation diagram, the system obtains the target financial data, laying a solid foundation for subsequent steps.

[0030] The preprocessed target financial data is input into a knowledge-based capital extraction algorithm (KBICE) for financial semantic understanding. KBICE leverages deep learning, natural language processing, and knowledge graphs to deeply analyze the semantic content of the data. In this process, the system first identifies key terms and numerical information in the financial data, extracting financial keywords, monetary values, and related time information through a financial dictionary and named entity recognition technology. This information is then mapped to a standard financial concept system, enabling higher-level semantic analysis. Through an ontology mapping module, financial terms are transformed into structured financial entities, and syntactic dependency parsing is used to identify logical relationships and semantic structures within the financial data. Building on this, the system performs event clustering, organizing financial events with semantic similarity and temporal relevance into event units, constructing a sequence of financial events. Furthermore, these events are classified and quantified, generating a financial data knowledge base rich in semantic information, providing robust data support for intelligent report generation.

[0031] Based on the obtained financial data knowledge base, intelligent report template matching and automatic financial information filling begin. First, the company's financial characteristic data, including asset size, liability structure, and profitability, are extracted as the basis for selecting the most suitable report template. By calculating the matching degree between each template in the template library and the company's financial characteristics, the financial report template that best meets the current needs can be selected. Next, the selected report template undergoes component evaluation, selecting table components suitable for the current company's data. By combining these components, a customized report structure that conforms to the company's financial characteristics is constructed. Then, the data in the financial data knowledge base is mapped to the fields in the report template, clarifying the filling rules for each field. Based on these rules, financial data is automatically filled into the report template, and financial indicators are calculated and the results are filled into the corresponding positions. The report undergoes format standardization processing, adjusting the numerical format and table style to ensure that the report meets professional standards, resulting in the first financial report. The first financial report undergoes financial logic verification through a causal relationship extraction model. This model uses techniques such as graph neural networks and dependency analysis to identify the causal relationships between different financial events in the report. By analyzing the logical sequence and relevance of financial events, the system can compare them against a pre-defined causal knowledge base to detect any abnormal causal relationships that do not conform to financial principles and mark these anomalies. It can reveal potential logical flaws in reports, ensuring internal consistency and rationality. The system generates a financial logic verification report, detailing the verified financial chains and the logical anomalies that need correction, and provides rectification suggestions.

[0032] Based on the financial logic verification report, the first financial report undergoes deep learning-based content generation and optimization. This process includes correcting logical flaws in the report, replacing unreasonable financial logic sections, supplementing the basis for corrections, and optimizing the text. Simultaneously, the system analyzes changes in key financial indicators, generating professional explanations for the reasons for indicator fluctuations, enriching the report content. This transforms the report content into more standardized and professional text, forming a second, optimized financial report. The second financial report undergoes multi-level automatic review and intelligent updates. It involves two-way cross-review, reverse-verifying the consistency between overall financial indicators and segmented data, and comparing it with historical financial report data to identify any abnormal trend changes. The system also performs semantic framework analysis to ensure semantic matching between financial descriptions and corresponding data, marking inconsistencies. Through these review processes, a problem space map is generated, displaying all detected issues and providing corresponding correction solutions. Based on these correction solutions, the report undergoes regional targeted updates to ensure greater consistency and accuracy, ultimately generating the target financial report.

[0033] In this embodiment, multi-source financial data collection and preprocessing solves the problems of scattered data sources and inconsistent formats in traditional financial reporting, significantly improving data processing efficiency. The knowledge-based capital extraction algorithm overcomes the limitations of traditional methods in understanding the surface of financial data, achieving intelligent parsing of the deep semantics of financial data and accurately identifying the intellectual capital elements implicit in the financial data, enabling the financial report to comprehensively reflect corporate value. The intelligent report template matching and automatic filling technology overcomes the rigidity of traditional template use, dynamically selecting and adjusting the optimal template based on corporate characteristics and financial data features, achieving personalized report customization. The application of the causal relationship extraction model is the core innovation of this invention. This model, by integrating Chinese language characteristics and graph attention network technology, accurately identifies… The causal logic between financial events effectively solves the problems of vague causal relationships and logical contradictions in traditional financial reports, making the report content logically rigorous and clearly organized. Deep learning-based content generation and optimization technology breaks through the limitations of traditional report generation methods, which often feature monotonous content and mechanical expression. It intelligently generates professional and accurate explanatory text based on the characteristics of financial data and verification results, enhancing the readability and professional depth of the report. Multi-level automatic review and intelligent update technology constructs a comprehensive quality assurance system. Through multi-dimensional detection such as cross-validation, trend analysis, and semantic matching, it ensures the accuracy, completeness, and standardization of the report, significantly reducing the error rate, improving the efficiency of financial report generation, and further guaranteeing the quality and professional standards of the report. This provides a fully intelligent solution for the entire process of corporate financial report preparation.

[0034] In one specific embodiment, the process of performing step S101 may specifically include the following steps:

[0035] Regularly collect data from the enterprise's core financial system, audit report database, historical financial statements, and external regulatory data to obtain raw financial datasets;

[0036] The original financial dataset is converted into a standardized structure by performing format conversion processing on the data in PDF reports, spreadsheets, and text documents to obtain the initial financial data.

[0037] Perform a data integrity check on the initial financial data, identify missing items in key financial fields and fill them in according to industry standards to obtain complete financial data;

[0038] Perform consistency verification on complete financial data, check the reconciliation between the balance sheet, income statement and cash flow statement, mark data items that do not conform to accounting standards, and obtain verified financial data.

[0039] Sensitive information processing is performed on the verified financial data, and customer identifiers and employee pay details are desensitized in a hierarchical manner to obtain desensitized financial data.

[0040] A correlation diagram of financial indicators is constructed based on anonymized financial data to obtain the target financial data.

[0041] Specifically, the system periodically collects data from a company's core financial system, audit report repository, historical financial statements, and external regulatory data to obtain raw financial datasets. The core of this process lies in automatically acquiring the company's financial data from various data sources through multi-source data acquisition technology. A company's core financial system typically stores its daily financial transaction records; the audit report repository provides audit information on past financial data; historical financial statements are detailed records of the company's past financial situation; and external regulatory data may include financial regulations and industry financial trends from government agencies or industry regulators. To ensure the timeliness and accuracy of the data, the system needs to periodically acquire data from these sources to ensure that the data used in the preparation of financial reports is the latest and most comprehensive. This process often involves interface connections with these data sources, using APIs or file uploads to achieve automatic data collection, enabling the automatic acquisition of various financial datasets.

[0042] The original financial dataset undergoes format conversion. Raw data typically exists in various formats such as PDF reports, spreadsheets, and text documents. Data in different formats needs to be transformed into a standardized data structure through a unified conversion process. This process often involves parsing files of different formats and extracting the financial data. Tables in PDF reports, cell data in spreadsheets, and descriptive information in text documents all require processing using data extraction techniques. For example, PDF reports typically require OCR (Optical Character Recognition) technology to recognize numbers and text content, spreadsheets require reading the values ​​and attributes of each cell, and financial information in text documents may require natural language processing technology to extract financial keywords. This conversion process transforms the originally inconsistent data into a standardized structure, making subsequent data processing more convenient and efficient.

[0043] After format conversion, the initial financial data undergoes a data integrity check. The integrity of financial data is crucial for report accuracy, therefore, missing items in key financial fields are automatically identified and filled. For example, total assets and total liabilities in the balance sheet may be missing, or certain items in historical reports may be incomplete. These missing items are filled according to industry standards and accounting principles. The filling process typically involves estimations based on financial data from similar companies, historical data, and industry trends, and algorithms automatically generate reasonable fill values ​​to ensure the integrity and consistency of the financial data. The filling algorithms and rules used in this process usually rely on industry standards and financial accounting norms, such as determining the proportional relationships between different items according to accounting principles, or inferring reasonable values ​​for missing data based on historical growth trends. After completing the data integrity check, consistency verification is performed to check the reconciliation relationships between different financial statements. The reconciliation relationships in financial statements refer to the interrelationships between the balance sheet, income statement, and cash flow statement. For example, total assets in the balance sheet should equal the sum of liabilities and owner's equity, and net profit in the income statement should match operating cash flow in the cash flow statement. The system automatically checks the logical consistency between these financial statements using algorithms, marking data items that do not comply with accounting principles. This process ensures the accuracy and compliance of financial data, avoiding data errors and inconsistencies in reports. For example, if the total assets on the balance sheet are inconsistent with the sum of liabilities and owner's equity, the data will be automatically marked as incorrect, indicating the need for further correction. After data consistency verification is completed, sensitive information processing is performed on the verified financial data. Sensitive information, such as customer personal identifiers and employee salary details, must be anonymized to ensure data security. Anonymization typically involves encrypting or replacing sensitive information to prevent data leakage. For example, the system may replace customer names and contact information with symbols, or encrypt specific employee salary amounts, retaining only the salary range or category information. This allows for subsequent financial analysis and report generation while ensuring data privacy and preventing the leakage of sensitive data. Based on the anonymized financial data, the system constructs a financial indicator correlation graph, laying the foundation for the subsequent generation of target financial data. The financial indicator correlation graph is a map containing various financial data items and their interrelationships. The system uses algorithms to analyze the correlations between different financial data and builds a structured financial data map based on these relationships. For example, there is a certain relationship between current assets and current liabilities in the balance sheet, and there is also a correlation between operating revenue and net profit in the income statement. This kind of graph can intuitively show the inherent connections between different financial data.

[0044] In one specific embodiment, the process of performing step S102 may specifically include the following steps:

[0045] The target financial data is input into the terminology recognition module of the knowledge-based capital extraction algorithm for processing. Financial keywords, monetary values, and time information are extracted through a financial professional dictionary and named entity recognition technology to obtain a set of financial terms.

[0046] The set of financial terms is input into the ontology mapping module of the knowledge-based capital extraction algorithm for processing. The terms are mapped to the standard financial concept system using the pre-set knowledge ontology in the financial field to obtain structured financial entities.

[0047] Structured financial entities are input into the relation extraction module of the knowledge-based capital extraction algorithm for financial semantic association analysis. The logical relationship between semantic subjects and objects is identified through syntactic dependency analysis to obtain a financial transaction relationship graph.

[0048] The financial event relationship diagram is input into the event clustering module of the knowledge-based capital extraction algorithm for processing. Based on the financial semantic similarity and temporal relevance, the related events are organized into event units to obtain the financial event sequence.

[0049] The sequence of financial events is input into the capital classification module of the knowledge-based capital extraction algorithm for intellectual capital classification. The financial events are categorized and quantified according to predefined capital category standards to obtain a classified financial knowledge structure.

[0050] The categorized financial knowledge structure is input into the weight calculation module of the knowledge-based capital extraction algorithm for processing. By calculating the importance weight of each knowledge node and constructing a multi-level index structure, a financial data knowledge base is obtained.

[0051] Specifically, the target financial data is input into the terminology recognition module of the knowledge-based capital extraction algorithm. This module extracts key information from the financial data using a financial lexicon and Named Entity Recognition (NER) technology. Financial data contains a large number of technical terms, monetary values, and time information. The terminology recognition module's task is to identify these key elements from the raw data. For example, in a financial statement, the terminology recognition module might identify terms such as accounts receivable, total assets, net profit, and the first quarter of 2023. Monetary values ​​such as 10,000,000 yuan will also be extracted, and time information such as December 31, 2023 will be identified and marked. Through the financial lexicon, various financial terms and values ​​can be accurately identified, forming a set containing a wide range of financial terms.

[0052] The set of financial terms will be input into the ontology mapping module of the knowledge-based capital extraction algorithm. The ontology mapping process maps the identified terms to a standard knowledge system in the financial domain, ensuring consistency and standardization. A financial domain knowledge ontology is typically a pre-defined standard framework that defines the meaning of financial terms and their interrelationships. Through the ontology mapping module, terms such as accounts receivable will be transformed into a unified representation within the standard financial concept system. For example, accounts receivable might be mapped to "Accounts Receivable" and defined as the amount a company receives from customers for the sale of goods or the provision of services. This mapping process gives financial terms a unified semantic meaning, enabling a deeper understanding of these terms through the ontology knowledge base, eliminating ambiguities arising from different data sources and formats, and providing structured financial entities for further analysis. After ontology mapping, the financial data will be input into the relation extraction module of the knowledge-based capital extraction algorithm. This module uses syntactic dependency analysis to identify semantic relationships between different entities in the financial data and establish logical connections between financial matters. For example, in the balance sheet, there is a certain correspondence between assets and liabilities, and in the income statement, operating revenue and net profit are also closely related. Dependency analysis can identify these relationships, such as the balance between assets and liabilities, and the causal relationship between revenue and profit, and represent them through a graph structure to form a financial event relationship diagram. This graph structure provides the foundation for subsequent financial event clustering, showing the correlation between various financial events and helping the system understand the logical connections and influence relationships between different financial data items.

[0053] The financial event relationship diagram will be input into the event clustering module of the knowledge-based capital extraction algorithm for processing. At this stage, related events are organized and clustered based on semantic similarity and temporal relevance. For example, in financial statements, some events, such as revenue growth and profit increase, may have semantic similarity, while other events, such as capital expenditure and debt repayment, may be temporally related. The event clustering module will categorize these related events according to semantic similarity and temporal relevance, forming a sequence of financial events. For example, if a company's operating revenue increases at multiple points in time, these growth events are categorized as a revenue growth series event, arranged chronologically to form an ordered sequence of financial events. This allows for the extraction of ordered event sequences from complex financial data.

[0054] The sequence of financial events is input into the capital classification module of the knowledge-based capital extraction algorithm for intellectual capital classification. Financial events are categorized and quantified according to predefined capital category standards. The classification criteria for financial events may include financial elements such as a company's assets, liabilities, and revenue. For example, the system can classify all asset-related financial events into the capital category and liability-related events into debt capital. This classification further refines financial data, forming a hierarchical financial knowledge structure. This structured financial knowledge helps the system quickly locate relevant data and information when generating reports and provides more accurate support during the analysis process.

[0055] The categorized financial knowledge structure is then input into the weight calculation module of the knowledge-based capital extraction algorithm. The weight calculation module calculates the importance weight of each knowledge node and constructs a multi-level index structure. Each financial event or indicator has a different degree of correlation with other events or indicators, and these strengths are quantified by the weight calculation module according to a specific algorithm. The system can assess the relative importance of each financial data item in the report and adjust the priority of subsequent report generation based on its importance. For example, if a financial data item has changed significantly over the past few reporting periods, the system may assign it a higher weight to highlight it in the report. By establishing a multi-level index structure, relevant data can be retrieved and accessed more efficiently, improving the efficiency and accuracy of report generation.

[0056] In one specific embodiment, the process of executing step S103 may specifically include the following steps:

[0057] Extract data on the company’s asset size, liability structure, and profitability indicators from the financial data knowledge base to obtain the company’s financial characteristic data.

[0058] Based on the enterprise's financial characteristic data, the template matching degree in the template library is calculated, and the template with the highest matching degree is selected to obtain the preferred report template;

[0059] The preferred report template is evaluated for its components. Table components suitable for the company's data are selected and combined to obtain a customized report structure.

[0060] By establishing a mapping relationship between the fields of the customized report structure and the data nodes in the financial data knowledge base, the field filling rules are obtained;

[0061] The data from the financial data knowledge base is populated into the customized report structure according to the field population rules, financial indicators are calculated and the results are populated to obtain a complete report;

[0062] The complete report is formatted and standardized, with adjustments made to numerical formats and table styles, and areas requiring supplementary explanations are marked to obtain the first financial report.

[0063] Specifically, the financial data knowledge base is processed to extract the company's financial characteristic data. This characteristic data mainly includes key financial information such as the company's asset size, liability structure, and profitability indicators. After standardization and semantic understanding, the data in the financial data knowledge base already possesses structured financial information. By extracting this information, the system can accurately obtain the company's financial status. For example, asset size can be calculated from the asset items listed in the balance sheet, the liability structure can be derived by analyzing liability items, and profitability indicators such as net profit and gross profit margin can be calculated from the data in the income statement. This extracted financial characteristic data provides crucial data support for subsequent report generation. After obtaining the company's financial characteristic data, the system then matches this data with various report templates in the template library, calculating the matching degree of each template. The template library contains various financial report templates, each with different structures, content, and formats for companies of different industries and sizes. By comparing the company's financial characteristic data with the structural requirements of the templates, the system can assess the suitability of each template. Specifically, information such as the company's asset size, liability structure, and profitability indicators will be used as input. The system uses a preset algorithm to calculate the matching degree between each template and the company's financial data, and finally selects the report template that best matches the company's financial data. For example, if the company has a large asset size, the system may select a template that includes a detailed balance sheet, while if the company's profitability indicators are outstanding, the system may select a template that focuses more on financial analysis and trend display. Through this matching, the system ensures that the selected template matches the company's financial characteristics, thereby providing a more accurate reporting framework.

[0064] After selecting the optimal report template, the system enters the component evaluation phase. The template typically consists of multiple table components, which may include a balance sheet, income statement, and cash flow statement. In this step, each table component is evaluated to determine its suitability for the company's needs based on its financial data characteristics. For example, if the company has a complex debt structure, the system might select a more detailed balance sheet component to display the composition and changes in liabilities; if the company has a simpler asset structure, the system might choose a concise balance sheet component to display the company's asset size. Component selection is not merely an adjustment to the template, but rather a customization of the report structure based on the company's financial characteristics, making the report content more aligned with actual needs. By combining different table components, a customized report structure can be generated, ensuring that the report content covers all aspects of the company's financial situation. The fields of the customized report structure are mapped to data nodes in the financial data knowledge base to obtain field filling rules. Each data item in the financial data knowledge base has its corresponding field position, and each field in the report template also needs to be filled with corresponding data. During this process, the system establishes a mapping relationship between data nodes and fields to ensure that each piece of financial data is accurately filled into the corresponding report field. For example, the total assets field in the balance sheet should be filled with asset data from the company's balance sheet, and the net profit field in the income statement should be filled with net profit data extracted from the income statement. This mapping relationship ensures that the financial reporting process is both accurate and efficient.

[0065] Once the field filling rules are clear, the system can fill the customized report structure with data from the financial data knowledge base according to these rules, calculate financial indicators, and ultimately generate a complete financial report. At this stage, the system not only fills the raw data into the designated locations in the report, but also calculates key financial indicators such as net profit, current ratio, and gross profit margin. These financial indicators are typically derived through a series of formulas; for example, net profit can be obtained by subtracting costs and expenses from revenue, and the current ratio can be obtained by dividing current assets by current liabilities. The system will perform necessary calculations on the financial data according to preset calculation rules and automatically fill the results into the corresponding locations in the report. After these operations, a complete report containing all necessary financial data and indicators will be obtained. The generated complete report will undergo format standardization processing, adjusting the numerical format and table style to ensure that the report meets professional standards. For example, the display format of numbers will be standardized to ensure that all monetary items are displayed in a consistent currency unit and with consistent decimal places; for data in tables, the table layout and column width will be adjusted according to preset styles to ensure that the report's visual effect is clear and neat. In addition, the system will mark areas that require supplementary explanations. For example, some important financial data may need additional annotations or explanations. Blank spaces will be left in these areas with a mark, prompting the user to manually add the information. After these formatting adjustments, the system generates the first financial report that meets the standards, ready for subsequent review and optimization stages.

[0066] In one specific embodiment, the process of executing step S104 may specifically include the following steps:

[0067] The first financial report is decomposed into multiple financial event units. Key financial indicator changes and business activities are extracted through word segmentation, syntactic parsing and named entity recognition to obtain a set of financial events.

[0068] Construct a financial event graph for the set of financial events, treat each financial event as a node, initialize the potential connection relationships between nodes, and obtain the initial financial event network.

[0069] The initial financial event network is input into the Chinese dictionary enhancement layer of the causal relationship extraction model. The event description is semantically enhanced by a financial professional dictionary to obtain a semantically enhanced event representation.

[0070] For semantically enhanced event representation, financial event dependency analysis is performed through the graph attention network layer in the causal relationship extraction model. Multi-head attention weights between nodes are calculated to identify the strength of causal associations between events and obtain a weighted causal relationship graph.

[0071] The weighted causal relationship graph is compared with a pre-set causal knowledge base in the financial field to detect abnormal causal links and logical contradictions, and causal relationships that do not conform to financial rules are marked to obtain financial logic anomaly markers.

[0072] A financial logic verification report is generated based on the financial logic anomaly markers. It includes verified causal chains and abnormal causal relationships that need to be corrected, and provides correction suggestions for each anomaly, thus obtaining the financial logic verification report.

[0073] Specifically, the first financial report is broken down into multiple financial event units. Natural language processing technologies, such as word segmentation, syntactic parsing, and named entity recognition (NER), are used to extract key financial indicator changes and business activities from the report. The goal of this process is to treat each financial change and business activity in the report as an independent financial event, forming a set of financial events. For example, if the report mentions that the company's net profit increased by 10% in the first quarter of 2023, this change will be extracted as an independent event unit and labeled as "net profit growth." Through word segmentation, the system can accurately segment sentences and identify relevant financial terms, numerical changes, and time points. Furthermore, named entity recognition technology helps the system identify key financial terms in the events, such as net profit, growth, and the relevant time information "first quarter of 2023," ensuring accurate extraction of financial events. These financial event units constitute a set. After obtaining the set of financial events, the system then constructs a financial event graph, treating each financial event as a node and initializing the potential connections between nodes. The financial event graph is a graph structure where each node represents a financial event, and the connections between nodes represent potential relationships between events. For example, if a company report mentions two events: revenue growth and cost reduction, the system might infer a causal or temporal relationship between them through business logic and initialize an edge between these two event nodes. Financial event graphs provide the graph structure foundation for subsequent causal relationship analysis, visually demonstrating the connections between different financial events.

[0074] The financial event graph is then fed into a causal relationship extraction model for processing. The model first enhances the semantics of the event descriptions through a Chinese dictionary enhancement layer. This process improves the understanding of financial terminology using a financial dictionary. For example, for the event of a 10% increase in net profit, the system identifies the semantic relationship between net profit and the 10% increase using a financial dictionary, enhancing the semantic representation of the event and ensuring an understanding of the financial context and meaning of this change. This enhanced event representation is further processed. Through syntactic dependency analysis and graph neural network (GNN) techniques, the system can deeply analyze the dependencies between events and identify causal relationships and semantic associations based on syntactic structure.

[0075] Following dependency analysis, a Graph Attention Network (GAT) is used to conduct an in-depth analysis of the financial event graph, calculating the multi-head attention weights between nodes to identify the strength of causal relationships between events. The GAT assesses the strength of relationships between nodes by calculating the attention between each node. For example, if a revenue growth event and a net profit growth event have a strong causal relationship, their weights will be larger, while events with weaker relationships, such as management changes and revenue growth, may have smaller weights. This process identifies the strength of causal relationships between financial events and ranks them according to their weights, providing a basis for subsequent financial logic verification.

[0076] The weighted causal relationship graph is compared with a pre-defined causal knowledge base in the financial domain to detect any abnormal causal links and logical contradictions. For example, the financial causal knowledge base may contain standard causal rules, such as revenue growth usually leading to net profit growth. If a financial event graph shows a connection where revenue growth leads to a decrease in net profit, it indicates a potential logical error. The system automatically compares these causal relationships, marking causal links that do not conform to financial principles, thus creating financial logic anomaly markers. These markers help the system identify potential abnormal financial logic in the report, ensuring that the generated report is logically consistent.

[0077] Based on these financial logic anomaly flags, a financial logic verification report is generated. This report contains verified causal chains and details any abnormal causal relationships that need correction. Each abnormal causal relationship in the report includes a suggested correction to help reviewers identify and correct potential errors. For example, if a report shows a causal relationship where revenue growth leads to a decrease in net profit, this anomaly will be flagged in the financial logic verification report, and a correction will be recommended. Possible correction methods include reassessing the relationship between revenue and costs or reviewing the calculation methods. The system not only detects potential problems in financial reports but also provides specific suggestions for report correction.

[0078] In one specific embodiment, the process of executing step S105 may specifically include the following steps:

[0079] The financial logic verification report is mapped to the first financial report, and the location markers and corresponding correction suggestions for each area that needs to be corrected are extracted to obtain a correction instruction dataset.

[0080] The correction instruction dataset is sorted according to the anomaly degree quantification value, and the priority score of each correction task is calculated to obtain a weighted correction task sequence.

[0081] The content of the first financial report is reconstructed based on a weighted sequence of correction tasks, the text of unreasonable financial logic areas is replaced, and the explanation of the basis for correction is added to obtain a draft logic correction report.

[0082] The draft logic correction report generates key indicator explanations, analyzes the reasons for abnormal fluctuations in financial indicators, adds professional background analysis text, and obtains an enhanced indicator explanation report.

[0083] The text expression of the enhanced indicator explanation report is transformed to obtain a standardized text report.

[0084] The text standardization report was reorganized and its chapters adjusted to obtain the second financial report.

[0085] Specifically, the process of mapping financial logic verification reports to the primary financial report and extracting a correction instruction dataset effectively identifies and corrects potential problems in the financial reports. During this process, the financial logic verification report identifies all areas requiring correction through comparative checks. By comparing financial events and logic verification results in the report, areas needing correction can be extracted from the logic verification report, including causal relationships that do not conform to financial principles, logical anomalies, and numerical errors. Each correction area is marked as requiring correction and accompanied by corresponding correction suggestions. These suggestions may involve adjusting a financial indicator or value, or modifying causal relationships. The location markers of these correction areas and their corresponding correction suggestions form the correction instruction dataset. The correction instruction dataset is sorted according to the degree of anomaly quantification, and the priority of each correction task is calculated. The quantification is based on factors such as the impact range of the correction area, the weight of the financial indicator, and the resources and time required for correction. For example, a major error that may affect the overall financial situation of the company (such as a significant deviation in net profit) will have a higher priority; while minor errors (such as small changes in accounts receivable) will have a lower priority. This process is actually based on the hierarchical structure of financial data and impact analysis, ensuring that high-impact errors are addressed first, while low-impact errors are adjusted later. In calculating the priority of corrective tasks, the weight relationships of financial data are used for ranking; indicators with higher weights are assigned higher corrective priority, ensuring that the most critical financial issues are resolved as early as possible.

[0086] Once the correction tasks are sorted, the system enters the content reconstruction phase. Based on the weighted sequence of correction tasks, the first financial report undergoes item-by-item reconstruction, replacing text in areas of unreasonable financial logic and supplementing explanations of the correction basis. For example, if the report mentions an unreasonable logic such as revenue growth leading to a decrease in net profit, this logical error will be marked in the report and replaced with a description more consistent with financial principles, such as revenue growth and cost control jointly leading to net profit growth. Simultaneously, corresponding correction basis will be added to each correction area, such as supplementing financial analysis data, industry standards, or accounting principles as support, ensuring the accuracy and rationality of the report's logic. After completing the content reconstruction of the first financial report, the system enters the key indicator explanation generation phase. Based on the key indicators involved in the financial report, the system analyzes the reasons for abnormal fluctuations. For example, if the company's net profit fluctuates abnormally during the reporting period, an in-depth analysis will be conducted based on the changing trends in the financial data and other relevant data (such as revenue, costs, taxes, etc.) to identify the root causes of the fluctuations, such as rising costs, declining revenue, or the impact of external economic factors. Based on these analysis results, professional background analysis text will be automatically generated, explaining in detail the reasons for these financial indicator fluctuations, and added to the report. This process not only provides accurate financial analysis, but also helps management understand the business activities behind the financial data, enhancing the practical value of the reports.

[0087] The report on the enhanced explanation of indicators will undergo text transformation to conform to standardized report formats. This transformation primarily involves unifying language style, refining expression, and standardizing terminology. The language will be adjusted to be more concise and clear, avoiding lengthy descriptions and ensuring high readability and professionalism. Standardization of terminology is also crucial; the use of financial terminology will be standardized to ensure all financial indicators, accounts, and terms comply with the latest accounting standards and industry norms. The goal of this stage is to ensure that the report is not only accurate and detailed but also highly standardized, complying with International Financial Reporting Standards (IFRS) or other relevant financial regulations.

[0088] The standardized financial report undergoes a chapter-by-chapter reorganization to ensure a logical and clear structure. This process involves dividing the report into different chapters based on its logic and content, and optimizing each chapter to ensure a clear hierarchical structure. For example, a financial report might include chapters on financial overview, financial analysis, risk assessment, and future outlook. Each chapter will be appropriately adjusted to conform to the standardized financial report structure. After these adjustments are completed, the final financial report (secondary financial report) is formed.

[0089] Taking a manufacturing company as an example, suppose its first financial report contains a logical error: revenue growth leads to a decrease in net profit. Through system-wide financial logic verification, the system identifies this problem and generates a correction task using data mapping and correction instructions, providing suggested adjustments. Since the relationship between revenue and net profit is crucial to a company's financial reporting, this correction task has a high priority. The system prioritizes this task through a weighted sequence of correction tasks, replacing the erroneous logic with a statement that revenue growth and cost optimization jointly contribute to increased net profit. In the corrected report, the system further provides detailed justification for this adjustment, incorporating industry-standard cost control methods and supporting financial data from similar companies. Simultaneously, the system generates an analysis of net profit fluctuations, highlighting cost control as a key factor in net profit growth. The report's text is also standardized to conform to standard financial reporting formats. Through this series of data processing and corrections, the company obtains a structurally complete, logically consistent, and industry-compliant second financial report, providing decision-makers with more accurate financial analysis.

[0090] In one specific embodiment, the process of executing step S106 may specifically include the following steps:

[0091] A two-way cross-audit was performed on the second financial report, which involved back-verifying the overall financial indicators with the segment data to obtain the cross-verification results.

[0092] The second financial report is input into the knowledge base comparison mechanism, compared with the historical financial report database, to identify abnormal trend change points and obtain a set of trend anomalies.

[0093] Semantic framework analysis is performed on the second financial report to detect the semantic mapping consistency between the financial descriptions and the corresponding data, and to mark the areas where the descriptions and data do not match, thus obtaining a semantic matching degree report.

[0094] Based on the cross-validation results, trend anomaly set, and semantic matching degree report, a problem space map is generated. The detected problem points are located in three-dimensional space and the problem evolution path is constructed to obtain the corrected map.

[0095] Based on the revised map, the second financial report is updated in a regionally targeted manner, and the financial indicators and their explanatory texts that affect each other are updated according to spatial correlation, resulting in a correlation update report;

[0096] Perform intelligent integrity reassessment on the associated update report, establish an index structure for the report content, and obtain the target financial report.

[0097] Specifically, in the process of implementing financial report verification and optimization, the second financial report will be processed through a series of steps, including two-way cross-verification, knowledge base comparison, semantic framework analysis, problem space map generation, targeted updates, and integrity reassessment. These steps ensure the accuracy and logical consistency of the financial report. Two-way cross-verification involves reverse-verifying overall financial indicators with segment data. Specifically, the cross-verification process compares and verifies the data in each part of the financial report to ensure consistency between the overall and segment data. For example, suppose a company reports overall revenue of 50 million yuan, and a subsidiary's revenue in the segment data is 15 million yuan. During cross-verification, the auditors need to verify whether the sum of the overall revenue and the segment revenue matches. If there is a discrepancy, this issue will be marked as an anomaly, and its location and cause will be recorded. The second financial report will be input into the knowledge base comparison mechanism, compared with the historical financial report database, identifying abnormal trend changes, and generating a set of trend anomalies. In this process, financial indicators involved in the report, such as revenue, gross profit, and net profit, will be extracted and compared with similar data in historical financial reports. This comparison identifies anomalous fluctuations in certain financial data. For example, if historical data indicates that an industry's average gross profit margin is typically 40%, while the current report shows 10%, this constitutes an anomaly. The identified set of trend anomalies provides a basis for subsequent financial analysis and report optimization, helping to analyze whether changes in the company's financial situation are reasonable. Semantic frame analysis checks the semantic consistency between financial statements and data. In financial reports, much data and its explanation are linked through textual descriptions; the task of semantic frame analysis is to ensure that these descriptions are consistent with the actual data. For example, if the report states that the company's net profit increased by 20% compared to last year, but the financial data shows a net profit increase of only 5%, then there is a mismatch between this description and the actual data. Semantic frame analysis technology compares financial statements and data, identifies all inconsistencies, and ultimately generates a semantic matching report, revealing potential linguistic or data errors in the report.

[0098] Based on cross-validation results, trend anomaly sets, and semantic matching reports, a problem space map is generated to locate problem points in the report and construct problem evolution paths. The problem space map is a three-dimensional coordinate system where each problem point is located within a corresponding spatial region. This map visually illustrates the relationships between different problems in the financial report; for example, some problems may be caused by data errors, while others may be caused by inconsistent financial descriptions. Through this visualization, financial reviewers can more clearly see the evolution of problems—how a small error expands into logical inconsistencies throughout the report. These problem points and their evolution paths help reviewers quickly identify and correct key errors in the report. The second financial report is then updated regionally based on the correction map, updating financial indicators and their explanatory text according to spatial correlations. Each problem point in the correction map represents an area in the report that needs correction, and other related data and text are intelligently adjusted based on the correlations between these areas. For example, if an inconsistency is found in the description of the revenue growth data item, the system will not only modify the description but also adjust the explanatory text of other sections, such as net profit or costs, based on related data on revenue growth. Through targeted updates, the system ensures that every part of the report is adjusted in a timely and accurate manner, thereby achieving efficient correction of the financial report. After the related report update is completed, an intelligent integrity reassessment is performed, and an index structure for the report content is established. The purpose of this process is to ensure that every part of the report has been reasonably corrected, and that no new errors have been introduced during the updates of all data and text. By establishing an index structure for the report content, every piece of financial data, every financial term and its description can be indexed, ensuring that the interrelationships and logical consistency between data are not disrupted during any modifications to the report. After the integrity reassessment, the second financial report will become more accurate and standardized, forming the final target financial report.

[0099] The above describes the AI-based financial report generation method in the embodiments of this application. The following describes the AI-based financial report generation system in the embodiments of this application. Please refer to [link / reference]. Figure 2 One embodiment of the artificial intelligence-based financial report generation system in this application includes:

[0100] The data acquisition module 201 is used to collect and preprocess multi-source financial data to obtain target financial data;

[0101] Input module 202 is used to input the target financial data into a knowledge-based capital extraction algorithm for financial semantic understanding to obtain a financial data knowledge base;

[0102] The filling module 203 is used to perform intelligent report template matching and automatic filling of financial information based on the financial data knowledge base to obtain the first financial report;

[0103] Verification module 204 is used to perform financial logic verification on the first financial report through a causal relationship extraction model, identify the causal relationship between financial events and compare it with the expected model to obtain a financial logic verification report.

[0104] The generation module 205 is used to generate and optimize the first financial report based on the financial logic verification report using deep learning to generate a second financial report.

[0105] The update module 206 is used to perform multi-level automatic auditing and intelligent updating of the second financial report to obtain the target financial report.

[0106] Through the collaborative efforts of the aforementioned components, and by collecting and preprocessing multi-source financial data, the problems of scattered data sources and inconsistent formats in the traditional financial reporting process are solved, significantly improving data processing efficiency. The knowledge-based capital extraction algorithm overcomes the limitations of traditional methods in understanding the surface of financial data, achieving intelligent parsing of the deep semantics of financial data and accurately identifying the intellectual capital elements implicit in the financial data, enabling financial reports to comprehensively reflect corporate value. The intelligent report template matching and automatic filling technology overcomes the rigidity of traditional template use, dynamically selecting and adjusting the optimal template based on corporate characteristics and financial data features, achieving personalized report customization. The application of the causal relationship extraction model is the core innovation of this invention. This model integrates the characteristics of the Chinese language and graph attention network technology. It accurately identifies the causal relationships between financial events, effectively solving the problems of vague causal relationships and logical contradictions in traditional financial reports, making the report content logically rigorous and clearly organized. Based on deep learning-based content generation and optimization technology, it breaks through the limitations of traditional report generation methods, which are characterized by monotonous content and mechanical expression. It intelligently generates professional and accurate explanatory text based on the characteristics of financial data and verification results, enhancing the readability and professional depth of the report. Multi-level automatic review and intelligent update technology builds a comprehensive quality assurance system. Through multi-dimensional detection such as cross-validation, trend analysis, and semantic matching, it ensures the accuracy, completeness, and standardization of the report, significantly reducing the error rate, improving the efficiency of financial report generation, and further guaranteeing the quality and professional standards of the report, providing a full-process intelligent solution for corporate financial report preparation.

[0107] above Figure 2 The AI-based financial report generation system in this embodiment of the invention will be described in detail from the perspective of modular functional entities. The AI-based financial report generation device in this embodiment of the invention will be described in detail from the perspective of hardware processing.

[0108] Figure 3 This is a schematic diagram of the structure of an AI-based financial report generation device 300 provided in an embodiment of the present invention. The AI-based financial report generation device 300 can vary significantly due to different configurations or performance. It may include one or more central processing units (CPUs) 310 (e.g., one or more processors) and a memory 320, and one or more storage media 330 (e.g., one or more mass storage devices) for storing application programs 333 or data 332. The memory 320 and storage media 330 can be temporary or persistent storage. The program stored in the storage media 330 may include one or more modules (not shown in the diagram), each module including a series of instruction operations on the AI-based financial report generation device 300. Furthermore, the processor 310 may be configured to communicate with the storage media 330 and execute the series of instruction operations in the storage media 330 on the AI-based financial report generation device 300 to implement the steps of the aforementioned AI-based financial report generation method.

[0109] The AI-based financial report generation device 300 may also include one or more power supplies 340, one or more wired or wireless network interfaces 350, one or more input / output interfaces 360, and / or one or more operating systems 331, such as Windows Server, Mac OS X, Unix, Linux, FreeBSD, etc. Those skilled in the art will understand that... Figure 3 The illustrated structure of the AI-based financial report generation device does not constitute a limitation on the AI-based financial report generation device provided by this invention. It may include more or fewer components than illustrated, or combine certain components, or have different component arrangements.

[0110] The present invention also provides a computer-readable storage medium, which may be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium, wherein the computer-readable storage medium stores instructions that, when executed on a computer, cause the computer to perform the steps of the artificial intelligence-based financial report generation method.

[0111] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0112] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause an artificial intelligence-based financial report generation device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0113] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for generating financial reports based on artificial intelligence, characterized in that, The method includes: Collect and preprocess multi-source financial data to obtain the target financial data; The target financial data is input into a knowledge-based capital extraction algorithm for financial semantic understanding, resulting in a financial data knowledge base. Based on the aforementioned financial data knowledge base, intelligent report template matching and automatic filling of financial information are performed to obtain the first financial report; The first financial report is subjected to financial logic verification using a causal relationship extraction model. This process identifies causal relationships between financial events and compares them against an expected model to obtain a financial logic verification report. The process includes: decomposing the first financial report into multiple financial event units; extracting key financial indicator changes and business activities through word segmentation, syntactic parsing, and named entity recognition to obtain a set of financial events; constructing a financial event graph for the set of financial events, using each financial event as a node, and initializing potential connections between nodes to obtain an initial financial event network; and inputting the initial financial event network into the Chinese dictionary enhancement layer of the causal relationship extraction model to semantically enhance the event descriptions using a financial professional dictionary to obtain a final financial event network. The semantically enhanced event representation is processed by performing financial event dependency analysis on the graph attention network layer in the causal relationship extraction model. Multi-head attention weights between nodes are calculated to identify the strength of causal relationships between events, resulting in a weighted causal relationship graph. This weighted causal relationship graph is then compared with a pre-defined causal knowledge base in the financial domain to detect abnormal causal links and logical contradictions. Causal relationships that do not conform to financial rules are marked, resulting in financial logic anomaly markers. Based on these financial logic anomaly markers, a financial logic verification report is generated, containing verified causal chains and abnormal causal relationships that need correction. Correction suggestions are provided for each anomaly, resulting in the financial logic verification report. Based on the financial logic verification report, the first financial report is generated and optimized using deep learning content to generate a second financial report. The second financial report is subjected to multi-level automatic review and intelligent updates to obtain the target financial report.

2. The method for generating financial reports based on artificial intelligence according to claim 1, characterized in that, The process of collecting and preprocessing multi-source financial data to obtain target financial data includes: Regularly collect data from the enterprise's core financial system, audit report database, historical financial statements, and external regulatory data to obtain raw financial datasets; The original financial dataset is converted into a standardized structure by performing format conversion processing on the data in PDF reports, spreadsheets, and text documents to obtain the initial financial data. The initial financial data is subjected to a data integrity check to identify missing items in key financial fields and fill them in according to industry standards to obtain complete financial data; Perform consistency verification on the complete financial data, check the reconciliation between the balance sheet, income statement and cash flow statement, mark data items that do not conform to accounting standards, and obtain the verified financial data. Sensitive information processing is performed on the verified financial data, and customer identifiers and employee salary details are desensitized in a hierarchical manner to obtain desensitized financial data. Based on the anonymized financial data, a correlation diagram of financial indicators is constructed to obtain the target financial data.

3. The method for generating financial reports based on artificial intelligence according to claim 1, characterized in that, The step involves inputting the target financial data into a knowledge-based capital extraction algorithm for financial semantic understanding to obtain a financial data knowledge base, including: The target financial data is input into the terminology recognition module of the knowledge-based capital extraction algorithm for processing. Financial keywords, monetary values, and time information are extracted through a financial professional dictionary and named entity recognition technology to obtain a set of financial terms. The set of financial terms is input into the ontology mapping module in the knowledge-based capital extraction algorithm for processing. The terms are mapped to the standard financial concept system using a pre-set knowledge ontology in the financial field to obtain structured financial entities. The structured financial entity is input into the relation extraction module of the knowledge-based capital extraction algorithm for financial semantic association analysis. The logical relationship between the semantic subject and object is identified through syntactic dependency analysis to obtain the financial transaction relationship graph. The financial event relationship diagram is input into the event clustering module of the knowledge-based capital extraction algorithm for processing. The related events are organized into event units according to financial semantic similarity and temporal relevance to obtain a financial event sequence. The sequence of financial events is input into the capital classification module of the knowledge-based capital extraction algorithm for intellectual capital classification. The financial events are categorized and quantified according to predefined capital category standards to obtain a classified financial knowledge structure. The categorized financial knowledge structure is input into the weight calculation module of the knowledge-based capital extraction algorithm for processing. By calculating the importance weight of each knowledge node and constructing a multi-level index structure, the financial data knowledge base is obtained.

4. The method for generating financial reports based on artificial intelligence according to claim 1, characterized in that, The first financial report is generated by intelligently matching report templates and automatically filling in financial information based on the financial data knowledge base, including: Extract enterprise asset size, liability structure, and profitability data from the financial data knowledge base to obtain enterprise financial characteristic data; Based on the enterprise's financial characteristic data, the template matching degree in the template library is calculated, and the template with the highest matching degree is selected to obtain the preferred report template. The preferred report template is evaluated for its components, and table components suitable for the enterprise's data are selected and combined to obtain a customized report structure. The fields of the customized report structure are mapped to the data nodes in the financial data knowledge base to obtain the field filling rules; According to the field filling rules, the data in the financial data knowledge base is filled into the customized report structure, financial indicators are calculated and the results are filled to obtain a complete report; The complete report is formatted and standardized by adjusting the numerical format and table style, and marking the areas that need supplementary explanation to obtain the first financial report.

5. The method for generating financial reports based on artificial intelligence according to claim 1, characterized in that, The process of generating and optimizing the first financial report based on the financial logic verification report using deep learning to generate the second financial report includes: The financial logic verification report is mapped to the first financial report, and the location markers and corresponding correction suggestions for each area that needs to be corrected are extracted to obtain a correction instruction dataset. The correction instruction dataset is sorted according to the anomaly degree quantification value, and the priority score of each correction task is calculated to obtain a weighted correction task sequence. Based on the weighted correction task sequence, the content of the first financial report is reconstructed, unreasonable financial logic area text is replaced, and the correction basis explanation is supplemented to obtain a draft logic correction report. The draft logic correction report is used to generate key indicator explanations, analyze the reasons for abnormal fluctuations in financial indicators, add professional background analysis text, and obtain an enhanced indicator explanation report. The enhanced report on the explanation of the aforementioned indicators is transformed into a textually standardized report. The text standardization report is reorganized and its chapters adjusted to obtain the second financial report.

6. The method for generating financial reports based on artificial intelligence according to claim 1, characterized in that, The process of performing multi-level automatic auditing and intelligent updating of the second financial report to obtain the target financial report includes: A two-way cross-audit was performed on the second financial report, and the overall financial indicators were back-verified with the segment data to obtain the cross-verification results; The second financial report is input into the knowledge base comparison mechanism, compared with the historical financial report database, to identify abnormal trend change points and obtain a set of trend anomalies. Semantic framework analysis is performed on the second financial report to detect the semantic mapping consistency between the financial description and the corresponding data, and to mark the areas where the description and data do not match, thus obtaining a semantic matching degree report. Based on the cross-validation results, the set of trend anomalies, and the semantic matching degree report, a problem space map is generated. The detected problem points are located in three-dimensional space and the problem evolution path is constructed to obtain a corrected map. Based on the revised map, the second financial report is updated in a regionally targeted manner, and the financial indicators and their explanatory texts that affect each other are updated according to spatial correlation to obtain a correlation update report; A smart integrity reassessment is performed on the associated update report to establish a report content index structure and obtain the target financial report.

7. A financial report generation system based on artificial intelligence, characterized in that, For implementing the AI-based financial report generation method as described in any one of claims 1 to 6, the AI-based financial report generation system comprises: The data acquisition module is used to collect and preprocess multi-source financial data to obtain the target financial data; The input module is used to input the target financial data into a knowledge-based capital extraction algorithm for financial semantic understanding, thereby obtaining a financial data knowledge base; The filling module is used to perform intelligent report template matching and automatic filling of financial information based on the financial data knowledge base to obtain the first financial report; The verification module is used to perform financial logic verification on the first financial report through a causal relationship extraction model, identify the causal relationship between financial events and compare it with the expected model to obtain a financial logic verification report. The generation module is used to generate and optimize the first financial report based on the financial logic verification report using deep learning content to generate and optimize the second financial report. The update module is used to perform multi-level automatic review and intelligent updates on the second financial report to obtain the target financial report.

8. A financial report generation device based on artificial intelligence, characterized in that, The system includes a memory and a processor, the memory storing a computer program that can run on the processor, and the processor executing the computer program to implement the artificial intelligence-based financial report generation method according to any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is run by a processor, it causes the processor to perform the artificial intelligence-based financial report generation method as described in any one of claims 1 to 6.

Images