Industrial report automatic generation system based on large language model

Abstract

The invention relates to the technical field of financial risk assessment, in particular to an industry report automatic generation system based on a large language model, and the system comprises the steps: obtaining structured and unstructured data of each real estate enterprise through a dynamic multi-source data collection module, and constructing a first multi-source credibility coefficient; constructing a first comprehensive data set based on the coefficient and a data collection strategy; the risk assessment model construction module constructs an enterprise financial risk assessment model to identify the first comprehensive data set, finance risk anomaly features are obtained, and then a multi-source financial risk coefficient is generated; the financial risk assessment and report generation module generates a comprehensive financial risk score according to the multi-source financial risk coefficient, and outputs an automatic report according to the comprehensive financial risk score sequence of the real estate enterprise; according to the invention, by constructing the enterprise financial risk assessment model, intelligent analysis and report automatic generation of real estate enterprise financial risks can be effectively realized.

Description

Technical Field

[0001] The present invention relates to the technical field of financial risk assessment, and in particular to an automatic industry report generation system based on a large language model. Background Art

[0002] Traditional real estate financial analysis reports have a single data source and do not fully consider financial risk report analysis based on multi-source data (financial data, market data, external regulatory data, and real estate development progress data), resulting in limited accuracy and comprehensiveness of financial risk report generation; therefore, there is an urgent need for an industry report automatic generation system that can dynamically integrate multi-source data, deeply mine multi-modal associations, and accurately assess risks.

[0003] Therefore, an automatic industry report generation system based on a large language model is proposed. Summary of the invention

[0004] The purpose of the present invention is to provide an automatic industry report generation system based on a large language model, which obtains structured and unstructured data of each real estate enterprise through a dynamic multi-source data collection module, and constructs a first comprehensive data set based on a multi-source credibility coefficient and a data collection strategy; constructs an enterprise financial risk assessment model to identify the first comprehensive data set, obtains abnormal characteristics of financial risks, and then generates a multi-source financial risk coefficient; the financial risk assessment and report generation module generates a comprehensive financial risk score according to the multi-source financial risk coefficient, and outputs an automated report according to the ranking of the comprehensive financial risk scores of the real estate enterprises; the present invention can effectively realize the intelligent analysis of the financial risks of real estate enterprises and the automatic generation of reports through the enterprise financial risk assessment model.

[0005] To achieve the above object, the present invention provides the following technical solutions: An automatic industry report generation system based on a large language model, comprising: A dynamic multi-source data collection module, which obtains first structured data and first unstructured data of each real estate enterprise, calculates a first multi-source credibility coefficient, and obtains a first comprehensive data set based on the first multi-source credibility coefficient and a data collection strategy; The risk assessment model construction module is used to construct an enterprise financial risk assessment model to identify the first comprehensive data set, which passes through the first input layer, the first preprocessing layer, the financial risk feature extraction layer, the financial risk feature analysis layer and the first output layer of the model in sequence; the financial risk feature extraction layer is used to extract abnormal financial risk features; the abnormal financial risk features include abnormal financial features, abnormal market features, abnormal external regulatory features and abnormal real estate development features; the financial risk feature analysis layer obtains multi-source financial risk coefficients according to the financial risk features; The financial risk assessment and report generation module is used to generate a comprehensive financial risk score based on multi-source financial risk coefficients, rank each enterprise based on the comprehensive financial risk score, and output a report.

[0006] Preferably, the first structured data includes corporate balance sheets, corporate cash flow statements, corporate income statements, housing sales transaction data, housing rental transaction data, land transaction databases and supply chain contracts; the first unstructured data includes external regulatory document reports, real estate development progress data and project payment settlement status.

[0007] Preferably, the first multi-source credibility coefficient is: ; in, is the first multi-source credibility coefficient; is the data freshness attenuation factor influence coefficient; is the data freshness attenuation factor, reflecting the freshness of the data; is the influence coefficient of the platform authority coefficient; is the platform authority coefficient, which is used to evaluate the authority of the data source platform; Scoring impact coefficient for data completeness; Provides a data integrity score to assess the completeness of the data.

[0008] Preferably, based on the first multi-source credibility coefficient and the preset threshold, the credibility level of each data is obtained, and the credibility levels include high credibility, medium credibility and low credibility; the high-credibility data is fully collected; the medium-credibility data is sampled and collected; the low-credibility data is not collected; and then the collected data are merged to obtain the first comprehensive data set.

[0009] Preferably, the enterprise financial risk assessment model is constructed based on LLM (Large Language Model), including a first input layer, a first preprocessing layer, a financial risk feature extraction layer, a financial risk feature analysis layer and a first output layer; The first input layer is used to input the first comprehensive data set into the enterprise financial analysis and evaluation model; The first preprocessing layer performs unified representation modeling on the structured data and unstructured data in the first comprehensive data set based on the natural language processing pre-training paradigm, including data encoding, text standardization and semantic embedding construction, to obtain a second comprehensive data set; The financial risk feature extraction layer is based on the Transformer architecture and uses a multi-head self-attention mechanism to perform context modeling and semantic association analysis on the second comprehensive data set to identify abnormal financial risk features; abnormal financial risk features include abnormal financial features, abnormal market features, abnormal external regulatory features, and abnormal real estate development features; The financial risk characteristic analysis layer obtains multi-source financial risk coefficients according to the four abnormal characteristics; the multi-source financial risk coefficients include financial risk coefficients, market risk coefficients, external regulatory risk coefficients and real estate development risk coefficients; The first output layer is used to output multi-source financial risk coefficients.

[0010] Preferably, the abnormal financial characteristics include year-on-year debt-to-asset ratio, month-on-month debt-to-asset ratio, year-on-year cash flow volatility, month-on-month cash flow volatility and net profit margin; the abnormal market characteristics include house price volatility, year-on-year sales volume ratio, month-on-month sales volume ratio, year-on-year rental volume ratio, month-on-month rental volume ratio and supply contract default rate; the abnormal external regulatory characteristics include financial operation regulatory matching and tax violation rate; the abnormal real estate development characteristics include progress lag rate and delayed payment time of project funds.

[0011] Preferably, the comprehensive financial risk score is: ; in, Indicates Comprehensive financial risk score of real estate enterprises; Indicates The financial risk coefficient of a real estate enterprise is used to reflect the impact of financial data on real estate financial risks; Indicates The market risk coefficient of a real estate enterprise; Indicates The external regulatory risk coefficient of each real estate enterprise is used to determine the impact of external regulatory data on real estate financial risks; Indicates The property development risk coefficient of a real estate enterprise is used to reflect the impact of property development progress on real estate financial risks.

[0012] Compared with the prior art, the present invention has the following beneficial effects: 1. The present invention obtains structured data and unstructured data of each real estate enterprise, calculates a first multi-source credibility coefficient according to a data freshness decay factor, a platform authority coefficient and a data integrity score, and obtains a first comprehensive data set based on the first multi-source credibility coefficient and a data collection strategy; the data collection strategy includes: obtaining a credibility level of each data according to the first multi-source credibility coefficient and a preset threshold, the credibility level including high credibility, medium credibility and low credibility; collecting all high-credibility data; sampling and collecting medium-credibility data; not collecting low-credibility data; and then merging the collected data to obtain a first comprehensive data set; the present invention effectively screens and selects the report source data through the first multi-source credibility coefficient and the data collection strategy, improves the quality of model input data, and thus effectively improves the accuracy and comprehensiveness of financial risk report generation.

[0013] 2. The present invention identifies and analyzes the first comprehensive data set by constructing an enterprise financial risk assessment model to obtain abnormal financial risk characteristics; abnormal financial risk characteristics include abnormal financial characteristics, abnormal market characteristics, abnormal external regulatory characteristics and abnormal real estate development characteristics; a multi-source financial risk coefficient is obtained based on these four abnormal characteristics; the multi-source financial risk coefficient includes a financial risk coefficient, a market risk coefficient, an external regulatory risk coefficient and a real estate development risk coefficient; a comprehensive financial risk score is generated based on the multi-source financial risk coefficient, and each enterprise is comprehensively ranked based on the comprehensive financial risk score, and an automated report is generated; by conducting financial analysis of real estate enterprises based on finance, market, external supervision and real estate development, the accuracy and comprehensiveness of the financial analysis of real estate enterprises can be effectively achieved, thereby effectively improving the accuracy of generating financial risk reports. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 A schematic diagram of the structure of an automatic industry report generation system based on a large language model provided by an embodiment of the present invention; Figure 2 A schematic diagram of the structure of an enterprise financial risk assessment model provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0015] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0016] Embodiment 1 In order to improve the accuracy and comprehensiveness of the financial analysis reports on real estate enterprises, an automatic industry report generation system based on a large language model was applied; Reference Figure 1 A structural diagram of an industry report automatic generation system based on a large language model provided by an embodiment of the present invention specifically includes: An automatic industry report generation system based on a large language model, comprising: A dynamic multi-source data collection module, which obtains first structured data and first unstructured data of each real estate enterprise, calculates a first multi-source credibility coefficient, and obtains a first comprehensive data set based on the first multi-source credibility coefficient and a data collection strategy; The risk assessment model construction module is used to construct an enterprise financial risk assessment model to identify the first comprehensive data set, which passes through the first input layer, the first preprocessing layer, the financial risk feature extraction layer, the financial risk feature analysis layer and the first output layer of the model in sequence; the financial risk feature extraction layer is used to extract abnormal financial risk features; the abnormal financial risk features include abnormal financial features, abnormal market features, abnormal external regulatory features and abnormal real estate development features; the financial risk feature analysis layer obtains multi-source financial risk coefficients according to the financial risk features; Furthermore, the first structured data includes corporate balance sheets, corporate cash flow statements, corporate income statements, housing sales transaction data, housing leasing transaction data, land transaction databases and supply chain contracts; the first unstructured data includes external regulatory document reports, real estate development progress data and project payment settlement status.

[0017] Further, ; in, is the first multi-source credibility coefficient; is the data freshness attenuation factor influence coefficient; is the data freshness attenuation factor, reflecting the freshness of the data; is the influence coefficient of the platform authority coefficient; is the platform authority coefficient, which is used to evaluate the authority of the data source platform; Scoring impact coefficient for data completeness; Provides a data integrity score to assess the completeness of the data.

[0018] Furthermore, the data freshness decay factor is: ; in, is the data freshness decay factor; represents the attenuation coefficient; Indicates the number of days since the data was generated; Indicates An exponential function with base ; The platform authority coefficient influence coefficient is obtained based on the authority of the data acquisition source platform and based on expert experience; the data integrity score is used to evaluate the integrity of the data, the more missing data, the lower the score; it is used to represent the overall integrity of structured data and unstructured data; Furthermore, based on the first multi-source credibility coefficient and the preset threshold, the credibility level of each data is obtained, and the credibility levels include high credibility, medium credibility and low credibility; the high-credibility data is fully collected; the medium-credibility data is sampled and collected; the low-credibility data is not collected; and the collected data is then merged to obtain the first comprehensive data set.

[0019] This embodiment obtains structured data and unstructured data of each real estate enterprise, calculates a first multi-source credibility coefficient according to a data freshness decay factor, a platform authority coefficient and a data integrity score, and obtains a first comprehensive data set based on the first multi-source credibility coefficient and a data collection strategy; the data collection strategy includes: obtaining a credibility level of each data according to the first multi-source credibility coefficient and a preset threshold, the credibility level including high credibility, medium credibility and low credibility; collecting all high-credibility data; sampling and collecting medium-credibility data; not collecting low-credibility data; and then merging the collected data to obtain a first comprehensive data set; the present invention effectively screens and selects the report source data through the first multi-source credibility coefficient and the data collection strategy, thereby effectively improving the accuracy and comprehensiveness of the generation of financial risk reports.

[0020] Further, the enterprise financial risk assessment model is constructed based on LLM, including a first input layer, a first preprocessing layer, a financial risk feature extraction layer, a financial risk feature analysis layer and a first output layer; The first input layer is used to input the first comprehensive data set into the enterprise financial analysis and evaluation model; The first preprocessing layer performs unified representation modeling on the structured data and unstructured data in the first comprehensive data set based on the natural language processing pre-training paradigm, including data encoding, text standardization and semantic embedding construction, to obtain a second comprehensive data set; The financial risk feature extraction layer is based on the Transformer architecture and uses a multi-head self-attention mechanism to perform context modeling and semantic association analysis on the second comprehensive data set to identify abnormal financial risk features; abnormal financial risk features include abnormal financial features, abnormal market features, abnormal external regulatory features, and abnormal real estate development features; The financial risk characteristic analysis layer obtains multi-source financial risk coefficients according to the four abnormal characteristics; the multi-source financial risk coefficients include financial risk coefficients, market risk coefficients, external regulatory risk coefficients and real estate development risk coefficients; The first output layer is used to output multi-source financial risk coefficients.

[0021] The structural diagram of the enterprise financial risk assessment model is as follows: Figure 2 As shown; Furthermore, the financial abnormal characteristics include the year-on-year debt-to-asset ratio, the month-on-month debt-to-asset ratio, the year-on-year cash flow volatility, the month-on-month cash flow volatility and the net profit margin; the market abnormal characteristics include the housing price volatility, the year-on-year sales volume ratio, the month-on-month sales volume ratio, the year-on-year rental volume ratio, the month-on-month rental volume ratio and the supply contract default rate; the external regulatory abnormal characteristics include the financial operation regulatory matching degree and the tax violation rate; the real estate development abnormal characteristics include the progress lag rate and the delayed payment time of the project funds; Net profit margin is used to indicate the net profit margin of a real estate enterprise in the current year; Furthermore, the financial risk coefficient is: ; in, Indicates The financial risk factor of a real estate enterprise; Indicates The year-on-year debt-to-asset ratio of real estate enterprises; Indicates The month-on-month asset-liability ratio of real estate enterprises; Indicates The year-on-year cash flow volatility of a real estate company; Indicates The month-on-month cash flow volatility of real estate companies; Indicates The net profit margin of a real estate enterprise; , , , and Indicates the weight of each influence; Furthermore, the market risk coefficient is: ; in, Indicates The market risk coefficient of a real estate enterprise; Indicates The volatility of housing prices of real estate companies; Indicates The year-on-year sales volume ratio of real estate enterprises; Indicates The month-on-month sales volume ratio of real estate enterprises; Indicates The year-on-year leasing volume ratio of real estate enterprises; Indicates The month-on-month leasing volume ratio of real estate companies; Indicates Supply contract default rate of real estate enterprises; , , , , and Indicates the weight of each influence; The year-on-year sales volume ratio refers to the ratio of the sales volume of the same month last year to the sales volume of the current month; the month-on-month sales volume ratio refers to the ratio of the sales volume of the previous month to the sales volume of the current month; the year-on-year rental volume ratio refers to the ratio of the rental volume of the same month last year to the rental volume of the current month; the month-on-month rental volume ratio refers to the ratio of the rental volume of the previous month to the rental volume of the current month; Furthermore, the external regulatory risk coefficient is: ; in, Indicates External regulatory risk factor of a real estate enterprise; Indicates The matching degree of financial operation supervision of real estate enterprises; Indicates The tax violation rate of real estate enterprises; and Indicates the weight of each influence; Furthermore, the risk coefficient of real estate development is:

[0022] in, Indicates The risk factor of real estate development of a real estate enterprise; Indicates The progress lag rate of real estate enterprises; Indicates Delayed payment of project funds for real estate enterprises; and Indicates the weight of each influence; This embodiment constructs an enterprise financial risk assessment model to identify and analyze the first comprehensive data set to obtain abnormal financial risk characteristics; abnormal financial risk characteristics include abnormal financial characteristics, abnormal market characteristics, abnormal external regulatory characteristics and abnormal real estate development characteristics; a multi-source financial risk coefficient is obtained based on these four abnormal characteristics; the multi-source financial risk coefficient includes a financial risk coefficient, a market risk coefficient, an external regulatory risk coefficient and a real estate development risk coefficient; by conducting financial analysis of real estate enterprises based on finance, market, external supervision and real estate development, the accuracy and comprehensiveness of financial analysis of real estate enterprises can be effectively achieved, thereby effectively improving the accuracy of generating financial risk reports.

[0023] The financial risk assessment and report generation module is used to generate a comprehensive financial risk score based on multi-source financial risk coefficients, rank each enterprise based on the comprehensive financial risk score, and output a report.

[0024] Furthermore, the comprehensive financial risk score is: ; in, Indicates Comprehensive financial risk score of real estate enterprises; Indicates The financial risk coefficient of a real estate enterprise is used to reflect the impact of financial data on real estate financial risks; Indicates The market risk coefficient of a real estate enterprise; Indicates The external regulatory risk coefficient of each real estate enterprise is used to determine the impact of external regulatory data on real estate financial risks; Indicates The property development risk coefficient of a real estate enterprise is used to reflect the impact of property development progress on real estate financial risks.

[0025] Furthermore, the report includes data presentation of real estate enterprises ranked from high to low in terms of risk scores, including comprehensive financial risk score, financial risk factor, market risk factor, external regulatory risk factor, property development risk factor, property development risk factor and related structured and unstructured data; This embodiment generates a comprehensive financial risk score based on multi-source financial risk coefficients, comprehensively ranks each enterprise based on the comprehensive financial risk score, and generates an automated report. By conducting financial analysis of real estate enterprises based on finance, market, external supervision and property development, it can effectively achieve the accuracy and comprehensiveness of financial analysis of real estate enterprises, thereby effectively improving the accuracy of financial risk report generation.

[0026] In order to verify the effectiveness of the industry report automatic generation system based on a large language model provided by this embodiment, the generation accuracy of the real estate financial risk analysis reports of City A and City B is compared by comparing different systems, including scoring the accuracy of reports generated by different systems according to users, and then obtaining the average accuracy scores of the financial risk analysis reports of real estate enterprises in City A and City B. The specific data are given as shown in Tables 1 and 2 respectively; System 1 is an industry report automatic generation system based on a large language model provided by this embodiment, and System 2 is based on System 1 without considering the first multi-source credibility coefficient and data collection strategy; System 3 is based on System 1 without considering the construction of the enterprise financial analysis and evaluation model; Table 1 Comparison of the generation accuracy of reports from different systems in City A system Average accuracy score System One 96 System 2 89 System Three 91 It can be seen from Table 1 that the automatic industry report generation system based on a large language model provided in this embodiment has certain advantages in the accuracy of report generation in City A.

[0027] Table 2 Comparison of report generation accuracy of different systems in City B system Average accuracy score System One 95 System 2 87 System Three 90 It can be seen from Table 2 that the automatic industry report generation system based on a large language model provided in this embodiment has certain advantages in the accuracy of report generation in City B.

[0028] This embodiment obtains structured and unstructured data of each real estate enterprise through a dynamic multi-source data collection module, and constructs a first multi-source credibility coefficient, and constructs a first comprehensive data set based on the coefficient and the data collection strategy; the risk assessment model construction module constructs an enterprise financial risk assessment model to identify the first comprehensive data set, obtains abnormal characteristics of financial risks, and then generates a multi-source financial risk coefficient; the financial risk assessment and report generation module generates a comprehensive financial risk score according to the multi-source financial risk coefficient, and outputs an automated report according to the ranking of the comprehensive financial risk scores of the real estate enterprises; the present invention can effectively realize the intelligent analysis of the financial risks of real estate enterprises and the automatic generation of reports by constructing an enterprise financial risk assessment model.

[0029] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. An automatic industry report generation system based on a large language model, characterized in that: include: A dynamic multi-source data collection module, which obtains first structured data and first unstructured data of each real estate enterprise, calculates a first multi-source credibility coefficient, and obtains a first comprehensive data set based on the first multi-source credibility coefficient and a data collection strategy; The risk assessment model construction module is used to construct an enterprise financial risk assessment model to identify the first comprehensive data set, which passes through the first input layer, the first preprocessing layer, the financial risk feature extraction layer, the financial risk feature analysis layer and the first output layer of the model in sequence; the financial risk feature extraction layer is used to extract abnormal financial risk features; the abnormal financial risk features include abnormal financial features, abnormal market features, abnormal external regulatory features and abnormal real estate development features; the financial risk feature analysis layer obtains multi-source financial risk coefficients according to the financial risk features; The financial risk assessment and report generation module is used to generate a comprehensive financial risk score based on multi-source financial risk coefficients, rank each enterprise based on the comprehensive financial risk score, and output a report.

2. The automatic industry report generation system based on a large language model according to claim 1 is characterized by: The first structured data includes the enterprise balance sheet, enterprise cash flow statement, enterprise income statement, house sales transaction data, house leasing transaction data, land transaction database and supply chain contract; the first unstructured data includes external regulatory document reports, real estate development progress data and project payment settlement status.

3. The automatic industry report generation system based on a large language model according to claim 2 is characterized by: The first multi-source credibility coefficient is: ； in, is the first multi-source credibility coefficient; is the data freshness attenuation factor influence coefficient; is the data freshness attenuation factor, reflecting the freshness of the data; is the influence coefficient of the platform authority coefficient; is the platform authority coefficient, which is used to evaluate the authority of the data source platform; Scoring impact coefficient for data completeness; Provides a data integrity score to assess the completeness of the data.

4. The automatic industry report generation system based on a large language model according to claim 3 is characterized by: According to the first multi-source credibility coefficient and the preset threshold, the credibility level of each data is obtained, and the credibility level includes high credibility, medium credibility and low credibility; the high-credibility data is fully collected; the medium-credibility data is sampled and collected; the low-credibility data is not collected; and the collected data is then merged to obtain the first comprehensive data set.

5. The automatic industry report generation system based on a large language model according to claim 4 is characterized by: The enterprise financial risk assessment model is constructed based on LLM, and includes a first input layer, a first preprocessing layer, a financial risk feature extraction layer, a financial risk feature analysis layer and a first output layer; The first input layer is used to input the first comprehensive data set into the enterprise financial analysis and evaluation model; The first preprocessing layer performs unified representation modeling on the structured data and unstructured data in the first comprehensive data set based on the natural language processing pre-training paradigm, including data encoding, text standardization and semantic embedding construction, to obtain a second comprehensive data set; The financial risk feature extraction layer is based on the Transformer architecture and uses a multi-head self-attention mechanism to perform context modeling and semantic association analysis on the second comprehensive data set to identify abnormal financial risk features; abnormal financial risk features include abnormal financial features, abnormal market features, abnormal external regulatory features, and abnormal real estate development features; The financial risk characteristic analysis layer obtains multi-source financial risk coefficients according to the four abnormal characteristics; the multi-source financial risk coefficients include financial risk coefficients, market risk coefficients, external regulatory risk coefficients and real estate development risk coefficients; The first output layer is used to output multi-source financial risk coefficients.

6. The automatic industry report generation system based on a large language model according to claim 5 is characterized by: The abnormal financial characteristics include the year-on-year debt-to-asset ratio, the month-on-month debt-to-asset ratio, the year-on-year cash flow volatility, the month-on-month cash flow volatility and the net profit margin; the abnormal market characteristics include the housing price volatility, the year-on-year sales volume ratio, the month-on-month sales volume ratio, the year-on-year rental volume ratio, the month-on-month rental volume ratio and the supply contract default rate; the abnormal external regulatory characteristics include the financial operation supervision matching degree and the tax violation rate; the abnormal real estate development characteristics include the progress lag rate and the delayed payment time of the project funds.

7. The automatic industry report generation system based on a large language model according to claim 6 is characterized by: The comprehensive financial risk score is: ； in, Indicates Comprehensive financial risk score of real estate enterprises; Indicates The financial risk coefficient of a real estate enterprise is used to reflect the impact of financial data on real estate financial risks; Indicates The market risk coefficient of a real estate enterprise; Indicates The external regulatory risk coefficient of each real estate enterprise is used to determine the impact of external regulatory data on real estate financial risks; Indicates The property development risk coefficient of a real estate enterprise is used to reflect the impact of property development progress on real estate financial risks.

Images