An intelligent question-answering method and device based on financial indicators and financial statements

Abstract

The application relates to the technical field of the cross of artificial intelligence and financial technology, and particularly discloses an intelligent question and answer method and device based on financial indicators and financial statements. Through collection of structured, semi-structured and unstructured financial data of enterprises, structured processing is carried out according to a financial indicator standard system, and a financial indicator knowledge base is constructed; semantic analysis and intention recognition are carried out on a natural language query of a user to obtain structured query elements, a data query instruction is generated by matching a question and answer template, financial data is searched, and derivative index calculation is completed, the data and the question and answer template are input into a large language model, a natural language form financial analysis result is output, and the semantic library and the template library are iteratively optimized based on user feedback. The application is adapted to enterprise self-defined financial indicator caliber, reduces the problem of large model illusion, realizes deep reasoning by relying on professional financial analysis logic, supports natural language fuzzy questioning, reduces the use threshold, and improves the accuracy and professionalism of financial data analysis.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence and financial technology, and more specifically, to an intelligent question-answering method, device, electronic device, and computer-readable storage medium that utilizes large language model technology to perform deep analysis, accurate retrieval, and natural language analysis results of structured, semi-structured, and unstructured financial data. Background Technology

[0002] With the deepening of digital transformation in domestic enterprises, the amount of unstructured documents and data generated in daily operations, such as financial statements, periodic operational analysis reports, internal and external audit working papers, and special financial statements, is experiencing explosive growth. This type of financial data comes in a variety of formats, including PDF files, Excel spreadsheets, scanned copies, and graphic reports, and is fragmented with low standardization. In traditional enterprise financial analysis and data retrieval scenarios, accounting personnel, financial analysts, and business managers need to expend significant time and effort to sift through massive amounts of financial documents, manually screening and retrieving core operational indicators and financial data. They also rely on their personal experience to perform complex analysis tasks such as financial ratio calculations, data comparison and verification, and business trend assessment. The overall workflow is cumbersome, inefficient, and prone to oversights and errors due to the high likelihood of human error.

[0003] Currently, some financial document retrieval systems based on keyword-based precise matching and basic financial question-and-answer tools developed using general-purpose language models exist in the market, which can initially meet simple financial data query needs. However, after practical application verification, these existing technical solutions have several significant and unavoidable defects when adapting to professional and sophisticated enterprise financial analysis scenarios. First, there are serious data interpretation biases and model illusions. General-purpose language models are not connected to the enterprise's specific financial accounting rules and indicator definition system. The accounting methods for similar financial indicators (such as gross profit, net profit margin, and operating costs) differ across different business segments and subsidiaries. General-purpose models cannot accurately adapt to the enterprise's personalized indicator rules, which easily leads to data interpretation errors, the generation of false data out of thin air, and calculation logic deviations, ultimately outputting incorrect analysis results and misleading enterprise business decisions. Secondly, there is a lack of professional and in-depth financial analysis logic. Existing search and question-and-answer tools can only perform basic text data extraction and numerical display functions. They cannot replicate the analytical thinking of professional financial analysts and cannot perform multi-level, in-depth data reasoning, problem attribution, and business judgment according to industry-standard and mature professional financial logic such as DuPont analysis, factor analysis, chain substitution method, and trend attribution analysis. The analysis dimensions are singular and lack depth. Thirdly, the human-computer interaction experience is poor and the usage threshold is high. Existing tools have extremely high requirements for the standardization of user input queries. Users must accurately input standard financial professional terms and standardized query sentences in order to retrieve corresponding data results. They cannot adapt to the interactive scenarios of everyday natural language vague questions and conversational consultations. Non-professional financial personnel find it difficult to use, and the practicality and popularity of the tools are extremely low.

[0004] In summary, current financial data question-and-answer and analysis technologies are ill-suited to meet the high-precision, in-depth, and lightweight financial data analysis needs of enterprises. The industry urgently needs to develop a financial intelligent question-and-answer method that can deeply integrate standardized financial indicator systems, industry-specific financial analysis logic, and intelligent reasoning capabilities of large language models. This method would thoroughly address the pain points of existing technologies, such as data bias, superficial analysis, and complex interactions, and achieve accurate, efficient, in-depth, and easy-to-use intelligent financial data analysis for enterprises. Summary of the Invention

[0005] The purpose of this invention is to provide an intelligent question-and-answer method based on financial indicators and financial statements, which aims to solve the problems of difficulty in retrieving financial data, insufficient analysis depth, and low accuracy of answers in the prior art.

[0006] To achieve the above objectives, the first aspect of the present invention provides an intelligent question-answering method based on financial indicators and financial statements, comprising the following steps: Step S1: Obtain the structured, semi-structured and unstructured financial data of the enterprise, and parse and structure the structured, semi-structured and unstructured financial data based on the preset financial indicator standard system to build a standardized financial indicator knowledge base; Step S2: Receive the natural language query request input by the user, and use the pre-built financial question-and-answer semantic database to perform semantic parsing and intent recognition on the natural language query request to obtain structured query elements; Step S3: Based on the structured query elements, match the target question and answer template from the pre-built question and answer template library, and generate a data query instruction according to the target question and answer template and the structured query elements; Step S4: Based on the data query instruction, retrieve the target financial data from the financial indicator knowledge base, and input the target financial data and the target question-and-answer template into the large language model for reasoning analysis, generating and outputting financial analysis results in natural language form.

[0007] Further, in step S1, the financial indicator standard system includes dimension standards, measurement standards, and attribute standards; the parsing and structuring of the structured, semi-structured, and unstructured financial data based on the preset financial indicator standard system includes: identifying key information in the structured, semi-structured, and unstructured financial data, and mapping the key information to the standardized fields corresponding to the dimension standards, measurement standards, and attribute standards; wherein, the dimension standards include at least time dimension, organizational dimension, and business segment dimension, and the measurement standards include at least beginning balance, ending balance, increase / decrease amount, and increase / decrease rate.

[0008] Further, in step S2, the step of using a pre-built financial question-and-answer semantic library to perform semantic parsing and intent recognition on the natural language query request includes: performing word segmentation on the natural language query request to obtain multiple query terms; performing entity alignment and disambiguation on the query terms based on the synonym and ambiguous word mapping relationship in the financial question-and-answer semantic library; identifying intent keywords in the processed query terms, and determining the user's query intent based on the intent keywords.

[0009] Further, in step S3, the question-and-answer templates in the question-and-answer template library are associated with specific financial analysis logic, which includes at least one of percentage analysis, trend analysis, and industry benchmarking analysis; generating a data query instruction based on the target question-and-answer template and the structured query elements includes: filling the structured query elements into the query statement frame corresponding to the target question-and-answer template to generate the data query instruction.

[0010] Furthermore, after step S4, the method further includes: Step S5: Obtain user feedback on the financial analysis results; Step S6: Based on the feedback information, update and optimize the financial question-and-answer semantic library and / or the question-and-answer template library.

[0011] Furthermore, in step S1, the construction of a standardized financial indicator knowledge base further includes: constructing an indicator system knowledge base based on industry expert knowledge and historical financial statements, wherein the indicator system knowledge base stores indicator names, indicator calculation formulas, and the relationships between indicators; in step S4, before inputting the target financial data into the large language model for reasoning analysis, the method further includes: performing derivative calculations on the target financial data based on the indicator calculation formulas in the indicator system knowledge base to obtain derivative indicator data; and inputting the target financial data and the derivative indicator data together into the large language model.

[0012] Furthermore, in step S4, the step of inputting the target financial data and the target question-and-answer template into a large language model for reasoning analysis includes: constructing analysis logic prompts containing the target question-and-answer template and input prompts for the target financial data; inputting the input prompts into a large language model fine-tuned with financial domain data to generate the financial analysis results.

[0013] A second aspect of the present invention provides an intelligent question-and-answer device based on financial indicators and financial statements, comprising: The knowledge base construction module is used to acquire structured, semi-structured and unstructured financial data of enterprises, and to parse and structure the structured, semi-structured and unstructured financial data based on a preset financial indicator standard system to build a standardized financial indicator knowledge base. The semantic parsing module is used to receive natural language query requests input by users, and to perform semantic parsing and intent recognition on the natural language query requests using a pre-built financial question-and-answer semantic database to obtain structured query elements; The instruction generation module is used to match a target question-and-answer template from a pre-built question-and-answer template library based on the structured query elements, and generate a data query instruction based on the target question-and-answer template and the structured query elements. The answer generation module is used to retrieve target financial data from the financial indicator knowledge base based on the data query instruction, and input the target financial data and the target question-and-answer template into the large language model for reasoning analysis, generating and outputting financial analysis results in natural language form.

[0014] A third aspect of the present invention provides an electronic device including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the program, implements the method as described in any of the preceding claims.

[0015] A fourth aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon that, when executed by a processor, implements the method as described in any of the preceding claims.

[0016] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are: This technical solution pre-builds a standardized financial indicator system to transform various unstructured financial documents into structured ones and constructs a financial indicator knowledge base. It leverages a financial question-and-answer semantic library to perform semantic parsing, entity disambiguation, and intent recognition for natural language queries. Combined with a question-and-answer template library, it matches corresponding professional financial analysis logic to generate data query commands. After retrieving target financial data from the knowledge base, it uses indicator calculation formulas to perform derivative indicator calculations. Then, it links a finely tuned large language model in the financial domain with the built-in analysis logic of the templates to perform reasoning and output financial analysis results. Simultaneously, it introduces a user feedback mechanism to continuously iterate and optimize the semantic and template libraries. This effectively overcomes the shortcomings of general-purpose large models, such as mismatched indicator definitions, generation illusions, and calculation errors. It breaks through the limitations of traditional tools that can only extract data superficially and cannot perform in-depth professional reasoning such as DuPont analysis and trend benchmarking. It lowers the barrier to entry for natural language queries, enabling intelligent financial question-and-answer that adapts to enterprise-defined financial indicator definitions, allows for in-depth professional deduction, and supports colloquial and ambiguous questions. This ensures accurate and reliable financial data analysis and provides high-quality financial analysis support for enterprise business decisions. Attached Figure Description

[0017] The present invention will be described by way of example and with reference to the accompanying drawings, wherein: Figure 1 A flowchart illustrating an intelligent question-answering method based on financial indicators and financial statements, provided for an embodiment of the present invention; Figure 2 A module structure diagram of the intelligent question-answering device provided in this embodiment of the invention. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of the embodiments. The components of the embodiments of this application described and shown in the accompanying drawings can be arranged and designed in various different configurations. Therefore, the following description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely represents selected embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0019] Example 1: This embodiment provides an intelligent question-answering method based on financial indicators and financial statements. This method relies on a standardized financial indicator system to process structured, semi-structured, and unstructured financial data. It combines a dedicated financial semantic database and a question-answering template library to achieve accurate parsing of natural language requirements. Furthermore, it leverages a large language model to complete professional financial reasoning and analysis, effectively solving the problems of inefficiency in traditional financial data queries, distortion in general model analysis, and insufficient analytical depth. Figure 1 As shown, the specific steps include: Step S1: Acquire the enterprise's structured, semi-structured, and unstructured financial data, and parse and structure the data based on a pre-defined financial indicator standard system to build a standardized financial indicator knowledge base. Specifically, this step comprehensively collects the enterprise's full-domain financial data, covering various formats of raw data such as monthly / quarterly / annual financial statements, business analysis reports, audit working papers, special financial reports, and business-finance linkage reports. For example, Excel spreadsheets can correspond to structured data or directly to data from other financial systems; PDFs and graphic reports correspond to semi-structured data; and scanned paper documents correspond to unstructured data. After acquiring the raw data, we strictly rely on the pre-built financial indicator standard system adapted to the enterprise's business scenarios to analyze the messy structured, semi-structured and unstructured financial data frame by frame, extract information and standardize and transform it into structured data. We eliminate invalid and redundant data, correct non-standard data formats, and organize the fragmented and non-standard raw financial information into unified and standardized structured data. Finally, we complete the construction of a standardized financial indicator knowledge base, which provides accurate, unified and complete data support for subsequent intelligent question answering and data analysis.

[0020] Step S2: Receive the user's natural language query request. Utilize a pre-built financial question-and-answer semantic database to perform semantic parsing and intent recognition on the natural language query request, obtaining structured query elements. After the user inputs a colloquial, ambiguous, or conventional natural language financial query request through the front-end interactive interface, the system captures the complete query text content in real time. It calls upon a pre-trained and built financial question-and-answer semantic database adapted to all scenarios of financial consulting to perform comprehensive semantic decomposition, semantic parsing, and accurate identification of the core query intent of the user's natural language statement. Invalid modifiers are removed, and core query information is filtered out. The unstructured natural language query statement is transformed into structured query elements with clear dimensions, complete elements, and a unified format, accurately identifying the user's query object, query dimension, query needs, and analytical requirements.

[0021] Step S3: Based on the structured query elements, match the target question-and-answer template from the pre-built question-and-answer template library, and generate a data query instruction based on the target question-and-answer template and the structured query elements. In this step, the system uses the parsed structured query elements as the matching basis, accurately traverses the pre-built professional financial question-and-answer template library, and combines the core information such as the corresponding financial analysis type, indicator type, and analysis dimension to accurately match a unique target question-and-answer template that is highly adapted to the user's query needs. After obtaining the target question-and-answer template, the core elements such as time, subject, business, indicator, and analysis method in the structured query elements are accurately filled into the pre-set standardized query statement framework of the target question-and-answer template, automatically assembling and generating a standardized data query instruction that can be directly called and adapted to knowledge base retrieval, realizing the automated transformation of query needs into data retrieval instructions.

[0022] Step S4: Based on the data query command, retrieve target financial data from the financial indicator knowledge base, and input the target financial data and the target question-and-answer template into the large language model for reasoning analysis, generating and outputting financial analysis results in natural language form. After receiving the generated data query command, the system accurately traverses the standardized financial indicator knowledge base, precisely filters and retrieves target financial data that perfectly matches the query command, avoiding interference from invalid data. Subsequently, the retrieved accurate target financial data and the target question-and-answer template carrying professional analysis logic are simultaneously input into the large language model. Relying on the semantic understanding and intelligent reasoning capabilities of the large language model, combined with the professional financial analysis logic embedded in the template, in-depth data analysis, data comparison, trend judgment, problem attribution, and other reasoning operations are completed. Finally, the professional and data-driven analysis results are transformed into easy-to-understand, logically clear, and complete natural language text form, completing the output display of the analysis results for users to directly view and use.

[0023] In a further implementation, in step S1, the financial indicator standard system is a dedicated system adapted to the standardized processing of enterprise-wide financial data, specifically including three core standard modules: dimension standards, measurement standards, and attribute standards. The specific process of parsing and structuring the structured, semi-structured, and unstructured financial data based on the preset financial indicator standard system is as follows: accurately identifying all key information contained in the structured, semi-structured, and unstructured financial data, such as the name, value, statistical time, attribution entity, business scenario, and data change status of the financial indicators; and accurately mapping each identified key information to the standardized fields corresponding to the dimension standards, measurement standards, and attribute standards, thus completing the standardized classification and organization of the original information. Among these, the dimension standards are the basic division dimensions for financial data analysis, including at least time, organizational, and business segment dimensions, which can accurately distinguish financial data from different statistical periods, different subsidiaries, and different business segments; the measurement standards are the core measurement and statistical dimensions of financial data, including at least beginning balances, ending balances, increases / decreases, and increases / decreases rates, which can comprehensively cover the statistical needs of both static and dynamic changes in financial data.

[0024] In a further implementation, step S2, the specific process of semantic parsing and intent recognition of the natural language query request using a pre-built financial question-and-answer semantic library, is as follows: First, the complete natural language query request input by the user is subjected to refined word segmentation, splitting it into several independent query terms, modifiers, and limiting words, completing the basic text decomposition. Then, the complete mapping relationship of pre-collected financial professional synonyms, industry-wide common aliases, and polysemous words in the financial question-and-answer semantic library is retrieved. All the split query terms are then subjected to precise entity alignment and semantic disambiguation processing to resolve the problems of inconsistent financial terminology and lexical ambiguity, ensuring the uniqueness and accuracy of terminology understanding. Finally, the intent keywords contained in the aligned and disambiguated query terms are accurately identified. Combined with financial scenario-specific intent determination rules, the user's core query intent is accurately locked and determined, providing accurate demand basis for subsequent template matching and data retrieval.

[0025] In a further implementation, in step S3, each question-and-answer template in the question-and-answer template library is pre-bound with a corresponding exclusive financial analysis logic, achieving one-to-one adaptation between the template and the analysis logic. The financial analysis logic includes at least one of percentage analysis, trend analysis, and industry benchmarking analysis, covering most of the daily financial analysis scenarios for enterprises. The specific method for generating data query instructions based on the target question-and-answer template and the structured query elements is as follows: The core parameter elements contained in the structured query elements, such as time range, organizational entity, business segment, target financial indicator, analysis type, and comparison dimension, are accurately filled into the standardized query statement framework preset in the target question-and-answer template, completing the precise adaptation and assembly of elements and framework, and ultimately generating a data query instruction that is formatted correctly, logically complete, and directly executable.

[0026] In a further embodiment, after the financial analysis results are output in step S4, an iterative optimization step is also provided, specifically including: Step S5: Obtain user feedback on the financial analysis results. After the system outputs the natural language financial analysis results to the user, it opens a user feedback interaction port to actively collect various types of user feedback on the analysis results. Specifically, this includes feedback on the accuracy of the results, satisfaction with the depth of the analysis, query suitability, existing problems, and optimization suggestions, etc., and retains complete user interaction data.

[0027] Step S6: Based on the feedback information, update and optimize the financial Q&A semantic database and / or the Q&A template database. The system categorizes, intelligently analyzes, and verifies the compliance of the collected user feedback information. Addressing issues mentioned in the feedback, such as semantic recognition bias, template matching errors, missing analysis logic, and insufficient Q&A adaptability, the system dynamically updates, corrects, and optimizes the synonym mapping, ambiguous word rules, and intent recognition model of the financial Q&A semantic database, as well as the template content, analysis logic, and adaptability scenarios of the Q&A template database. This continuously improves the accuracy of the system's semantic parsing, the accuracy of template matching, and the professionalism of financial analysis, achieving continuous iterative upgrades of the model and knowledge base.

[0028] In a further implementation, step S1, the process of constructing a standardized financial indicator knowledge base, also includes a specialized knowledge base construction process: relying on the professional knowledge and mature financial analysis experience of senior financial experts in the industry, combined with the company's historical financial statements, financial accounting rules, and business-financial linkage data, a specialized indicator system knowledge base is constructed. This indicator system knowledge base is specifically used to systematically store complete information on various financial indicators, including indicator standard names, indicator-specific calculation formulas, indicator statistical definitions, applicable scenarios for indicators, and the relationships, linkage logic, and calculation levels between different indicators. In step S4, before inputting the target financial data into the large language model for reasoning analysis, a data derivation calculation step is also set up: retrieving the calculation formulas and operation rules of various indicators stored in the indicator system knowledge base, and automatically completing multi-level and multi-dimensional derivation calculations based on the retrieved original target financial data to accurately calculate various derivation indicator data and make up for the lack of dimensions of the original data; finally, the original target financial data and the calculated derivation indicator data are input into the large language model together to provide more comprehensive, multi-dimensional and in-depth data support for the deep financial reasoning analysis of the large model, and ensure the integrity and professionalism of the analysis results.

[0029] In a further implementation, step S4, specifically the process of inputting the target financial data and the target question-and-answer template into a large language model for reasoning analysis, is as follows: First, combining the professional financial analysis logic embedded in the target question-and-answer template, targeted analysis logic prompts adapted to the needs of this query analysis are constructed. Simultaneously, standardized target financial data is integrated and organized to generate complete input prompt content. Then, the constructed input prompts are input into a large language model that has been specifically fine-tuned and optimized using massive amounts of financial domain-specific data and professional financial analysis cases. Relying on the financial domain-specific reasoning capabilities of the fine-tuned large model, the illusion problem and logical defects of general models are avoided, accurately completing operations such as data interpretation, computational analysis, trend comparison, and attribution judgment. Finally, a professional financial analysis result that is logically rigorous, data-accurate, and closely aligned with the actual operating conditions of the enterprise is generated.

[0030] Example 2: This embodiment provides an intelligent question-answering device based on financial indicators and financial statements. This device is a functional entity adapted to the intelligent question-answering method of Embodiment 1 above. Through modular division of labor and collaboration, it achieves fully automated operation of the entire process, including standardized processing of structured, semi-structured, and unstructured financial data, parsing of user natural language requirements, generation of query commands, intelligent financial analysis, and result output. Each module is functionally independent yet interconnected and collaborative, efficiently completing the intelligent financial question-answering analysis work. Figure 2 As shown, it specifically includes the following core functional modules: The knowledge base construction module is the core module for basic data processing and storage of this device. It is primarily used to comprehensively collect various structured, semi-structured, and unstructured financial data from enterprises, covering original financial documents such as multi-format financial statements, operating reports, and audit working papers. Simultaneously, it strictly adheres to a pre-set enterprise-specific financial indicator standard system, performing comprehensive information analysis, key information extraction, format standardization, and structure conversion on the collected messy structured, semi-structured, and unstructured financial data. This unifies data statistical standards and output formats, ultimately building a standardized, regulated, and efficiently searchable financial indicator knowledge base. This provides stable, accurate, and comprehensive underlying data support for all subsequent queries, analyses, and reasoning operations, mitigating analytical errors caused by non-standard data and inconsistent standards from the outset.

[0031] The semantic parsing module, the core module for demand analysis in this device, is mainly used to receive various natural language financial query requests input by users through the interactive terminal in real time, adapting to various user input formats such as colloquial, ambiguous, and conventional. Simultaneously, it calls upon the system's pre-built professional financial question-and-answer semantic library to perform a full-process processing on the user's input natural language query text, including fine-grained word segmentation, semantic parsing, entity alignment, ambiguity resolution, and core intent recognition. This accurately removes invalid information, identifies core query elements, and ultimately outputs structured query elements that are complete in dimensions, clear in elements, and uniform in format. This achieves accurate conversion of user natural language needs into standardized, machine-recognizable needs, lowering the user's barrier to entry.

[0032] The instruction generation module, the core module for instruction conversion in this device, primarily receives structured query elements output by the semantic parsing module. Using these elements as the core matching criterion, it quickly traverses the system's pre-built specialized question-and-answer template library. Combining the corresponding financial indicator type, analysis scenario, and analysis logic, it accurately matches and obtains a highly suitable target question-and-answer template. Simultaneously, relying on the standardized query statement framework inherent in the target question-and-answer template, it accurately fills in and adapts various core parameters from the structured query elements, automatically assembling and generating standardized data query instructions that can be directly retrieved and called by the knowledge base. This automates and accurately transforms user needs into data retrieval instructions, ensuring the efficiency and accuracy of the retrieval.

[0033] The answer generation module, the core module for intelligent analysis and result output of this device, primarily receives data query commands from the command generation module. Based on these commands, it accurately retrieves standardized financial indicator knowledge bases, quickly filters and retrieves the target financial data with the highest matching degree, avoiding interference from redundant and invalid data. Simultaneously, the retrieved precise target financial data and the target question-and-answer template carrying professional financial analysis logic are input into a finely tuned financial-specific large-scale language model. Relying on the intelligent reasoning capabilities and professional financial analysis logic of the large model, it completes in-depth data calculations, trend analysis, proportion analysis, benchmarking analysis, problem attribution, and other reasoning operations. Ultimately, it transforms the complex data analysis results into easily understandable, logically clear, and detailed natural language financial analysis results, and automatically outputs and displays them for users to directly view, reference, and use for decision-making.

[0034] Example 3: This embodiment provides an electronic device that can be deployed independently or integrated into an enterprise's financial digitization system or business management platform. It can stably execute the entire process of the intelligent question-and-answer method based on financial indicators and financial statements described in Embodiment 1, achieving automated operation of intelligent question-and-answer and in-depth analysis of financial data. Specifically, the electronic device comprises three core hardware and software architectures: a memory, a processor, and a computer program stored in the memory and capable of running on the processor.

[0035] The memory, as the core storage unit of the device, is used to permanently or temporarily store various structured, semi-structured, and unstructured financial data of the enterprise, including raw financial data, standardized financial indicator knowledge base data, financial question-and-answer semantic database data, question-and-answer template database data, indicator system knowledge base data, as well as the complete computer program, algorithm model, fine-tuning parameters, and operation logs for implementing the intelligent question-and-answer method. This ensures stable storage, rapid retrieval, and secure retention of various data and programs. The processor, as the core computing and control unit of the device, establishes a stable data interaction link with the memory. It can retrieve the computer program stored in the memory in real time. During program execution, it sequentially executes all steps, including financial data acquisition and structured processing, user query request reception and semantic parsing, question-and-answer template matching and query command generation, financial data retrieval and large-scale model inference analysis, result output, and iterative optimization. This accurately implements all the technical processes and functions of the intelligent question-and-answer method described in any of the above embodiments.

[0036] Meanwhile, this electronic device is compatible with various deployment modes such as local deployment, cloud deployment, and hybrid deployment. It has stable computing power, data processing power, and model reasoning capabilities, and can adapt to the processing needs of massive financial data and high-frequency intelligent question-and-answer query scenarios of enterprises. It has strong operational stability and high data processing accuracy, and can effectively support the implementation of routine financial intelligent analysis work of enterprises.

[0037] Example 4: This embodiment provides a computer-readable storage medium, which is a software carrier adapted to the application of financial intelligent question answering methods. It can be stored independently, ported and reused, and has a computer program that can be recognized and executed by the processor of an electronic device.

[0038] When the computer program is retrieved and executed by the processor of the electronic device, it can accurately implement all the technical steps and core functions of the intelligent question-answering method based on financial indicators and financial statements described in any of the above embodiments. Specifically, it can realize the full-process functions such as standardized and structured processing of enterprise structured, semi-structured and unstructured financial data, construction of financial indicator knowledge base, semantic parsing and intent recognition of user natural language query requests, question-answer template matching and data query instruction generation, accurate retrieval of financial data, deep financial reasoning analysis of large models, output of natural language analysis results, collection of user feedback and iterative optimization of knowledge base.

[0039] The computer-readable storage medium can take the form of various conventional readable storage media such as USB flash drives, portable hard drives, read-only memory, random access memory, optical discs, and cloud storage media. It has the characteristics of strong versatility, high portability, stable storage, and convenient reading. It can be used by various smart terminals, servers, industrial control equipment and other hardware devices, and can quickly realize the batch deployment and application of financial intelligent question answering technology solutions, effectively reducing the technical implementation threshold and improving the reusability and adaptability of the solution.

[0040] The above-described 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 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. An intelligent question-answering method based on financial indicators and financial statements, characterized in that, Includes the following steps: Step S1: Obtain the structured, semi-structured and unstructured financial data of the enterprise, and parse and structure the financial data based on the preset financial indicator standard system to build a standardized financial indicator knowledge base; Step S2: Receive the natural language query request input by the user, and use the pre-built financial question-and-answer semantic database to perform semantic parsing and intent recognition on the natural language query request to obtain structured query elements; Step S3: Based on the structured query elements, match the target question and answer template from the pre-built question and answer template library, and generate a data query instruction according to the target question and answer template and the structured query elements; Step S4: Based on the data query instruction, retrieve the target financial data from the financial indicator knowledge base, and input the target financial data and the target question-and-answer template into the large language model for reasoning analysis, generating and outputting financial analysis results in natural language form.

2. The method according to claim 1, characterized in that, In step S1, the financial indicator standard system includes dimensional standards, measurement standards, and attribute standards. The method of parsing and structuring the structured, semi-structured, and unstructured financial data based on a preset financial indicator standard system includes: Identify key information in the structured, semi-structured, and unstructured financial data, and map the key information to the standardized fields corresponding to the dimension standards, measurement standards, and attribute standards; The dimensional standards include at least time, organization, and business segment dimensions, and the measurement standards include at least beginning balance, ending balance, increase / decrease amount, and increase / decrease rate.

3. The method according to claim 2, characterized in that, In step S2, the step of using a pre-built financial question-and-answer semantic database to perform semantic parsing and intent recognition on the natural language query request includes: The natural language query request is segmented to obtain multiple query terms; Based on the synonym and ambiguous word mapping relationship in the financial question and answer semantic database, entity alignment and disambiguation processing are performed on the query terms; Identify intent keywords in the processed query terms and determine the user's query intent based on the intent keywords.

4. The method according to claim 3, characterized in that, In step S3, the question-and-answer templates in the question-and-answer template library are associated with specific financial analysis logic, which includes at least one of percentage analysis, trend analysis, and industry benchmarking analysis. The step of generating data query instructions based on the target question-and-answer template and the structured query elements includes: The structured query elements are populated into the query statement frame corresponding to the target question and answer template to generate the data query instruction.

5. The method according to claim 4, characterized in that, Following step S4, the method further includes: Step S5: Obtain user feedback on the financial analysis results; Step S6: Based on the feedback information, update and optimize the financial question-and-answer semantic library and / or the question-and-answer template library.

6. The method according to claim 5, characterized in that, In step S1, the construction of a standardized financial indicator knowledge base further includes: Based on industry expert knowledge and historical financial statements, an indicator system knowledge base is constructed, which stores indicator names, indicator calculation formulas, and the relationships between indicators. In step S4, before inputting the target financial data into the large language model for reasoning analysis, the method further includes: Based on the indicator calculation formulas in the indicator system knowledge base, the target financial data is derived to obtain derived indicator data. The target financial data and the derived indicator data are input together into the large language model.

7. The method according to claim 6, characterized in that, In step S4, inputting the target financial data and the target question-answer template into the large language model for reasoning analysis includes: Construct analytical logic prompts containing the target question-and-answer template and input prompts for the target financial data; The input prompts are fed into a large language model that has been fine-tuned with financial domain data to generate the financial analysis results.

8. An intelligent question-and-answer device based on financial indicators and financial statements, characterized in that, include: The knowledge base construction module is used to acquire structured, semi-structured and unstructured financial data of enterprises, and to parse and structure the structured, semi-structured and unstructured financial data based on a preset financial indicator standard system to build a standardized financial indicator knowledge base. The semantic parsing module is used to receive natural language query requests input by users, and to perform semantic parsing and intent recognition on the natural language query requests using a pre-built financial question-and-answer semantic database to obtain structured query elements; The instruction generation module is used to match a target question-and-answer template from a pre-built question-and-answer template library based on the structured query elements, and generate a data query instruction based on the target question-and-answer template and the structured query elements. The answer generation module is used to retrieve target financial data from the financial indicator knowledge base based on the data query instruction, and input the target financial data and the target question-and-answer template into the large language model for reasoning analysis, generating and outputting financial analysis results in natural language form.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the method as described in any one of claims 1 to 7.

Images