A deep analysis method of financial data based on large language model

Abstract

The present invention relates to the technical field of financial data analysis, specifically a method for deep analysis of financial data based on a large language model; collecting historical financial data and historical news data of a target enterprise; identifying the historical financial data to obtain associated enterprises that have financial ties with the target enterprise; and deeply analyzing the historical financial data to identify and obtain financial correlation feature data between the associated enterprises and the target enterprise; identifying news data involving the target enterprise and the associated enterprises in the historical news data to obtain financial correlation news data; identifying the financial correlation news data to obtain financial news feature data; and training a model through financial correlation feature data, financial news feature data and historical financial data to obtain a financial correlation analysis model. The present invention collects and identifies real-time news data through a financial correlation analysis model to accurately obtain financial forecast data.

Description

Technical Field

[0001] The present invention relates to the technical field of financial data analysis, and in particular to a method for deep analysis of financial data based on a large language model. Background Art

[0002] With the continuous advancement of artificial intelligence technology and natural language processing, large language models have become an important tool for processing and analyzing complex data. Especially in the financial field, in-depth analysis and prediction of financial data has gradually become an inevitable trend, and news data, as an important source of information, can reflect market trends, corporate dynamics and changes in the economic environment, and thus have an important impact on corporate financial status and stock market performance.

[0003] Traditional financial analysis systems mostly rely on structured data, such as balance sheets and transaction records; they assess a company's financial risks through statistical models or shallow machine learning algorithms.

[0004] However, this method has two major flaws: first, it does not include unstructured news data, resulting in insufficient sensitivity to news events, such as policy changes, industry public opinion, and corporate announcements; second, the extraction of financial correlation features is limited to transaction frequency and amount, lacking in-depth quantification of capital flow patterns. In summary, existing technologies have not yet achieved the coordinated identification of the intrinsic correlation features of financial data and the external influencing factors of news.

[0005] To this end, a deep analysis method of financial data based on a large language model is proposed. Summary of the invention

[0006] The purpose of the present invention is to provide a financial data in-depth analysis method based on a large language model, which collects historical financial data and historical news data of a target enterprise; identifies the historical financial data to obtain associated enterprises that have financial ties with the target enterprise; and deeply analyzes the historical financial data to identify and obtain financial correlation feature data between the associated enterprises and the target enterprise; identifies news data involving the target enterprise and associated enterprises in the historical news data to obtain financial correlation news data; identifies the financial correlation news data to obtain financial news feature data; and performs model training through financial correlation feature data, financial news feature data and historical financial data to obtain a financial correlation analysis model. The present invention collects and identifies real-time news data through a financial correlation analysis model to accurately obtain financial forecast data.

[0007] To achieve the above object, the present invention provides the following technical solutions:

[0008] A method for deep analysis of financial data based on a large language model, comprising:

[0009] S10. Collect historical financial data and historical news data of the target enterprise;

[0010] S20. Identify the historical financial data to obtain the associated enterprises that have financial ties with the target enterprise; and perform in-depth analysis on the historical financial data to identify the financial correlation feature data between the associated enterprises and the target enterprise; the financial correlation feature data includes capital density data, transaction activity data, term coupling data, and chain conduction data;

[0011] S30. Identify the news data related to the target enterprise and its related enterprises in the historical news data to obtain financial related news data; identify the financial related news data to obtain financial news feature data; the financial news feature data includes news content feature data, news dissemination feature data and news announcement feature data;

[0012] S40. Performing model training through financial correlation feature data, financial news feature data and historical financial data to obtain a financial correlation analysis model;

[0013] S50. Collect and identify real-time news data through a financial correlation analysis model to obtain financial forecast data.

[0014] The identification process of the financial association feature data is as follows:

[0015] Identify the historical financial data to obtain related enterprises; identify the financial data of the target enterprise and related enterprises in the historical financial data to obtain capital density data, transaction activity data, term coupling data and chain conduction data;

[0016] The capital density data is calculated through the working capital of the target enterprise and the working capital between the target enterprise and its affiliated enterprises during the period;

[0017] The transaction activity data is calculated by the number of transactions between the target enterprise and its affiliated enterprises within the period, the total number of transactions of the target enterprise within the period, and the average number of periodic transactions of the industry;

[0018] The term coupling data is obtained by identifying the time data of the time transaction process between the target enterprise and the associated enterprises;

[0019] The chain conduction data is obtained by identifying the cash flow nodes of the transaction;

[0020] Financial correlation characteristic data are obtained through capital density data, transaction activity data, term coupling data and chain conduction data.

[0021] The process of obtaining the financial related news data:

[0022] Identify and obtain news keywords based on the target enterprise and related enterprises; retrieve the historical news data based on the news keywords to obtain first news retrieval data;

[0023] Identifying and dividing the content of the first news retrieval data, dividing news with the same content together, and obtaining second news retrieval data;

[0024] Identifying the second news retrieval data to obtain finance-related news data;

[0025] The financial related news data includes news content data, news sending information, news forwarding information, news comment information and news announcement information.

[0026] The process of identifying financial news related data and obtaining financial news feature data is as follows:

[0027] Identify news content data by identifying positive sentiment words, negative sentiment words, policy-sensitive words, and industry terminology words to obtain news content feature data;

[0028] By identifying news sending information, news forwarding information and news comment information, news dissemination characteristic data is obtained; the news dissemination characteristic data includes dissemination heat data characteristics and dissemination path data characteristics;

[0029] The news announcement information is identified to obtain news announcement feature data; the news announcement information includes announcement content data and announcement comment data.

[0030] In the process of training the financial correlation analysis model, the identification process of financial news feature data and financial correlation feature data is as follows:

[0031] Obtain the release timeline data of news announcements in the financial news feature data;

[0032] The news content feature data and the news dissemination feature data are divided according to the release timeline data to obtain announcement division subsets;

[0033] The news impact characteristic data is obtained by identifying the news content characteristic data and the news dissemination characteristic data in the announcement division subsets; and the changes of the news content characteristic data and the news dissemination characteristic data between the announcement division subsets;

[0034] The financial correlation analysis model is trained by using the influence rules of news impact feature data on financial correlation feature data.

[0035] The financial relevance analysis model includes a real-time news acquisition layer, a news feature recognition layer, a direct impact prediction layer, a news feature comparison layer and a financial impact calibration layer;

[0036] The real-time news acquisition layer is used to acquire real-time news data and initial financial data within a period, and identify and divide the news content to obtain a real-time news data set;

[0037] The news feature recognition layer recognizes the features of the real-time news data set to obtain news content feature data, news dissemination feature data and news announcement feature data;

[0038] The direct impact prediction layer obtains news impact feature data by identifying news content feature data, news dissemination feature data and news announcement feature data; obtains first predicted financial data by identifying news impact feature data and initial financial data;

[0039] The news feature comparison layer filters out similar data from historical data based on the similarity of news features;

[0040] The financial impact calibration layer corrects the first predicted financial data according to the variation rules of the financial data in the corresponding period of historical news to obtain the financial forecast data.

[0041] Compared with the prior art, the present invention has the following beneficial effects:

[0042] 1. This application identifies the financial data between the target enterprise and its related enterprises, and obtains the financial correlation feature data through the identification of capital density data, transaction activity data, term coupling data and chain transmission data; among them, the transaction activity data is combined with the industry average to support cross-enterprise and cross-industry horizontal comparative analysis and improve the objectivity of the assessment; the introduction of chain transmission data can identify potential vulnerable links in cash flow nodes and improve the sensitivity of data identification; it can effectively quantify the complex financial connections between enterprises and enhance the ability to identify risk transmission and correlation impacts.

[0043] 2. The present invention obtains news keywords by identifying target enterprises and related enterprises, and retrieves and divides historical news data through news keywords, including financial related news data, including news content data, news issuance information, news forwarding information, news commentary information and news announcement information; then the financial related news data is identified to obtain news content feature data, news dissemination feature data and news announcement feature data, so as to accurately identify relevant news data.

[0044] 3. This application obtains the release timeline data of news announcements in the financial news feature data; divides the news content feature data and the news dissemination feature data according to the release timeline data to obtain the announcement subsets; obtains the news impact feature data by identifying the news content feature data and the news dissemination feature data within the announcement subsets; and the changes in the news content feature data and the news dissemination feature data between the announcement subsets; and obtains the financial correlation analysis model by training the impact rules of the news impact feature data on the financial correlation feature data.

[0045] 4. The present application collects and identifies real-time news data through a financial correlation analysis model to obtain real-time news data sets and initial financial data within a period, identifies the characteristics of the real-time news data sets, obtains news content characteristic data, news dissemination characteristic data and news announcement characteristic data, and further identifies news impact characteristic data; identifies the first predicted financial data through the news impact characteristic data and the initial financial data; and accurately obtains financial forecast data through comparison and correction of historical data. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 A flowchart of a method for deep analysis of financial data based on a large language model according to the present invention;

[0047] Figure 2 It is a schematic diagram of the training process of the financial association analysis model of the present invention;

[0048] Figure 3 It is a schematic diagram of the structure of the financial association analysis model of the present invention. DETAILED DESCRIPTION

[0049] 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.

[0050] Embodiment 1

[0051] The present invention proposes a financial data in-depth analysis method based on a large language model, the process of which is as follows: Figure 1 As shown, including:

[0052] S10. Collect and obtain historical financial data and historical news data of the target enterprise.

[0053] S20. Identify the historical financial data to obtain the associated enterprises that have financial ties with the target enterprise; and conduct an in-depth analysis of the historical financial data to identify the financial correlation feature data between the associated enterprises and the target enterprise; the financial correlation feature data includes capital density data, transaction activity data, term coupling data and chain conduction data.

[0054] The identification process of the financial association feature data is as follows:

[0055] Identify the historical financial data to obtain related enterprises; identify the financial data of the target enterprise and related enterprises in the historical financial data to obtain capital density data, transaction activity data, term coupling data and chain conduction data;

[0056] The capital density data is calculated through the working capital of the target enterprise and the working capital between the target enterprise and its affiliated enterprises during the period; the capital density index can be calculated through the capital density data, and its calculation formula is:

[0057] ;

[0058] in, is the capital density index, A is the target enterprise, and B is the associated enterprise; Represents the hyperbolic tangent function, which is used to compress the capital density index to between 0 and 1, and can effectively avoid interference from extreme values; It represents the total absolute value of the transaction amount between target enterprise A and related enterprise B during the period; It represents the total outflow of funds of target enterprise A; Represents the total inflow of funds from affiliated company B.

[0059] The capital density index can effectively quantify the concentration of capital flows and identify the degree of capital correlation between enterprises.

[0060] The transaction activity data is calculated by the number of transactions between the target enterprise and its affiliated enterprises within the period, the total number of transactions of the target enterprise within the period, and the average number of periodic transactions of the industry;

[0061] The transaction activity index can be calculated through the transaction activity data. The calculation formula of the transaction activity index is:

[0062] ;

[0063] in, Represents the transaction activity index between target enterprise A and associated enterprise B; It is a limit function, which limits the trading activity index between 0 and 1; Indicates the number of transactions between target enterprise A and associated enterprise B; Indicates the average monthly number of transactions in the industry; is the standard deviation of the number of transactions.

[0064] The transaction activity index can be used to identify abnormal fluctuations in transaction frequency and reflect the closeness of the business.

[0065] The term coupling data is obtained by identifying the time data of the time transaction process between the target enterprise and the associated enterprises;

[0066] The term coupling index can be calculated through the term coupling data, and its calculation formula is:

[0067] ;

[0068] in, It represents the term coupling index between target enterprise A and associated enterprise B; Indicates the time interval between transactions between target enterprise A and associated enterprise B; Indicates the standard deviation of the trading time interval; Indicates the standard deviation threshold.

[0069] The term coupling index can be used to assess the regularity of transactions and the stability of cooperation between enterprises.

[0070] The chain conduction data is obtained by identifying the cash flow nodes of the transaction; the chain conduction coefficient can be identified through the chain conduction data; the calculation formula of the chain conduction coefficient is:

[0071] ;

[0072] in, It represents the chain transmission coefficient between target enterprise A and related enterprise B; represents the logarithmic function; represents the path attenuation coefficient, which is determined based on data verification; It indicates the length of the funds transmission path, which is determined by the funds transfer path; Represents a natural constant.

[0073] The chain transmission coefficient can reflect the path risk of transactions between enterprises.

[0074] Financial correlation characteristic data are obtained through capital density data, transaction activity data, term coupling data and chain conduction data.

[0075] This application identifies the financial data between the target enterprise and its affiliated enterprises, and obtains financial correlation feature data through the identification of capital density data, transaction activity data, term coupling data and chain transmission data; among them, the transaction activity data is combined with the industry average to support cross-enterprise and cross-industry horizontal comparative analysis and improve the objectivity of the assessment; the introduction of chain transmission data can identify potential vulnerable links in cash flow nodes and improve the sensitivity of data identification; it can effectively quantify the complex financial connections between enterprises and enhance the ability to identify risk transmission and correlation impacts.

[0076] S30. Identify the news data related to the target enterprise and related enterprises in the historical news data to obtain financial related news data; identify the financial related news data to obtain financial news feature data; the financial news feature data includes news content feature data, news dissemination feature data and news announcement feature data.

[0077] The process of obtaining the financial related news data:

[0078] Identify and obtain news keywords based on the target enterprise and related enterprises; retrieve the historical news data based on the news keywords to obtain first news retrieval data;

[0079] Identifying and dividing the content of the first news retrieval data, dividing news with the same content together, and obtaining second news retrieval data;

[0080] Identifying the second news retrieval data to obtain finance-related news data;

[0081] The financial related news data includes news content data, news sending information, news forwarding information, news comment information and news announcement information.

[0082] The process of identifying financial news related data and obtaining financial news feature data is as follows:

[0083] Identify news content data by identifying positive sentiment words, negative sentiment words, policy-sensitive words, and industry terminology words to obtain news content feature data;

[0084] By identifying news sending information, news forwarding information and news comment information, news dissemination characteristic data is obtained; the news dissemination characteristic data includes dissemination heat data characteristics and dissemination path data characteristics;

[0085] The news announcement information is identified to obtain news announcement feature data; the news announcement information includes announcement content data and announcement comment data.

[0086] The present invention obtains news keywords by identifying target enterprises and related enterprises, and retrieves and divides historical news data through news keywords, including financial related news data, including news content data, news issuance information, news forwarding information, news commentary information and news announcement information; then the financial related news data is identified to obtain news content feature data, news dissemination feature data and news announcement feature data, so as to accurately identify related news data.

[0087] S40. Perform model training through financial correlation feature data, financial news feature data and historical financial data to obtain a financial correlation analysis model.

[0088] The training of financial correlation analysis model is as follows Figure 2 As shown in the figure, the identification process of financial news feature data and financial related feature data is as follows:

[0089] Obtain the release timeline data of news announcements in the financial news feature data;

[0090] The news content feature data and the news dissemination feature data are divided according to the release timeline data to obtain announcement division subsets;

[0091] The news impact characteristic data is obtained by identifying the news content characteristic data and the news dissemination characteristic data in the announcement division subsets; and the changes of the news content characteristic data and the news dissemination characteristic data between the announcement division subsets;

[0092] The changes in news content feature data and news dissemination feature data between announcement subsets include the differences in forwarded content, the number of forwarding times, the forwarding platform, and the forwarding time after the announcement is released, as well as the changes in the number of comments and the emotional color.

[0093] The financial correlation analysis model is trained by using the influence rules of news impact feature data on financial correlation feature data.

[0094] This application obtains the release timeline data of news announcements in the financial news feature data; divides the news content feature data and the news dissemination feature data according to the release timeline data to obtain the announcement division subsets; obtains the news impact feature data by identifying the news content feature data and the news dissemination feature data within the announcement division subsets; and the changes in the news content feature data and the news dissemination feature data between the announcement division subsets; and obtains the financial correlation analysis model through training through the influence rules of the news impact feature data on the financial correlation feature data.

[0095] S50. Collect and identify real-time news data through a financial correlation analysis model to obtain financial forecast data.

[0096] The financial association analysis model is constructed based on the large language model, and its structure is as follows: Figure 3 As shown, it includes a real-time news acquisition layer, a news feature recognition layer, a direct impact prediction layer, a news feature comparison layer and a financial impact calibration layer;

[0097] The real-time news acquisition layer is used to acquire real-time news data and initial financial data within a period, and identify and divide the news content to obtain a real-time news data set;

[0098] The news feature recognition layer recognizes the features of the real-time news data set to obtain news content feature data, news dissemination feature data and news announcement feature data;

[0099] The direct impact prediction layer obtains news impact feature data by identifying news content feature data, news dissemination feature data and news announcement feature data; obtains first predicted financial data by identifying news impact feature data and initial financial data;

[0100] The news feature comparison layer filters out similar data from historical data based on the similarity of news features;

[0101] The financial impact calibration layer corrects the first predicted financial data according to the variation rules of the financial data in the corresponding period of historical news to obtain the financial forecast data.

[0102] The present application collects and identifies real-time news data through a financial correlation analysis model, which is used to obtain real-time news data sets and initial financial data within a period, identify the characteristics of the real-time news data sets, obtain news content characteristic data, news dissemination characteristic data and news announcement characteristic data, and further identify news impact characteristic data; through the news impact characteristic data and the initial financial data, the first predicted financial data is identified; and then by comparing and correcting historical data, the financial forecast data is accurately obtained.

[0103] The present invention collects historical financial data and historical news data of a target enterprise; identifies the historical financial data to obtain associated enterprises that have financial ties with the target enterprise; and deeply analyzes the historical financial data to identify and obtain financial correlation feature data between the associated enterprises and the target enterprise; identifies news data involving the target enterprise and associated enterprises in the historical news data to obtain financial correlation news data; identifies the financial correlation news data to obtain financial news feature data; and performs model training through financial correlation feature data, financial news feature data and historical financial data to obtain a financial correlation analysis model. Real-time news data is collected and identified through the financial correlation analysis model to accurately obtain financial forecast data.

[0104] Embodiment 2

[0105] The present invention proposes a financial data in-depth analysis method based on a large language model, the process of which is as follows: Figure 1 As shown, including:

[0106] S10. Collect and obtain historical financial data and historical news data of the target enterprise.

[0107] S20. Identify the historical financial data to obtain the associated enterprises that have financial ties with the target enterprise; and conduct an in-depth analysis of the historical financial data to identify the financial correlation feature data between the associated enterprises and the target enterprise; the financial correlation feature data includes capital density data, transaction activity data, term coupling data and chain conduction data.

[0108] The identification process of the financial association feature data is as follows:

[0109] Identify the historical financial data to obtain related enterprises; identify the financial data of the target enterprise and related enterprises in the historical financial data to obtain capital density data, transaction activity data, term coupling data and chain conduction data;

[0110] The capital density data is calculated through the working capital of the target enterprise and the working capital between the target enterprise and its affiliated enterprises during the period;

[0111] The transaction activity data is calculated by the number of transactions between the target enterprise and its affiliated enterprises within the period, the total number of transactions of the target enterprise within the period, and the average number of periodic transactions of the industry;

[0112] The term coupling data is obtained by identifying the time data of the time transaction process between the target enterprise and the associated enterprises;

[0113] The chain conduction data is obtained by identifying the cash flow nodes of the transaction;

[0114] Financial correlation characteristic data are obtained through capital density data, transaction activity data, term coupling data and chain conduction data.

[0115] This application identifies the financial data between the target enterprise and its affiliated enterprises, and obtains financial correlation feature data through the identification of capital density data, transaction activity data, term coupling data and chain transmission data; among them, the transaction activity data is combined with the industry average to support cross-enterprise and cross-industry horizontal comparative analysis and improve the objectivity of the assessment; the introduction of chain transmission data can identify potential vulnerable links in cash flow nodes and improve the sensitivity of data identification; it can effectively quantify the complex financial connections between enterprises and enhance the ability to identify risk transmission and correlation impacts.

[0116] S30. Identify the news data related to the target enterprise and related enterprises in the historical news data to obtain financial related news data; identify the financial related news data to obtain financial news feature data; the financial news feature data includes news content feature data, news dissemination feature data and news announcement feature data.

[0117] The process of obtaining the financial related news data:

[0118] Identify and obtain news keywords based on the target enterprise and related enterprises; retrieve the historical news data based on the news keywords to obtain first news retrieval data;

[0119] Identifying and dividing the content of the first news retrieval data, dividing news with the same content together, and obtaining second news retrieval data;

[0120] Identifying the second news retrieval data to obtain finance-related news data;

[0121] The financial related news data includes news content data, news sending information, news forwarding information, news comment information and news announcement information.

[0122] The news content data includes the first published news content and the forwarded news content;

[0123] The news release information includes the release time, release platform and release author of the first released news;

[0124] The news forwarding information includes the forwarding time, forwarding platform and forwarding author during the news forwarding process;

[0125] The news comment data refers to the comment data on the publishing platform and forwarding platform;

[0126] The news announcement information refers to the relevant announcements issued by the related companies after the news is released, as well as the comment data after the announcement is issued;

[0127] The news sending and forwarding data in the historical news data are collected, and the obtained data are shown in Table 1.

[0128] Table 1 News forwarding data table

[0129]

[0130] The process of identifying financial news related data and obtaining financial news feature data is as follows:

[0131] Identify news content data by identifying positive sentiment words, negative sentiment words, policy-sensitive words, and industry terminology words to obtain news content feature data;

[0132] By identifying news sending information, news forwarding information and news comment information, news dissemination characteristic data is obtained; the news dissemination characteristic data includes dissemination heat data characteristics and dissemination path data characteristics;

[0133] The news announcement information is identified to obtain news announcement feature data; the news announcement information includes announcement content data and announcement comment data.

[0134] The present invention obtains news keywords by identifying target enterprises and related enterprises, and retrieves and divides historical news data through news keywords, including financial related news data, including news content data, news issuance information, news forwarding information, news commentary information and news announcement information; then the financial related news data is identified to obtain news content feature data, news dissemination feature data and news announcement feature data, so as to accurately identify related news data.

[0135] S40. Perform model training through financial correlation feature data, financial news feature data and historical financial data to obtain a financial correlation analysis model.

[0136] During the training of the financial correlation analysis model, the identification process of financial news feature data and financial correlation feature data is as follows:

[0137] Obtain the release timeline data of news announcements in the financial news feature data;

[0138] The news content feature data and the news dissemination feature data are divided according to the release timeline data to obtain announcement division subsets;

[0139] The news impact characteristic data is obtained by identifying the news content characteristic data and the news dissemination characteristic data in the announcement division subsets; and the changes of the news content characteristic data and the news dissemination characteristic data between the announcement division subsets;

[0140] The financial correlation analysis model is trained by using the influence rules of news impact feature data on financial correlation feature data.

[0141] In order to verify the recognition effect of the financial correlation analysis model, model validation is performed on it, including validation model 1, validation model 2 and validation model 3;

[0142] The verification model 1 is a financial correlation analysis model obtained by training in this application, including dividing the news content feature data and the news dissemination feature data according to the release timeline data to obtain the announcement division subsets; obtaining the news impact feature data by identifying the news content feature data and the news dissemination feature data within the announcement division subsets; and the changes in the news content feature data and the news dissemination feature data between the announcement division subsets; and training the influence rules of the news impact feature data on the financial correlation feature data.

[0143] The verification model 2 is based on the verification model 1 and does not consider the data differences between the announcement subsets; it includes dividing the news content feature data and the news dissemination feature data according to the release timeline data to obtain the announcement subsets; obtaining the news impact feature data by identifying the news content feature data and the news dissemination feature data in the announcement subsets; and training the influence rules of the news impact feature data on the financial correlation feature data.

[0144] Based on verification model one, verification model three directly identifies news content feature data and news dissemination feature data to obtain news impact feature data without considering the division of announcement data; and trains the influence rules of news impact feature data on financial correlation feature data.

[0145] Verification model 1, verification model 2 and verification model 3 are verified, and the data similarity between the recognition data and the standard data is used as the recognition accuracy to obtain data table 2.

[0146] Table 2 Financial correlation analysis model verification data table

[0147]

[0148] Through model verification, it can be found that the recognition effect of verification model 1 is the best.

[0149] This application obtains the release timeline data of news announcements in the financial news feature data; divides the news content feature data and the news dissemination feature data according to the release timeline data to obtain the announcement division subsets; obtains the news impact feature data by identifying the news content feature data and the news dissemination feature data within the announcement division subsets; and the changes in the news content feature data and the news dissemination feature data between the announcement division subsets; and obtains the financial correlation analysis model through training through the influence rules of the news impact feature data on the financial correlation feature data.

[0150] S50. Collect and identify real-time news data through a financial correlation analysis model to obtain financial forecast data.

[0151] The financial relevance analysis model includes a real-time news acquisition layer, a news feature recognition layer, a direct impact prediction layer, a news feature comparison layer and a financial impact calibration layer;

[0152] The real-time news acquisition layer is used to acquire real-time news data and initial financial data within a period, and identify and divide the news content to obtain a real-time news data set;

[0153] The news feature recognition layer recognizes the features of the real-time news data set to obtain news content feature data, news dissemination feature data and news announcement feature data;

[0154] The direct impact prediction layer obtains news impact feature data by identifying news content feature data, news dissemination feature data and news announcement feature data; obtains first predicted financial data by identifying news impact feature data and initial financial data;

[0155] The news feature comparison layer filters out similar data from historical data based on the similarity of news features;

[0156] The financial impact calibration layer corrects the first predicted financial data according to the variation rules of the financial data in the corresponding period of historical news to obtain the financial forecast data.

[0157] The present application collects and identifies real-time news data through a financial correlation analysis model, which is used to obtain real-time news data sets and initial financial data within a period, identify the characteristics of the real-time news data sets, obtain news content characteristic data, news dissemination characteristic data and news announcement characteristic data, and further identify news impact characteristic data; through the news impact characteristic data and the initial financial data, the first predicted financial data is identified; and then by comparing and correcting historical data, the financial forecast data is accurately obtained.

[0158] 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. A method for deep analysis of financial data based on a large language model, characterized in that: include: S10. Collect historical financial data and historical news data of the target enterprise; S20. Identify the historical financial data to obtain the associated enterprises that have financial ties with the target enterprise; and perform in-depth analysis on the historical financial data to identify the financial correlation feature data between the associated enterprises and the target enterprise; the financial correlation feature data includes capital density data, transaction activity data, term coupling data, and chain conduction data; The identification process of the financial association feature data is as follows: Identify the historical financial data to obtain related enterprises; identify the financial data of the target enterprise and related enterprises in the historical financial data to obtain capital density data, transaction activity data, term coupling data and chain conduction data; The capital density data is calculated through the working capital of the target enterprise and the working capital between the target enterprise and its affiliated enterprises during the period; The transaction activity data is calculated by the number of transactions between the target enterprise and its affiliated enterprises within the period, the total number of transactions of the target enterprise within the period, and the average number of periodic transactions of the industry; The term coupling data is obtained by identifying the time data of the time transaction process between the target enterprise and the associated enterprises; The chain conduction data is obtained by identifying the cash flow nodes of the transaction; Financial correlation characteristic data are obtained through capital density data, transaction activity data, term coupling data and chain transmission data; S30. Identify news data related to the target enterprise and its related enterprises in the historical news data to obtain financial related news data; Identify financial related news data to obtain financial news feature data; the financial news feature data includes news content feature data, news dissemination feature data and news announcement feature data; S40. Performing model training through financial correlation feature data, financial news feature data and historical financial data to obtain a financial correlation analysis model; In the process of training the financial correlation analysis model, the identification process of financial news feature data and financial correlation feature data is as follows: Get the release timeline data of news announcements in the financial news feature data; The news content feature data and the news dissemination feature data are divided according to the release timeline data to obtain announcement division subsets; The news impact characteristic data is obtained by identifying the news content characteristic data and the news dissemination characteristic data in the announcement division subsets; and the changes of the news content characteristic data and the news dissemination characteristic data between the announcement division subsets; Through the influence of news impact feature data on financial correlation feature data, a financial correlation analysis model is trained; S50. Collect and identify real-time news data through a financial correlation analysis model to obtain financial forecast data.

2. The method for deep analysis of financial data based on a large language model according to claim 1, characterized in that: The acquisition process of the financial related news data: Identify and obtain news keywords based on the target enterprise and related enterprises; retrieve the historical news data based on the news keywords to obtain first news retrieval data; Identifying and dividing the content of the first news retrieval data, dividing news with the same content together, and obtaining second news retrieval data; Identifying the second news retrieval data to obtain finance-related news data; The financial related news data includes news content data, news sending information, news forwarding information, news comment information and news announcement information.

3. The method for deep analysis of financial data based on a large language model according to claim 2, characterized in that: The process of identifying financial related news data and obtaining financial news feature data is as follows: Identify news content data by identifying positive sentiment words, negative sentiment words, policy-sensitive words, and industry terminology words to obtain news content feature data; By identifying news sending information, news forwarding information and news comment information, news dissemination characteristic data is obtained; the news dissemination characteristic data includes dissemination heat data characteristics and dissemination path data characteristics; The news announcement information is identified to obtain news announcement feature data; the news announcement information includes announcement content data and announcement comment data.

4. The method for deep analysis of financial data based on a large language model according to claim 1, characterized in that: The financial relevance analysis model includes a real-time news acquisition layer, a news feature recognition layer, a direct impact prediction layer, a news feature comparison layer and a financial impact calibration layer; The real-time news acquisition layer is used to acquire real-time news data and initial financial data within a period, and identify and divide the news content to obtain a real-time news data set; The news feature recognition layer recognizes the features of the real-time news data set to obtain news content feature data, news dissemination feature data and news announcement feature data; The direct impact prediction layer obtains news impact feature data by identifying news content feature data, news dissemination feature data and news announcement feature data; obtains first predicted financial data by identifying news impact feature data and initial financial data; The news feature comparison layer filters out similar data from historical data based on the similarity of news features; The financial impact calibration layer corrects the first predicted financial data according to the variation rules of the financial data in the corresponding period of historical news to obtain the financial forecast data.

Images