Analysis report generation method and device, electronic equipment and storage medium

Abstract

The invention discloses an analysis report generation method and apparatus, an electronic device and a storage medium, and realizes that an analysis system can adaptively adjust processing logic in a complex scene in which heterogeneous conflicts or quality defects exist in original data by establishing a closed-loop feedback mechanism of a service scheme and data quality, thereby improving the analysis efficiency. Therefore, the continuity of the automatic analysis process and the robustness of the output report conclusion are ensured.

Description

Technical Field

[0001] This invention relates to the field of analytical report generation technology, and in particular to an analytical report generation method, an analytical report generation device, an electronic device, and a readable storage medium. Background Technology

[0002] As financial markets become increasingly complex, financial analysis (such as credit risk assessment and industry research) relies more and more on multi-source heterogeneous data. However, two core technological challenges remain in achieving automated data analysis: First, the semantic heterogeneity of multi-source data makes it difficult to establish a seamless automated processing chain. Financial data is distributed across internal enterprise databases, third-party financial terminals, and various external interfaces. Different data sources often have significant differences in the definition, calculation methods, numerical units, and even currency for the same indicator (such as "credit rating" or "debt ratio"). Furthermore, many business labels exist in the form of enumeration codes. Existing automated analysis tools often struggle to accurately align these semantic ambiguities, resulting in the system being unable to perform direct logical operations after acquiring multi-source data due to conflicting definitions.

[0003] Secondly, defects in the quality of raw data can easily lead to interruptions in automated processes or distorted conclusions. In real-world business scenarios, due to missing disclosures, data entry delays, or system errors, the collected raw indicator data often contains varying degrees of missing fields or logically abnormal values. The automated analysis processes of related technologies are mostly linearly executed, lacking real-time awareness and adaptive adjustment capabilities regarding data quality. This results in the final analytical conclusions often deviating significantly from the actual business situation, posing a significant risk of misleading conclusions when generating financial analysis. Summary of the Invention

[0004] The present invention provides an analysis report generation method, apparatus, electronic device, and readable storage medium to overcome or at least partially solve the above-mentioned problems.

[0005] To solve the above-mentioned technical problems, this application is implemented as follows: In a first aspect, embodiments of this application provide a method for generating an analysis report, including: In response to receiving a natural language task sent by a user, perform semantic parsing on the natural language task to determine the analysis dimensions, business scope, and expected goals of the natural language task; The analytical dimensions, business criteria, and expected objectives are encapsulated into a structured task solution; By using the mapping relationship of the structured task scheme, the original structured indicator data used to perform the natural language task is obtained; Eliminate the heterogeneity of the data source of the original structured indicator data and output the dataset to be evaluated without semantic ambiguity; Obtain data quality features that characterize the health status of the dataset to be evaluated, and revise the structured task scheme based on the data quality features to generate the final data processing scheme; A complete analysis report is generated using the final data processing scheme.

[0006] Optionally, the step of obtaining the original structured indicator data for performing the natural language task through the mapping relationship of the structured task scheme includes: The structured task scheme is read, and by retrieving a preset semantic dictionary or metadata mapping table, data tags and field requirements for representing the analysis dimension, the business scope, and the expected goal are generated. Based on the data tags and field requirements, access the enterprise's internal database and / or metadata center corresponding to the data tags to obtain the original structured indicator data.

[0007] Optionally, the step of eliminating the heterogeneity of the data source of the original structured indicator data and outputting a semantically ambiguous dataset to be evaluated includes: Obtain the metrics used to characterize the business attributes and financial performance of the target analysis object from the original structured index data; The metric values ​​are standardized and aligned using semantic rules to output a dataset to be evaluated with semantic ambiguity eliminated.

[0008] Optionally, the step of obtaining data quality features for characterizing the health status of the dataset to be evaluated includes: Perform integrity checks on the dataset to be evaluated to identify missing fields; The data to be evaluated is subjected to quality inspection to identify abnormal values.

[0009] Optionally, the step of modifying the structured task scheme based on the data quality characteristics to generate the final data processing scheme includes: Based on the missing proportion of the missing fields, the analysis dimensions and expected goals of the structured task solution are prioritized and adjusted, or the structured task solution is logically modified based on the severity of the anomalies expressed by the abnormal values, and a verified final data processing solution is output.

[0010] Optionally, before the step of generating a complete analysis report through the final data processing scheme, the method further includes: Identify the modified dataset that is associated with the final data processing scheme, and the task adjustment record for the modified dataset; Feature analysis code is automatically generated and run in an isolated sandbox environment to identify the spatial distribution of anomaly markers in the corrected dataset and generate an analysis report reflecting the health status of the dataset to be evaluated.

[0011] Optionally, the step of generating a complete analysis report through the final data processing scheme includes: Using the analysis report as a basis for decision-making, numerical imputation is performed on fields with a missing rate below a preset threshold using a predetermined industry median or regression estimation method. Abnormal values ​​with logical conflicts are removed or smoothed, and high-quality analytical sample data that is cleaned and statistically significant is output. Generate analysis code and perform statistical calculations based on high-quality analysis sample data to complete the quantitative analysis of financial analysis information for the target analysis object and generate structured calculation results that reflect the real business status of the target analysis object. Based on the structured calculation results and the task adjustment records, a complete analysis report is output, including explanations of data limitations, derivation of business conclusions, and multi-dimensional statistical charts.

[0012] Secondly, embodiments of this application provide an analysis report generation apparatus, characterized in that it includes: The semantic parsing module is used to respond to receiving a natural language task sent by a user, perform semantic parsing on the natural language task, and determine the analysis dimension, business scope and expected goal of the natural language task; A structured task solution encapsulation module is used to encapsulate the analysis dimensions, the business scope, and the expected goals into a structured task solution; The raw structured indicator data acquisition module is used to acquire raw structured indicator data for performing the natural language task through the mapping relationship of the structured task scheme. The semantic ambiguity elimination module is used to eliminate the heterogeneity of the data source of the original structured index data and output the dataset to be evaluated with semantic ambiguity eliminated. The final data processing scheme generation module is used to obtain data quality characteristics that characterize the health status of the dataset to be evaluated, and to modify the structured task scheme based on the data quality characteristics to generate the final data processing scheme. The complete analysis report generation module is used to generate a complete analysis report through the final data processing scheme.

[0013] Thirdly, embodiments of this application provide an electronic device including a processor, a memory, and a program or instructions stored in the memory and executable on the processor, wherein the program or instructions, when executed by the processor, implement the steps of the method described in the first aspect.

[0014] Fourthly, embodiments of this application provide a readable storage medium on which a program or instructions are stored, which, when executed by a processor, implement the steps of the method described in the first aspect.

[0015] Fifthly, embodiments of this application provide a chip, the chip including a processor and a communication interface, the communication interface being coupled to the processor, the processor being used to run programs or instructions to implement the method as described in the first aspect.

[0016] The embodiments of the present invention have the following advantages: In this embodiment of the invention, in response to receiving a natural language task sent by a user, semantic parsing is performed on the natural language task to determine its analysis dimensions, business scope, and expected goals. These dimensions, scope, and goals are then encapsulated into a structured task scheme. Through the mapping relationship of the structured task scheme, raw structured indicator data for executing the natural language task is obtained. The heterogeneity of the data source of the raw structured indicator data is eliminated, and a dataset to be evaluated with semantic ambiguity resolved is output. Data quality characteristics characterizing the health status of the dataset to be evaluated are obtained, and the structured task scheme is modified based on these characteristics to generate a final data processing scheme. A complete analysis report is generated through the final data processing scheme, thereby establishing a closed-loop feedback mechanism between the business scheme and data quality. This enables the analysis system to adaptively adjust its processing logic even in complex scenarios where the raw data exhibits heterogeneous conflicts or quality defects, ensuring the continuity of the automated analysis process and the robustness of the output report conclusions. Attached Figure Description

[0017] Figure 1 This is a flowchart of the steps of an analysis report generation method provided in an embodiment of the present invention; Figure 2 This is a flowchart illustrating an analysis report generation method provided in an embodiment of the present invention; Figure 3 This is a flowchart illustrating a data acquisition method provided in an embodiment of the present invention; Figure 4 This is a flowchart illustrating a data evaluation method provided in an embodiment of the present invention; Figure 5 This is a flowchart illustrating a data processing method provided in an embodiment of the present invention; Figure 6 This is a flowchart illustrating a data analysis method provided in an embodiment of the present invention; Figure 7 This is a structural block diagram of an analysis report generation device provided in an embodiment of the present invention; Figure 8 This is a hardware structure block diagram of an electronic device provided in an embodiment of the present invention; Figure 9 This is a schematic diagram of a computer-readable medium provided in an embodiment of the present invention. Detailed Implementation

[0018] 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 some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0019] The terms "first," "second," etc., used in the specification and claims of this application are used to distinguish similar objects and not to describe a specific order or sequence. It should be understood that such use of data can be interchanged where appropriate so that embodiments of this application can be implemented in orders other than those illustrated or described herein. Furthermore, in the specification and claims, "and / or" indicates at least one of the connected objects, and the character " / " generally indicates that the preceding and following objects are in an "or" relationship.

[0020] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0021] Reference Figure 1 The diagram illustrates a flowchart of an analysis report generation method provided in an embodiment of the present invention, which may specifically include the following steps: Step 101: In response to receiving a natural language task sent by the user, perform semantic parsing on the natural language task to determine the analysis dimension, business scope and expected goal of the natural language task; Step 102: Encapsulate the analysis dimensions, the business scope, and the expected goals into a structured task scheme; Step 103: Obtain the original structured indicator data for performing the natural language task through the mapping relationship of the structured task scheme; Step 104: Eliminate the heterogeneity of the data source of the original structured index data and output the dataset to be evaluated with semantic ambiguity eliminated; Step 105: Obtain data quality features to characterize the health status of the dataset to be evaluated, and revise the structured task scheme based on the data quality features to generate the final data processing scheme; Step 106: Generate a complete analysis report using the final data processing scheme.

[0022] In a specific implementation, the embodiments of the present invention can be applied to an analysis report generation system (hereinafter referred to as the system). This system can be applied in large financial institutions to assist analysts in performing in-depth investment research analysis, credit risk assessment, and automated analysis report generation based on massive structured data (such as financial indicators, market conditions, valuations, business scores, and product performance). The system can automatically complete data capture, quality verification, and logic correction through multi-agent collaboration, thereby transforming complex raw data into structured calculation results and chart-based reports with business insights.

[0023] In this embodiment of the invention, in response to receiving a natural language task sent by a user, semantic parsing is performed on the natural language task to determine the analysis dimensions, business scope, and expected goals of the natural language task, so as to transform the user's unstructured language intent (intention) into standardized elements (logic) that can be understood and computed by a computer, and to establish the boundaries and blueprint of the analysis work.

[0024] Natural language tasks can be financial business requests made by users in spoken or written form (such as "analyze the debt repayment pressure of a certain group").

[0025] Semantic parsing can be the process of extracting keywords and identifying their logical relationships using natural language processing techniques.

[0026] The analytical dimensions can be the entry points for observing data (such as corporate credit scores, debt-to-equity ratios, interest coverage ratios, and cash flow volatility).

[0027] Business scope can be the specific criteria for calculating indicators (e.g., whether interest expense includes capitalized interest).

[0028] The expected goal can be the result pursued by the task (such as: indicator numerical results, warning level, score, benchmarking report).

[0029] In this embodiment of the invention, the analysis dimensions, the business scope, and the expected goals can be encapsulated into a structured task scheme to integrate the scattered analytical elements into a unified, parameterized "execution instruction set," enabling it to be transmitted and modified as an independent object within the system.

[0030] A structured task scheme can be a machine instruction document that includes all parameters, logical paths, and preset weights required for task execution.

[0031] In this embodiment of the invention, the original structured indicator data for performing the natural language task is obtained through the mapping relationship of the structured task scheme. By using a preset bridge (mapping table), the physical location of the business indicators in the underlying database is accurately located, thereby realizing the extraction from "business concept" to "actual data".

[0032] For example, the mapping relationship can be, but is not limited to, a correspondence between business terms (such as “total liabilities”) and database storage names (such as “TOTAL_LIABILITIES”).

[0033] The raw structured index data can be raw values ​​extracted directly from the database that have not yet been aligned to units or cleaned of anomalies.

[0034] In this embodiment of the invention, the heterogeneity of the original structured indicator data source can be eliminated, and the dataset to be evaluated with semantic ambiguity eliminated can be output. This solves the problem that different multi-source data express independent meanings, and unifies data from different sources under the same units, currencies and definitions, making them comparable.

[0035] Data source heterogeneity can refer to conflicts in format, unit, and definition of data from different vendors.

[0036] Eliminating semantic ambiguity can be achieved by standardizing operations to ensure that the same numerical value has a unique and definite meaning in business understanding.

[0037] The dataset to be evaluated can be a collection of data that has undergone preliminary cleaning and alignment and is ready to undergo quality review.

[0038] In this embodiment of the invention, data quality characteristics used to characterize the health status of the dataset to be evaluated can be obtained, and the structured task scheme can be modified based on the data quality characteristics to generate the final data processing scheme. In order to establish a feedback mechanism, the analysis logic can be dynamically adjusted according to the current status of the data (whether there is data and whether the data is accurate) to prevent the program from crashing or the conclusion from being abnormal.

[0039] Health status can refer to the performance of a dataset in terms of completeness, accuracy, and logical consistency.

[0040] Data quality characteristics can be quantitative indicators that describe the health status (such as missing percentage, outlier rate, and logical conflict markers).

[0041] The final data processing solution can be the final execution version after "downgrading" or "logically switching" the initial solution based on data quality.

[0042] According to the embodiments of the present invention, a complete analysis report can be generated through the final data processing scheme, so as to perform the final calculation and integrate the context information under the guidance of the revised logic, and output a final document with decision-making basis.

[0043] A complete analysis report can be a comprehensive output that includes descriptions of raw data, traceability of the calculation process, business conclusions, and visualization results.

[0044] In this embodiment of the invention, in response to receiving a natural language task sent by a user, semantic parsing is performed on the natural language task to determine its analysis dimensions, business scope, and expected goals. These dimensions, scope, and goals are then encapsulated into a structured task scheme. Through the mapping relationship of the structured task scheme, raw structured indicator data for executing the natural language task is obtained. The heterogeneity of the data source of the raw structured indicator data is eliminated, and a dataset to be evaluated with semantic ambiguity resolved is output. Data quality characteristics characterizing the health status of the dataset to be evaluated are obtained, and the structured task scheme is modified based on these characteristics to generate a final data processing scheme. A complete analysis report is generated through the final data processing scheme, thereby establishing a closed-loop feedback mechanism between the business scheme and data quality. This enables the analysis system to adaptively adjust its processing logic even in complex scenarios where the raw data exhibits heterogeneous conflicts or quality defects, ensuring the continuity of the automated analysis process and the robustness of the output report conclusions.

[0045] Based on the above embodiments, modified embodiments of the above embodiments are proposed. It should be noted that, in order to keep the description brief, only the differences from the above embodiments are described in the modified embodiments.

[0046] refer to Figure 2 , Figure 2 This is a flowchart illustrating an analysis report generation method provided in an embodiment of the present invention; In an optional embodiment of the present invention, in response to receiving a natural language task sent by a user, a data analysis agent is invoked to perform semantic parsing on the natural language task to determine the analysis dimension, business scope and expected goal of the natural language task. The analytical dimensions, business criteria, and expected objectives are encapsulated into a structured task solution; The data acquisition agent reads the structured task plan and generates data labels and field requirements to characterize the analysis dimension, the business scope, and the expected goal by retrieving a preset semantic dictionary or metadata mapping table. The data acquisition agent, based on the data tags and the field requirements, accesses the enterprise's internal database and / or metadata center corresponding to the data tags to obtain raw structured indicator data. The data evaluation agent retrieves metrics from the raw structured indicator data to characterize the business attributes and financial performance of the target analysis object. The data evaluation agent uses semantic rules to perform standardization transformation and caliber alignment operations on the metric values, and outputs the evaluation dataset with semantic ambiguity eliminated. Perform integrity checks on the dataset to be evaluated to identify missing fields; The data to be evaluated is subjected to quality inspection to identify abnormal values.

[0047] Based on the missing proportion of the missing fields, the priority of the analysis dimensions and expected goals of the structured task solution is adjusted, or, based on the severity of the anomalies expressed by the abnormal values, the structured task solution is logically modified, and a verified final data processing solution is output. Identify the modified dataset that is associated with the final data processing scheme, and the task adjustment record for the modified dataset; The data processing agent receives the final data processing scheme and automatically generates and runs feature analysis code in an isolated sandbox environment to identify the spatial distribution of anomaly markers in the corrected dataset and generate an analysis report reflecting the health status of the dataset to be evaluated. The data processing agent is invoked to use the analysis report as a basis for decision-making. For fields in the decision-making basis with a missing rate lower than a preset threshold, numerical imputation is performed using a predetermined industry median or regression estimation method. Abnormal values ​​with logical conflicts are removed or smoothed, and high-quality analysis sample data that is cleaned and meets statistical significance is output. The data processing agent is invoked to generate analysis code and perform statistical calculations on high-quality analysis sample data, complete the quantitative analysis of financial analysis information for the target analysis object, and generate structured calculation results that reflect the real business status of the target analysis object. The natural language generation engine is invoked to output a complete analysis report based on the structured computation results and the task adjustment records. This report includes explanations of data limitations, derivations of business conclusions, and multi-dimensional statistical charts.

[0048] In its specific implementation, this invention deconstructs the complex analysis process into five specialized functional modules: intent understanding, resource scheduling, quality feedback, code execution, and report synthesis. This decouples business logic from the computing environment, ensuring that the system can automatically correct the execution algorithm based on the real-time captured data quality characteristics. As a result, while improving the professional depth of the analysis report and the robustness of the conclusions, this invention completely solves the technical problems of task interruption caused by data defects and the lack of traceability of black-box computation that are common in the financial analysis process.

[0049] Specifically, in the task parsing stage executed by the data analysis agent, a model with deep language understanding capabilities can be used to accurately deconstruct the vague "business language" of humans into a "logical framework" that can be executed by computers, thus solving the problem of inaccurate understanding of intent leading to deviations in subsequent steps.

[0050] Data analytics agents are specialized models responsible for intent recognition and logical modeling, and are adept at handling semantic parsing and solution encapsulation.

[0051] In the data acquisition phase performed by the data acquisition agent, the model can be freed from heavy computational tasks, and a model specifically responsible for resource retrieval can accurately translate and capture "business indicators" into "physical fields", solving the problem of data sources not being found or not being found correctly.

[0052] The data acquisition agent is a specialized model that is specifically responsible for metadata mapping, database SQL generation, and cross-source data retrieval.

[0053] In the quality assessment phase performed by the data evaluation agent, an independent model with statistical auditing capabilities can be introduced to conduct a "third-party check" on the raw materials retrieved. The initial plan can be forcibly corrected through data feature feedback, thus solving the "garbage in, garbage out (GIGO)" problem caused by flawed data.

[0054] Data evaluation agents are specialized models focused on integrity verification, outlier detection, and data health status quantification.

[0055] In the code execution and computation phase using data processing agents, high-intensity quantitative computation and data cleaning can be performed in a secure, isolated environment by a model specifically responsible for code generation and environment operation, thus solving the problems of environmental safety and computational accuracy in the automated computation process.

[0056] A data processing agent is a model that has the ability to write code (such as Python code generation) and invoke mathematical calculation engines.

[0057] During the report generation stage of the natural language generation engine, models skilled in text polishing and logical connection can transform dry, structured calculation results into professional analysis reports that are readable by humans, thus solving the problems of high threshold for interpreting conclusions and inconsistent logic.

[0058] A natural language generation engine is a model component responsible for transforming quantitative data and task adjustment records into text that conforms to financial writing standards.

[0059] Optionally, the step of obtaining the original structured indicator data for performing the natural language task through the mapping relationship of the structured task scheme includes: The structured task scheme is read, and by retrieving a preset semantic dictionary or metadata mapping table, data tags and field requirements for representing the analysis dimension, the business scope, and the expected goal are generated. Based on the data tags and field requirements, access the enterprise's internal database and / or metadata center corresponding to the data tags to obtain the original structured indicator data.

[0060] In this embodiment of the invention, the structured task scheme can be read, and by retrieving a preset semantic dictionary or metadata mapping table, data tags and field requirements for representing the analysis dimensions, business scope, and expected goals can be generated. This translates the high-level "business logic requirements" into the low-level "data retrieval operation guidelines." By establishing a semantic bridge, it clarifies where the system needs to go and which specific data items to retrieve, thereby solving the problem of the disconnect between business requirements and the physical storage structure of the database.

[0061] A semantic dictionary can be a knowledge base that records the correspondence between financial business terms and their synonyms and near-synonyms.

[0062] A metadata mapping table can be a mapping matrix that records the mapping relationship between business metrics (logical names) and database table names and field names (physical names).

[0063] Data tags can be business attribute identifiers assigned to raw data, used for rapid classification and indexing of massive amounts of data.

[0064] Field requirements can be specific data retrieval instructions down to the physical database level, including field names, data types, and the data tables they belong to.

[0065] In this embodiment of the invention, based on the data tags and the field requirements, the enterprise's internal database and / or metadata center corresponding to the data tags can be accessed to obtain raw structured indicator data. According to the precise instructions generated in the previous step, cross-system and cross-database data crawling actions can be performed to gather raw materials scattered in different physical locations into the analysis environment.

[0066] An enterprise's internal database can be a physical storage system that stores core data such as the enterprise's actual business transactions and financial records.

[0067] The metadata center is a standardized index library used within an enterprise to store the mapping relationship between business metrics and underlying physical fields. By providing data definitions, calculation rules, and data source location information, it provides logical support for intelligent agents to achieve automated translation from "business semantics" to "physical data retrieval".

[0068] The difference between a metadata center and an internal enterprise database lies in the fact that a metadata center is a system that stores "data descriptions." It is a dictionary and map of the data, storing the calculation methods, physical table locations, and field definitions corresponding to business indicators (such as "credit scores"). Essentially, it is a logical navigation layer that guides the intelligent agent "where to retrieve data and how to retrieve it." An internal enterprise database, on the other hand, is a system that stores "the data itself." It is the actual data warehouse, storing real values ​​such as specific financial flows, transaction records, and customer details. It is the physical layer where the intelligent agent ultimately executes the data retrieval action.

[0069] The original structured indicator data can be numerical and text data extracted directly from the underlying library, retaining the original storage format.

[0070] For example, in a scenario assessing the solvency of a life insurance company, the system first reads a structured task plan containing "Analysis Dimension: Q4 2023", "Business Scope: Core Solvency Adequacy Ratio", and "Expected Target: Default Risk Assessment". Then, by retrieving the metadata mapping table, the system translates the business term "Core Solvency Adequacy Ratio" into a physical field name named CORE_SOLVENCY_RATIO in the underlying database, and associates it with the corresponding "Data Label: Solvency Indicator" and "Field Requirements: Numeric, Two Decimal Places". Based on these precise labels and field instructions, the system automatically accesses the company's internal financial details table, accurately extracting the raw value corresponding to Q4 2023 (e.g., "215.42%)", thus completing the automated acquisition process from business intent to physical raw data.

[0071] This embodiment introduces a semantic dictionary and a metadata mapping table as the transformation hub, realizing automated and accurate mapping from abstract business solutions to specific physical fields. Its beneficial effects are that it not only significantly improves the accuracy and efficiency of data acquisition in complex and heterogeneous data environments, but also eliminates semantic ambiguity in the data retrieval process through data tagging management, laying a reliable data foundation for subsequent standardized processing.

[0072] Optionally, the step of eliminating the heterogeneity of the data source of the original structured indicator data and outputting a semantically ambiguous dataset to be evaluated includes: Obtain the metrics used to characterize the business attributes and financial performance of the target analysis object from the original structured index data; The metric values ​​are standardized and aligned using semantic rules to output a dataset to be evaluated with semantic ambiguity eliminated.

[0073] In this embodiment of the invention, measures that characterize the business attributes and financial performance of the target analysis object can be obtained from the original structured indicator data. This allows for the precise extraction of core financial and business indicators from the messy original data stream, the removal of irrelevant metadata or redundant system information, and the identification of the "original numerical material" required for subsequent analysis.

[0074] The target of analysis can refer to the entity to which this analysis task is directed, such as a specific company, industry, bond, or asset-backed security.

[0075] Business attributes can refer to non-financial information that characterizes the basic features of an object, such as industry classification, enterprise nature, and geographical attributes.

[0076] Financial performance can refer to quantitative information that characterizes the operating status of an entity, such as total assets, debt-to-equity ratio, and net profit growth rate.

[0077] A metric can be a numerical or quantitative identifier in the raw data that can be used for statistics, calculations, or comparisons (e.g., the number 100,000,000 or the rating code AAA).

[0078] In this embodiment of the invention, semantic rules can be used to perform standardization transformation and caliber alignment operations on the metric values, outputting a dataset to be evaluated with semantic ambiguity eliminated. This eliminates the "understanding gap" caused by different definitions, different units, or different currencies in multi-source data, unifies the data to the same benchmark level, and ensures that data from different sources are comparable.

[0079] Semantic rules can be a set of pre-defined logical instructions that specify how to process specific words or units (e.g., if the unit is "ten thousand yuan", multiply by 10,000; if the currency is "dollars", convert according to the exchange rate).

[0080] Standardization conversion can be the process of converting non-standard formats (such as different ways of expressing dates or different units of measurement) into a unified and standardized format.

[0081] Aligning metrics can be a process of ensuring consistency in data calculation logic. For example, adjusting the "profit" indicator in different financial reports, which may or may not include government subsidies, according to a unified business logic.

[0082] The dataset to be evaluated, after semantic ambiguity has been eliminated, can be a set of data that has undergone unified transformation so that all fields have unique and clear meanings and is ready to enter the quality inspection stage.

[0083] This embodiment introduces semantic rules to enforce standardization and caliber alignment of multi-source metrics, thereby achieving the underlying automatic elimination of the heterogeneity of financial data. Its beneficial effects are that it not only solves the logical calculation errors caused by inconsistent dimensions and semantic ambiguity, but also significantly improves the accuracy and objectivity of the analysis system's conclusions when processing cross-source and cross-temporal data.

[0084] Optionally, the step of obtaining data quality features for characterizing the health status of the dataset to be evaluated includes: Perform integrity checks on the dataset to be evaluated to identify missing fields; The data to be evaluated is subjected to quality inspection to identify abnormal values.

[0085] This invention allows for integrity checks on the dataset to be evaluated to identify missing fields. By scanning the dataset, it identifies metrics that should exist under preset business criteria but are actually empty, thereby assessing the current data's support for the analysis task and preventing computational crashes due to missing data.

[0086] Integrity checks can be an auditing process for the existence of data. By comparing the preset field requirements with the actual fields retrieved, it can identify whether there are empty values ​​or unentered items.

[0087] Missing fields can refer to data items in the dimensions required by the analysis plan that have not been able to obtain valid values ​​due to missing disclosures, data entry delays, or data collection failures (e.g., a company's "interest expense" field is empty).

[0088] In this embodiment of the invention, the data to be evaluated can be subjected to quality inspection to identify abnormal values. Through statistical feature analysis or logical correlation verification, erroneous items in the dataset that have numerical values ​​but are logically unreasonable can be found, and interference factors can be eliminated to provide a basis for decision-making for subsequent logical correction.

[0089] Quality inspection can be a deep scan of the accuracy, rationality and consistency of data values, including range verification, logical consistency and trend anomaly identification.

[0090] Abnormal values ​​can refer to values ​​that deviate significantly from the reasonable range, violate the accounting equation, or exhibit extremely outlier characteristics compared to similar data (e.g., a debt-to-asset ratio exceeding 500%, or negative operating revenue recorded under non-zero reporting).

[0091] This embodiment achieves digital modeling of the "health status" of financial data by scanning the standardized data in multiple dimensions for completeness and logical rationality. Its beneficial effects are that it not only provides accurate quantitative basis for subsequent adaptive logical correction, but also avoids the risk of process interruption and distortion of analysis conclusions caused by flawed data by identifying data defects in advance.

[0092] Optionally, the step of modifying the structured task scheme based on the data quality characteristics to generate the final data processing scheme includes: Based on the missing proportion of the missing fields, the analysis dimensions and expected goals of the structured task solution are prioritized and adjusted, or the structured task solution is logically modified based on the severity of the anomalies expressed by the abnormal values, and a verified final data processing solution is output.

[0093] In this embodiment of the invention, the analysis dimensions and expected goals of the structured task scheme can be prioritized based on the proportion of missing fields, thereby achieving adaptive reconstruction of the analysis logic. When the system detects missing data in the original data, it no longer rigidly executes the preset path, but instead dynamically adjusts the contribution weight of each analysis dimension by quantifying the proportion of missing fields, and rearranges the logical priorities for achieving the expected goals (such as switching from "cash flow orientation" to "asset quality orientation"), thereby ensuring that even with incomplete data, the analysis conclusions closest to the truth can still be derived.

[0094] Missing percentage can refer to the ratio of the number of actual missing data items to the total number of required data items for a specific analytical dimension.

[0095] In practical applications, the execution order of analysis dimensions refers to the logical sequence in which each analysis module (dimension) is activated and processed during the process of achieving the expected goal. Its function is to prioritize processing the dimension with the most complete data and the most reliable logic when data is incomplete, and then decide whether to trigger subsequent lower-priority dimensions based on the calculation results of the preceding dimensions.

[0096] The weighting of indicators refers to the percentage contribution (as a percentage) of each specific indicator to the score or conclusion of the same analytical dimension. Its function is to quantify the influence of different data points. In the financial field, this represents the "importance coefficient" of the indicator.

[0097] Priority adjustment can refer to the execution order of analytical dimensions in the analytical model or the weight allocation method of the calculated indicators of the expected goals. Specifically, it refers to the real-time modification of the importance ranking and achievement path of analytical dimensions and goals. For example, if too much data is missing in the "financial dimension," the system will automatically reduce the evaluation weight of this dimension and increase the priority of the "industry status" or "external guarantee" dimensions to maintain the output quality of the expected goals.

[0098] In this embodiment of the invention, the structured task scheme can be logically modified based on the severity of the anomaly expressed by the abnormal values ​​to establish a qualitative logic switching mechanism. When it is found that the indicator has numerical values ​​but the logic is extremely flawed (which may lead to completely different conclusions), the system no longer simply adjusts the weights, but directly changes the calculation model or analysis path, fundamentally preventing the conclusion from being distorted.

[0099] The severity of anomalies can be quantified and classified according to the multiple by which the value deviates from the industry benchmark, the degree of violation of accounting reconciliation, or the distance from outliers, and the level of data error (such as minor outliers, logical conflicts, extreme anomalies).

[0100] Logical correction can refer to replacing the original analysis algorithm or calculation formula. For example, when the "balance sheet" data is severely abnormal, the system abandons the original "asset look-through model" and corrects it to an alternative analysis model based on "bank transaction characteristics".

[0101] In this embodiment of the invention, a verified final data processing solution can be output to solidify the adaptively adjusted execution logic, forming a "customized execution instruction manual" that conforms to the current data status and has business rationality, serving as the sole criterion for subsequent automated calculations.

[0102] The final data processing solution can be a deterministic instruction set that includes the latest calculation formula, indicator weights, and processing paths, after missing weight compensation and abnormal logic correction.

[0103] For example, when conducting credit risk analysis on a local government financing vehicle (LGFV), the system discovered that the company's "non-operating interbank receivables" indicator was missing in as much as 70% of cases. It determined that this field could no longer support the original "asset penetration analysis logic," and thus automatically triggered a priority adjustment, reducing the weight of the missing field and increasing the weight of complete data such as the "interest coverage ratio." Simultaneously, the system detected that the company's entered "asset-liability ratio" was -150%, an extremely serious abnormal value and logical conflict. Therefore, the system performed a logic correction, automatically switching the analysis path from the original "asset quality assessment model" to an alternative assessment logic based on "operating cash flow stability," and recorded this task adjustment. Finally, it output a validated final data processing solution that avoided interference from flawed data.

[0104] This embodiment achieves a technological leap from "linear rigidity" to "dynamic self-healing" in the analysis scheme by establishing an adaptive correction closed loop based on data quality characteristics. Its beneficial effects are that it can not only autonomously ensure the continuity of the analysis process according to the data missing situation, but also switch the calculation path in real time for logically abnormal data. It eliminates the financial assessment bias caused by flawed data from the underlying architecture, and ensures the robustness and reliability of the final analysis conclusion.

[0105] Optionally, before the step of generating a complete analysis report through the final data processing scheme, the method further includes: Identify the modified dataset that is associated with the final data processing scheme, and the task adjustment record for the modified dataset; Feature analysis code is automatically generated and run in an isolated sandbox environment to identify the spatial distribution of anomaly markers in the corrected dataset and generate an analysis report reflecting the health status of the dataset to be evaluated.

[0106] This invention can identify corrected datasets related to the final data processing scheme, as well as task adjustment records for the corrected datasets, to establish a strong audit trail linking data status and logical changes. By recording the exact state of data after cleaning, filling, or removing data, and the corresponding weight adjustments or model switching actions performed by the system, a deterministic traceability basis is provided for the generation of subsequent analysis reports. This ensures that each final conclusion has a complete logical support path, thereby automatically deriving accurate data limitation explanations based on these structured records when generating the final report, improving the transparency of the analysis process and the reliability of the conclusions.

[0107] A revised dataset can refer to a dataset that, based on the results of previous quality checks, has undergone anomaly labeling, removal, or preliminary processing, and is ready to enter the formal analysis and calculation stage.

[0108] Task adjustment records can be a detailed audit log that records the specific changes made by the system to the initial plan (e.g., the weight was adjusted from 0.3 to 0.1 due to the absence of a certain indicator), reflecting the transparency of the analysis process.

[0109] This invention enables the automatic generation and execution of feature analysis code in an isolated sandbox environment, ensuring the security and independence of the computing environment. Since the feature analysis code is dynamically generated by the system based on current data characteristics, the sandbox mechanism prevents unknown scripts from posing security risks or interfering with the main system during execution.

[0110] An isolated sandbox environment can be a virtual, restricted computing space where code running will not affect the security of the external operating system or database.

[0111] Profiling code is a program script that is automatically written by the system to deeply understand the internal patterns of data (such as statistical data distribution, detecting numerical correlations, etc.).

[0112] This invention can identify the spatial distribution of outlier markers in the corrected dataset and generate an analysis report reflecting the health status of the dataset to be evaluated, thus creating a "deep profile" of data quality from both macro and micro dimensions. By identifying the clustering of outliers across different dimensions (spatial distribution), it determines whether the data problem is systemic or accidental, and transforms these conclusions into a human-readable health diagnosis.

[0113] Anomaly markers can be “suspicious” or “error” labels affixed during preceding quality inspection steps.

[0114] Spatial distribution identification can be used to analyze the distribution patterns of outliers across different subjects, industries, time periods, or indicators (e.g., discovering logical inconsistencies in the financial data of a specific industry).

[0115] An analysis report can be an in-depth technical document about the "health check results" of a dataset, describing the completeness, reliability, and potential risks of the data.

[0116] For example, when performing risk penetration analysis on the investment assets of an insurance company, the system first determines the corrected dataset after alignment with the "core asset classification" standard, and simultaneously generates a task adjustment record containing "the weight of the indicator is reduced to 0.1 due to the 40% missing rate of 'net real estate value'". Subsequently, the system automatically writes and runs a set of statistical distribution verification scripts in an independent and constrained computing space to identify the cross-dimensional distribution of financial values ​​in the dataset labeled "out of range" (such as finding outliers concentrated in specific subsidiaries). Finally, it produces a health status analysis report that details the data completeness, logical deviation, and abnormal clustering patterns, providing accurate decision-making basis for subsequent automated cleaning and filling.

[0117] This embodiment achieves in-depth quantitative diagnosis and full-process traceability of the health status of flawed data by introducing dynamic code execution and data distribution feature analysis under the isolation sandbox mechanism. Its beneficial effect is that it not only ensures the system security of the automated analysis process, but also completely solves the technical problem of "black box in the conclusion generation process and imperceptible data limitations" in financial analysis by generating visualized analysis reports and task adjustment records.

[0118] Optionally, the step of generating a complete analysis report through the final data processing scheme includes: Using the analysis report as a basis for decision-making, numerical imputation is performed on fields with a missing rate below a preset threshold using a predetermined industry median or regression estimation method. Abnormal values ​​with logical conflicts are removed or smoothed, and high-quality analytical sample data that is cleaned and statistically significant is output. Generate analysis code and perform statistical calculations based on high-quality analysis sample data to complete the quantitative analysis of financial analysis information for the target analysis object and generate structured calculation results that reflect the real business status of the target analysis object. Based on the structured calculation results and the task adjustment records, a complete analysis report is output, including explanations of data limitations, derivation of business conclusions, and multi-dimensional statistical charts.

[0119] In this embodiment of the invention, the analysis report can be used as a decision-making basis to perform numerical imputation or anomaly processing on fields, outputting high-quality analytical sample data. This allows for the scientific repair or purification of flawed data using the "health check conclusion" (analysis report) output from previous steps. Financial algorithms are used to fill in missing values ​​and correct conflicting values, ensuring that the data entering the calculation stage is both complete and conforms to statistical distribution patterns.

[0120] The preset threshold can be a set tolerance boundary (e.g., missing rate below 5%), used to determine whether the field is worth "repairing" through the algorithm or "abandoning" it directly.

[0121] The median imputation method can be to use the median value of the indicator of similar companies in the same industry to replace the missing values, so as to ensure that the data does not deviate from the industry benchmark.

[0122] Regression estimation methods can be used to predict and fill in missing values ​​by building a mathematical model using other strongly correlated indicators in the dataset.

[0123] Smoothing can be a mathematical adjustment of highly volatile or noisy values ​​to make them more consistent with actual business trends.

[0124] High-quality analytical sample data can be standardized datasets that have been cleaned, imputed, and corrected for logical consistency using algorithms, and can be directly used for quantitative modeling.

[0125] This invention enables the generation of analytical code and the execution of statistical calculations from high-quality analytical sample data. This allows for the quantitative analysis of financial information related to the target analysis object, generating structured calculation results that reflect the actual business situation of the target object. This achieves a quantitative transformation from "standardized data" to "professional business conclusions." By dynamically generating analytical scripts with financial logic from cleaned sample data using a large language model, and ensuring the reproducibility of the calculation process, the automated output of structured evaluation indicators can accurately quantify the enterprise's risk level and its relative position in the industry, providing objective quantitative support for final business decisions.

[0126] High-quality analytical sample data refers to standardized datasets that have undergone data acquisition, quality assessment, and cleaning, completion, and alignment operations by data processing agents, eliminating analytical ambiguity and conforming to statistical significance, and can be directly used for modeling and computation.

[0127] Analysis code can be a program script (such as Python or SQL) dynamically generated by a data processing agent (based on a large language model) according to the current task objective and data characteristics, used to perform specific mathematical model calculations and logical derivations.

[0128] For example, financial analysis information may include, but is not limited to, credit risk indices and industry risk percentiles; The credit risk index is a comprehensive score derived by multi-factor modeling of a company's financial performance (such as debt-to-equity ratio and interest coverage ratio) and business attributes, used to quantitatively assess the likelihood of a target entity experiencing credit risk.

[0129] Industry risk percentile can be determined by placing the risk indicators of a target company in a sample pool of the same industry for horizontal comparison, and determining the percentile ranking (such as the 28th percentile) to reflect the company's position in the overall risk distribution of the industry.

[0130] Quantitative analysis can be achieved by using mathematical and statistical methods to transform qualitative business problems into a calculation process based on precise numerical values, thereby eliminating the subjective bias of human judgment.

[0131] Structured computation results can refer to a set of data (such as scores, rankings, and category labels) that has a fixed format and clear meaning after computation is completed. These results can directly drive the generation of charts or the filling of report text.

[0132] This invention significantly reduces the risk of calculation errors caused by human error or logical mistakes by automatically generating code instead of manually writing it. The system can instantly perform benchmarking calculations (such as quantile analysis) on massive amounts of industry data, freeing researchers from tedious mathematical calculations and allowing them to focus on higher-value decision derivations. Quantitative analysis based on high-quality samples ensures that the conclusions are based on an objective mapping of the company's real financial data to industry benchmarks, rather than "superficial numerical guesses" from large models, significantly improving the business credibility of the analysis results.

[0133] This invention, in its embodiments, can output a complete analysis report based on the structured calculation results and the task adjustment records, thereby completing the final result transformation. By combining the raw numerical calculation results with "logical adjustment explanations" due to data flaws, a professional text report is output that includes conclusions, supporting evidence, and accurately reflects the limitations of the data.

[0134] Data limitation statements can be declarations of operations such as weight reduction or imputation taken by the system during analysis due to missing or abnormal original data, reminding report users to pay attention to the boundaries of the relevant conclusions.

[0135] Business conclusions can be derived by translating quantitative indicators into business judgments that are understandable to humans (e.g., "The company has tight liquidity and it is recommended to pay attention to short-term debt repayment risks").

[0136] A complete analysis report can be a final decision-making reference document that integrates textual descriptions, data charts, and limitations and disclaimers.

[0137] For example, when conducting a risk assessment of a financial leasing company, the system uses the missing rate of "asset-liability ratio" (15%, below the preset threshold of 20%) identified in the analysis report as the basis for decision-making. It automatically calls the median of the same industry to scientifically fill in the missing information and performs smoothing on the negative value of the incorrectly entered "interest coverage ratio," thereby producing a high-quality analysis sample that conforms to statistical laws. Subsequently, the system automatically writes a quantitative analysis script for this sample, calculates the company's credit risk index as 0.42 by performing multi-factor weighted calculation, and determines that it is in the low-risk percentile of the top 15% in the industry, forming structured data that reflects the current business situation. Finally, the system integrates previously recorded task adjustment information such as "weight reduction" and automatically generates a complete analysis report that includes an explanation that "due to the use of industry extrapolation for some indicators, the conclusion has certain limitations in terms of liquidity prediction," business risk rating recommendations, and a risk distribution histogram.

[0138] This embodiment establishes an intelligent data repair and fully automated analysis mechanism based on the analysis report, realizing the final closed loop from flawed datasets to professional-grade analysis reports. Its beneficial effects are that it not only utilizes industry algorithms to maximize the remaining value of incomplete data, but also deeply integrates task adjustment records with quantitative results to produce an in-depth analysis report that is logically traceable, risk-aware, and can truly reflect the current business status of the enterprise.

[0139] For example, in financial research scenarios, analysts often need to assess the credit risk of target companies and compare them with companies in the same industry. The multi-agent collaborative data analysis system described in this invention can automatically complete the entire process from data acquisition to result output, ensuring that the analysis results are interpretable, reproducible, and stable.

[0140] When the data analysis agent receives the natural language task "Assess Company A's credit risk in 2024 and compare it with companies in the same industry", it first performs semantic parsing of the task objective to identify key analysis dimensions (such as corporate credit score, debt-to-equity ratio, interest coverage ratio, and cash flow volatility). It then transforms the natural language instructions into a structured task definition, clarifies data dependencies and execution logic, and finally schedules the data acquisition agent to perform data collection operations.

[0141] After receiving the task instruction, the data acquisition agent matches key data fields with corresponding data sources according to the structured task definition, locates the industry code of Company A (such as the insurance industry) through the industry mapping table, and automatically generates a list of companies in the same industry.

[0142] The system invokes multiple data sources according to the task definition, including enterprise databases (credit scoring tables, financial statements) and external interfaces (such as Wind, peer rating platforms, etc.), to acquire the required indicator data in parallel. During this process, the system performs semantic matching and metadata verification. After acquisition, the system automatically performs data standardization and fusion operations, including format unification, time period alignment, field mapping, and currency conversion. For example, when the definition of "Credit Score A" in data source A is inconsistent with that of "Credit Rating B" in data source B, the agent automatically matches the differences in scoring criteria and performs value domain conversion through built-in semantic rules, making "Credit Score A=1" semantically aligned with "Credit Rating B=AAA". When processing research conclusion data, the dimension field values ​​returned by the interface are numeric codes (such as 1, 2, 3). The agent automatically replaces these with corresponding semantic labels (such as "macroeconomic aspect", "valuation aspect", "policy aspect") according to the mapping table, thereby significantly improving the large model's understanding of the data meaning and the efficiency of subsequent processing. Through the above mechanism, the system achieves automatic mapping from multi-source heterogeneous data to unified structured data, and the output dataset covers all target indicators, field definitions, and missing identifiers.

[0143] After receiving the output from the data acquisition phase, the data evaluation agent enters the dynamic availability assessment process. This agent checks the data's completeness, timeliness, and task relevance item by item. If all core fields (such as credit score or debt ratio) are missing, the system determines the task is unexecutable, immediately outputs a failure signal, and terminates the process. At this point, the main agent prompts the user that "core data is missing; the task needs to be redefined or the data source supplemented." If some fields are missing but do not affect the core analysis objectives, the system automatically assesses the missing percentage, indicator importance, and data distribution stability. For example, if the "cash flow volatility" missing percentage is as high as 60%, but the credit score and interest coverage ratio have a completeness of over 90%, the system will determine that although this field is missing, it does not constitute the basis of the main conclusion. The agent will then generate optimization instructions, prompting the data processing agent to focus on the complete data for credit score and debt ratio, and reduce the analysis weight of cash flow volatility. Simultaneously, the system outputs an optimized task objective and data. If all fields are complete and the data quality meets the standards, the system confirms that the task will continue as planned. Finally, the data evaluation agent outputs a clear feasibility assessment report and an optimized task plan, which are directly handed over to the downstream data processing agent for execution.

[0144] After receiving the evaluated and confirmed data and task plan, the data processing agent automatically performs data processing and analysis in a sandbox environment. The process is as follows: First, the agent automatically generates profiling code to identify and mark data distribution, missing rates, and outliers (such as "debt ratio > 200%)". Then, it enters the cleaning phase, using methods such as industry median or regression estimation to fill in missing data fields with missing rates below a threshold (e.g., 30%), and fully records the filling logic and source to ensure traceability. If there are differences in accounting standards (e.g., between individual financial statements and consolidated financial statements), the system prioritizes retaining the consolidated standard and removes inconsistent samples. After data cleaning, the agent further generates analysis code to calculate the credit risk index and evaluate Company A's risk percentile and ranking among industry samples. Throughout the process, if a calculation anomaly or execution error occurs, the system will automatically roll back to the previous stable version and re-execute, effectively ensuring process continuity and result reliability. The final output includes cleaned structured data and a result description. The data analysis agent integrates the results from the data processing module, generating two types of output: Textualized results can be analytical conclusions automatically generated by the system through a natural language generation engine. For example: "Company A's credit risk index for 2024 was 0.37, ranking 28th in the industry, with overall risk lower than the industry average. Its main advantages stem from a low debt-to-equity ratio and a high interest coverage ratio. Insufficient disclosure of cash flow volatility has a limited impact on the conclusions." The graphical results can be generated automatically by the system inputting the cleaned data into the visualization module, including various charts such as credit score box plots, industry risk distribution curves, and financial indicator scatter plots.

[0145] This example demonstrates the specific application process of the system of this invention in a financial credit research scenario. Through the collaborative efforts of three intelligent agents—data acquisition, evaluation, and processing—the system achieves automatic interpretation and usability correction of structured data. When faced with complex situations such as missing fields, inconsistent definitions, or data anomalies, the system can automatically adjust the task scope and indicator weights, output optimization solutions, and ensure process continuity. The final generated report possesses consistency, traceability, and interpretability, significantly improving the large model's ability to understand data and the reliability of business analysis.

[0146] To enable those skilled in the art to better understand the embodiments of the present invention, an example is used below to illustrate the embodiments of the present invention.

[0147] refer to Figure 3 , Figure 3 This is a flowchart illustrating a data acquisition method provided in an embodiment of the present invention; 1. The data acquisition agent, as the starting module of the entire analysis process, aims to solve the problem that large models cannot directly understand the semantics of enterprise-specific data. This agent replaces the traditional manual process that requires customized development or maintenance of explanatory metadata tables through intelligent data source matching and semantic mapping mechanisms, thereby achieving automatic standardization of data semantics.

[0148] The input to this intelligent agent comes from the structured and semantically clear analysis objectives generated by the data analysis agent after parsing the user's task. Its execution is carried out by the data acquisition agent itself, which, based on received instructions, efficiently and accurately acquires the required information from diverse data sources through an intelligent scheduling mechanism. The final output is a structured dataset that has undergone preliminary cleaning and standardization, for use by the data evaluation agent.

[0149] Its core execution process includes: Receive structured analysis objectives from the data analysis agent, deeply analyze task requirements, and accurately identify analysis dimensions, indicator fields, and data source types; The process involves matching available data fields and sources using a data dictionary. If no available data can be found for the current data requirement, the process terminates, and for matching data requirements, the process proceeds to the next step of the query. Based on its built-in decision-making logic, the intelligent agent intelligently schedules enterprise databases, API interfaces, and externally authorized data services, and executes data query tasks according to the real-time requirements of the data source. For the acquired raw data, the AI ​​performs key integration and standardization operations, including format unification, field mapping, and value transformation, to address the inherent semantic heterogeneity of multi-source data and ensure a consistent and usable dataset. This is based on two core mechanisms: By developing predefined semantic rules through customization; State queries and matching of external explanatory metadata tables; For example, when an analysis task requires "comparing a specific stock with other stocks in the same industry," the agent automatically generates a list of stocks in the same industry and uses this list to iteratively call detailed data interfaces such as "Stock360" to fully construct the sample set required for the analysis. When processing "credit score" data, the agent automatically matches the company's internal definitions and analysis descriptions of the differences between "Score A" and "Score B," and performs the corresponding value range conversion. Through built-in multi-source parallel access and dynamic fault tolerance mechanisms, this module achieves automatic mapping from unstructured business problems to structured, computable data, laying the data foundation for subsequent analysis.

[0150] refer to Figure 4 , Figure 4 This is a flowchart illustrating a data evaluation method provided in an embodiment of the present invention; 2. The data evaluation agent aims to establish a dynamic data quality evaluation mechanism oriented towards task objectives, addressing the bias in analysis results caused by data quality issues. The agent's input sources are the structured raw dataset output by the upstream data acquisition agent and the initial task definition context. Its execution is performed by the data evaluation agent itself, which comprehensively evaluates the applicability and completeness of the data based on task requirements as the core standard.

[0151] Based on the original task requirements, we assess whether the acquired data meets the basic requirements for analysis. The specific assessment logic is as follows: If the data cannot meet the task requirements at all (such as the core analysis fields being completely missing, or the data content being seriously inconsistent with the requirements), the system will determine that the task is not executable, output a failure signal, and terminate the process. If the data partially meets the requirements, the agent will assess the degree of missing data and its relevance to the core objectives of the task. Based on the assessment results, it will automatically adjust the analysis strategy, which may narrow the scope of analysis, adjust the time period, or focus on analyzable dimensions. The system will then output an optimized task solution. For data that perfectly meets the task objectives, the system confirms that the original analysis plan will continue. Ultimately, the data evaluation agent outputs a clear feasibility judgment and a calibrated, operable task plan, which is directly delivered to the downstream data processing agent as the basis for execution.

[0152] refer to Figure 5 , Figure 5 This is a flowchart illustrating a data processing method provided in an embodiment of the present invention; The data processing agent is the core execution module for the system's automated analysis, undertaking the crucial task of transforming raw data into high-quality analytical results. This agent receives usability-calibrated structured data and clearly defined data processing objectives from the data evaluation agent. Its execution is carried out by a built-in large language model and a secure sandbox environment, ultimately outputting data analysis results and corresponding descriptions.

[0153] The execution process includes: 1) Initialize an isolated sandbox environment for executing code that will be used to generate the large model later; 2) Receive calibration data and processing objectives from the data evaluation agent; 3) Automatically generate and execute profiling code to analyze data distribution, outliers, and missing data ratios; 4) Generate cleaning code based on the profiling results, and perform operations such as anomaly handling, missing data filling, and field standardization to ensure that the data quality meets the analysis requirements; 5) Generate analysis code based on task requirements, perform calculation tasks such as statistical analysis and feature engineering, and generate analysis results; 6) A large language model describes the analysis results, and finally outputs structured analysis results and corresponding text descriptions, providing reliable input for subsequent in-depth analysis and report generation.

[0154] For example, when processing financial data, the AI ​​first identifies outlier distributions through profiling, then generates targeted processing code, and finally performs financial ratio calculations. For time series data, it first checks for missing patterns, then generates appropriate interpolation code, and finally performs trend analysis.

[0155] This intelligent agent, through this phased and verifiable processing flow, not only ensures the processing effect but also significantly reduces code execution failures caused by outliers and missing values, thereby improving the reliability of the analysis process.

[0156] refer to Figure 6 , Figure 6 This is a flowchart illustrating a data analysis method provided in an embodiment of the present invention; After the data processing flow is complete, the data analysis agent will receive the output from the data processing agent and transform it into a structured output that can be used to generate the final report. Its core output includes two parts: Textualized results description: This section contains the analysis conclusions, key insights, and logical explanations in natural language form, which are directly fed into the report's text generation module to automatically generate the narrative report body.

[0157] Processed data results: This part consists of cleaned and normalized structured data, which is directly sent to the chart generation module as input to the visualization engine to automatically generate charts, graphs and other visualization content.

[0158] It should be noted that, for the sake of simplicity, the method embodiments are all described as a series of actions. However, those skilled in the art should understand that the embodiments of the present invention are not limited to the described order of actions, because according to the embodiments of the present invention, some steps can be performed in other orders or simultaneously. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions involved are not necessarily essential to the embodiments of the present invention.

[0159] Reference Figure 7 The diagram shows a structural block diagram of an analysis report generation device provided in an embodiment of the present invention, which may specifically include the following modules: The semantic parsing module 701 is used to perform semantic parsing on the natural language task in response to receiving a natural language task sent by the user, and to determine the analysis dimension, business scope and expected goal of the natural language task. The structured task solution encapsulation module 702 is used to encapsulate the analysis dimensions, the business scope, and the expected goals into a structured task solution; The raw structured indicator data acquisition module 703 is used to acquire raw structured indicator data for performing the natural language task through the mapping relationship of the structured task scheme. The semantic ambiguity elimination module 704 is used to eliminate the heterogeneity of the data source of the original structured index data and output the dataset to be evaluated with semantic ambiguity eliminated. The final data processing scheme generation module 705 is used to obtain data quality characteristics that characterize the health status of the dataset to be evaluated, and to modify the structured task scheme based on the data quality characteristics to generate the final data processing scheme. The complete analysis report generation module 706 is used to generate a complete analysis report through the final data processing scheme.

[0160] As the device embodiment is basically similar to the method embodiment, the description is relatively simple, and relevant parts can be found in the description of the method embodiment.

[0161] In addition, embodiments of the present invention also provide an electronic device, such as... Figure 8As shown, it includes a processor 801, a communication interface 802, a memory 803, and a communication bus 804. The processor 801, communication interface 802, and memory 803 communicate with each other via the communication bus 804. Memory 803 is used to store computer programs; When the processor 801 executes the program stored in the memory 803, it implements any of the analysis report generation methods described in the above embodiments: The communication bus mentioned above can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. This communication bus can be divided into address bus, data bus, control bus, etc. For ease of illustration, only one thick line is used to represent it in the diagram, but this does not mean that there is only one bus or one type of bus.

[0162] The communication interface is used for communication between the aforementioned terminal and other devices.

[0163] The memory may include random access memory (RAM) or non-volatile memory, such as at least one disk storage device. Optionally, the memory may also be at least one storage device located remotely from the aforementioned processor.

[0164] The processors mentioned above can be general-purpose processors, including central processing units (CPUs), network processors (NPs), etc.; they can also be digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0165] like Figure 9 As shown, in another embodiment of the present invention, a computer-readable storage medium 901 is also provided, which stores instructions that, when run on a computer, cause the computer to execute the analysis report generation method described in the above embodiment.

[0166] This application embodiment also provides a chip, which includes a processor and a communication interface. The communication interface is coupled to the processor. The processor is used to run programs or instructions to implement the various processes of the above-described analysis report generation method embodiment and can achieve the same technical effect. To avoid repetition, it will not be described again here.

[0167] It should be understood that the chip mentioned in the embodiments of this application may also be referred to as a system-on-a-chip, system chip, chip system, or system-on-a-chip, etc.

[0168] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element. Furthermore, it should be noted that the scope of the methods and apparatuses in the embodiments of this application is not limited to performing functions in the order shown or discussed, but may also include performing functions substantially simultaneously or in the reverse order, depending on the functions involved. For example, the described methods may be performed in a different order than described, and various steps may be added, omitted, or combined. Additionally, features described with reference to certain examples may be combined in other examples.

[0169] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal (which may be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0170] The embodiments of this application have been described above with reference to the accompanying drawings. However, this application is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of this application without departing from the spirit and scope of the claims, and all of these forms are within the protection scope of this application.

Claims

1. A method for generating analysis reports, characterized in that, include: In response to receiving a natural language task sent by a user, perform semantic parsing on the natural language task to determine the analysis dimensions, business scope, and expected goals of the natural language task; The analytical dimensions, business criteria, and expected objectives are encapsulated into a structured task solution; By using the mapping relationship of the structured task scheme, the original structured indicator data used to perform the natural language task is obtained; Eliminate the heterogeneity of the data source of the original structured indicator data and output the dataset to be evaluated without semantic ambiguity; Obtain data quality features that characterize the health status of the dataset to be evaluated, and revise the structured task scheme based on the data quality features to generate the final data processing scheme; A complete analysis report is generated using the final data processing scheme.

2. The method according to claim 1, characterized in that, The step of obtaining the original structured indicator data for performing the natural language task through the mapping relationship of the structured task scheme includes: The structured task scheme is read, and by retrieving a preset semantic dictionary or metadata mapping table, data tags and field requirements for representing the analysis dimension, the business scope, and the expected goal are generated. Based on the data tags and field requirements, access the enterprise's internal database and / or metadata center corresponding to the data tags to obtain the original structured indicator data.

3. The method according to claim 2, characterized in that, The step of eliminating the heterogeneity of the data source of the original structured index data and outputting a dataset to be evaluated without semantic ambiguity includes: Obtain the metrics used to characterize the business attributes and financial performance of the target analysis object from the original structured index data; The metric values ​​are standardized and aligned using semantic rules to output a dataset to be evaluated with semantic ambiguity eliminated.

4. The method according to claim 3, characterized in that, The step of obtaining data quality characteristics to characterize the health status of the dataset to be evaluated includes: Perform integrity checks on the dataset to be evaluated to identify missing fields; The data to be evaluated is subjected to quality inspection to identify abnormal values.

5. The method according to claim 4, characterized in that, The step of modifying the structured task scheme based on the data quality characteristics to generate the final data processing scheme includes: Based on the missing proportion of the missing fields, the analysis dimensions and expected goals of the structured task solution are prioritized and adjusted, or the structured task solution is logically modified based on the severity of the anomalies expressed by the abnormal values, and a verified final data processing solution is output.

6. The method according to claim 5, characterized in that, Prior to the step of generating a complete analysis report through the final data processing scheme, the method further includes: Identify the modified dataset that is associated with the final data processing scheme, and the task adjustment record for the modified dataset; Feature analysis code is automatically generated and run in an isolated sandbox environment to identify the spatial distribution of anomaly markers in the corrected dataset and generate an analysis report reflecting the health status of the dataset to be evaluated.

7. The method according to claim 6, characterized in that, The steps for generating a complete analysis report using the final data processing scheme include: Using the analysis report as a basis for decision-making, numerical imputation is performed on fields with a missing rate below a preset threshold using a predetermined industry median or regression estimation method. Abnormal values ​​with logical conflicts are removed or smoothed, and high-quality analytical sample data that is cleaned and statistically significant is output. Generate analysis code and perform statistical calculations based on high-quality analysis sample data to complete the quantitative analysis of financial analysis information for the target analysis object and generate structured calculation results that reflect the real business status of the target analysis object. Based on the structured calculation results and the task adjustment records, a complete analysis report is output, including explanations of data limitations, derivation of business conclusions, and multi-dimensional statistical charts.

8. An analytical report generation device, characterized in that, include: The semantic parsing module is used to respond to receiving a natural language task sent by a user, perform semantic parsing on the natural language task, and determine the analysis dimension, business scope and expected goal of the natural language task; A structured task solution encapsulation module is used to encapsulate the analysis dimensions, the business scope, and the expected goals into a structured task solution; The raw structured indicator data acquisition module is used to acquire raw structured indicator data for performing the natural language task through the mapping relationship of the structured task scheme. The semantic ambiguity elimination module is used to eliminate the heterogeneity of the data source of the original structured index data and output the evaluation dataset with semantic ambiguity eliminated. The final data processing scheme generation module is used to obtain data quality characteristics that characterize the health status of the dataset to be evaluated, and to modify the structured task scheme based on the data quality characteristics to generate the final data processing scheme. The complete analysis report generation module is used to generate a complete analysis report through the final data processing scheme.

9. An electronic device, characterized in that, It includes a processor, a memory, and a program or instructions stored in the memory and executable on the processor, wherein the program or instructions, when executed by the processor, implement the method as described in claims 1-7.

10. A readable storage medium, characterized in that, The readable storage medium stores a program or instructions that, when executed by a processor, implement the method as described in claims 1-7.

Images