Intelligent invoicing method and system based on finance and tax industry large model

Abstract

The invention provides an intelligent invoicing method based on a finance and taxation industry large model, and the method comprises the following steps: S1, constructing a finance and taxation industry knowledge base, and storing a text vector of finance and taxation industry invoice data; s2, decomposing the input data through a data preprocessing module, and uniformly splitting the input data into data in a text format and data in a picture format; s3, converting the text data into a text vector by using an intelligent invoicing module, performing retrieval in a finance and taxation industry knowledge base through the text vector, inputting the obtained finance and taxation industry knowledge into the finance and taxation industry large model, and inputting the picture data into the finance and taxation industry large model at the same time; and S4, performing manual verification and correction on the invoicing information extracted by the finance and tax industry large model through a manual verification module. The billing efficiency is improved, and manpower is liberated.

Description

Technical Field

[0001] The present invention relates to the field of deep learning, and in particular to a method and system for intelligent invoicing based on a large model of the finance and taxation industry. Background Art

[0002] When companies or users issue invoices, the types and formats of invoicing materials are complex and varied, including text, pictures, chat screenshots, word documents, Excel documents, PDF documents, contract documents, etc. In the traditional invoicing method, users need to manually enter the company information and invoicing information in these materials, which is extremely inefficient and prone to errors. Currently, there are image text recognition technology OCR and text entity extraction technology NER. Although they can extract information from images and texts, since OCR technology recognizes text by line, the text of a cell will be recognized as multiple lines for line breaks in images and tables, resulting in inaccurate results. NER technology requires a large amount of training corpus and has poor results for digital entity extraction. Summary of the invention

[0003] The present invention provides a method and system for intelligent invoicing based on a large model of the finance and taxation industry, which solves the problem that the traditional invoicing method requires manual input of enterprise information and product information, and the existing technology cannot well extract information from pictures and texts, resulting in low invoicing efficiency. The technical solution is as follows:

[0004] A method for intelligent invoicing based on a large model of the finance and taxation industry includes the following steps:

[0005] S1: Build a knowledge base for the finance and taxation industry to store text vectors of invoice data for the finance and taxation industry;

[0006] S2: Decompose the input data through the data preprocessing module and split it into two formats: text and image.

[0007] S3: Use the intelligent invoicing module to convert text data into text vectors, search the finance and taxation industry knowledge base through the text vectors, input the acquired finance and taxation industry knowledge into the finance and taxation industry big model, and input the image data into the finance and taxation industry big model;

[0008] S4: Through the manual verification module, the invoicing information extracted from the large model of the finance and taxation industry is manually verified and corrected.

[0009] Furthermore, in step S1, building a finance and taxation industry knowledge base includes the following steps:

[0010] S11: Collect invoice data of the finance and taxation industry, including text content of text, paragraphs, and documents;

[0011] S12: Processing the text content into a set length through a slicing algorithm and converting it into text data;

[0012] S13: The text data is converted into a text vector containing semantic information through a text embedding model and stored in a vector database.

[0013] Furthermore, in step S2, the data preprocessing module determines the data format of the input data, decomposes the input data through a word parser, an excel parser, and a PDF parser, and splits the input data into data in two formats: text and image.

[0014] Furthermore, in step S3, the text data is converted into a text vector using the intelligent invoicing module, and the text vector is used to search the finance and taxation industry knowledge base, and the acquired finance and taxation industry knowledge is input into the finance and taxation industry big model, and the image data is input into the finance and taxation industry big model at the same time, including the following steps:

[0015] S21: The intelligent invoicing module converts the text data processed by the data preprocessing module into a text vector through a text embedding model;

[0016] S22: Search the finance and taxation industry knowledge base through the text vector, and obtain the finance and taxation industry knowledge related to the text data through retrieval and recall;

[0017] S23: Feed the recalled industry knowledge to the pre-trained large language model of the base to form a large model of the finance and taxation industry with finance and taxation industry knowledge;

[0018] S24: The image data is directly input into the big model of the finance and taxation industry.

[0019] Furthermore, in step S22, the recall method takes data whose cosine distance is greater than a threshold value, and the formula of the cosine distance is: Where cos(A,B) represents cosine similarity, A represents the input text vector, and B represents the knowledge base text vector.

[0020] The finance and taxation industry model is designed through prompt engineering and can automatically extract the enterprise information and product information contained in the text and pictures according to the set format.

[0021] Furthermore, the data preprocessing module determines the format of the input data. If the input data is a word file, the content of the word file is extracted in two formats, text and picture, through the word parser; if the input data is an excel file, the content of the excel file is extracted in two formats, text and picture, through the excel parser; if the input data is a PDF file, the content of the PDF file is extracted in two formats, text and picture, through the PDF parser; if the input data is text and picture, the text and picture formats are kept unchanged.

[0022] An intelligent invoicing system based on a large model of the finance and taxation industry, comprising:

[0023] Finance and taxation industry knowledge base: For finance and taxation industry invoice data, the corresponding text content is converted into text data through the slicing algorithm, and the text data is converted into a high-dimensional dense text vector containing semantic information through the text embedding model, and stored in the vector database;

[0024] Data preprocessing module: determines the data format of the input data, decomposes the input data through word parser, excel parser, and PDF parser, and uniformly splits it into two formats: text and image.

[0025] Intelligent invoicing module: used to convert text data into text vectors, search the finance and taxation industry knowledge base through the text vectors, and input the acquired finance and taxation industry knowledge into the finance and taxation industry big model;

[0026] Manual verification module: Manually verify and correct the invoicing information extracted from the large model of the finance and taxation industry.

[0027] The finance and taxation industry model is designed through prompt engineering and can automatically extract the enterprise information and product information contained in the text and pictures according to the set format.

[0028] The image data obtained through the data preprocessing module is directly input into the large model of the finance and taxation industry.

[0029] The method and system for intelligent invoicing based on the big model of the finance and taxation industry improves the efficiency of invoicing and frees up manual work. The data preprocessing module reduces the calculation steps required for model reasoning by uniformly processing the data, improves the reasoning speed and effect of the model, introduces knowledge enhancement of the finance and taxation industry knowledge base, increases the model's understanding of industry knowledge, and improves the reasoning effect of the big model. The manual verification module avoids and corrects the problem of inaccurate extraction results caused by probability problems in the big model through manual verification. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1It is a flowchart of the intelligent invoicing method based on the big model of the finance and taxation industry;

[0031] Figure 2 It is a flow chart for constructing the finance and taxation industry knowledge base;

[0032] Figure 3 is a processing schematic diagram of the data preprocessing module;

[0033] Figure 4 It is a processing diagram of the intelligent invoicing module. DETAILED DESCRIPTION

[0034] like Figure 1 As shown, the intelligent invoicing method based on the big model of the finance and taxation industry includes the following steps:

[0035] S1: Build a knowledge base for the finance and taxation industry;

[0036] like Figure 2 As shown in the figure, building a knowledge base for the finance and taxation industry includes the following steps:

[0037] S11: Collect invoice data of the finance and taxation industry, including text content in various formats such as text, paragraphs, and documents;

[0038] S12: Process the text content into a suitable length through a slicing algorithm and convert it into text data;

[0039] S13: The text data is converted into a high-dimensional dense text vector containing semantic information through a text embedding model and stored in a vector database.

[0040] S2: Decompose the input data through the data preprocessing module and split it into two formats: text and image.

[0041] like Figure 3 As shown in the figure, when issuing invoices, users will input data in various formats, including text, pictures, and documents in various formats. The data preprocessing module determines the data format of the input data, and decomposes the input data through the word parser, excel parser, and PDF parser, and uniformly splits it into two formats of data: text and picture. This can reduce the reasoning calculation steps of the large model, because when the large model performs reasoning calculations, it must not only understand the semantics of the text, but also understand the format of the text and data. Simplifying the data input format can reduce the reasoning calculation steps of the large model, thereby improving the reasoning efficiency of the large model.

[0042] S3: Use the intelligent invoicing module to convert text data into text vectors, search the finance and taxation industry knowledge base through the text vectors, input the acquired finance and taxation industry knowledge into the finance and taxation industry big model, and input the image data into the finance and taxation industry big model.

[0043] like Figure 4 As shown, the specific steps include:

[0044] S21: The intelligent invoicing module converts the pre-processed text data into text vectors through a text embedding model;

[0045] S22: Search the finance and taxation industry knowledge base through the text vector, and obtain the finance and taxation industry knowledge related to the text data through retrieval and recall;

[0046] Furthermore, the recall method takes the data whose cosine distance is greater than the threshold. The formula of cosine distance is: Where cos(A,B) represents cosine similarity, A represents the input text vector, and B represents the knowledge base text vector.

[0047] S23: Feed the recalled industry knowledge to the pre-trained large language model on the base to form a large model with finance and taxation industry knowledge;

[0048] S24: The image data is directly input into the big model of the finance and taxation industry.

[0049] The finance and taxation industry large model can automatically extract the enterprise information and product information contained in the text and pictures according to the set format through prompt engineering design.

[0050] S4: Through the manual verification module, the invoicing information extracted from the large model of the finance and taxation industry is manually verified and corrected.

[0051] The manual verification module allows the user to manually verify and correct the invoicing information extracted from the large model, and after confirmation, the invoicing can be carried out with one click.

[0052] An intelligent invoicing system based on the big model of the finance and taxation industry includes the following modules:

[0053] Finance and taxation industry knowledge base: For finance and taxation industry invoice data, the corresponding text content is converted into text data through the slicing algorithm, and the text data is converted into a high-dimensional dense text vector containing semantic information through the text embedding model, and stored in the vector database;

[0054] Data preprocessing module: determines the data format of the input data, decomposes the input data through word parser, excel parser, and PDF parser, and divides it into two formats: text and image.

[0055] Intelligent invoicing module: used to convert text data into text vectors, search the finance and taxation industry knowledge base through the text vectors, and input the acquired finance and taxation industry knowledge into the finance and taxation industry big model; the finance and taxation industry big model can automatically extract the company information and product information contained in the text and pictures according to the set format through prompt engineering design.

[0056] Manual verification module: Manually verify and correct the invoicing information extracted from the large model of the finance and taxation industry.

[0057] In one embodiment, the present invention includes the following:

[0058] (1) Construction of the knowledge base of finance and taxation industry:

[0059] Collect text data from the finance and taxation industry, including text, paragraphs, and documents. Use the slicing algorithm to convert the text data into text blocks with a certain length range. Then use the text embedding model to perform text vectorization on the text blocks, and convert each text block into a 1024-dimensional text vector. For example, the text content is: "Case study on the identification of R&D expense additional deduction project". The converted text vector is:

[0060] “[-0.0010660919,-0.031123366,-0.025069363,0.020800227,-0.012491459,-0.0035816291,-0.014876229,0.017138708,0.006494726,0.014357794,......,0.019406851,-0.008294031, -0.019862482,0.008888464,-0.033391386,0.041231737,0.014283797,-0.0043961103,0.025307167,-0.0098932,-0.005779804,-0.0119812125,0.014142047,0.03495005,-0.02790379]”

[0061] (2) Data preprocessing:

[0062] Data preprocessing processes the data in different formats input by the user, and finally converts the input data into two formats: text data and image data. The specific operation process includes:

[0063] First, the input data format is judged. If the input data is a word file, the content of the word file is extracted in two formats, text and picture, through the word parser; if the input data is an excel file, the content of the excel file is extracted in two formats, text and picture, through the excel parser; if the input data is a PDF file, the content of the PDF file is extracted in two formats, text and picture, through the PDF parser; if the input data is text and picture, the text and picture formats are kept unchanged.

[0064] (3) Intelligent invoicing module:

[0065] For the text data after data preprocessing, the text will be converted into text vectors through the text embedding model. Through retrieval and recall, the finance and taxation industry knowledge related to the text data is obtained from the finance and taxation industry knowledge base, and these data contents are fed to the base pre-trained large language model. The base pre-trained large language model obtains certain finance and taxation industry knowledge by learning finance and taxation industry data, and forms a finance and taxation industry large model. Through the prompt project, the extraction task and the output of the specified format are designed. The large model performs reasoning calculations based on general prior knowledge and finance and taxation industry knowledge, and finally extracts the corporate information and product information we want.

[0066] (4) Manual verification module:

[0067] Users can verify and check the invoicing information extracted from the large model. If the information is accurate, they can click one button to issue the invoice. If inaccurate information is extracted, users can modify and correct it, thereby ensuring the accuracy of the final invoicing information results.

[0068] The present invention solves the problem that the traditional invoice issuance method requires manual entry of enterprise invoice information and commodity information, and the invoice issuance efficiency is low, thereby improving the invoice issuance efficiency and freeing up manual labor. The data preprocessing module reduces the calculation steps required for model reasoning by uniformly processing the data, improves the reasoning speed and effect of the model, introduces knowledge enhancement of the financial and tax industry knowledge base, increases the model's understanding of industry knowledge, and improves the reasoning effect of the large model. The manual verification module avoids and corrects the problem of inaccurate extraction results caused by probability problems in the large model through manual verification.

Claims

1. A method for intelligent invoicing based on a large model of the finance and taxation industry, comprising the following steps: S1: Build a knowledge base for the finance and taxation industry to store text vectors of invoice data for the finance and taxation industry; S2: Decompose the input data through the data preprocessing module and split it into two formats: text and image. S3: Use the intelligent invoicing module to convert text data into text vectors, search the finance and taxation industry knowledge base through the text vectors, input the acquired finance and taxation industry knowledge into the finance and taxation industry big model, and input the image data into the finance and taxation industry big model; S4: Through the manual verification module, the invoicing information extracted from the large model of the finance and taxation industry is manually verified and corrected.

2. The intelligent invoicing method based on the finance and taxation industry big model according to claim 1 is characterized by: In step S1, building a finance and taxation industry knowledge base includes the following steps: S11: Collect invoice data of the finance and taxation industry, including text content of text, paragraphs, and documents; S12: Processing the text content into a set length through a slicing algorithm and converting it into text data; S13: The text data is converted into a text vector containing semantic information through a text embedding model and stored in a vector database.

3. The intelligent invoicing method based on the finance and taxation industry big model according to claim 1 is characterized by: In step S2, the data preprocessing module determines the data format of the input data, decomposes the input data through a word parser, an excel parser, and a PDF parser, and splits the input data into data in two formats: text and image.

4. The intelligent invoicing method based on the finance and taxation industry big model according to claim 1 is characterized by: In step S3, the text data is converted into a text vector using the intelligent invoicing module, and the text vector is used to search the finance and taxation industry knowledge base, and the acquired finance and taxation industry knowledge is input into the finance and taxation industry big model, and the image data is input into the finance and taxation industry big model at the same time, including the following steps: S21: The intelligent invoicing module converts the text data processed by the data preprocessing module into a text vector through a text embedding model; S22: Search the finance and taxation industry knowledge base through the text vector, and obtain the finance and taxation industry knowledge related to the text data through retrieval and recall; S23: Feed the recalled industry knowledge to the pre-trained large language model of the base to form a large model of the finance and taxation industry with finance and taxation industry knowledge; S24: The image data is directly input into the big model of the finance and taxation industry.

5. The intelligent invoicing method based on the finance and taxation industry big model according to claim 4 is characterized by: In step S22, the recall method takes the data whose cosine distance is greater than the threshold value. The formula of cosine distance is: Where cos(A,B) represents cosine similarity, A represents the input text vector, and B represents the knowledge base text vector.

6. The intelligent invoicing method based on the finance and taxation industry big model according to claim 5 is characterized by: The finance and taxation industry model is designed through prompt engineering and can automatically extract the enterprise information and product information contained in the text and pictures according to the set format.

7. The intelligent invoicing method based on the finance and taxation industry big model according to claim 3 is characterized by: The data preprocessing module determines the format of the input data. If the input data is a word file, the content of the word file is extracted in two formats, text and picture, through the word parser; if the input data is an excel file, the content of the excel file is extracted in two formats, text and picture, through the excel parser; if the input data is a PDF file, the content of the PDF file is extracted in two formats, text and picture, through the PDF parser; if the input data is text and picture, the text and picture formats are kept unchanged.

8. An intelligent invoicing system based on a large model of the finance and taxation industry, characterized in that: include: Finance and taxation industry knowledge base: For finance and taxation industry invoice data, the corresponding text content is converted into text data through the slicing algorithm, and the text data is converted into a high-dimensional dense text vector containing semantic information through the text embedding model, and stored in the vector database; Data preprocessing module: determines the data format of the input data, decomposes the input data through word parser, excel parser, and PDF parser, and uniformly splits it into two formats: text and image. Intelligent invoicing module: used to convert text data into text vectors, search the finance and taxation industry knowledge base through the text vectors, and input the acquired finance and taxation industry knowledge into the finance and taxation industry big model; Manual verification module: Manually verify and correct the invoicing information extracted from the large model of the finance and taxation industry.

9. The intelligent invoicing system based on the finance and taxation industry big model according to claim 8 is characterized by: The finance and taxation industry model is designed through prompt engineering and can automatically extract the enterprise information and product information contained in the text and pictures according to the set format.

10. The intelligent invoicing system based on the finance and taxation industry big model according to claim 8 is characterized by: The image data obtained through the data preprocessing module is directly input into the large model of the finance and taxation industry.

Images