Document segmentation processing method and device, computer equipment and readable storage medium
By determining the document type and element type and performing targeted segmentation processing, the problem of multimodal information extraction in intelligent document processing is solved, and the accuracy of intelligent document processing and information utilization is improved.
Patent Information
- Application Number
- CN202411547514.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-11-01
- Publication Date
- 2024-12-03
- Estimated Expiration
- 2044-11-01
AI Technical Summary
The prior art cannot effectively extract multimodal information in documents, resulting in insufficient information utilization in document intelligent processing.
By obtaining the document type of the to-process document, it is routed to the corresponding document processing system, pre-processing is performed to determine the type of the target document element, and the slicing strategy is determined according to the element type for slicing processing, extracting multimodal information.
It realizes accurate segmentation of different document types and effective extraction of multimodal information, is compatible with different document types, and improves the accuracy and information utilization of intelligent document processing.
Smart Images

Figure CN119066037A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of artificial intelligence technology, and in particular to a document segmentation processing method, device, computer equipment and readable storage medium. Background Art
[0002] With the development of large language models and multimodal large language models, the field of intelligent documents has begun to experience explosive growth. Document intelligence mainly refers to the understanding, classification, extraction, and information summarization of the text and layout information contained in scanned documents through artificial intelligence technology. The structured data obtained based on document intelligence technology can also be used for higher-level intelligent applications, such as intelligent question and answer.
[0003] In the related art, the document segmentation processing method is based on the use of natural language processing technology to identify natural dividing points in the text and segment the document into smaller text blocks containing independent semantic information. However, the related technology cannot extract useful multimodal information from the document. Therefore, how to extract useful multimodal information from the document has become a difficulty at the current stage. Summary of the invention
[0004] Based on this, it is necessary to provide a document segmentation processing method, device, computer equipment, computer-readable storage medium and computer program product that can extract useful multimodal information and ensure the accuracy of document segmentation in response to the above technical problems.
[0005] In a first aspect, the present application provides a document segmentation processing method, comprising: Obtaining a document to be processed, and determining the document type of the document to be processed; Routing to a corresponding document processing system according to the document type, and determining a document preprocessing strategy corresponding to the document type; Preprocessing the document to be processed according to the document preprocessing strategy to determine the element type of the target document element in the document to be processed; A corresponding segmentation strategy is determined according to the element type, and segmentation processing is performed on the document to be processed based on the segmentation strategy to obtain a segmentation processing result.
[0006] In one embodiment, preprocessing the document to be processed according to the document preprocessing strategy to determine the element type of the target document element in the document to be processed includes: When the document type is the first document type and the target document element includes at least one of a picture and a table, locating the position of the target document element in the document to be processed, extracting the document to be processed based on the position, and obtaining an extracted image; The content in the extracted image is identified to determine the element type of the target document element.
[0007] In one embodiment, The preprocessing of the document to be processed according to the document preprocessing strategy to determine the element type of the target document element in the document to be processed includes: When the document type is the second document type and the target document element includes only a table, determining the table as the element type of the target document element in the document to be processed; When the document type is the third document type, and the target document element includes only text, and the language type corresponding to the text includes at least one, determining that the element type of the target document element in the document to be processed is multilingual text; When the document type is the fourth document type and the target document element exists in the form of a programming language, the document to be processed is parsed to determine the element type of the target document element in the document to be processed.
[0008] In one embodiment, determining a corresponding segmentation strategy according to the element type, segmenting the document to be processed based on the segmentation strategy, and obtaining a segmentation result includes: In the case where the element type is a table, determining the number of split rows of the table to be processed in the document to be processed; Starting from the first row of the table to be processed, the table to be processed is segmented in a sliding window manner based on the segmentation window length until the segmentation of the table to be processed is completed, and the table segments obtained after each segmentation are formatted to obtain table segments in csv format; In each segmentation starting from the second switching, the starting position of the next segmentation for retaining the adjacent row information is determined according to the number of segmentation rows, and segmentation is performed from the starting position based on the segmentation window length.
[0009] In one embodiment, before determining the number of rows to be segmented into for the table in the to-be-processed document, the method further includes: In the case where the element type is a table, if the document type is a first document type and the target document element is a picture table, the picture table is identified to obtain table information of the picture table; The image table is converted into a table to be processed.
[0010] In one embodiment, determining a corresponding segmentation strategy according to the element type, segmenting the document to be processed based on the segmentation strategy, and obtaining a segmentation result includes: In the case where the element type is a picture, the document to be processed is parsed to determine a picture-text pair in the document to be processed and position information of the picture-text pair, wherein the picture-text pair includes the picture to be processed and a title text associated with the picture to be processed; Determine a first center point position according to a first position of the image to be processed, and determine a second center point position according to a second position of the title text; Determine a center point distance according to the first center point position and the second center point position; The target title text corresponding to the minimum center point distance is determined, and the description text associated with the target title text is determined as the target description text matching the image to be processed.
[0011] In one embodiment, the method further comprises: If there are at least two target title texts corresponding to the center point distance within the preset range, determine the candidate description texts associated with each of the target title texts; Obtaining a prompt text for determining a target description text that matches the image to be processed; Based on the prompt text and the large language model, the image to be processed is matched with each of the candidate description texts respectively, and a target description text matching the image to be processed is determined from the candidate description texts.
[0012] In one embodiment, the target document element further includes a formula, and the corresponding segmentation strategy is determined according to the element type, and the document to be processed is segmented based on the segmentation strategy to obtain a segmentation result, including: In the case where the element type is a formula type, determining a formula to be processed in the document to be processed and context information associated with the formula to be processed; A formula description corresponding to the formula to be processed is determined according to the context information.
[0013] In one embodiment, determining a corresponding segmentation strategy according to the element type, segmenting the document to be processed based on the segmentation strategy, and obtaining a segmentation result includes: In the case where the element type is multilingual text, translating the text to be processed in the document to be processed, and determining a target text in a preset language that matches the text to be processed from a preset language library; The target text is segmented based on a preset text processing scheme to obtain a segmentation result.
[0014] In a second aspect, the present application also provides a document segmentation processing device, comprising: A data processing module is used to obtain a document to be processed and determine the document type of the document to be processed; A routing module, used for routing to a corresponding document processing system according to the document type, and determining a document preprocessing strategy corresponding to the document type; A preprocessing module, used to preprocess the document to be processed according to the document preprocessing strategy, and determine the element type of the target document element in the document to be processed; The segmentation module is used to determine a corresponding segmentation strategy according to the element type, and perform segmentation processing on the document to be processed based on the segmentation strategy to obtain a segmentation processing result.
[0015] In a third aspect, the present application further provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the following steps are implemented: Obtaining a document to be processed, and determining the document type of the document to be processed; Routing to a corresponding document processing system according to the document type, and determining a document preprocessing strategy corresponding to the document type; Preprocessing the document to be processed according to the document preprocessing strategy to determine the element type of the target document element in the document to be processed; A corresponding segmentation strategy is determined according to the element type, and segmentation processing is performed on the document to be processed based on the segmentation strategy to obtain a segmentation processing result.
[0016] In a fourth aspect, the present application further provides a computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the following steps are implemented: Obtaining a document to be processed, and determining the document type of the document to be processed; Routing to a corresponding document processing system according to the document type, and determining a document preprocessing strategy corresponding to the document type; Preprocessing the document to be processed according to the document preprocessing strategy to determine the element type of the target document element in the document to be processed; A corresponding segmentation strategy is determined according to the element type, and segmentation processing is performed on the document to be processed based on the segmentation strategy to obtain a segmentation processing result.
[0017] In a fifth aspect, the present application further provides a computer program product, including a computer program, which implements the following steps when executed by a processor: Obtaining a document to be processed, and determining the document type of the document to be processed; Routing to a corresponding document processing system according to the document type, and determining a document preprocessing strategy corresponding to the document type; Preprocessing the document to be processed according to the document preprocessing strategy to determine the element type of the target document element in the document to be processed; A corresponding segmentation strategy is determined according to the element type, and segmentation processing is performed on the document to be processed based on the segmentation strategy to obtain a segmentation processing result.
[0018] The above-mentioned document segmentation processing method, device, computer equipment, computer-readable storage medium and computer program product, when segmenting a document, route it to a document processing system corresponding to the document type according to the document type, process documents of different document types according to the document preprocessing strategy of the document processing system, and determine the element type of the target document element in the document to be processed; determine the corresponding segmentation strategy according to the element type, and segment the document to be processed based on the segmentation strategy to obtain the segmentation processing result. This method is compatible with documents of different document types, that is, it can process different types of documents, and at the same time, perform corresponding segmentation processing according to the segmentation strategy corresponding to each target document element. This effective processing method ensures the accuracy of segmentation, and can also effectively extract target document element information of different modalities. BRIEF DESCRIPTION OF THE DRAWINGS
[0019] In order to more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the drawings required for use in the embodiments of the present application or related technical descriptions will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.
[0020] Figure 1 An application environment diagram of a document segmentation processing method in one embodiment; Figure 2 A schematic diagram of a flow chart of a document segmentation processing method in one embodiment; Figure 3 A schematic diagram of a flow chart of a table segmentation processing method in one embodiment; Figure 4 A schematic diagram of a process of image segmentation method in one embodiment; Figure 5 is a structural block diagram of a document segmentation processing device in one embodiment; Figure 6 FIG. 4 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION
[0021] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.
[0022] layoutLM: An algorithm for layout detection based on text position and text background.
[0023] LLM (Large Language Model): Large language model, a currently popular language model based on generative algorithms such as chatgpt.
[0024] RAG (Retieve Augmented Generate): An algorithmic model that combines external text for retrieval and superimposes a large language model for understanding and generation.
[0025] With the development of large language models and multimodal large language models, the field of intelligent documents has begun to usher in an explosive growth trend. In the field of related business consulting, consulting services can be provided to consultants by building an intelligent question-and-answer system. Building an intelligent question-and-answer system involves document segmentation and building an intelligent question-and-answer database. In related technologies, document segmentation is a text-based LLM (Large Language Model) solution. With the continuous upgrading of business needs and user needs, the output information of a single modality cannot meet the needs of users. Therefore, how to extract useful multimodal information from documents has become a difficulty at the current stage. Among them, multimodal information includes but is not limited to pictures, formulas, and tables.
[0026] For example, the processing method of Excel documents is rather special. Direct input into LLM may result in data that is too long, and the header information is missing. How to effectively input the table data is a problem. For word / pdf documents containing pictures / formulas / tables, how to effectively extract the correspondence between these media and the corresponding explanatory text is a difficulty. Secondly, for multilingual document information, languages of different regions will be involved in international scenarios. For documents in multiple languages, how to translate and align semantic information becomes another difficulty.
[0027] In order to solve the technical problem of how to extract useful multimodal information from documents, the document to be processed is obtained to determine the document type; the document is routed to the corresponding document processing system according to the document type, and the document preprocessing strategy corresponding to the document type is determined; the document to be processed is preprocessed according to the document preprocessing strategy, and the element type of the target document element in the document to be processed is determined; the corresponding segmentation strategy is determined according to the element type, and the document to be processed is segmented based on the segmentation strategy to obtain the segmentation processing result.
[0028] The document segmentation processing method provided in the embodiment of the present application can be applied to Figure 1 In the application environment shown. Among them, the terminal 102 communicates with the server 104 through the network. The data storage system can store the data that the server 104 needs to process. The data storage system can be integrated on the server 104, or it can be placed on the cloud or other network servers. Document processing systems corresponding to different document types are deployed on the terminal. By obtaining the document to be processed, the document type of the document to be processed is determined; according to the document type, it is routed to the corresponding document processing system to determine the document preprocessing strategy corresponding to the document type; the document to be processed is preprocessed according to the document preprocessing strategy to determine the element type of the target document element in the document to be processed; according to the element type, the corresponding segmentation strategy is determined, and the document to be processed is segmented based on the segmentation strategy to obtain the segmentation processing result. Among them, the terminal 102 can be, but is not limited to, various personal computers, laptops, smart phones, tablet computers, Internet of Things devices and portable wearable devices. The Internet of Things devices can be smart speakers, smart TVs, smart air conditioners, smart car devices, projection devices, etc. The server 104 can be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides cloud computing services.
[0029] In an exemplary embodiment, Figure 2 As shown, a document segmentation processing method is provided, which is applied to Figure 1 The terminal in is taken as an example to illustrate, including the following steps 202 to 206. Among them: Step 202: Obtain the document to be processed and determine the document type of the document to be processed.
[0030] The document to be processed is determined based on the document access entry. If the document to be processed is a web page type, the document access entry can be a web page link address or a graphic code. For example, if the document to be processed is web page data such as html, it is necessary to call the web page acquisition system to obtain page information, save it locally in the form of html, and download the image to local storage. If the document to be processed is not a web page type, the document access entry can be a document identifier.
[0031] Exemplarily, in response to a document processing instruction, a corresponding document to be processed is obtained according to a document access entry carried in the document processing instruction.
[0032] Step 204 , routing the document to a corresponding document processing system according to the document type, and determining a document pre-processing strategy corresponding to the document type.
[0033] Among them, the terminal deploys document processing systems corresponding to different document types, and different document processing systems are equipped with document preprocessing strategies corresponding to the document types. That is to say, the document segmentation processing method in this example can process different types of documents.
[0034] The document preprocessing strategy includes: for a document type of the first document type, the document to be processed corresponding to the first document type includes at least target document elements such as tables, pictures, formulas and titles, determining the positions of the tables, pictures, formulas and titles in the document corresponding to the first document type, obtaining an extracted image of the first document type and identifying the content in the extracted image, and segmenting according to the identification result. The first document type can be but is not limited to being represented as ppt, word or pdf, etc.; for a second document type including only pure tables, such as an excel file, the document corresponding to the second document type can be directly segmented; for a third document type including only text, the language type corresponding to the text includes at least one, and the text of the document corresponding to the third document type needs to be translated first to obtain text in a preset language, and on this basis, the translated text is segmented; for a fourth document type in the form of a programming language, the document corresponding to the fourth document type can be directly segmented.
[0035] Exemplarily, the document is routed to a corresponding document processing system according to the document type, and a document preprocessing strategy corresponding to the document type is obtained.
[0036] Step 206: pre-process the document to be processed according to the document pre-processing strategy, and determine the element type of the target document element in the document to be processed.
[0037] The document to be processed includes at least one modality of target document elements, which can also be called element types. Element types include but are not limited to one or more of tables, pictures, formulas, and text. Target document elements include but are not limited to one or more of tables, pictures, formulas, text, and titles. Target document elements can also be called media. For example, the media in the document to be processed include tables, pictures, formulas, and titles.
[0038] Exemplarily, if the document to be processed is a first document type including tables, pictures, formulas and titles, obtain a document preprocessing strategy corresponding to the first document type, determine the locations of the tables, pictures, formulas and titles in the document corresponding to the first document type, obtain an extracted image of the first document type, identify the content in the extracted image, and determine the element type in the extracted image.
[0039] Step 208, determining a corresponding segmentation strategy according to the element type, and performing segmentation processing on the document to be processed based on the segmentation strategy to obtain a segmentation processing result.
[0040] Among them, there are corresponding table segmentation strategies for tables, corresponding formula segmentation strategies for formulas, corresponding image segmentation strategies for pictures, and corresponding text segmentation strategies for text in pure text documents. The segmentation strategy corresponding to the table can be to add a header and segment the number of rows according to the maximum number of input tokens, which can reduce the number of tokens while maintaining the field information of the table; and for the relationship information of adjacent rows of the table, the sequential interception of the sliding window scheme can effectively maintain the position relationship of adjacent rows. The segmentation strategy corresponding to the picture can be to determine the description text matching the picture based on the matching relationship between the shortest distance picture / title and the description text, and obtain the most matching picture-text relationship pair. The segmentation strategy corresponding to the formula can be to determine the context information of the text position of the formula and determine the formula description matching the formula. The segmentation strategy corresponding to multilingual text can be to segment the document according to a preset text processing scheme.
[0041] Exemplarily, the corresponding segmentation strategy is determined according to the element type, and the document to be processed is segmented based on the segmentation strategy to obtain the segmentation processing result. Furthermore, the segmentation processing result can be processed according to the data processing method corresponding to each element type to obtain the processing result, and the processing result is stored in the vector database according to each corresponding storage format. That is, the segmentation processing result corresponding to the element type is subjected to data processing to obtain the processing result corresponding to the segmentation processing result, and according to the storage format corresponding to the element type, the processing result is stored in the vector database according to the storage format. Among them, data processing includes vector processing and format conversion, and vector processing can be vector processing for the segmentation processing results of element types such as text, pictures and formulas to obtain the corresponding vector representations. Format conversion can be for a table, and the segmentation processing result of the table is format converted to obtain a table in csv format.
[0042] For example, the segmentation processing results corresponding to the text class are directly embedded through the Embedding algorithm to obtain vectors and then stored in the vector database. The meta information saves the document location, title, page, number of rows and other information; the segmentation processing results corresponding to the table class are embedded in the header + data row csv format and stored in the database; the segmentation processing results of the image class are stored in the form of image vectors and description text vectors respectively; the segmentation processing results corresponding to the formula class are embedded in the form of formula screenshots and corresponding formula descriptions and then stored in the database; among them, the vector database can use the Chromdb database, and the open source solution of this database is more scalable.
[0043] It should be noted that the above method can be applied to but not limited to the RAG system. The RAG system is compatible with various document processing solutions and can effectively process various media such as pictures, texts, formulas, tables, etc., so that multiple document formats can be processed at the same time, and the dialogue system of the RAG and LLM models can provide more data forms to improve the user experience.
[0044] In the above document segmentation and processing method, when segmenting a document, it is routed to a document processing system corresponding to the document type according to the document type, and documents of different document types are processed according to the document preprocessing strategy of the document processing system to determine the element type of the target document element in the document to be processed; the corresponding segmentation strategy is determined according to the element type, and the document to be processed is segmented based on the segmentation strategy to obtain a segmentation processing result. This method is compatible with documents of different document types, that is, it can process different types of documents, and at the same time, the corresponding segmentation processing is performed according to the segmentation strategy corresponding to each target document element. This effective processing method ensures the accuracy of the segmentation, and can also effectively extract target document element information of different modalities.
[0045] In an exemplary embodiment, preprocessing the document to be processed according to the document preprocessing strategy to determine the element type of the target document element in the document to be processed includes the following situations: Case 1: When the document type is the first document type and the target document element includes at least one of a picture and a table, locate the position of the target document element in the document to be processed, extract the document to be processed based on the position, and obtain an extracted image; identify the content in the extracted image to determine the element type of the target document element.
[0046] The target document element of the first document type may include at least one or more of a picture, a table, a formula and a title.
[0047] Exemplarily, according to the document preprocessing strategy corresponding to the first document type, the position of the target document element in the document to be processed is located, the position interval of each target document element in the text is determined based on the position, and the content corresponding to the target document element is screenshotted according to the position interval to obtain at least one screenshot, and the extracted image is determined according to the at least one screenshot, and the content in the extracted image is identified to determine the element type of the target document element. Among them, the position of the target document element can be determined by calling the LayoutLMv3 model, for example, calling the LayoutLMv3 model to obtain the position of the medium (picture, table, formula, title), and taking a screenshot of it. Determining the element type of the target document element can be achieved by existing methods. For example, if the target document element is a table and the table is a picture, an OCR (Optical Character Recognition) (such as Tesseract-OCR or PaddlePaddleOCR, etc.) recognition system can be called to identify the content in the extracted image.
[0048] Case 2: When the document type is the second document type and the target document element includes only a table, it is determined that the element type of the target document element in the document to be processed is a table.
[0049] The second document type is a pure table file, for example, an Excel file.
[0050] Case three: when the document type is the third document type, and the target document element includes only text, and the language type corresponding to the text includes at least one, it is determined that the element type of the target document element in the document to be processed is multilingual text.
[0051] The third document type is a document including only text, the language type of the text is not a preset language type, and the language type of the text of the document corresponding to the third document type is not a preset language type and includes at least one language type.
[0052] Case 4: When the document type is the fourth document type and the target document element includes at least one of a picture, a table, a formula and a title, if the target document element exists in the form of a programming language, the document to be processed is parsed to determine the element type of the target document element in the document to be processed.
[0053] The programming language may be, but is not limited to, HTML. The element types of the target document elements in the document to be processed include at least one or more of a picture, a table, a formula, and a title.
[0054] In the above method, different document types are routed to corresponding document processing systems, and preprocessing processes of different document formats are implemented based on each system to be compatible with documents of different formats.
[0055] It is understandable that the processing method of Excel documents is quite special. Directly inputting LLM data may result in too long data, and the header information is missing. How to effectively input the table data is a problem. Therefore, in an exemplary embodiment, Figure 3 As shown, a table segmentation processing method is provided, including steps 302 to 304, wherein: Step 302: When the element type is a table, determine the number of rows of the table to be processed in the document to be processed.
[0056] Among them, considering that the number of table rows is relatively large, the text window that the LLM model context can process is limited, that is, there is a context window. In order to avoid data loss and low segmentation accuracy during table segmentation, and to ensure that there is enough space to process other information (for example, system prompt words, top-k recall context, etc.), it is necessary to determine the preset number of segmentation rows. For example, since the maximum processing character of the LLM model is 8192, half of the default length, that is, 4096 characters, is chosen as the processing window. In actual processing, the maximum number of characters of each input segment is limited to 1024 characters, that is, the maximum number of segmentation rows allowed is determined. The purpose is to ensure that when processing long tables, each segment will not exceed the processing capacity of the LLM model, and also to leave space for other necessary context information.
[0057] That is, determining the number of split rows in this way can divide long table data into small chunks suitable for LLM model processing, while ensuring that there is sufficient contextual information to support the model's understanding and generation capabilities.
[0058] Step 304 , starting from the first row of the table to be processed, the table to be processed is segmented in a sliding window manner based on the segmentation window length until the segmentation of the table to be processed is completed, and the table segments obtained after each segmentation are formatted and converted to obtain table segments in csv format.
[0059] In each splitting starting from the second splitting, the starting position of the next splitting for retaining the adjacent row information is determined according to the number of split rows, and the splitting is performed from the starting position based on the splitting window length.
[0060] It is understandable that the table in the related technology has been sorted in a certain order. At this time, if it is split in order or randomly, the information of adjacent rows may be lost. For this reason, we randomly select a starting position in the original table in order and according to a certain window (number of rows) to start intercepting again.
[0061] Exemplarily, the table is segmented according to the maximum number of split rows determined in step 302, and the first row of the table is used as the table header schema, that is, the meaning of each column field, that is, a table header is added for each split, and the table to be processed is segmented in a sliding window manner from the first row of the table to be processed based on the split window length. The first split starts from the first row of data in the table, and each split starting from the second split determines the starting position of the next split for retaining adjacent row information according to the number of split rows, and splits from the starting position based on the split window length to obtain table fragments for each split until the table to be processed is split. The table fragments obtained after each split are formatted and converted to obtain table fragments in csv format. For example, the number of split rows is 4, the first split is from the first row to the fourth row, the starting position of the second split is determined to be the second row, and the second split is from the second row to the fifth row.
[0062] The table format conversion can be converted into a matrix using the pandas dataFrame library, and then converted into a segmented csv format using to_csv. The starting position of the second segmentation is determined from the table row of the last segmentation, so that each segmentation can maintain the adjacent row information and avoid the loss of table information leading to inaccurate table segmentation. Furthermore, the LLM model is used to summarize the paragraphs before and after the table to obtain the subheadings of the table.
[0063] It is understandable that the tables in both the first document type and the second document type can be segmented in the above manner. However, in an exemplary embodiment, when the element type is a table, if the document type is the first document type and the target document element is a picture table, the picture table is identified to obtain the table information of the picture table; the picture table is converted into a table to be processed. In other words, it is necessary to perform OCR recognition on the table in the picture format to obtain the corresponding text information, and then convert the obtained text information into a table, and then perform the above table segmentation steps. Furthermore, if the table in the document corresponding to the first document type is in a chart format, it does not need to be identified, and the table segmentation can be performed directly in the above manner.
[0064] In this embodiment, by segmenting the table, adding table headers respectively, and segmenting according to the maximum number of segmentation rows, the field information of the table is maintained while reducing the number of tokens; further, the relationship information of adjacent rows of the table is intercepted sequentially using a sliding window solution, which can effectively maintain the position relationship of adjacent rows, avoid missing header information, and determine valid input for table data. It should be noted that this table segmentation processing method can be applied to but not limited to table segmentation of document types such as ppt / word / excel / pdf.
[0065] In the related art, it is difficult to effectively extract the corresponding relationship between the image and the corresponding caption. Therefore, in an exemplary embodiment, Figure 4 As shown, a method for segmenting an image is provided, including steps 402 to 408, wherein: Step 402, when the element type is a picture, the document to be processed is parsed to determine the position information of the image-text pair in the document to be processed, wherein the image-text pair includes the image to be processed and the title text associated with the image to be processed.
[0066] The document to be processed may be a first document type, and the document parsing method may be to parse the model document using the LayoutLM model to obtain the first position information of the image and the second position information of the title text. The number of title texts associated with the image may be one or more.
[0067] Step 404, determining the first center point position according to the first position of the image to be processed, and determining the second center point position according to the second position of the title text.
[0068] Among them, the method of determining the first center position and the second center position can be achieved through existing methods, which will not be described in detail here.
[0069] Step 406, determining the center point distance according to the first center point position and the second center point position.
[0070] Among them, the method for confirming the center point distance can be achieved through existing methods, which will not be described in detail here.
[0071] Step 408, determining the target title text corresponding to the minimum center point distance, and determining the description text associated with the target title text as the target description text that matches the image to be processed.
[0072] Among them, the method for determining the description text associated with the title text can be to parse the document to be processed, determine the position information of the title text and the description text in the document to be processed, determine the respective center point positions according to the position information of the title text and the description text, determine the center point distance between the title text and the description text based on the center point position, and determine the description text corresponding to the smallest center point distance as the description text associated with the title text. Furthermore, if the language type corresponding to the title text and the description text is not a preset language type, a long text model such as the Giraffe long-distance memory model is called for translation to obtain the title text and description text of the preset language type.
[0073] Exemplarily, the title text corresponding to the smallest center point distance is determined as the target title text, the description text associated with the target title text is determined as the target description text that matches the image to be processed, and the corresponding target image-text pair is determined based on the target description text and the image to be processed.
[0074] Furthermore, in an exemplary embodiment, if there are at least two target title texts corresponding to the center point distance within a preset range, the candidate description texts associated with each target title text are determined; a prompt text for determining the target description text matching the image to be processed is obtained; based on the prompt text and the large language model, the image to be processed is matched with each candidate description text respectively, and the target description text matching the image to be processed is determined from the candidate description texts.
[0075] Among them, if there are at least two target title texts corresponding to the center point distance within the preset range, it can be understood that there are at least two target title texts within the center distance range, and the distance difference between the two target title texts does not exceed the preset distance, and the preset distance can be the distance corresponding to 20 pixels. For example, there are title texts above and below a picture at the same time, and the distance difference between the title texts does not exceed the distance corresponding to 20 pixels. The prompt text is preset, and the text content of the prompt text includes "determine the caption text with the highest matching degree with the picture to be processed from the provided picture-text pairs."
[0076] Exemplarily, if there are at least two target title texts for the image to be processed, determine the candidate description texts associated with each target title text, input the correspondence between each candidate description text and the image to be processed and the prompt text into the large language model, and output the target description text matching the image to be processed.
[0077] In the above embodiment, by using the matching relationship between the shortest distance image / title and the caption text, when there are multiple matching relationships, the LVM visual language model is used for evaluation to obtain the most matching image-text relationship pair. It should be noted that this image segmentation processing method can be applied to but not limited to the image segmentation of document types such as ppt / word / excel / pdf.
[0078] Regarding the segmentation strategy for documents including formulas, optionally, in an exemplary embodiment, when the element type is a formula type, the formula to be processed in the document to be processed and the context information associated with the formula to be processed are determined; and the formula description corresponding to the formula to be processed is determined based on the context information.
[0079] It is understandable that the formula category is generally relatively clear in content, so a summary can be obtained by combining the context information where the formula is mentioned, that is, a formula description of the formula is obtained. Among them, determining the context information associated with the formula to be processed can be achieved through existing methods, which will not be elaborated here.
[0080] For a document that is a multilingual document, optionally, in an exemplary embodiment, when the element type is multilingual text, the text to be processed in the document to be processed is translated, and a target text in a preset language that matches the text to be processed is determined from a preset language library; the target text is segmented based on a preset text processing scheme to obtain a segmentation processing result.
[0081] Among them, translation can be performed by calling long text models such as Long-LLAM to translate the full text, and semantically aligning the translated language to make the text before and after translation consistent in semantics. By translating multilingual documents, it is possible to segment documents in different languages and realize multilingual processing, which can improve user experience in practical applications.
[0082] In an exemplary embodiment, a document segmentation processing method is provided, wherein a first document type is any one of ppt, word or pdf including pictures, tables, formulas and titles, a second document type is excel including only tables, a third document type is any one of ppt, word or pdf including only text, and a fourth document type is html including at least one of pictures, tables, formulas and titles.
[0083] Obtain the document to be processed and determine the document type of the document to be processed; route to the corresponding document processing system according to the document type, and determine the document preprocessing strategy corresponding to the document type; When the document type is the first document type, and the target document element includes at least one of a picture and a table, the position of the target document element in the document to be processed is located, and the document to be processed is extracted based on the position to obtain an extracted image; the content in the extracted image is identified to determine the element type of the target document element. When the document type is the second document type, and the target document element only includes a table, the table is determined as the element type of the target document element in the document to be processed; when the document type is the third document type, and the target document element only includes text, and the language type corresponding to the text includes at least one, the element type of the target document element in the document to be processed is determined to be multilingual text; when the document type is the fourth document type, and the target document element exists in the form of a programming language, the document to be processed is parsed to determine the element type of the target document element in the document to be processed.
[0084] For a table in a first document type or a second document type, determine the number of split rows of a table to be processed in the document to be processed; split the table to be processed in a sliding window manner based on the split window length from the first row of the table to be processed until the splitting of the table to be processed is completed, and convert the table fragments obtained after each splitting to obtain table fragments in csv format; wherein, for each splitting starting from the second switch, determine the starting position of the next splitting for retaining adjacent row information according to the number of split rows, and split from the starting position based on the split window length.
[0085] When the element type in the to-be-processed document corresponding to the first document type is a picture, the to-be-processed document is parsed to determine a picture-text pair and position information of the picture-text pair in the to-be-processed document, wherein the picture-text pair includes the to-be-processed picture and a title text associated with the to-be-processed picture; Determine the first center point position according to the first position of the image to be processed, and determine the second center point position according to the second position of the title text; determine the center point distance according to the first center point position and the second center point position; determine the target title text corresponding to the minimum center point distance, and determine the description text associated with the target title text as the target description text matching the image to be processed. If there are at least two target title texts corresponding to the center point distance within the preset range, determine the candidate description texts associated with each target title text; obtain the prompt text for determining the target description text matching the image to be processed; match the image to be processed with each candidate description text based on the prompt text and the large language model, and determine the target description text matching the image to be processed from the candidate description texts.
[0086] In the case where the element type in the first document type is a formula type, a formula to be processed in the document to be processed and context information associated with the formula to be processed are determined; and a formula description corresponding to the formula to be processed is determined according to the context information.
[0087] For the document to be processed corresponding to the third document type, the text to be processed in the document to be processed is translated, and a target text in a preset language matching the text to be processed is determined from a preset language library; the target text is segmented based on a preset text processing scheme to obtain a segmentation processing result.
[0088] For the document to be processed corresponding to the fourth document type, the document to be processed can be directly parsed, and based on the programming language format of the document to be processed, the target document element type can be identified, and the document to be processed can be segmented to obtain a segmentation result.
[0089] For different types of documents, the text class is directly embedded through the Embedding algorithm to obtain the vector and then stored in the vector database. The meta information saves the document location, title, page, number of rows and other information; the table data is embedded in the form of header + data row and stored in the database; the picture class is stored in the form of picture embedding + description embedding; the formula class is embedded in the form of formula screenshot and corresponding summary and then stored in the database to obtain a vector database.
[0090] It should be noted that the specific implementation of this embodiment can be achieved through the above-mentioned limitation method, which will not be elaborated here.
[0091] The above-mentioned document segmentation and processing method, when segmenting documents, routes them to the document processing system corresponding to the document type according to the document type, processes documents of different document types according to the document preprocessing strategy of the document processing system, and determines the element type of the target document element in the document to be processed; determines the corresponding segmentation strategy according to the element type, and performs segmentation processing on the document to be processed based on the segmentation strategy to obtain the segmentation processing result. This method segments the document into a suitable granularity and ensures the accuracy of the segmentation, solves the problem of automatic warehousing of various types of documents, and retains information other than text in the document, such as tables, pictures, flow charts, etc., provides multiple modal input materials for both RAG retrieval and production stages, and provides support for the input and output of graphic and text results of the dialogue system.
[0092] In actual applications, for documents published externally in word, ppt, html, pdf, etc., the above document segmentation processing method can quickly enter knowledge points into the knowledge base system, and at the same time can help complete the deduplication detection of the knowledge base; in internal and external customer service systems, the results can be directly displayed to customers in the form of graphics and text tables, which can quickly respond, improve user experience and reduce labor costs.
[0093] It should be understood that, although the steps in the flowcharts involved in the above embodiments are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear explanation in this article, the execution of these steps is not strictly limited in order, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above embodiments may include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the steps or stages in other steps.
[0094] Based on the same inventive concept, the embodiment of the present application also provides a document segmentation processing device for implementing the document segmentation processing method involved above. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme recorded in the above method, so the specific limitations in the one or more document segmentation processing device embodiments provided below can refer to the limitations of the document segmentation processing method above, and will not be repeated here.
[0095] In an exemplary embodiment, Figure 5 As shown, a document segmentation processing device is provided, including: a data processing module 502, a routing module 504, a pre-processing module 506 and a segmentation module 508, wherein: The data processing module 502 is used to obtain the document to be processed and determine the document type of the document to be processed.
[0096] The routing module 504 is used to route the document to the corresponding document processing system according to the document type and determine the document pre-processing strategy corresponding to the document type.
[0097] The preprocessing module 506 is used to preprocess the document to be processed according to the document preprocessing strategy, and determine the element type of the target document element in the document to be processed.
[0098] The segmentation module 508 is used to determine a corresponding segmentation strategy according to the element type, and perform segmentation processing on the document to be processed based on the segmentation strategy to obtain a segmentation processing result.
[0099] The above-mentioned document segmentation and processing device, when segmenting a document, routes the document to a document processing system corresponding to the document type according to the document type, processes documents of different document types according to the document preprocessing strategy of the document processing system, and determines the element type of the target document element in the document to be processed; determines the corresponding segmentation strategy according to the element type, and performs segmentation processing on the document to be processed based on the segmentation strategy to obtain a segmentation processing result. This method is compatible with documents of different document types, that is, it can process different types of documents, and at the same time, executes corresponding segmentation processing according to the segmentation strategy corresponding to each target document element. This effective processing method ensures the accuracy of segmentation, and can also effectively extract target document element information of different modalities.
[0100] In an exemplary embodiment, the preprocessing module 506 is further used to locate the position of the target document element in the to-be-processed document when the document type is the first document type and the target document element includes at least one of a picture and a table, and extract the to-be-processed document based on the position to obtain an extracted image; The content in the extracted image is identified to determine the element type of the target document element.
[0101] In an exemplary embodiment, the pre-processing module 506 is further configured to determine the table as the element type of the target document element in the document to be processed when the document type is the second document type and the target document element only includes a table; When the document type is the third document type, and the target document element includes only text, and the language type corresponding to the text includes at least one, it is determined that the element type of the target document element in the document to be processed is multilingual text; When the document type is the fourth document type and the target document element includes at least any one of a picture, a table, a formula and a title, if the target document element exists in the form of a programming language, the document to be processed is parsed to determine the element type of the target document element in the document to be processed.
[0102] In an exemplary embodiment, the segmentation module 508 is further used to determine the number of segmentation rows of the table to be processed in the document to be processed when the element type is a table; Starting from the first row of the table to be processed, the table to be processed is segmented in a sliding window manner based on the segmentation window length until the segmentation of the table to be processed is completed, and the table segments obtained after each segmentation are formatted and converted to obtain table segments in csv format; In each segmentation starting from the second switching, the starting position of the next segmentation for retaining the adjacent row information is determined according to the number of segmented rows, and segmentation is performed from the starting position based on the segmentation window length.
[0103] In an exemplary embodiment, the document segmentation processing device further includes a format conversion module, the format conversion module being used for, when the element type is a table, if the document type is a first document type and the target document element is a picture table, identifying the picture table to obtain table information of the picture table; Convert the image table to a table to be processed.
[0104] In an exemplary embodiment, the segmentation module 508 is further used to parse the document to be processed when the element type is a picture, and determine the image-text pair and the position information of the image-text pair in the document to be processed, wherein the image-text pair includes the image to be processed and the title text associated with the image to be processed; Determine a first center point position according to a first position of the image to be processed, and determine a second center point position according to a second position of the title text; Determine the center point distance according to the first center point position and the second center point position; The target title text corresponding to the minimum center point distance is determined, and the description text associated with the target title text is determined as the target description text matching the image to be processed.
[0105] In an exemplary embodiment, the segmentation module 508 is further configured to determine candidate description texts associated with each target title text if there are at least two target title texts corresponding to the center point distance within a preset range; Get the hint text used to determine the target description text that matches the image to be processed; Based on the prompt text and the large language model, the image to be processed is matched with each candidate description text respectively, and the target description text matching the image to be processed is determined from the candidate description texts.
[0106] In an exemplary embodiment, the segmentation module 508 is further configured to determine the formula to be processed in the document to be processed and context information associated with the formula to be processed when the element type is a formula type; A formula description corresponding to the formula to be processed is determined according to the context information.
[0107] In an exemplary embodiment, the segmentation module 508 is further used to translate the text to be processed in the document to be processed when the element type is a multi-language text, and determine a target text in a preset language that matches the text to be processed from a preset language library; The target text is segmented based on a preset text processing solution to obtain a segmentation result.
[0108] Each module in the document segmentation processing device can be implemented in whole or in part by software, hardware, or a combination thereof. Each module can be embedded in or independent of a processor in a computer device in the form of hardware, or can be stored in a memory in a computer device in the form of software, so that the processor can call and execute operations corresponding to each module.
[0109] In an exemplary embodiment, a computer device is provided. The computer device may be a terminal, and its internal structure diagram may be as shown in FIG. Figure 6 As shown. The computer device includes a processor, a memory, an input / output interface, a communication interface, a display unit and an input device. The processor, the memory and the input / output interface are connected through a system bus, and the communication interface, the display unit and the input device are connected to the system bus through the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be implemented through WIFI, a mobile cellular network, near field communication (NFC) or other technologies. When the computer program is executed by the processor, a document segmentation processing method is implemented. The display unit of the computer device is used to form a visually visible picture, which can be a display screen, a projection device or a virtual reality imaging device. The display screen can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covering the display screen, or a button, trackball or touchpad set on the computer device shell, or an external keyboard, touchpad or mouse.
[0110] Those skilled in the art will understand that Figure 6 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.
[0111] In one embodiment, a computer device is further provided, including a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the steps in the above method embodiments when executing the computer program.
[0112] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps in the above-mentioned method embodiments are implemented.
[0113] In one embodiment, a computer program product is provided, including a computer program, which implements the steps in the above method embodiments when executed by a processor.
[0114] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations.
[0115] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in this application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited to this. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, an artificial intelligence (AI) processor, etc., but are not limited to this.
[0116] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.
[0117] The above embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.
Claims
1. A document segmentation processing method, characterized in that: The method comprises: Obtaining a document to be processed, and determining the document type of the document to be processed; Routing to a corresponding document processing system according to the document type, and determining a document preprocessing strategy corresponding to the document type; Preprocessing the document to be processed according to the document preprocessing strategy to determine the element type of the target document element in the document to be processed; A corresponding segmentation strategy is determined according to the element type, and segmentation processing is performed on the document to be processed based on the segmentation strategy to obtain a segmentation processing result.
2. The method according to claim 1, characterized in that The preprocessing of the document to be processed according to the document preprocessing strategy to determine the element type of the target document element in the document to be processed includes: When the document type is the first document type and the target document element includes at least one of a picture and a table, locating the position of the target document element in the document to be processed, extracting the document to be processed based on the position, and obtaining an extracted image; The content in the extracted image is identified to determine the element type of the target document element.
3. The method according to claim 1, characterized in that The preprocessing of the document to be processed according to the document preprocessing strategy to determine the element type of the target document element in the document to be processed includes: When the document type is the second document type and the target document element includes only a table, determining the table as the element type of the target document element in the document to be processed; When the document type is the third document type, and the target document element includes only text, and the language type corresponding to the text includes at least one, determining that the element type of the target document element in the document to be processed is multilingual text; When the document type is the fourth document type and the target document element exists in the form of a programming language, the document to be processed is parsed to determine the element type of the target document element in the document to be processed.
4. The method according to any one of claim 2 or claim 3, characterized in that: The step of determining a corresponding segmentation strategy according to the element type, segmenting the document to be processed based on the segmentation strategy, and obtaining a segmentation result includes: In the case where the element type is a table, determining the number of split rows of the table to be processed in the document to be processed; Starting from the first row of the table to be processed, the table to be processed is segmented in a sliding window manner based on the segmentation window length until the segmentation of the table to be processed is completed, and the table segments obtained after each segmentation are formatted to obtain table segments in csv format; In each segmentation starting from the second switching, the starting position of the next segmentation for retaining the adjacent row information is determined according to the number of segmentation rows, and segmentation is performed from the starting position based on the segmentation window length.
5. The method according to claim 4, characterized in that Before determining the number of rows to be segmented into for the table in the to-be-processed document, the method further includes: In the case where the element type is a table, if the document type is a first document type and the target document element is a picture table, the picture table is identified to obtain table information of the picture table; The image table is converted into a table to be processed.
6. The method according to claim 2, characterized in that The step of determining a corresponding segmentation strategy according to the element type, segmenting the document to be processed based on the segmentation strategy, and obtaining a segmentation result includes: In the case where the element type is a picture, the document to be processed is parsed to determine a picture-text pair in the document to be processed and position information of the picture-text pair, wherein the picture-text pair includes the picture to be processed and a title text associated with the picture to be processed; Determine a first center point position according to a first position of the image to be processed, and determine a second center point position according to a second position of the title text; Determine a center point distance according to the first center point position and the second center point position; The target title text corresponding to the minimum center point distance is determined, and the description text associated with the target title text is determined as the target description text matching the image to be processed.
7. The method according to claim 6, characterized in that The method further comprises: If there are at least two target title texts corresponding to the center point distance within the preset range, determine the candidate description texts associated with each of the target title texts; Obtaining a prompt text for determining a target description text that matches the image to be processed; Based on the prompt text and the large language model, the image to be processed is matched with each of the candidate description texts respectively, and a target description text matching the image to be processed is determined from the candidate description texts.
8. The method according to claim 2, characterized in that: The target document element also includes a formula, wherein the corresponding segmentation strategy is determined according to the element type, and the document to be processed is segmented based on the segmentation strategy to obtain a segmentation result, including: In the case where the element type is a formula type, determining a formula to be processed in the document to be processed and context information associated with the formula to be processed; A formula description corresponding to the formula to be processed is determined according to the context information.
9. The method according to claim 3, characterized in that: The step of determining a corresponding segmentation strategy according to the element type, segmenting the document to be processed based on the segmentation strategy, and obtaining a segmentation result includes: In the case where the element type is multilingual text, translating the text to be processed in the document to be processed, and determining a target text in a preset language that matches the text to be processed from a preset language library; The target text is segmented based on a preset text processing scheme to obtain a segmentation result.
10. A document segmentation processing device, characterized in that: The device comprises: A data processing module is used to obtain a document to be processed and determine the document type of the document to be processed; A routing module, used for routing to a corresponding document processing system according to the document type, and determining a document preprocessing strategy corresponding to the document type; A preprocessing module, used to preprocess the document to be processed according to the document preprocessing strategy, and determine the element type of the target document element in the document to be processed; The segmentation module is used to determine a corresponding segmentation strategy according to the element type, and perform segmentation processing on the document to be processed based on the segmentation strategy to obtain a segmentation processing result.
11. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 9 are implemented.
12. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 9 are implemented.
13. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 9 are implemented.
Citation Information
Patent Citations
Extraction method for characters in form document image
CN103258198A
Document segmentation method and device and electronic equipment
CN111832258A
Document analysis method, system and device based on image recognition and storage medium
CN112989786A
Long text webpage generation method and device, electronic equipment and storage medium
CN113536182A
Document identification method and device based on multiple modes, equipment and storage medium
CN115131801A
Cited By
Data extraction method and device, electronic equipment and storage medium
CN119761358A
Multistage document segmentation method adopting self-adaptive dynamic partitioning algorithm
CN120951941A
Document processing method and device based on multi-modal large model, equipment and medium
CN122715171A
