An intelligent question and answer and analysis method and system based on artificial intelligence and a medium

Abstract

This invention belongs to the field of data analysis and processing technology. Specifically, it relates to an artificial intelligence-based intelligent question-answering and analysis method, system, and medium. The method includes acquiring and integrating internal financial data and external market data; constructing a financial knowledge graph based on the processed data; when receiving a user's natural language question, analyzing and identifying key elements in the question and linking them to the knowledge graph to generate a structured query; retrieving relevant data from the knowledge graph based on this query and calling pre-set financial analysis rules for calculation to generate interpretable analysis results containing data sources, calculation processes, and reasoning paths; furthermore, this invention can proactively monitor key indicators based on historical data, identify anomalies or trends, and push warnings to users. This invention effectively lowers the professional threshold for financial analysis, improves analysis efficiency and the interpretability of results, and achieves proactive intelligent warnings.

Description

Technical Field

[0001] This invention belongs to the field of data analysis and processing technology, specifically, it relates to an intelligent question answering and analysis method, system and medium based on artificial intelligence. Background Technology

[0002] In the field of corporate financial management and decision support, efficient and accurate data analysis and insight capabilities are crucial. However, there are currently many technical challenges: (1) Data heterogeneity and semantic barriers are prominent. Financial data comes from a wide range of sources, including internal structured transaction data (such as ERP system data), semi-structured and unstructured documents (such as financial reports and audit opinions), as well as external market conditions, industry benchmarks, and other data. These data have different formats and naming conventions, and there are a lot of synonyms (such as sales revenue, operating revenue, and main operating revenue pointing to the same indicator), which makes data integration difficult; (2) The threshold for information acquisition and analysis is high. If business personnel or managers want to obtain specific analysis results (such as comparing the sales gross profit margins of companies A and B in East China over the past three years), they usually need to make a request to the IT department, and technical personnel will write complex query scripts or build analysis models. This process is time-consuming, slow to respond, and heavily relies on the professional knowledge of technical personnel, which cannot meet the needs of real-time decision-making; (3) The interpretability of analysis results is insufficient. Traditional reporting or BI tools primarily provide final numerical values, lacking a clear display of data origins, calculation logic, and reasoning processes. When anomalies occur, users struggle to quickly understand the results, reducing the credibility of the analysis and its value in supporting decision-making. Summary of the Invention

[0003] The purpose of this invention is to provide an intelligent question-answering and analysis method based on artificial intelligence to solve the technical problems existing in the prior art.

[0004] To achieve the above objectives, the technical solution adopted by the present invention is as follows: An AI-based intelligent question answering and analysis method includes the following steps: Step S1: Obtain financial data, which includes internal financial data and external market data, and perform semantic alignment and standardization processing. Step S2: Based on the semantically aligned financial data, construct a financial knowledge graph to provide a unified representation of the financial data; Step S3: Receive the user's natural language question, parse the natural language question, identify the financial entities, indicators, time and analysis dimensions, and generate a structured query; Step S4: Based on the structured query, retrieve relevant information from the financial knowledge graph and perform corresponding financial analysis calculations; Step S5: Generate analysis results, which include data sources, calculation processes, and reasoning paths; Step S6: Based on historical queries and financial data characteristics, proactively identify financial anomalies or trends, and push relevant information to users.

[0005] Preferably, the specific method of step S1 is as follows: Step S1.1: Obtain the company's internal structured transaction data, unstructured financial documents, and external market and industry data; Step S1.2: Parse the unstructured financial documents to extract financial entities, numerical indicators, accounting periods, and business events; perform semantic annotation on the structured data fields; Step S1.3: Establish semantic mapping relationships to align entities and indicators with the same semantics but different expressions from different data sources; Step S1.4: Convert the data from different accounting periods and currencies to the standard reporting period and the designated functional currency.

[0006] Preferably, in step S1.2, a natural language processing model optimized with financial domain knowledge is used for document parsing and information extraction.

[0007] Preferably, in step S1.4, the method for converting to the standard reporting period is as follows: identify the original accounting period of the data, and aggregate the non-standard time point data to the most recent standard reporting time point by means of forward filling or moving average.

[0008] Preferably, the specific method for constructing the financial knowledge graph in step S2 includes: Step S2.1: Define the financial entities, numerical indicators, accounting periods, and business events processed in step S1 as graph nodes; wherein, financial entities include companies, departments, products, customers, and suppliers; numerical indicators include revenue, profit, cost, and various financial ratios; Step S2.2: Based on financial rules, accounting standards, and business logic, predefine the relationship types between nodes, including membership relationships, computational relationships, comparative relationships, and time series relationships; establish relationship edges of corresponding types for node instances according to the actual associations in the data; Step S2.3: Populate the triples of the knowledge graph using relations directly mapped from structured data and relation assertions parsed from unstructured text.

[0009] Preferably, the specific method of step S3 is as follows: Step S3.1: Determine the type of analytical intent for the natural language question, including fact query, indicator calculation, time comparison, trend analysis, and anomaly attribution; Step S3.2: Identify financial entities, numerical indicators, time phrases, and keywords for analysis dimensions from the natural language question; Step S3.3: Associate and match the keywords with the corresponding nodes in the financial knowledge graph constructed in step S2; Step S3.4: Structured query generation. Based on the analysis intent type and the successfully linked graph nodes, a structured query statement containing the target entity, indicators, time range, and analysis operation is generated to drive the retrieval and analysis in step S4.

[0010] Preferably, in step S4, performing financial analysis calculations means: according to the analysis intent type of the structured query, calling preset financial analysis rules and calculation functions to perform calculations on the relevant data retrieved from the financial knowledge graph.

[0011] Preferably, the specific implementation steps of step S6 are as follows: Step S6.1: Based on historical query frequency and business importance, select a set of key financial indicators; extract the historical time series data of the key financial indicators from the data after standardization processing in step S1; Step S6.2: Analyze the historical time series data to identify its periodicity and trend characteristics, and detect abnormal data points that deviate from the normal historical fluctuation range or preset business rules; Step S6.3: When the abnormal data point is detected or a significant trend change is identified, an early warning message containing an anomaly description, possible causes, and related data is automatically generated; Step S6.4: Push the warning information to the relevant business manager or analysis user through a preset message interface.

[0012] To achieve the above objectives, the present invention also provides an intelligent question-answering and analysis system based on artificial intelligence, comprising: The data acquisition module acquires financial data, including internal corporate financial data and external market data, and performs semantic alignment and standardization processing. The financial knowledge graph module constructs a financial knowledge graph based on semantically aligned financial data to provide a unified representation of the financial data. The question parsing module receives users' natural language questions, parses the natural language questions, identifies the financial entities, indicators, time periods, and analysis dimensions, and generates structured queries. The financial analysis and calculation module retrieves relevant information from the financial knowledge graph based on the structured query and performs corresponding financial analysis and calculations. The analysis result generation module generates analysis results, which include data sources, calculation processes, and reasoning paths. The early warning module proactively identifies financial anomalies or trends based on historical queries and financial data characteristics, and pushes relevant information to users.

[0013] To achieve the above objectives, the present invention also provides a computer-readable storage medium having a computer program stored thereon, the computer program being executed by a processor to implement the artificial intelligence-based intelligent question answering and analysis method described above.

[0014] Compared with the prior art, the present invention has the following beneficial effects: (1) This invention solves the semantic gap between financial data from different sources and in different formats by semantic alignment and standardization, and unifies structured and unstructured data.

[0015] (2) The present invention allows users to ask questions directly using natural language to obtain complex financial analysis results. Users do not need to master professional query languages ​​or programming skills, which greatly reduces the threshold for data analysis and improves decision-making efficiency.

[0016] (3) The analysis results of this invention not only include the final values, but also clearly list the data sources, detailed calculation steps and business logic reasoning paths, which improves the transparency of the analysis process and the credibility of the results, making it easier for users to verify and understand.

[0017] (4) By constructing a financial knowledge graph, this invention connects scattered data points into a network rich in semantic relationships, which not only provides rich context for natural language understanding, but also lays a solid foundation for complex association analysis and influence reasoning. Attached Figure Description

[0018] Figure 1 This is a flowchart of Embodiment 1 of the present invention.

[0019] Figure 2 This is a principle block diagram of Embodiment 3 of the present invention. Detailed Implementation

[0020] To enable those skilled in the art to have a clearer understanding of the present invention, the present invention will be further described in detail below with reference to embodiments. It should be understood that the specific embodiments described below are merely for illustrative purposes and to facilitate understanding. The technical solutions provided by the present invention are not limited to those provided in the following embodiments, nor should they limit the scope of protection of the present invention.

[0021] It should be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of the present invention. Therefore, the drawings only show the components related to the present invention and are not drawn according to the actual number, shape and size of the components in the actual implementation. In the actual implementation, the shape, quantity and proportion of each component can be arbitrarily changed, and the layout of the components may also be more complex. Example 1

[0022] like Figure 1 As shown, this embodiment provides an intelligent question-answering and analysis method based on artificial intelligence, which includes the following steps: Step S1: Obtain financial data and perform semantic alignment and standardization processing.

[0023] The specific implementation of this step is as follows: Step S1.1: Obtain the enterprise's internal structured transaction data, unstructured financial documents, and external market and industry data; specifically, obtain structured transaction details and general ledger data from the enterprise's internal financial system (such as SAP, Oracle) through tools such as database connectors, API interfaces, and file parsers; obtain unstructured annual reports, quarterly reports, board resolutions, and other PDF or Word documents from the document management system; at the same time, obtain external data such as industry average profit margins, stock market prices, and macroeconomic indicators through web scraping or purchased financial data APIs.

[0024] Step S1.2: Information parsing and semantic annotation; For unstructured documents, a natural language processing model optimized with financial domain knowledge is used for parsing. This model can be based on a general model such as BERT architecture, fine-tuned using a large amount of financial reports and accounting standards text. The model's task is to identify and extract key information from the document, such as: financial entities (e.g., XX Co., Ltd., Consumer Electronics Division), numerical indicators (e.g., net profit: 10 million yuan), accounting period (e.g., Q1 of fiscal year 2023), and business events (e.g., completion of the acquisition of Company X); For structured data, semantic annotation is performed on its database table names and field names, for example, the field rev_amt is annotated as revenue amount.

[0025] Step S1.3: Establish semantic mapping relationships to align entities and indicators with the same semantics but different expressions from different data sources. Specifically, establish a thesaurus and ontology library for the financial domain. For example, map sales revenue, revenue, and main business revenue to the standard concept of operating revenue; map the Shanghai Stock Exchange and Shanghai Stock Exchange to the entity Shanghai Stock Exchange. Using this mapping table, unify entities and indicators with different expressions but the same substance from different data sources to achieve semantic alignment.

[0026] Step S1.4: Convert data from different accounting periods and currencies to the standard reporting period and the designated local currency. Specifically, identify the original timestamps of all data points (such as daily and monthly data), define a standard reporting period sequence (such as the end of each calendar month), and for non-standard time point data (such as data from a certain week), use methods such as forward filling (using the data from the last day of the week to represent the month) or moving average (calculating the average of all weekly data within the month) to aggregate them to the most recent standard reporting time point; identify the original currency of the data, and convert all currency amounts to the local currency according to the business-specified local currency (such as RMB) using the historical exchange rate or the period-end exchange rate of the corresponding accounting period.

[0027] Step S2: Construct a financial knowledge graph based on the semantically aligned financial data.

[0028] Based on the unified data processed in step S1, a financial knowledge graph is constructed to form a global, interconnected data view. The specific steps are as follows: Step S2.1: Define the extracted and labeled financial entities, numerical indicators, accounting periods, business events, etc., as nodes of the knowledge graph. For example, node types include: company, product line, accounting subject, financial indicator, reporting period, business event, etc.

[0029] Step S2.2: Based on financial rules, accounting standards, and business logic, predefine the relationship types between nodes, as follows: Subordination relationship: such as (Business Unit A, belongs to, Company X); Calculation relationship: such as (Gross Profit Margin, calculation formula is, (Operating Revenue - Operating Costs) / Operating Revenue), this relationship can be associated with an attribute to store the formula; Time series relationship: such as (Company X, in 2023Q1, Operating Revenue, 1000), this is a relationship edge with a temporal attribute; Comparison relationship: such as (Company A's Gross Profit Margin, industry comparison, industry average gross profit margin).

[0030] Step S2.3: Utilize the relationships directly mapped from foreign key associations in structured tables, and the relationship assertions obtained from unstructured text through a relationship extraction model (e.g., extracting (merger event, cause, goodwill increase) from the goodwill increase resulting from mergers and acquisitions), to form a large number of (head entity, relationship, tail entity) triples, which are then populated into the graph database. At the same time, set up regular (e.g., daily) and triggered (e.g., new financial report release) update mechanisms to ensure the timeliness of the graph.

[0031] Step S3: Receive and parse the natural language question to generate a structured query.

[0032] Suppose a user enters a question on the front-end interface such as "Compare the R&D expense ratio trends of companies A and B over the past three years," the processing step is as follows: Step S3.1: Determine the type of analytical intent for the natural language question. The types of analytical intent include fact query, indicator calculation, time comparison, trend analysis, and anomaly attribution. Taking the above assumption as an example, the analytical intent for this question is determined to be "time comparison" and "trend analysis".

[0033] Step S3.2: Identify financial entities, numerical indicators, time phrases, and analytical dimension keywords from the natural language question; taking the above assumption as an example, using named entity recognition and keyword extraction technology, identify the key elements in the question: entities [Company A, Company B], indicators [R&D expense ratio], time [past three years], and dimensions [trends]; for possible polysemous words, such as "apple", combine the context (financial field) to disambiguate it as "Apple Inc." instead of fruit.

[0034] Step S3.3: Match the keywords with the corresponding nodes in the financial knowledge graph constructed in step S2; taking the above assumptions as an example, for instance, link the R&D expense ratio to the standard indicator node and obtain its calculation definition; link Company A and Company B to the corresponding company nodes.

[0035] Step S3.4: Structured query generation. Specifically, based on the analysis intent type and successfully linked graph nodes, a structured query statement containing the target entity, indicators, time range, and analysis operations is generated. For example: { "intent":"trend_comparison", “entities”:["Company A_ID", "Company B_ID"], “metric”: “R&D Expense Ratio_ID”, "time_range":{"start":"2020-01-01","end":"2022-12-31"}, "granularity": "quarterly" / / Analyze granularity }

[0036] Step S4: Perform intelligent financial analysis based on structured queries.

[0037] Based on the analytical intent type of the structured query, pre-defined financial analysis rules and calculation functions are invoked to perform calculations on the relevant data retrieved from the financial knowledge graph. For example, based on the entity ID, indicator ID, and time range in the query, the financial knowledge graph is quickly traversed to retrieve the raw data of R&D expenses and operating revenue for Company A and Company B for each quarter (or month) over the past three years (because the R&D expense ratio needs to be calculated). Based on the query intent `trend_comparison`, pre-defined trend calculation and comparative analysis functions are invoked: First, according to the definition formula of the R&D expense ratio (R&D expenses / operating revenue), the ratio values ​​of the two companies at each time point are calculated. Then, the time series analysis function is invoked to fit two trend lines respectively. Finally, the slope, distance, and other comparative indicators of the two trend lines are calculated.

[0038] Step S5: Generate analysis results.

[0039] The generated analysis report not only provides conclusions (such as the R&D expense ratio of Company A showing an upward trend and being higher than that of Company B), but also includes the following: (1) Data source: List the original data items on which the calculation is based, such as Company A's R&D expense in Q1 2021: 500,000, operating income: 5 million, taken from page X of the Q1 2021 financial report; (2) Calculation process: Clearly show each step of the calculation, such as the R&D expense ratio in Q1 2021 = 50 / 500 = 10%; (3) Reasoning path: Explain the analysis logic, such as the fact that the query intention is to compare trends, so the ratio is calculated by quarter first, then linear regression is used to fit the trend, and finally the slope of the trend line is compared.

[0040] Step S6: Proactively identify financial anomalies or trends and push notifications.

[0041] The specific implementation of this step is as follows: Step S6.1: Based on historical query frequency and business importance, select a set of key financial indicators; extract historical time series data of key financial indicators from the data after standardization in step S1; for example, determine the list of key indicators to be monitored (such as sales expense ratio, accounts receivable turnover days) based on historical query popularity and business rules, and extract the long-term series of these indicators from the standardized data pool. Step S6.2: Analyze historical time series data to identify its periodicity and trend characteristics, and detect abnormal data points that deviate from the normal fluctuation range of history or preset business rules; for example, using statistical process control methods, calculate the dynamic control upper and lower limits of indicators based on historical data. When the data point of the latest period exceeds the control limit, or shows a significant change in the same direction for multiple consecutive periods (such as a decrease in gross profit margin of more than 1 percentage point for three consecutive months), it is marked as an abnormal or trend signal. Step S6.3: When an abnormal data point is detected or a significant trend change is identified, an early warning message containing an anomaly description, possible causes, and related data is automatically generated; for example, Alert: Company Y's sales expense ratio has abnormally increased to 15% this quarter, exceeding the historical normal range (8%-12%); Possible related event: A large-scale promotional campaign was launched in XX district this month (see Marketing Department Report 123).

[0042] Step S6.4: Push the warning information to relevant business leaders or analysis users through a preset message interface. For example, by integrating with systems such as WeChat Work, DingTalk, and email, the warning information can be pushed to preset recipients such as the CFO and relevant business unit leaders in real time. Example 2

[0043] This embodiment provides a computer-readable storage medium on which a computer program is stored. The computer program is executed by a processor to implement the artificial intelligence-based intelligent question-answering and analysis method provided in Embodiment 1. Those skilled in the art will understand that all or part of the steps of the method provided in Embodiment 1 can be implemented by hardware related to the computer program. The aforementioned computer program can be stored in a computer-readable storage medium. When the program is executed, it performs the steps of the method provided in Embodiment 1. The aforementioned storage medium includes various media capable of storing program code, such as ROM, RAM, magnetic disks, or optical disks. Example 3

[0044] like Figure 2 As shown, this embodiment provides an intelligent question-answering and analysis system based on artificial intelligence, including: The data acquisition module acquires financial data, including internal corporate financial data and external market data, and performs semantic alignment and standardization processing. The financial knowledge graph module constructs a financial knowledge graph based on semantically aligned financial data to provide a unified representation of the financial data. The question parsing module receives users' natural language questions, parses the natural language questions, identifies the financial entities, indicators, time periods, and analysis dimensions, and generates structured queries. The financial analysis and calculation module, based on structured queries, retrieves relevant information from the financial knowledge graph and performs corresponding financial analysis and calculations. The analysis results generation module generates analysis results, which include data sources, calculation processes, and reasoning paths. The early warning module proactively identifies financial anomalies or trends based on historical queries and financial data characteristics, and pushes relevant information to users.

[0045] The functions of each module correspond one-to-one with the steps described above, and will not be repeated here.

[0046] The above embodiments are merely illustrative of the principles and effects of the present invention and are not intended to limit the invention. Any person skilled in the art can modify or alter the above embodiments without departing from the spirit and scope of the present invention. Therefore, all equivalent modifications or alterations made by those skilled in the art without departing from the spirit and technical concept disclosed in the present invention should still be covered by the claims of the present invention.

Claims

1. An intelligent question-answering and analysis method based on artificial intelligence, characterized in that, Includes the following steps: Step S1: Obtain financial data, which includes internal financial data and external market data, and perform semantic alignment and standardization processing. Step S2: Based on the semantically aligned financial data, construct a financial knowledge graph to provide a unified representation of the financial data; Step S3: Receive the user's natural language question, parse the natural language question, identify the financial entities, indicators, time and analysis dimensions, and generate a structured query; Step S4: Based on the structured query, retrieve relevant information from the financial knowledge graph and perform corresponding financial analysis calculations; Step S5: Generate analysis results, which include data sources, calculation processes, and reasoning paths; Step S6: Based on historical queries and financial data characteristics, proactively identify financial anomalies or trends, and push relevant information to users.

2. The intelligent question-answering and analysis method based on artificial intelligence according to claim 1, characterized in that, The specific method for step S1 is as follows: Step S1.1: Obtain the company's internal structured transaction data, unstructured financial documents, and external market and industry data; Step S1.2: Parse the unstructured financial documents to extract financial entities, numerical indicators, accounting periods, and business events; perform semantic annotation on the structured data fields; Step S1.3: Establish semantic mapping relationships to align entities and indicators with the same semantics but different expressions from different data sources; Step S1.4: Convert the data from different accounting periods and currencies to the standard reporting period and the designated functional currency.

3. The intelligent question-answering and analysis method based on artificial intelligence according to claim 2, characterized in that, In step S1.2, a natural language processing model optimized with financial domain knowledge is used for document parsing and information extraction.

4. The intelligent question-answering and analysis method based on artificial intelligence according to claim 3, characterized in that, In step S1.4, the method for converting to the standard reporting period is as follows: identify the original accounting period of the data, and aggregate the non-standard time point data to the most recent standard reporting time point by means of forward filling or moving average.

5. The intelligent question-answering and analysis method based on artificial intelligence according to claim 4, characterized in that, The specific methods for constructing the financial knowledge graph in step S2 include: Step S2.1: Define the financial entities, numerical indicators, accounting periods, and business events processed in step S1 as graph nodes; wherein, financial entities include companies, departments, products, customers, and suppliers; numerical indicators include revenue, profit, cost, and various financial ratios; Step S2.2: Based on financial rules, accounting standards, and business logic, predefine the relationship types between nodes, including membership relationships, computational relationships, comparative relationships, and time series relationships; establish relationship edges of corresponding types for node instances according to the actual associations in the data; Step S2.3: Populate the triples of the knowledge graph using relations directly mapped from structured data and relation assertions parsed from unstructured text.

6. The intelligent question-answering and analysis method based on artificial intelligence according to claim 5, characterized in that, The specific method for step S3 is as follows: Step S3.1: Determine the type of analytical intent for the natural language question, including fact query, indicator calculation, time comparison, trend analysis, and anomaly attribution; Step S3.2: Identify financial entities, numerical indicators, time phrases, and keywords for analysis dimensions from the natural language question; Step S3.3: Associate and match the keywords with the corresponding nodes in the financial knowledge graph constructed in step S2; Step S3.4: Structured query generation. Based on the analysis intent type and the successfully linked graph nodes, a structured query statement containing the target entity, indicators, time range, and analysis operation is generated to drive the retrieval and analysis in step S4.

7. The intelligent question-answering and analysis method based on artificial intelligence according to claim 6, characterized in that, In step S4, performing financial analysis calculations means: according to the analysis intent type of the structured query, calling preset financial analysis rules and calculation functions to perform calculations on the relevant data retrieved from the financial knowledge graph.

8. The intelligent question-answering and analysis method based on artificial intelligence according to claim 7, characterized in that, The specific implementation steps of step S6 are as follows: Step S6.1: Based on historical query frequency and business importance, select a set of key financial indicators; extract the historical time series data of the key financial indicators from the data after standardization processing in step S1; Step S6.2: Analyze the historical time series data to identify its periodicity and trend characteristics, and detect abnormal data points that deviate from the normal historical fluctuation range or preset business rules; Step S6.3: When the abnormal data point is detected or a significant trend change is identified, an early warning message containing an anomaly description, possible causes, and related data is automatically generated; Step S6.4: Push the warning information to the relevant business manager or analysis user through a preset message interface.

9. An intelligent question-answering and analysis system based on artificial intelligence, characterized in that, include: The data acquisition module acquires financial data, including internal corporate financial data and external market data, and performs semantic alignment and standardization processing. The financial knowledge graph module constructs a financial knowledge graph based on semantically aligned financial data to provide a unified representation of the financial data. The question parsing module receives users' natural language questions, parses the natural language questions, identifies the financial entities, indicators, time periods, and analysis dimensions, and generates structured queries. The financial analysis and calculation module retrieves relevant information from the financial knowledge graph based on the structured query and performs corresponding financial analysis and calculations. The analysis result generation module generates analysis results, which include data sources, calculation processes, and reasoning paths. The early warning module proactively identifies financial anomalies or trends based on historical queries and financial data characteristics, and pushes relevant information to users.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, The computer program is executed by a processor to implement the artificial intelligence-based intelligent question answering and analysis method as described in any one of claims 1 to 8.

Images