An accounting method, device and readable medium based on a large language model

Abstract

The present invention discloses an accounting method, device and readable medium based on a large language model, which relates to the field of language processing, including: supervised fine-tuning of a large language model based on accounting profession through an instruction set and a prompt word library for accounting information understanding tasks and an instruction set and a prompt word library for accounting information processing tasks, respectively obtaining an accounting information understanding large language model and an accounting information processing large language model; obtaining input information, reading the input information through the accounting information understanding large language model, and outputting tasks to be executed; combining the tasks to be executed with the instruction set and the prompt word library for accounting information processing tasks, setting task goals, generating executable task plans, reading the executable task plans through the accounting information processing large language model, and generating preliminary processing results; and outputting final processing results after expert evaluation and repeated generation of the preliminary processing results.

Description

Technical Field

[0001] The present invention relates to the field of language processing, and in particular to an accounting method, device and readable medium based on a large language model. Background Art

[0002] With the development of modern information technology, accounting systems tend to be automated, integrated, and intelligent, and applications such as smart finance and smart investment advisors around accounting information have begun to appear in large numbers. However, accounting information has its own professional threshold, and the continuous updating of accounting standards has led to many barriers to the generation, dissemination, and use of accounting information, which is not conducive to the resource allocation function of accounting information. There are a large number of tasks around language information and accounting data in the current accounting system. The core is to integrate the logic of accounting practitioners into the processing flow of the accounting system based on the comprehensive business experience of accounting standards. However, the current accounting system has a low degree of intelligence and can only perform simple operations such as data calculation, merging, and integration. It cannot adapt to changes in complex tasks, cannot simulate the professional logical thinking of accounting practitioners, and cannot automatically meet the comprehensive requirements of the accounting system.

[0003] The large language model is a major breakthrough in the field of natural language processing in artificial intelligence. It is an artificial neural network based on the transformer infrastructure with billions or even hundreds of billions of parameters. It is developed by training on large-scale data sets and has excellent text understanding and text generation capabilities. It has excellent performance in language translation, sentiment analysis, natural language understanding and generation. Its "emergence ability" enables it to not only have text understanding capabilities, but also to produce personalized and creative outputs based on different contexts and data inputs, and has a wide range of applications. However, there is currently a lack of Chinese large language models in the accounting field, a lack of high-quality accounting data sets, and a lack of professional instruction sets and prompt word libraries constructed by experts in the accounting and artificial intelligence fields. Training only on data sets composed of ordinary financial news and announcements cannot achieve good results in processing accounting professional tasks.

[0004] Although large language models have excellent capabilities, they cannot directly complete complex tasks in accounting systems. For example, accounting information in accounting systems is diverse, comes from a wide range of sources and is updated quickly. Accounting business in accounting systems is diverse, and the processing procedures are professional and complex, involving multi-faceted and multi-modal information integration and processing. Summary of the invention

[0005] The purpose of this application is to propose an accounting method, device and readable medium based on a large language model in response to the above-mentioned technical problems. Starting from a large language model based on accounting profession, it fills the gap in the current field of Chinese accounting large language model, and develops a supporting accounting information comprehensive system to cope with large, complex and highly professional accounting system tasks.

[0006] In a first aspect, the present invention provides an accounting method based on a large language model, comprising the following steps:

[0007] S1, construct a large language model based on accounting major, task instruction set and prompt word library, classify the task instruction set and prompt word library to obtain the instruction set and prompt word library for accounting information understanding task and the instruction set and prompt word library for accounting information processing task, and use the instruction set and prompt word library for accounting information understanding task and the instruction set and prompt word library for accounting information processing task to perform supervised fine-tuning on the large language model based on accounting major, and obtain the large language model for accounting information understanding and the large language model for accounting information processing respectively;

[0008] S2, obtaining input information, reading the input information through the accounting information understanding large language model, and outputting the tasks to be performed, specifically including: selecting corresponding instructions and prompt words from the instruction set and prompt word library for accounting information understanding tasks based on the input information, splicing the input information with the instructions and prompt words and inputting them into the accounting information processing large language model, identifying the corresponding task requirements and refining them to obtain the tasks to be performed;

[0009] S3, combining the task to be executed with the instruction set and prompt word library for accounting information processing tasks, setting the task goal, generating an executable task plan, reading the executable task plan through the accounting information processing large language model, and generating preliminary processing results, specifically including: the executable task plan includes task goals, task steps, instructions and prompt words, determining the data type according to the task goals and task steps, splicing the task steps, instructions, prompt words, task goals and data types into the accounting information processing large language model, generating text data or selecting corresponding professional processing tools to generate other types of data, and the text data and other types of data constitute the preliminary processing results;

[0010] S4, input the preliminary processing results into the expert evaluation system for scoring, and obtain an evaluation score. If the evaluation score is lower than the preset threshold, repeat steps S3 and S4 to re-plan the task and select instructions and prompt words; if the evaluation score is higher than the preset threshold and the evaluation score is still higher than the preset threshold after multiple cycles of steps S3 and S4, output the final processing result.

[0011] As a preferred method, a large language model, task instruction set and prompt word library based on accounting profession are constructed, including:

[0012] Collect accounting-related data and clean the data to obtain a training data set;

[0013] The training data set is classified into different sub-data sets according to the file type and the company industry to which it belongs. Accounting experts design corresponding instruction sets and prompt word libraries based on the sub-data sets for the actual business of the accounting system;

[0014] Build a large language model, use the training data set to pre-train the large language model, and obtain a pre-trained large language model;

[0015] The pre-trained large language model is fine-tuned using accounting templates and reinforced learning supervised by accounting experts to obtain a large language model based on accounting expertise.

[0016] Preferably, the professional processing tools include chart generation scripts, machine learning-based listed company financial behavior identification scripts, machine learning-based impairment loss prediction scripts, and crawler scripts.

[0017] As a preferred embodiment, the final processing results have different output contents according to different tasks. The final processing result of the enterprise financial statement interpretation task is the analysis and interpretation of the financial statements of a listed company, the final processing result of the enterprise accounting indicator analysis task is the interpretation of various accounting indicators of a company, the final processing result of the enterprise news capture and interpretation task is the news summary, keyword extraction and unfavorable factors of a company; the final processing result of the enterprise financial real-time dialogue task is the real-time dialogue generated by the user, and the final processing result of the enterprise research report writing task is a research report on a company.

[0018] In a second aspect, the present invention provides an accounting device based on a large language model, comprising:

[0019] The model building module is configured to build a large language model based on accounting major, a task instruction set and a prompt word library, classify the task instruction set and the prompt word library to obtain an instruction set and a prompt word library for accounting information understanding tasks and an instruction set and a prompt word library for accounting information processing tasks, and fine-tune the large language model based on accounting major in a supervised manner through the instruction set and the prompt word library for accounting information understanding tasks and the instruction set and the prompt word library for accounting information processing tasks, so as to obtain a large language model for accounting information understanding and a large language model for accounting information processing respectively;

[0020] The task interpretation module is configured to obtain input information, read the input information through the accounting information understanding large language model, and output the task to be executed, specifically including: selecting corresponding instructions and prompt words from the instruction set and prompt word library for the accounting information understanding task based on the input information, splicing the input information with the instructions and prompt words and inputting them into the accounting information processing large language model, identifying the corresponding task requirements and refining them to obtain the task to be executed;

[0021] The information processing module is configured to combine the task to be executed with the instruction set and prompt word library for accounting information processing tasks, set the task goal, generate an executable task plan, read the executable task plan through the accounting information processing large language model, and generate a preliminary processing result, which specifically includes: the executable task plan includes task goals, task steps, instructions and prompt words, the data type is determined according to the task goals and task steps, the task steps, instructions, prompt words, task goals and data types are spliced ​​and input into the accounting information processing large language model, text data is generated or corresponding professional processing tools are selected to generate other types of data, and the text data and other types of data constitute the preliminary processing results;

[0022] The evaluation module is configured to input the preliminary processing results into the expert evaluation system for scoring to obtain an evaluation score. If the evaluation score is lower than a preset threshold, the information processing module and the evaluation module are repeatedly executed to re-plan the task and select instructions and prompt words; if the evaluation score is higher than the preset threshold and the evaluation score is still higher than the preset threshold after multiple cycles of executing the information processing module and the evaluation module, the final processing result is output.

[0023] In a third aspect, the present invention provides an electronic device comprising one or more processors; a storage device for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the method described in any implementation manner in the first aspect.

[0024] In a fourth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method described in any implementation manner in the first aspect.

[0025] In a fifth aspect, the present invention provides a computer program product, comprising a computer program, which, when executed by a processor, implements the method described in any implementation manner in the first aspect.

[0026] Compared with the prior art, the present invention has the following beneficial effects:

[0027] (1) The accounting method based on a large language model proposed in the present invention constructs a high-quality accounting professional data set, and constructs a rich instruction set and prompt vocabulary through an accounting expert system; through training with a high-quality data set, a large language model based on the accounting profession is obtained.

[0028] (2) The accounting method based on the big language model proposed in the present invention builds an accounting system task processing framework around the big language model based on the accounting profession, and realizes a professional big language model accounting information comprehensive system at the application layer. The system integrates the big language model for accounting information understanding and the big language model for accounting information processing, and can output different results for different accounting tasks. It also proposes a modern accounting information analysis system based on the big language model and artificial intelligence, which has professionalism and accuracy, can widely meet the needs of the majority of accounting information users, and promote the efficient use of the accounting information resource allocation function.

[0029] (3) The accounting method based on the large language model proposed in this invention can, under the premise of ensuring the accuracy and professionalism of the data, facilitate the generation, dissemination and use of accounting information, break the professional barriers of the accounting system, greatly improve the utilization efficiency of accounting information, and promote its role in resource allocation in the capital market. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0031] Figure 1 A flowchart of an accounting method based on a large language model according to an embodiment of the present application;

[0032] Figure 2 A schematic diagram of the training process of the accounting information understanding big language model and the accounting information processing big language model of the accounting method based on the big language model in the embodiment of the present application;

[0033] Figure 3 A flowchart of a specific application of an accounting method based on a large language model according to an embodiment of the present application;

[0034] Figure 4 A schematic diagram of the architecture of an accounting information comprehensive system based on a professional large language model for an accounting method based on a large language model according to an embodiment of the present application;

[0035] Figure 5 A schematic diagram of an accounting device based on a large language model according to an embodiment of the present application;

[0036] Figure 6 A schematic diagram of the hardware structure of an electronic device provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0037] In order to make the purpose, technical scheme and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all 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.

[0038] Figure 1 An accounting method based on a large language model provided by an embodiment of the present application is shown, comprising the following steps:

[0039] S1. Construct a large language model, task instruction set and prompt vocabulary based on accounting major. Classify the task instruction set and prompt vocabulary to obtain the instruction set and prompt vocabulary for accounting information understanding tasks and the instruction set and prompt vocabulary for accounting information processing tasks. Through the instruction set and prompt vocabulary for accounting information understanding tasks and the instruction set and prompt vocabulary for accounting information processing tasks, supervised fine-tuning the large language model based on accounting major is performed to obtain the large language model for accounting information understanding and the large language model for accounting information processing respectively.

[0040] In a specific embodiment, a large language model, task instruction set and prompt word library based on accounting profession are constructed, specifically including:

[0041] Collect accounting-related data and clean the data to obtain a training data set;

[0042] The training data set is classified into different sub-data sets according to the file type and the company industry to which it belongs. Accounting experts design corresponding instruction sets and prompt word libraries based on the sub-data sets for the actual business of the accounting system;

[0043] Build a large language model, use the training data set to pre-train the large language model, and obtain a pre-trained large language model;

[0044] The pre-trained large language model is fine-tuned using accounting templates and reinforced learning supervised by accounting experts to obtain a large language model based on accounting expertise.

[0045] Specifically, refer to Figure 2, construct an accounting information professional database, map the information in the database to the subdivided fields in the accounting system to obtain accounting information professional databases of different granularities; use the accounting expert system to perform fine-grained and high-quality screening and supplementation of the data; perform instruction engineering processing and prompt engineering processing on the high-quality accounting data to obtain an accounting professional training data set suitable for large language model training. The content of this accounting information professional database includes texts collected from listed company financial statements, company announcements, news, analyst reports, accounting policies, regulatory documents, etc., and screens and removes documents with format errors and missing characters.

[0046] Sources of professional databases of accounting information include:

[0047] A large amount of financial web content captured by web crawlers;

[0048] Public announcements, research reports, papers and other text data;

[0049] User-generated content such as forums and social media;

[0050] Specialized datasets in the field of accounting and finance.

[0051] Data processing methods include removing noise data and irrelevant content, and labeling and classifying some data. The data preprocessing process includes segmenting the text into words, subwords or characters, and converting them into numerical representations that the model can handle. Unifying the text format, such as lowercase conversion, removing special symbols, etc. Building a vocabulary based on the frequency of words or subwords appearing in the data and classifying them, for example, obtaining a Chinese financial accounting vocabulary.

[0052] Furthermore, the training data set is segmented and the instruction set and prompt word library are designed. Specifically, different sub-data sets can be classified according to the file type and the company industry to which they belong. Accounting experts design corresponding task instruction sets and prompt word libraries based on the sub-data sets for the actual business of the accounting system. The data input source of the model is constructed from the generation of accounting information; the specific task processing flow and logic of each link in the accounting system are integrated into the task instruction set and prompt word library.

[0053] Select a large language model of appropriate specifications as the base for pre-training. The mainstream large language model base based on the transformer architecture has several types, including Encoder-only, Decoder-only and Encoder-Decoder. According to the task type and complexity, the base of the appropriate type and parameter specifications can be selected for pre-training. Pre-training is performed through the above training data set, so that the base model has the initial natural language processing ability and reserves accounting knowledge, that is, a pre-trained large language model is obtained. The embodiment of the present application takes the large language model of the decoder-only architecture as an example. The Transformer decoder in the large language model is composed of a multi-layer stacked decoder block, and each decoder block contains two main sub-layers: a self-attention mechanism and a feedforward neural network. The self-attention mechanism allows the model to take into account all previously generated words when generating each word. The attention mechanism helps the model understand the contextual relationship. The feedforward neural network further processes the word representation of each position through nonlinear transformation. The large language model of the decoder-only architecture gradually generates new words based on the text sequence that has been generated. This means that when generating each word, the model only considers the words that have been generated before, and does not need to generate all words at the same time.

[0054] The pre-trained large language model is further fine-tuned based on the template provided by the accounting expert, so that the model has data processing bias similar to that of the accounting expert. Further reinforcement learning supervised by the accounting expert is carried out. Specifically, the accounting expert is asked to evaluate the processing results of the above model, eliminate the inferior and select the best, and further align the large language model with the logic and bias of the accounting expert to obtain a large language model based on the accounting profession.

[0055] Based on the task instruction set and prompt word library, the instruction set and prompt word library are classified and constructed into two categories, namely: instruction set and prompt word library for accounting information understanding tasks, and instruction set and prompt word library for accounting information processing tasks. The large language model based on accounting major is fine-tuned by accounting experts through the instruction set and prompt word library for accounting information understanding tasks and the instruction set and prompt word library for accounting information processing tasks, respectively, to obtain the accounting information understanding large language model and the accounting information processing large language model. After supervision and fine-tuning, the accounting information understanding large language model is better at extracting information from input data, the intention in the text, and making appropriate accounting business plans. The accounting information processing large language model is better at directly processing specific business implementation in the accounting system.

[0056] S2, obtains input information, reads the input information through the accounting information understanding large language model, and outputs the tasks to be executed, specifically including: based on the input information, selects corresponding instructions and prompt words from the instruction set and prompt word library for accounting information understanding tasks, splices the input information with the instructions and prompt words and inputs them into the accounting information processing large language model, identifies the corresponding task requirements and refines them to obtain the tasks to be executed.

[0057] Specifically, refer to Figure 3 After obtaining the input information, the accounting information understanding large language model is used to extract information, identify requirements, and further refine requirements to obtain tasks to be performed. Specifically, prompt words are used to allow the accounting information understanding large language model to identify the current task requirements and output the tasks to be performed.

[0058] S3, combines the task to be executed with the instruction set and prompt word library for accounting information processing tasks, sets the task goal, generates an executable task plan, reads the executable task plan through the accounting information processing large language model, and generates preliminary processing results, including: the executable task plan includes task goals, task steps, instructions and prompt words, determines the data type according to the task goals and task steps, splices the task steps, instructions, prompt words, task goals and data types into the accounting information processing large language model, generates text data or selects corresponding professional processing tools to generate other types of data, and the text data and other types of data constitute the preliminary processing results.

[0059] In a specific embodiment, the professional processing tools include a chart generation script, a listed company financial behavior identification script based on machine learning, an impairment loss prediction script based on machine learning, and a crawler script.

[0060] According to the task to be executed, combined with the instruction set and prompt word library for accounting information processing tasks, a detailed executable task plan is generated, the task goal is set, and the data package used for the next step is summarized. The executable task plan includes the task goal, task steps, instructions and prompt words. The example is as follows:

[0061] If the task goal is to collect financial data, the corresponding task steps are:

[0062] Obtain the company's financial statements for the last three years (including balance sheet, income statement, and cash flow statement).

[0063] Obtain the company's key financial ratios (such as current ratio, quick ratio, return on assets, return on equity, etc.).

[0064] If the task goal is data analysis, the corresponding task steps are:

[0065] Analyze the balance sheet and evaluate the company's asset and liability structure.

[0066] Analyze the income statement to evaluate the company's profitability and profit trends.

[0067] Analyze the cash flow statement to assess the company's cash flow health.

[0068] Calculate and interpret key financial ratios to assess a company's liquidity, profitability, and debt-paying ability.

[0069] If the task goal is industry comparison, the corresponding task steps are:

[0070] Collect industry average financial data and conduct industry comparisons.

[0071] Analyze how a company's financial performance compares to the industry average.

[0072] If the task objective is risk assessment, the corresponding task steps are:

[0073] Identify potential financial risks (such as excessive debt, unstable profits, etc.).

[0074] Evaluate the company's financial risk management measures.

[0075] If the task goal is a summary report, the corresponding task steps are:

[0076] Organize the analysis results and generate a detailed financial health report.

[0077] Include an executive summary, detailed analysis, charts, and conclusions in your report.

[0078] The accounting information processing large language model reads the above data packets, selects appropriate professional processing tools according to the task steps, instructions and prompt words, task objectives, and data types, and generates preliminary processing results. Due to the limited capabilities of the accounting information processing large language model, it is only suitable for processing text data and outputting text information. The scope of its capabilities is limited, resulting in the task form being text-based; therefore, when generating other types of data, it is necessary to call corresponding professional processing tools, such as chart generation scripts, machine learning-based listed company financial behavior identification scripts, machine learning-based impairment loss prediction scripts, and crawler scripts. These are all tools that help the accounting information processing large language model expand its capabilities and increase the scope of tasks that can be processed. Develop supporting edge tools for tasks that the large language model cannot directly handle; build reliable accounting information transmission pipelines, front-end platforms, and back-end processing centers.

[0079] S4, input the preliminary processing results into the expert evaluation system for scoring, and obtain an evaluation score. If the evaluation score is lower than the preset threshold, repeat steps S3 and S4 to re-plan the task and select instructions and prompt words; if the evaluation score is higher than the preset threshold and the evaluation score is still higher than the preset threshold after multiple cycles of steps S3 and S4, output the final processing result.

[0080] In a specific embodiment, the final processing result outputs different contents according to different tasks. The final processing result corresponding to the enterprise financial statement interpretation task is the analysis and interpretation of the financial statements of a listed company, the final processing result corresponding to the enterprise accounting indicator analysis task is the interpretation of various accounting indicators of a company, the final processing result corresponding to the enterprise news capture and interpretation task is the news summary, keyword extraction and unfavorable factors of a company; the final processing result corresponding to the enterprise financial real-time dialogue task is the real-time dialogue generated by the user, and the final processing result corresponding to the enterprise research report writing task is a research report on a company.

[0081] Specifically, the preliminary processing result generated in step S3 is input into the expert evaluation system for scoring to obtain an evaluation score. If the evaluation score is lower than the preset threshold, it will return to step S3 to re-plan the task, and reselect the instruction and prompt word to cycle the above process. If the evaluation score is higher than the preset threshold and is still higher than the preset threshold after multiple cycles, the final processing result will be output to the front end. The number of cycles can be set manually and adjusted flexibly. Different outputs are also different depending on the task, which is determined by the input information.

[0082] refer to Figure 4 , through the above accounting method based on the big language model, an accounting information comprehensive system based on the professional big language model is constructed. The accounting information comprehensive system based on the professional big language model can run on a high-performance computing network cluster. The equipment at the hardware layer includes but is not limited to: high-performance computing graphics cards, high-performance processing chips, data communication equipment, dedicated network equipment, embedded devices, programmable devices, smart mobile terminals, etc. The model layer includes the accounting information understanding big language model and the accounting information processing big language model. The architecture layer includes but is not limited to: accounting information data source, prompt word library instruction set, accounting task tool set, business application development framework, etc. The application layer includes but is not limited to: interpretation of corporate financial statements, analysis of corporate accounting indicators, interpretation of corporate news capture, real-time corporate financial dialogue, and writing of corporate research reports.

[0083] Further references Figure 5 As an implementation of the methods shown in the above figures, the present application provides an embodiment of an accounting device based on a large language model. Figure 1 Corresponding to the method embodiment shown, the device can be specifically applied to various electronic devices.

[0084] The present application embodiment provides an accounting device based on a large language model, including:

[0085] The model building module 1 is configured to build a large language model based on accounting major, a task instruction set and a prompt word library, classify the task instruction set and the prompt word library to obtain an instruction set and a prompt word library for accounting information understanding tasks and an instruction set and a prompt word library for accounting information processing tasks, and fine-tune the large language model based on accounting major in a supervised manner through the instruction set and the prompt word library for accounting information understanding tasks and the instruction set and the prompt word library for accounting information processing tasks, so as to obtain a large language model for accounting information understanding and a large language model for accounting information processing respectively;

[0086] The task interpretation module 2 is configured to obtain input information, read the input information through the accounting information understanding large language model, and output the task to be executed, specifically including: selecting corresponding instructions and prompt words from the instruction set and prompt word library for the accounting information understanding task based on the input information, splicing the input information with the instructions and prompt words and inputting them into the accounting information processing large language model, identifying the corresponding task requirements and refining them to obtain the task to be executed;

[0087] The information processing module 3 is configured to combine the task to be executed with the instruction set and prompt word library for accounting information processing tasks, set the task target, generate an executable task plan, read the executable task plan through the accounting information processing large language model, and generate a preliminary processing result, which specifically includes: the executable task plan includes task targets, task steps, instructions and prompt words, the data type is determined according to the task targets and task steps, the task steps, instructions, prompt words, task targets and data types are spliced ​​and input into the accounting information processing large language model, text data is generated or corresponding professional processing tools are selected to generate other types of data, and the text data and other types of data constitute the preliminary processing results;

[0088] The evaluation module 4 is configured to input the preliminary processing results into the expert evaluation system for scoring to obtain an evaluation score. If the evaluation score is lower than a preset threshold, the information processing module 3 and the evaluation module 4 are repeatedly executed to re-plan the task and select instructions and prompt words; if the evaluation score is higher than the preset threshold and the evaluation score is still higher than the preset threshold after multiple cycles of executing the information processing module 3 and the evaluation module 4, the final processing result is output.

[0089] Figure 6 Schematic diagram of the hardware structure of the electronic device provided by the embodiment of the present invention. Figure 6As shown, the electronic device of this embodiment includes: a processor 601 and a memory 602; wherein the memory 602 is used to store computer-executable instructions; the processor 601 is used to execute the computer-executable instructions stored in the memory to implement the various steps performed by the electronic device in the above embodiment. For details, please refer to the relevant description in the above method embodiment.

[0090] Optionally, the memory 602 may be independent or integrated with the processor 601 .

[0091] When the memory 602 is independently provided, the electronic device further includes a bus 603 for connecting the memory 602 and the processor 601 .

[0092] The embodiment of the present invention further provides a computer storage medium, in which computer execution instructions are stored. When the processor 601 executes the computer execution instructions, the above method is implemented.

[0093] The embodiment of the present invention further provides a computer program product, including a computer program. When the computer program is executed by the processor 601, the above method is implemented.

[0094] In the embodiments provided by the present invention, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic, for example, the division of modules is only a logical function division, and there may be other division methods in actual implementation, such as multiple modules can be combined or integrated into another system, or some features can be ignored or not executed. Another point, the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, devices or modules, which can be electrical, mechanical or other forms.

[0095] The modules described as separate components may or may not be physically separated, and the components shown as modules may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the modules may be selected according to actual needs to implement the solution of this embodiment.

[0096] In addition, each functional module in each embodiment of the present invention may be integrated into one processing unit, or each module may exist physically separately, or two or more modules may be integrated into one unit. The above-mentioned module-composed unit may be implemented in the form of hardware or in the form of hardware plus software functional units.

[0097] The above-mentioned integrated module implemented in the form of a software function module can be stored in a computer-readable storage medium. The above-mentioned software function module is stored in a storage medium, including a number of instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) or a processor 601 to perform some steps of the methods of various embodiments of the present application.

[0098] It should be understood that the processor 601 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), etc. A general-purpose processor may be a microprocessor or the processor 601 may be any conventional processor 601, etc. The steps of the method disclosed in the invention may be directly embodied in the execution of the hardware processor 601, or may be executed by a combination of hardware and software modules in the processor 601.

[0099] The memory 602 may include a high-speed RAM memory, and may also include a non-volatile storage NVM, such as at least one disk memory, and may also be a USB flash drive, a mobile hard disk, a read-only memory, a magnetic disk, or an optical disk.

[0100] The bus 603 may be an Industry Standard Architecture (ISA), a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. The bus 603 may be divided into an address bus, a data bus, a control bus, etc. For ease of representation, the bus 603 in the drawings of the present application is not limited to only one bus 603 or one type of bus 603.

[0101] The above storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk. The storage medium can be any available medium that can be accessed by a general or special purpose computer.

[0102] An exemplary storage medium is coupled to the processor 601, so that the processor 601 can read information from the storage medium and write information to the storage medium. Of course, the storage medium can also be a component of the processor 601. The processor 601 and the storage medium can be located in an application specific integrated circuit (ASIC). Of course, the processor 601 and the storage medium can also exist as discrete components in an electronic device or a main control device.

[0103] Those skilled in the art can understand that all or part of the steps of implementing the above-mentioned method embodiments can be completed by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When the program is executed, the steps of the above-mentioned method embodiments are executed; and the aforementioned storage medium includes: ROM, RAM, disk or optical disk and other media that can store program codes.

[0104] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. An accounting method based on a large language model, characterized in that: The following steps are involved: S1, build a large language model, task instruction set and prompt word library based on accounting profession, including: Collect accounting-related data and clean the data to obtain a training data set; The training data set is classified into different sub-data sets according to the file types and the industry to which they belong, and accounting experts design corresponding task instruction sets and prompt word libraries based on the sub-data sets for the actual business of the accounting system; Constructing a large language model, and pre-training the large language model using the training data set to obtain a pre-trained large language model; The pre-trained large language model is fine-tuned in a supervised manner and reinforced by accounting experts using an accounting template to obtain a large language model based on the accounting profession, the task instruction set and the prompt word library are classified and constructed to obtain an instruction set and prompt word library for the accounting information understanding task and an instruction set and prompt word library for the accounting information processing task, and the large language model based on the accounting profession is fine-tuned in a supervised manner using the instruction set and prompt word library for the accounting information understanding task and the instruction set and prompt word library for the accounting information processing task to obtain a large language model for accounting information understanding and a large language model for accounting information processing, respectively; S2, obtaining input information, reading the input information through the accounting information understanding large language model, and outputting the task to be executed, specifically including: selecting corresponding instructions and prompt words from the instruction set and prompt word library for accounting information understanding tasks based on the input information, splicing the input information with the instructions and prompt words and inputting them into the accounting information understanding large language model, identifying the corresponding task requirements and refining them to obtain the task to be executed; S3, combining the task to be executed with the instruction set and prompt word library for the accounting information processing task, setting the task goal, generating an executable task plan, reading the executable task plan through the accounting information processing large language model, and generating a preliminary processing result, specifically including: the executable task plan includes task goals, task steps, instructions and prompt words, determining the data type according to the task goals and task steps, splicing the task steps, instructions, prompt words, task goals and data types into the accounting information processing large language model, generating text data or selecting corresponding professional processing tools to generate other types of data, the text data and other types of data constitute the preliminary processing result; S4, input the preliminary processing results into the expert evaluation system for scoring to obtain an evaluation score. If the evaluation score is lower than the preset threshold, repeat steps S3 and S4 to re-plan the task and select instructions and prompt words; if the evaluation score is higher than the preset threshold and the evaluation score is still higher than the preset threshold after multiple cycles of steps S3 and S4, output the final processing result.

2. The accounting method based on a large language model according to claim 1, characterized in that: The professional processing tools include chart generation scripts, machine learning-based listed company financial behavior identification scripts, machine learning-based impairment loss prediction scripts and crawler scripts.

3. The accounting method based on a large language model according to claim 1, characterized in that: The final processing result has different output contents according to different tasks. The final processing result of the enterprise financial statement interpretation task is the analysis and interpretation of the financial statements of a listed company, the final processing result of the enterprise accounting indicator analysis task is the interpretation of various accounting indicators of a company, the final processing result of the enterprise news capture and interpretation task is the news summary, keyword extraction and unfavorable factors of a company; the final processing result of the enterprise financial real-time dialogue task is the real-time dialogue generated by the user, and the final processing result of the enterprise research report writing task is a research report on a company.

4. An accounting device based on a large language model, characterized in that: include: The model building module is configured to build a large language model, task instruction set and prompt word library based on accounting profession, including: Collect accounting-related data and clean the data to obtain a training data set; The training data set is classified into different sub-data sets according to the file types and the industry to which they belong, and accounting experts design corresponding task instruction sets and prompt word libraries based on the sub-data sets for the actual business of the accounting system; Constructing a large language model, and pre-training the large language model using the training data set to obtain a pre-trained large language model; The pre-trained large language model is fine-tuned in a supervised manner and reinforced by accounting experts using an accounting template to obtain a large language model based on the accounting profession, the task instruction set and the prompt word library are classified and constructed to obtain an instruction set and prompt word library for the accounting information understanding task and an instruction set and prompt word library for the accounting information processing task, and the large language model based on the accounting profession is fine-tuned in a supervised manner using the instruction set and prompt word library for the accounting information understanding task and the instruction set and prompt word library for the accounting information processing task to obtain a large language model for accounting information understanding and a large language model for accounting information processing, respectively; The task interpretation module is configured to obtain input information, read the input information through the accounting information understanding large language model, and output the task to be executed, specifically including: selecting corresponding instructions and prompt words from the instruction set and prompt word library for accounting information understanding tasks based on the input information, splicing the input information with the instructions and prompt words and inputting them into the accounting information understanding large language model, identifying the corresponding task requirements and refining them to obtain the task to be executed; The information processing module is configured to combine the task to be executed with the instruction set and prompt word library for the accounting information processing task, set the task goal, generate an executable task plan, read the executable task plan through the accounting information processing large language model, and generate a preliminary processing result, which specifically includes: the executable task plan includes task goals, task steps, instructions and prompt words, the data type is determined according to the task goals and task steps, the task steps, instructions, prompt words, task goals and data types are spliced ​​and input into the accounting information processing large language model, text data is generated or corresponding professional processing tools are selected to generate other types of data, and the text data and other types of data constitute the preliminary processing result; The evaluation module is configured to input the preliminary processing results into the expert evaluation system for scoring to obtain an evaluation score. If the evaluation score is lower than a preset threshold, the information processing module and the evaluation module are repeatedly executed to re-plan the task and select instructions and prompt words; if the evaluation score is higher than the preset threshold and the evaluation score is still higher than the preset threshold after multiple cycles of executing the information processing module and the evaluation module, the final processing result is output.

5. An electronic device comprising: one or more processors; a storage device for storing one or more programs, When the one or more programs are executed by the one or more processors, the one or more processors implement the method according to any one of claims 1 to 3.

6. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the method according to any one of claims 1 to 3 is implemented.

7. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the method according to any one of claims 1 to 3 is implemented.

Images