CN117235233A - Automatic financial report question-answering method and device based on large model - Google Patents
Automatic financial report question-answering method and device based on large model Download PDFInfo
- Publication number
- CN117235233A CN117235233A CN202311379778.3A CN202311379778A CN117235233A CN 117235233 A CN117235233 A CN 117235233A CN 202311379778 A CN202311379778 A CN 202311379778A CN 117235233 A CN117235233 A CN 117235233A
- Authority
- CN
- China
- Prior art keywords
- financial
- large model
- model
- financial report
- question
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 46
- 238000004458 analytical method Methods 0.000 claims abstract description 49
- 238000012549 training Methods 0.000 claims abstract description 25
- 238000005457 optimization Methods 0.000 claims abstract description 21
- 230000007246 mechanism Effects 0.000 claims abstract description 17
- 238000005516 engineering process Methods 0.000 claims description 19
- 238000004590 computer program Methods 0.000 claims description 12
- 238000010276 construction Methods 0.000 claims description 12
- 238000012360 testing method Methods 0.000 claims description 9
- 238000004364 calculation method Methods 0.000 claims description 8
- 238000012545 processing Methods 0.000 claims description 8
- 238000003860 storage Methods 0.000 claims description 7
- 230000008569 process Effects 0.000 claims description 4
- 239000000463 material Substances 0.000 claims description 3
- 230000003993 interaction Effects 0.000 claims description 2
- 230000002787 reinforcement Effects 0.000 claims description 2
- 230000006870 function Effects 0.000 description 6
- 230000000694 effects Effects 0.000 description 5
- 239000000284 extract Substances 0.000 description 5
- 238000000605 extraction Methods 0.000 description 5
- 238000012015 optical character recognition Methods 0.000 description 5
- 238000013528 artificial neural network Methods 0.000 description 4
- 238000004422 calculation algorithm Methods 0.000 description 4
- 238000010586 diagram Methods 0.000 description 3
- 230000009286 beneficial effect Effects 0.000 description 2
- 238000010801 machine learning Methods 0.000 description 2
- 238000013507 mapping Methods 0.000 description 2
- 238000003058 natural language processing Methods 0.000 description 2
- 230000002159 abnormal effect Effects 0.000 description 1
- 238000007792 addition Methods 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000006399 behavior Effects 0.000 description 1
- 238000010224 classification analysis Methods 0.000 description 1
- 238000007405 data analysis Methods 0.000 description 1
- 238000013480 data collection Methods 0.000 description 1
- 238000003066 decision tree Methods 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000005206 flow analysis Methods 0.000 description 1
- 238000007477 logistic regression Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 239000000203 mixture Substances 0.000 description 1
- 238000007781 pre-processing Methods 0.000 description 1
- 238000007637 random forest analysis Methods 0.000 description 1
- 230000004044 response Effects 0.000 description 1
- 238000012502 risk assessment Methods 0.000 description 1
- 230000011218 segmentation Effects 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 238000012706 support-vector machine Methods 0.000 description 1
- 230000007474 system interaction Effects 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Landscapes
- Financial Or Insurance-Related Operations Such As Payment And Settlement (AREA)
Abstract
The invention discloses a large model-based financial accounting automatic question-answering method and device, wherein the method comprises the following steps: pre-training and fine-tuning a basic open source large model based on a financial data set and calendar financial report data, and constructing a financial report large model with financial question-answering capability; uploading a financial report file by a user, inputting a problem to a financial report large model, analyzing the financial report file by the financial report large model, generating an answer and returning the answer to the user; constructing a database, wherein the database comprises the calendar financial report data, financial report file analysis results and scoring results for answers; carrying out iterative optimization on the financial report large model by using information in the database through a feedback iterative mechanism; and performing automated financial report question answering by using the optimized financial report big model. The method can effectively improve the accuracy and efficiency of the financial analysis, and is suitable for intelligent questioning and answering application scenes of financial data in the financial field.
Description
Technical Field
The invention belongs to the technical field of financial newspaper data processing, and particularly relates to a financial newspaper automatic question-answering method and device based on a large model.
Background
With the continued advancement of artificial intelligence and natural language processing technologies, more and more fields begin to apply these technologies to improve work efficiency and accuracy. In the financial field, a financial statement is a highly specialized document containing a large amount of financial data and information, in order to help users to better understand and analyze the financial statement, researchers develop a series of financial statement analysis methods, traditional financial statement analysis needs to manually read and analyze cumbersome financial statements, the method is low in efficiency and easy to cause errors, and at present, the financial statement analysis is developing to an automatic and intelligent direction so as to improve the accuracy and efficiency of analysis.
There are some automatic methods for analyzing financial reports, such as using machine learning algorithm to perform classification analysis on the financial reports, including decision trees, neural networks, support vector machines, etc., learn association rules, trends and abnormal points of various indexes from a large amount of financial report data, mine important features such as income composition, cost structure, stakeholder rights and interests, etc., and perform corresponding financial ratio calculation and analysis, including financial leverage ratio, profit margin, cash flow analysis, etc., so as to more comprehensively and accurately understand the business condition and trend of the enterprise, and be beneficial to making countermeasures and decisions. And the machine learning can be used for carrying out financial prediction and risk assessment, and the historical behavior data, financial data, social media data and other information of the client are analyzed through random forest, logistic regression, naive Bayesian algorithm and other algorithms to predict the future credit risk.
The Chinese patent application with publication number of CN113779940A discloses a method for generating analysis comment text based on financial data, which constructs a reference text and a text-form mapping vector of financial comment text according to mathematical characteristics of enterprise financial forms and word segmentation results of the financial comment text, builds a neural network text generation model, trains the neural network text generation model and uses the neural network text generation model to generate text. According to the method, financial comments can be automatically generated according to the financial report, the production efficiency of analysis comment texts in the financial field is improved, and investors are helped to know the main content of the financial report. However, the method can only extract data based on the preset reference text according to the mapping relation in the model to generate the final financial comment text, and cannot answer the specific questions presented by the user autonomously.
The Chinese patent application with the publication number of CN116542800A discloses an intelligent financial statement analysis system based on cloud AI technology, cloud storage financial statement data is built through a cloud architecture unit, voice instructions are sent by a user, a financial statement extraction unit extracts corresponding financial statement data, the control effect is improved through targeted analysis and calculation of financial indexes of the financial statement data corresponding to the voice instructions of the user by an index analysis unit, targeted analysis is facilitated, and meanwhile, visual presentation of the financial statement data by an intelligent decision unit is improved in accuracy. However, the method only extracts corresponding financial indexes for the voice instruction of the user, and cannot answer various user questions.
The above methods still have some challenges in processing complex and diversified financial statement data, such as data collection, data processing, data analysis, etc., which require a great deal of human intervention and expertise, and these methods generally require a great deal of data training, and cannot answer the specific questions of the user's diversification. In recent years, with the wide application of large model technology in the field of natural language processing, a corpus fine tuning model in a specific format is used to enable a basic large model to have a question-answering function, and on the basis of the question-answering function, a method and a device capable of intelligently question-answering financial reports are needed in the financial field.
Disclosure of Invention
In view of the above, the present invention aims to provide a large model-based automated financial questioning and answering method and apparatus, which obtain a large financial questioning and answering model through pre-training and fine tuning of a basic open source large model, extract information and search information by combining calendar financial data through a financial file analysis technology, generate answers according to user questioning, score answers, realize iterative optimization of the large financial models through a feedback iteration mechanism, improve accuracy and efficiency of financial analysis, and are suitable for intelligent questioning and answering application scenarios of financial data in the financial field.
In order to achieve the above purpose, the technical scheme provided by the invention is as follows:
in a first aspect, the present invention provides a large model-based financial accounting automated question-answering method, which includes the following steps:
pre-training and fine-tuning a basic open source large model based on a financial data set and calendar financial report data, and constructing a financial report large model with financial question-answering capability;
uploading a financial report file by a user, inputting a problem to a financial report large model, analyzing the financial report file by the financial report large model, generating an answer and returning the answer to the user;
constructing a database, wherein the database comprises the calendar financial report data, financial report file analysis results and scoring results for answers;
carrying out iterative optimization on the financial report large model by using information in the database through a feedback iterative mechanism;
and performing automated financial report question answering by using the optimized financial report big model.
Preferably, the pre-training and fine-tuning are performed on the basic open source large model based on the financial data set and the calendar data, and the constructing of the financial large model with the financial question-answering capability includes:
constructing a financial data set based on financial teaching materials and data, and pre-training a basic open-source large model to obtain a pre-training large model with financial knowledge;
collecting a calendar year financial report file, analyzing the file to extract financial information, and constructing fine tuning data based on the financial information, wherein the fine tuning data comprises a question-answer data set with relevant information and standard answers required for answering a question;
and taking fine adjustment data based on the financial information as input, and performing supervision fine adjustment on the pre-trained large model so as to obtain the financial report large model with the capability of financial question answering.
Preferably, the user uploads a financial report file and inputs a problem to a financial report big model, and the financial report file is subjected to file analysis and answer generation through the financial report big model and returned to the user, which comprises the following steps:
inputting the financial report file and the problems into a financial report large model, analyzing the financial report file by a file analysis module of the financial report large model by utilizing a file analysis technology, extracting graphic and text information in the financial report and storing the graphic and text information in a database;
the method comprises the steps of extracting information from image-text information in a database through a question and answer module of a financial report large model, and retrieving information from calendar financial report data in the database to obtain related information required by answering a question;
converting related information and problems into token marks which can be processed by a financial report large model through an encoder;
and processing the related information in different modes according to the question types, generating answers of questions proposed by the user, and returning the answers to the user.
Preferably, the file analysis module through the large financial newspaper model uses a file analysis technology to analyze the financial newspaper file, and at least includes: the OCR technology is utilized to identify and store the graphic information in the financial report file, and the Python technology is utilized to read and analyze various financial report formats and process the financial report data.
Preferably, the processing the related information in different manners according to the question category to generate an answer to the question posed by the user includes:
for searching the class questions, generating answers directly according to the returned information;
for the calculation class questions, calling a calculator by using an API, and generating answers by combining the questions;
for common sense class questions, the API is used to call a search engine, and answers are generated according to the contents returned by the search engine.
Preferably, the iterative optimization of the financial report big model by using the information in the database through a feedback iteration mechanism includes:
collecting scoring results, scoring answers generated by the financial report large model, and storing the scoring results in a database;
determining an error mode, finding out an error answer of the financial report large model, and analyzing error data details;
updating the model, namely correcting the data generating the wrong answer, updating the corrected data into a database, and performing reinforcement training on the financial report large model by utilizing the updated database to update the model;
the test model is used for testing the updated financial report large model;
repeating the iteration, continuously collecting scoring results, updating and testing the financial report large model, and optimizing the question and answer capability of the financial report large model through continuous iteration.
In a second aspect, in order to achieve the above object, an embodiment of the present invention further provides a large model-based automated financial accounting device, including: the system comprises a financial report large model construction module, a file analysis and question-answering module, a database construction module, a feedback iteration optimization module and an automatic question-answering module;
the financial accounting large model construction module is used for pre-training and fine-tuning a basic open source large model based on a financial data set and the annual financial accounting data to construct a financial accounting large model with financial questioning and answering capacity;
the file analysis and question answering module is used for uploading the financial report file to the financial report large model by the user, analyzing the file of the financial report file through the financial report large model, generating an answer and returning the answer to the user;
the database construction module is used for constructing a database, wherein the database comprises the annual financial report data, financial report file analysis results and scoring results for answers;
the feedback iterative optimization module is used for carrying out iterative optimization on the financial report large model through a feedback iterative mechanism by utilizing information in the database;
and the automatic question and answer module is used for carrying out automatic question and answer on the financial newspaper by utilizing the optimized financial newspaper large model.
Further, the system architecture of the device includes:
the front-end interface layer is divided into a user interface and an employee interface and is respectively used for interaction between a user and an employee and the device, and comprises the steps of uploading a financial report file by the user, inputting a question and scoring an answer by the employee;
the business logic layer is used for interacting with the front-end interface layer, comprising the steps of acquiring a financial report file and a question from a user interface and a scoring result from an employee interface, constructing a financial report large model, generating an answer to the user question through the financial report large model, constructing a database, performing iterative optimization on the financial report large model through a feedback iteration mechanism by utilizing the database, and performing financial report automatic question and answer by utilizing the optimized financial report large model;
and a server layer for providing hardware support for the whole device and computing force support for the model.
In order to achieve the above object, the embodiment of the present invention further provides a large model-based automated financial questioning and answering device, which includes a memory and a processor, where the memory is configured to store a computer program, and the processor is configured to implement, when executing the computer program, the large model-based automated financial questioning and answering method provided by the embodiment of the present invention in the first aspect.
In a fourth aspect, in order to achieve the above object, an embodiment of the present invention further provides a computer-readable storage medium, where a computer program is stored on the storage medium, and when the computer program uses a computer, the automated financial accounting method based on a big model provided in the first aspect of the present invention is implemented.
Compared with the prior art, the invention has the beneficial effects that at least the following steps are included:
(1) According to the invention, the basic open source large model is pre-trained and finely adjusted based on the financial data set and the calendar data, so that the financial report large model with financial question-answering capability is obtained, and the model effect is more fit with the actual business requirements.
(2) The method of the invention can extract the relevant information from the analysis result of the financial newspaper file according to the questions presented by the user, and screen the required information from the historical financial newspaper data in the database, thereby generating the answers of the questions presented by the user, wherein the related file analysis, information extraction and information retrieval technologies complete the automatic extraction of the relevant graphic and text information in the target financial newspaper file through the automatic flow design, simplify the complicated data preprocessing flow and reduce the time cost and the labor cost of information acquisition.
(3) The feedback iteration mechanism related in the invention can feed back the question and answer effect to the large financial newspaper model, thereby realizing the iterative optimization of the large financial newspaper model, and further continuously improving the accuracy and efficiency of the intelligent question and answer of the large financial newspaper model, so that the system is more suitable for the actual application scene.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings that are required in the embodiments or the description of the prior art will be briefly described, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.
FIG. 1 is a schematic flow chart of a large model-based automated financial question-answering method provided by an embodiment of the invention;
FIG. 2 is a schematic flow chart of a basic open source large model pre-training and fine tuning to generate a financial report large model according to the embodiment of the invention;
FIG. 3 is a schematic flow diagram of a question-answering module based on a large financial report model according to an embodiment of the present invention;
FIG. 4 is a schematic flow chart of a feedback iteration mechanism provided by an embodiment of the present invention;
FIG. 5 is a schematic structural diagram of a large model-based automated financial accounting device according to an embodiment of the present invention;
fig. 6 is a system architecture diagram of a large model-based automated financial accounting device according to an embodiment of the present invention.
Detailed Description
The present invention will be described in further detail with reference to the drawings and examples, in order to make the objects, technical solutions and advantages of the present invention more apparent. It should be understood that the detailed description is presented by way of example only and is not intended to limit the scope of the invention.
The invention is characterized in that: aiming at the problems that the traditional financial analysis accuracy and efficiency are low and the automatic financial analysis lacks an intelligent question-answering function for diversified user questions in the prior art, the embodiment of the invention provides a financial automatic question-answering method and device based on a large model, which are used for pre-training and fine-tuning the basic open-source large model to obtain a financial large model with financial question-answering capability, and generating answers to user questions through the financial large model in combination with file analysis and information extraction of financial files and information retrieval of historical financial data, and optimizing the financial large model through a feedback iteration mechanism, thereby realizing financial automatic question-answering and improving the accuracy and efficiency of question-answering.
Fig. 1 is a schematic flow chart of a large model-based automated financial accounting method according to an embodiment of the present invention. As shown in fig. 1, an embodiment provides a large model-based automated financial accounting method, which includes the following steps:
s1, pre-training and fine-tuning are carried out on a basic open source large model based on a financial data set and calendar data, and a financial report large model with financial question-answering capability is built.
Specifically, as shown in fig. 2, a financial data set is first constructed based on a large number of financial teaching materials and data, and a basic open-source large model is pre-trained to obtain a pre-trained large model with financial knowledge. And then collecting the calendar financial report files, analyzing the files through file analysis technologies such as OCR (optical character recognition), python and the like to extract financial information, and constructing fine adjustment data based on the financial information, wherein the fine adjustment data comprises question-answer data sets with relevant information and standard answers required by answering the questions. And finally, taking fine adjustment data based on the financial information as input, and performing supervision fine adjustment on the pre-trained large model so as to obtain the financial report large model with the capability of financial question answering.
It should be noted that, for finding class questions, the goal of fine-tuning the pre-trained large model is set directly to the correct answer. For the problems of calculation and common sense, the goal of the fine-tuning pre-training large model is not a simple answer to the problem, but the pre-training large model needs to be guided to call an API for numerical calculation and information retrieval, but the existing open-source basic large model does not have the capability, so that in the process of constructing model training data, an API call thinking chain needs to be constructed for the problems of calculation and common sense as the goal of fine-tuning pre-training large model.
S2, uploading the financial report file by the user, inputting the problem to the financial report large model, analyzing the financial report file by the financial report large model, generating an answer and returning the answer to the user.
Specifically, as shown in fig. 3, the financial report file and the problem are input into a financial report large model, and the financial report file is processed and analyzed by a file analysis module of the financial report large model by using a file analysis technology, which at least comprises: the graphic information in the financial report file is identified and stored by utilizing an OCR (Optical Character Recognition) technology, various financial report formats are read and analyzed by utilizing a Python technology, and financial report data are processed. Finally, the graphic information in the financial newspaper is extracted and stored in a database.
And extracting information from the graphic information in the database and retrieving information from the calendar financial data in the database by using a question and answer module of the financial large model to obtain related information required by answering the questions. The relevant information and questions are converted by an Encoder (Encoder) into tokens (tokens) that can be handled by the financial big model. Processing the related information in different modes according to the problem category, and directly generating an answer according to the returned information for searching the class of problems; for the calculation class problem, calling a calculator by using an API (Application Programming Interface, application program interface), and generating an answer by combining the problem requirement (such as reserving two bits after decimal point, returning an absolute value and the like); for common sense class questions, the API is used to call a search engine, and answers are generated according to the contents returned by the search engine. And finally generating an answer to the question posed by the user, and returning the answer to the user.
S3, constructing a database, wherein the database comprises the calendar financial report data, the financial report file analysis result and the scoring result of the answers.
Specifically, the database is used for providing a financial data information retrieval function, the database contains the historical financial report data, the financial report file analysis result and the scoring result of the answers, relevant information is extracted from the database when the financial report large model answers the questions, and information such as the scoring result is extracted from the database when the financial report large model is subjected to iterative optimization, so that the time cost and the labor cost of information acquisition are reduced.
S4, carrying out iterative optimization on the financial report large model through a feedback iterative mechanism by utilizing information in the database.
Specifically, as shown in fig. 4, the method comprises the following steps:
and collecting scoring results. Scoring answers generated by the financial reporting model by staff with financial expertise, and storing the scoring results in a database.
An error pattern is determined. In some cases, the model may incorrectly answer a particular type of question or may miss a particular type of detail, and thus may need to find out the wrong answer to the large financial model and analyze the wrong data details.
Updating the model. The method comprises the steps of changing an algorithm or adding new training data, overcoming an error mode, correcting data generating an error answer, updating the corrected data into a database, and carrying out intensive training on a financial report large model by utilizing the updated database to update the model, thereby improving the capability of answering a question.
And (5) testing the model. And testing the updated financial report large model to ensure that the model has better response effect on new data.
The iteration is repeated. And continuously collecting scoring results, updating and testing the financial reporting model, and optimizing the question-answering capability of the financial reporting model through continuous iteration.
S5, performing automated financial report question answering by using the optimized financial report big model.
Specifically, the optimized large financial newspaper model is used as a final application model, and required answers are generated according to target financial newspaper files and questions input by a user, so that automated financial newspaper questions and answers are realized.
In sum, according to the large model-based financial accounting automated questioning and answering method, the basic open-source large model is pre-trained and finely adjusted to obtain the financial accounting large model with financial questioning and answering capability, so that the model effect is more fit with the actual business requirements. Information extraction is carried out through a financial report file analysis technology, information retrieval is carried out by combining with the annual financial report data, answers are generated according to user questions, and multi-category financial information intelligent questions and answers such as searching categories, calculating categories, common sense categories and the like can be carried out in face of diversified user questions, so that the accuracy and efficiency of analysis are greatly improved, and the time cost and the labor cost of information acquisition are reduced. And scoring the answers and performing iterative optimization on the large financial newspaper model through a feedback iterative mechanism, so that the accuracy and the efficiency of the analysis of the financial newspaper are further improved, and the method is suitable for an intelligent question-answering application scene of the financial newspaper data in the financial field.
Based on the same inventive concept, the embodiment also provides a large model-based financial accounting automated question-answering device 500, as shown in fig. 5, comprising: the system comprises a financial report big model construction module 510, a file analysis and question-answering module 520, a database construction module 530, a feedback iteration optimization module 540 and an automatic question-answering module 550.
The financial accounting large model construction module 510 is used for pre-training and fine-tuning a basic open source large model based on a financial data set and the calendar year financial accounting data, and constructing a financial accounting large model with financial questioning and answering capability;
the file analysis and question-answering module 520 is used for uploading the financial report file to the user, inputting the questions to the financial report large model, analyzing the financial report file through the financial report large model, generating an answer and returning the answer to the user;
the database construction module 530 is configured to construct a database, which includes the calendar data, the analysis result of the financial report file, and the scoring result of the answer;
the feedback iterative optimization module 540 is configured to iteratively optimize the financial report big model by using information in the database through a feedback iterative mechanism;
the automated questioning and answering module 550 is used for performing automated questioning and answering of the finance report by using the optimized finance report large model.
The system architecture of the large model-based automated financial question-answering device 500 provided in the embodiment is shown in fig. 6, and includes:
the front interface layer is divided into a user interface and an employee interface and is respectively used for system interaction between a user and an employee and a device, and comprises the functions of uploading financial report files and inputting questions through the user interface, and the like, and the functions of scoring and feeding back answers output by the large financial report model through the employee interface.
The business logic layer is used for interacting with the front-end interface layer, comprising the steps of acquiring a financial report file and a question from a user interface and a scoring result from an employee interface, constructing a financial report large model, generating an answer to the user question through the financial report large model, constructing a database, performing iterative optimization on the financial report large model through a feedback iteration mechanism by utilizing the database, and performing financial report automatic question and answer by utilizing the optimized financial report large model.
And a server layer for providing hardware support for the whole device and computing force support for the model.
Based on the same inventive concept, the embodiment also provides a large model-based financial accounting automated questioning and answering device, which comprises a memory and a processor, wherein the memory is used for storing a computer program, and the processor is used for realizing the large model-based financial accounting automated questioning and answering method when executing the computer program.
Based on the same inventive concept, the embodiment also provides a computer-readable storage medium, wherein the storage medium is stored with a computer program, and the automatic financial accounting question answering method based on the large model is realized when the computer program uses a computer.
It should be noted that, the large model-based automatic financial accounting device and the computer-readable storage medium provided in the foregoing embodiments belong to the same concept as the large model-based automatic financial accounting method embodiment, and detailed implementation processes of the large model-based automatic financial accounting method embodiment are omitted here.
The foregoing detailed description of the preferred embodiments and advantages of the invention will be appreciated that the foregoing description is merely illustrative of the presently preferred embodiments of the invention, and that no changes, additions, substitutions and equivalents of those embodiments are intended to be included within the scope of the invention.
Claims (10)
1. The automatic financial accounting question-answering method based on the large model is characterized by comprising the following steps of:
pre-training and fine-tuning a basic open source large model based on a financial data set and calendar financial report data, and constructing a financial report large model with financial question-answering capability;
uploading a financial report file by a user, inputting a problem to a financial report large model, analyzing the financial report file by the financial report large model, generating an answer and returning the answer to the user;
constructing a database, wherein the database comprises the calendar financial report data, financial report file analysis results and scoring results for answers;
carrying out iterative optimization on the financial report large model by using information in the database through a feedback iterative mechanism;
and performing automated financial report question answering by using the optimized financial report big model.
2. The automated financial questioning and answering method based on large models according to claim 1, wherein the pre-training and fine-tuning of the basic open source large model based on financial data sets and calendar financial data, constructing a financial large model with financial questioning and answering capability, comprises:
constructing a financial data set based on financial teaching materials and data, and pre-training a basic open-source large model to obtain a pre-training large model with financial knowledge;
collecting a calendar year financial report file, analyzing the file to extract financial information, and constructing fine tuning data based on the financial information, wherein the fine tuning data comprises a question-answer data set with relevant information and standard answers required for answering a question;
and taking fine adjustment data based on the financial information as input, and performing supervision fine adjustment on the pre-trained large model so as to obtain the financial report large model with the capability of financial question answering.
3. The automated financial accounting method based on big model according to claim 1, wherein the steps of uploading financial accounting file by the user and inputting questions to the big model of financial accounting, analyzing the financial accounting file by the big model of financial accounting and generating answers and returning the answers to the user include:
inputting the financial report file and the problems into a financial report large model, carrying out file analysis on the financial report file by a file analysis module of the financial report large model by utilizing a file analysis technology, extracting graphic and text information in the financial report and storing the graphic and text information in a database;
the method comprises the steps of extracting information from image-text information in a database through a question and answer module of a financial report large model, and retrieving information from calendar financial report data in the database to obtain related information required by answering a question;
converting related information and problems into token marks which can be processed by a financial report large model through an encoder;
and processing the related information in different modes according to the question types, generating answers of questions proposed by the user, and returning the answers to the user.
4. The automated financial accounting method based on big model according to claim 3, wherein the file analysis module for analyzing the financial accounting file by using file analysis technology at least comprises: the OCR technology is utilized to identify and store the graphic information in the financial report file, and the Python technology is utilized to read and analyze various financial report formats and process the financial report data.
5. The automated financial accounting method based on big models according to claim 3, wherein the processing related information in different ways according to the category of the questions to generate answers to questions posed by the user comprises:
for searching the class questions, generating answers directly according to the returned information;
for the calculation class questions, calling a calculator by using an API, and generating answers by combining the questions;
for common sense class questions, the API is used to call a search engine, and answers are generated according to the contents returned by the search engine.
6. The automated financial accounting method based on big model according to claim 1, wherein the iterative optimization of the financial accounting big model by using information in the database through a feedback iteration mechanism comprises:
collecting scoring results, scoring answers generated by the financial report large model, and storing the scoring results in a database;
determining an error mode, finding out an error answer of the financial report large model, and analyzing error data details;
updating the model, namely correcting the data generating the wrong answer, updating the corrected data into a database, and performing reinforcement training on the financial report large model by utilizing the updated database to update the model;
the test model is used for testing the updated financial report large model;
repeating the iteration, continuously collecting scoring results, updating and testing the financial report large model, and optimizing the question and answer capability of the financial report large model through continuous iteration.
7. The utility model provides a financial accounting automated question answering device based on big model which characterized in that includes: the system comprises a financial report large model construction module, a file analysis and question-answering module, a database construction module, a feedback iteration optimization module and an automatic question-answering module;
the financial accounting large model construction module is used for pre-training and fine-tuning a basic open source large model based on a financial data set and the annual financial accounting data to construct a financial accounting large model with financial questioning and answering capacity;
the file analysis and question answering module is used for uploading the financial report file to the financial report large model by the user, analyzing the file of the financial report file through the financial report large model, generating an answer and returning the answer to the user;
the database construction module is used for constructing a database, wherein the database comprises the annual financial report data, financial report file analysis results and scoring results for answers;
the feedback iterative optimization module is used for carrying out iterative optimization on the financial report large model through a feedback iterative mechanism by utilizing information in the database;
and the automatic question and answer module is used for carrying out automatic question and answer on the financial newspaper by utilizing the optimized financial newspaper large model.
8. The automated large model-based financial instrument of claim 7, wherein the system architecture of the instrument comprises:
the front-end interface layer is divided into a user interface and an employee interface and is respectively used for interaction between a user and an employee and the device, and comprises the steps of uploading a financial report file by the user, inputting a question and scoring an answer by the employee;
the business logic layer is used for interacting with the front-end interface layer, comprising the steps of acquiring a financial report file and a question from a user interface and a scoring result from an employee interface, constructing a financial report large model, generating an answer to the user question through the financial report large model, constructing a database, performing iterative optimization on the financial report large model through a feedback iteration mechanism by utilizing the database, and performing financial report automatic question and answer by utilizing the optimized financial report large model;
and a server layer for providing hardware support for the whole device and computing force support for the model.
9. A large model based automated financial questioning and answering device comprising a memory for storing a computer program and a processor, wherein the processor is configured to implement the large model based automated financial questioning and answering method of any one of claims 1-6 when the computer program is executed.
10. A computer-readable storage medium, on which a computer program is stored, characterized in that the automated financial accounting method based on a large model according to any one of claims 1-6 is implemented when the computer program is used in a computer.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202311379778.3A CN117235233B (en) | 2023-10-24 | 2023-10-24 | Automatic financial report question-answering method and device based on large model |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202311379778.3A CN117235233B (en) | 2023-10-24 | 2023-10-24 | Automatic financial report question-answering method and device based on large model |
Publications (2)
Publication Number | Publication Date |
---|---|
CN117235233A true CN117235233A (en) | 2023-12-15 |
CN117235233B CN117235233B (en) | 2024-06-11 |
Family
ID=89088061
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202311379778.3A Active CN117235233B (en) | 2023-10-24 | 2023-10-24 | Automatic financial report question-answering method and device based on large model |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN117235233B (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN117807963A (en) * | 2024-03-01 | 2024-04-02 | 之江实验室 | Text generation method and device in appointed field |
CN118245588A (en) * | 2024-05-27 | 2024-06-25 | 烟台海颐软件股份有限公司 | Automatic construction method and construction system for field question and answer set |
Citations (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110489555A (en) * | 2019-08-21 | 2019-11-22 | 创新工场(广州)人工智能研究有限公司 | A kind of language model pre-training method of combination class word information |
US20200234373A1 (en) * | 2019-01-23 | 2020-07-23 | Steven F. Graver | Digital wealth management advisor |
CN112036145A (en) * | 2020-09-01 | 2020-12-04 | 平安国际融资租赁有限公司 | Financial statement identification method and device, computer equipment and readable storage medium |
CN112364150A (en) * | 2021-01-12 | 2021-02-12 | 南京云创大数据科技股份有限公司 | Intelligent question and answer method and system combining retrieval and generation |
CN112966097A (en) * | 2021-03-09 | 2021-06-15 | 华泰证券股份有限公司 | NLP-based marketing company financial news-express automatic generation method and system |
CN114936099A (en) * | 2022-07-25 | 2022-08-23 | 之江实验室 | Graph optimization method and device for neural network calculation |
CN116029835A (en) * | 2023-01-18 | 2023-04-28 | 之江实验室 | Method and device for determining investment portfolio based on reinforcement learning and electronic equipment |
KR20230077588A (en) * | 2021-11-25 | 2023-06-01 | 아일리스프런티어 주식회사 | Method of classifying intention of various question and searching answers of financial domain based on financial term language model and system impelemting thereof |
CN116842263A (en) * | 2023-07-10 | 2023-10-03 | 中国工商银行股份有限公司 | Training processing method and device for intelligent question-answering financial advisor model |
CN116910223A (en) * | 2023-08-09 | 2023-10-20 | 北京安联通科技有限公司 | Intelligent question-answering data processing system based on pre-training model |
-
2023
- 2023-10-24 CN CN202311379778.3A patent/CN117235233B/en active Active
Patent Citations (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20200234373A1 (en) * | 2019-01-23 | 2020-07-23 | Steven F. Graver | Digital wealth management advisor |
CN110489555A (en) * | 2019-08-21 | 2019-11-22 | 创新工场(广州)人工智能研究有限公司 | A kind of language model pre-training method of combination class word information |
CN112036145A (en) * | 2020-09-01 | 2020-12-04 | 平安国际融资租赁有限公司 | Financial statement identification method and device, computer equipment and readable storage medium |
CN112364150A (en) * | 2021-01-12 | 2021-02-12 | 南京云创大数据科技股份有限公司 | Intelligent question and answer method and system combining retrieval and generation |
CN112966097A (en) * | 2021-03-09 | 2021-06-15 | 华泰证券股份有限公司 | NLP-based marketing company financial news-express automatic generation method and system |
KR20230077588A (en) * | 2021-11-25 | 2023-06-01 | 아일리스프런티어 주식회사 | Method of classifying intention of various question and searching answers of financial domain based on financial term language model and system impelemting thereof |
CN114936099A (en) * | 2022-07-25 | 2022-08-23 | 之江实验室 | Graph optimization method and device for neural network calculation |
CN116029835A (en) * | 2023-01-18 | 2023-04-28 | 之江实验室 | Method and device for determining investment portfolio based on reinforcement learning and electronic equipment |
CN116842263A (en) * | 2023-07-10 | 2023-10-03 | 中国工商银行股份有限公司 | Training processing method and device for intelligent question-answering financial advisor model |
CN116910223A (en) * | 2023-08-09 | 2023-10-20 | 北京安联通科技有限公司 | Intelligent question-answering data processing system based on pre-training model |
Non-Patent Citations (1)
Title |
---|
吴冲;刘佳明;郭志达;: "基于改进粒子群算法的模糊聚类-概率神经网络模型的企业财务危机预警模型研究", 运筹与管理, no. 02, 25 February 2018 (2018-02-25), pages 110 - 118 * |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN117807963A (en) * | 2024-03-01 | 2024-04-02 | 之江实验室 | Text generation method and device in appointed field |
CN117807963B (en) * | 2024-03-01 | 2024-04-30 | 之江实验室 | Text generation method and device in appointed field |
CN118245588A (en) * | 2024-05-27 | 2024-06-25 | 烟台海颐软件股份有限公司 | Automatic construction method and construction system for field question and answer set |
Also Published As
Publication number | Publication date |
---|---|
CN117235233B (en) | 2024-06-11 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Zhang et al. | Transformer-based attention network for stock movement prediction | |
CN117235233B (en) | Automatic financial report question-answering method and device based on large model | |
Yadav et al. | Comparing the performance of naive bayes and decision tree classification using R | |
CN115357719B (en) | Power audit text classification method and device based on improved BERT model | |
CN116738959B (en) | Resume rewriting method and system based on artificial intelligence | |
Jeyaraman et al. | Data Engineering Evolution: Embracing Cloud Computing, Machine Learning, and AI Technologies | |
Wimmer et al. | Leveraging vision-language models for granular market change prediction | |
Diniz et al. | Github label embeddings | |
CN117764724A (en) | Intelligent credit rating report construction method and system | |
WO2023134676A1 (en) | Method and system for automatically formulating optimization problem using machine learning | |
CN116049376A (en) | Method, device and system for retrieving and replying information and creating knowledge | |
Sakarwala et al. | Use advances in data science and computing power to invest in stock market | |
Macanovic et al. | A systematic evaluation of text mining methods for short texts: Mapping individuals’ internal states from online posts | |
WO2019043380A1 (en) | Semantic parsing | |
Dai et al. | Intelligent audit question answering system based on knowledge graph and semantic similarity | |
CN117540004B (en) | Industrial domain intelligent question-answering method and system based on knowledge graph and user behavior | |
Chen et al. | Displacement or Augmentation? The Effects of AI on Workforce Dynamics and Firm Value | |
Pchelin et al. | Analysis of machine learning models by solving the text data classification problem | |
CN114818659B (en) | Text emotion source analysis method and system and storage medium | |
US11830081B2 (en) | Automated return evaluation with anomoly detection | |
CN117541044B (en) | Project classification method, system, medium and equipment based on project risk analysis | |
Siddiqui et al. | Predicting bankruptcy of firms using earnings call data and transfer learning | |
WO2024098282A1 (en) | Geometric problem-solving method and apparatus, and device and storage medium | |
Schlaubitz | Natural Language Processing in finance: analysis of sentiment and complexity of news and earnings reports of swiss SMEs and their relevance for stock returns | |
Charlier et al. | XtracTree for regulator validation of bagging methods used in retail banking |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |