A calibration-based PDF document page range matching method and system

Abstract

The application discloses a kind of based on calibration PDF document page range matching method and system, the method includes: obtaining the PDF tender document to be analyzed;PDF tender document is converted into editable tender document with the style structure of text content reservation, and the page number calibration and directory calibration are carried out to editable tender document, obtain standardized tender document;Acquire target review item for tender document, at least one search mode is searched in standardized tender document, obtain the page number range where the text associated with target review item is located;The search output content obtained in the directory of standardized tender document and / or text under at least one search mode is combined, and the effective review page number range of target review item is obtained.The application realizes the positioning effective review text content for expert review or for AI simulates expert review positioning, so that review process can focus on relevant content to generate review conclusion, improve review efficiency.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of PDF document processing technology, and particularly relates to a PDF document page range matching method and system based on calibration. BACKGROUND

[0002] In the field of bidding evaluation, under the traditional mode, the expert evaluators need to manually flip through the bidding documents, which are as few as 300 pages or as many as 2000 pages, and find the corresponding content for each evaluation item. This method completely relies on manpower and is extremely inefficient. The current information retrieval and positioning means for bidding PDF documents are limited, and some general document processing tools can achieve some functions, but there are obvious shortcomings. For example, when facing the problems of chaotic correspondence between bidding file directory and page number and missing page number, accurate PDF document evaluation item page positioning cannot be achieved, which makes it difficult to accurately lock the directory range related to the evaluation item in the huge document, and the return of a large amount of irrelevant information further increases the screening burden of the experts. SUMMARY

[0003] In view of the problems existing in the prior art, the present application provides a PDF document page range matching method and system based on calibration, which can provide effective evaluation page range for the evaluation process and improve the evaluation efficiency. The technical scheme is as follows:

[0004] In a first aspect, a PDF document page range matching method based on calibration is provided, including the following steps:

[0005] Obtaining a PDF bidding document to be analyzed;

[0006] Converting the PDF bidding document into an editable bidding document with the style structure of the main text being reserved, and performing page calibration and directory calibration on the editable bidding document to obtain a standardized bidding document;

[0007] Obtaining a target evaluation item for the bidding document, performing retrieval in the standardized bidding document based on at least one retrieval method, and obtaining the page range where the main text associated with the target evaluation item is located; wherein the retrieval method has a retrieval input content generation method corresponding to the retrieval method itself for generating retrieval input content with the original content of the target evaluation item as the data source, and the retrieval input content is used to execute the corresponding retrieval method to obtain the retrieval output content under the retrieval method; the retrieval input content is used to represent the key information of the target evaluation item; the retrieval output content includes the page range where the main text associated with the retrieval input content, i.e., the target evaluation item, is located;

[0008] Combining the retrieval output content obtained in the directory and / or the main text of the standardized bidding document under at least one retrieval method to obtain the effective evaluation page range of the target evaluation item.

[0009] In some implementations, the step of searching standardized tender documents based on at least one retrieval method to obtain the page range of the main text associated with the target review item includes:

[0010] Step A: Search the standardized tender document catalog using at least one search method to obtain the page range of the main text associated with the target review item, denoted as the catalog search page range;

[0011] And / or,

[0012] Step B involves searching the main text of the standardized tender document using at least one search method to obtain the page range of the main text associated with the target review item, denoted as the main text search page range; wherein the search method used in Step A may be the same as or different from the search method used in Step B, and the number of search methods used in Step A may be different from or the same as the number of search methods used in Step B.

[0013] In some implementations, after searching the standardized tender document catalog based on at least one search method to obtain the catalog search page range, the method further includes:

[0014] Step C: Based on at least one retrieval method, search the text within the range of page numbers in the directory retrieval to determine whether there is text in the text within the range of page numbers in the directory retrieval that is associated with the target review item; if so, proceed to step C1; otherwise, proceed to step C2; wherein, the retrieval method used in step A is different from or the same as the retrieval method used in step C.

[0015] Step C1: Based on the page range of the main text that is actually associated with the target review item within the page range of the directory search, the precise page range is used as the precise page range. The page range of the directory search and the precise page range are used as the search output content.

[0016] Step C2 determines that the page number range results from the directory search are invalid.

[0017] In some embodiments, the retrieval method includes a first retrieval method, which includes the following steps:

[0018] Based on the overall original content of the target review item, the first search input content is generated using a large language model. The content generated by the large language model includes the keywords, constraints, and prompt words of the target review item.

[0019] Based on the first search input content, a semantic association analysis is performed on the selected search location of the standardized bid document using a large language model to determine the first page code range where the text associated with the first search input content, i.e., the target review item, is located. The selected search location is the table of contents section and / or the text section of the bid document.

[0020] In some embodiments, the first search method includes the following steps:

[0021] Based on the overall original content of the target review item, a first search input content is generated using a large language model. The content generated by the large language model includes keywords, constraints, and prompts of the target review item.

[0022] Based on the first search input content, at least one of the following first method and second method is executed to determine the second page code range where the text associated with the target review item is located. The first page code range is determined based on the union of the execution results of the first method and the second method.

[0023] First method: Based on the first search input content, a semantic association analysis is performed on the table of contents section of the standardized bid document using a large language model to infer the directory title associated with the first search input content. Based on the page code of the associated directory title, the page code range where the text associated with the target review item is located is determined.

[0024] Second method: Based on the first search input content, a semantic matching analysis is performed on the text section of the standardized bid document using a large language model to determine the text content that semantically matches the first search input content. Based on the page code where the semantically matched text content is located, the page code range where the text associated with the target review item is located is determined.

[0025] In some embodiments, the search method has a second search method, which includes the following steps:

[0026] The tender document is obtained and parsed, and the content arrangement requirements for the bid document in the tender document are obtained.

[0027] Based on the content arrangement requirements, an association database of "review item - corresponding board - content element" is generated.

[0028] For the target review item, it is determined based on the association database whether there is a corresponding "review item - corresponding board - content element" data item. If it exists, the text content of the data item is used as the second search input content to search at the selected search location of the bid document, and the second page code range where the text associated with the target review item is located is obtained. The selected search location is the table of contents section and / or the text section of the bid document.

[0029] In some embodiments, after obtaining the second page code range where the text associated with the target review item is located, the method further comprises: checking whether the text in the second page code range is associated with the target review item, determining whether the content of the text associated with the review item in the bid document meets the content arrangement requirements of the bidding document, and if not, marking the format inconsistency information.

[0030] The checking method comprises: using a large language model to obtain semantic understanding features of the overall original content of the target review item and the text in the second page code range, respectively, and determining whether the two are associated.

[0031] In some embodiments, the retrieval method has a third retrieval method, which comprises the following steps:

[0032] Based on the target review item, a third retrieval keyword is extracted, the third retrieval keyword is used as third retrieval input content, and retrieval is performed at the selected retrieval position of the bid document to obtain a third page code range where the text associated with the target review item is located, and the retrieval comprises a text matching manner.

[0033] In some embodiments, the union of the retrieval output content obtained in the catalog and / or the text of the standardized bid document in at least one retrieval method comprises: determining the union of the retrieval output content obtained in at least one retrieval method as the effective review page code range of the target review item.

[0034] In a second aspect, a PDF document page code range matching system based on calibration is provided, comprising:

[0035] A PDF document acquisition unit is configured to acquire a PDF bid document to be analyzed;

[0036] A calibration unit is configured to convert the PDF bid document into an editable bid document with the style structure of the text content being preserved, and perform page code calibration and catalog calibration on the editable bid document to obtain a standardized bid document;

[0037] A retrieval unit is configured to acquire a target review item for the bid document, and perform retrieval in the standardized bid document based on at least one retrieval method to obtain a page code range where the text associated with the target review item is located; wherein the retrieval method has a retrieval input content generation method corresponding to the retrieval method itself for generating retrieval input content with the overall original content of the target review item as a data source, and the retrieval input content is used to execute the corresponding retrieval method to obtain retrieval output content in the retrieval method; the retrieval input content is used to represent the key information of the target review item; and the retrieval output content comprises a page code range where the text associated with the retrieval input content, i.e., the target review item, is located.

[0038] The page code range acquisition unit is used for combining the search output contents obtained in the catalog and / or the text of the standardized bidding document under at least one search mode to obtain the effective evaluation page code range of the target evaluation item.

[0039] The PDF document page code range matching method and system based on calibration provided by the present application have the following beneficial effects: the PDF document page code range matching method of the present application is compatible with various PDF formats through the standardization process of the bidding document, and effectively deals with the problems of missing or chaotic page codes through page code calibration and catalog calibration on the editable bidding document, so as to ensure that various types of bidding documents can be effectively parsed. Experts do not need to find information page by page, but can directly locate the relevant content through the accurate catalog range provided by the method, so that the efficiency of the whole evaluation process is greatly improved. The accurate catalog range provided by the present application provides a clear analysis boundary for expert evaluation or AI simulation of expert evaluation, so that the evaluation process can focus on the relevant content to generate evaluation conclusions, and the evaluation efficiency is improved. BRIEF DESCRIPTION OF DRAWINGS

[0040] Figure 1 is a flowchart of the PDF document page code range matching method based on calibration in the embodiments of the present application;

[0041] Figure 2 is a structural schematic diagram of the PDF document page code range matching system based on calibration in the embodiments of the present application. DETAILED DESCRIPTION

[0042] 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.

[0043] Referring to Figure 1 The present application provides a PDF document page code range matching method based on calibration, which comprises the following steps:

[0044] Step 1: obtaining a PDF bidding document to be analyzed;

[0045] Step 2: converting the PDF bidding document into an editable bidding document with the structure of the text content style reserved, and performing page code calibration and catalog calibration on the editable bidding document to obtain a standardized bidding document;

[0046] Step 3, obtaining a target review item for the bidding document, searching in the bidding document based on at least one search method, and obtaining a page range where the text associated with the target review item is located; wherein the search method has a search input content generation method corresponding to the search method itself for generating search input content with the target review item as a data source, and the search input content is used to represent the key information of the target review item; the search output content includes the page range where the text associated with the search input content, i.e., the target review item, is located.

[0047] Step 4, combining the search output content obtained in the directory and / or the text of the standardized bidding document under at least one search method to obtain the effective review page range of the target review item.

[0048] In the embodiments of the present application, the page range of the bidding document text for expert review of the target review item is determined based on different search methods, the page range of the review item in the PDF bidding document is effectively located, the effective review text content is provided for expert review or for AI simulation of expert review, the review process can focus on relevant content for review conclusion generation, and the review efficiency is improved. In the embodiments of the present application, for example, there are three search methods, which are denoted as A search method, B search method and C search method, respectively. The A search method can be used to search the text content associated with the review item in the full text of the bidding document to obtain an A page range result, the B search method can be used to search the text content associated with the review item in the full text of the bidding document to obtain a B page range result, and the C search method can be used to search the text content associated with the review item in the full text of the bidding document to obtain a C page range result. Finally, the union of the A page range result, the B page range result and the C page range result is used as the effective review page range of the target review item. Of course, the search input content used in the A search method, the B search method and the C search method should correspond to the corresponding search method.

[0049] In the embodiments of the present application, the PDF bidding document is converted into an editable bidding document, and existing PDF parsing conversion tools can be used. The parsed document can retain the formula, table, picture, multi-level title format structure and layout structure in the original PDF document. For pictures, the extracted text content and the original picture can be displayed simultaneously in the parsed document. The PDF parsing conversion tool can be converted into Markdown format or json format, retaining the original format result and layout structure, and the coordinates of each title can be identified and obtained. The present application uses OCR technology, Markdown conversion, json conversion and other technologies to solve the problem of processing pure scanned PDFs, and effectively deals with the problem of missing or chaotic page numbers through page number calibration and directory calibration, ensuring that various types of PDF bidding documents can be effectively parsed. The page number calibration can be achieved by combining virtual page number generation with Markdown title positioning technology. It should be noted that the text representation associated with the target evaluation item in the present application can be used to analyze the text content of the target evaluation item score of the bidding supplier.

[0050] In one embodiment, in step 3 above, the at least one retrieval method is used to retrieve the target evaluation item associated text in the bidding document to obtain the page number range where the target evaluation item associated text is located, including:

[0051] Step A: Based on at least one retrieval method, the directory of the bidding document is retrieved to obtain the page number range where the target evaluation item associated text is located, denoted as the directory retrieval page number range.

[0052] And / or,

[0053] Step B: Based on at least one retrieval method, the text of the bidding document is retrieved to obtain the page number range where the target evaluation item associated text is located, denoted as the text retrieval page number range. The retrieval method used in step A and the retrieval method used in step B are the same or different, and the number of types of retrieval methods used in step A and the number of types of retrieval methods used in step B are different or the same.

[0054] In the embodiments of the present application, the retrieval process is divided into directory retrieval and text retrieval. It can be understood that the retrieval in the directory is actually a correlation inference between the evaluation item information and each title in the directory. When a title is obtained that is associated with the evaluation item information (the correlation degree is greater than a preset threshold), the page number range between the page number where the title is located and the page number of the next title at the same level of the title can be used to obtain the page number range where the text associated with the evaluation item is located, i.e. the directory retrieval page number range result.

[0055] In the text retrieval, the evaluation item information is analyzed in correlation with the local area content in the text of the bidding document.

[0056] In the embodiments of the present application, on the basis of searching the bidding document based on different search modes, the search in the table of contents and / or the search in the main text are also considered, so that the search can be performed in combination based on different kinds of search modes and search positions (table of contents, main text), the search result is determined, and the union of different search execution results is taken as the valid evaluation page range of the target evaluation item. The bidding document is searched by using a search mode to obtain search output content, i.e., the page range where the main text associated with the target evaluation item is located. In this process, only the table of contents is searched, or only the main text is searched, or the table of contents and the main text of the bidding document, i.e., the full text of the bidding document, are searched. Then, the bidding document can be searched again by using other different search modes to obtain other search output content. Similarly, in the search process of the other different search modes, only the table of contents can be searched, or only the main text can be searched, or the table of contents and the main text of the bidding document, i.e., the full text of the bidding document, can be searched. It can be illustratively explained that, for example, there are two search modes, which are denoted as A search mode and B search mode. The table of contents can be searched based on the A search mode to obtain an A page range result. The table of contents can be searched based on the B search mode to obtain a B page range result. The A page range result and the B page range result are combined to obtain the valid evaluation page range. Optionally, the union of the A page range result and the B page range result is taken as the valid evaluation page range. The table of contents can be searched based on the A search mode to obtain an A page range result. The main text can be searched based on the B search mode to obtain a B page range result. The A page range result and the B page range result are combined to obtain the valid evaluation page range. Optionally, the union of the A page range result and the B page range result is taken as the valid evaluation page range. The main text can be searched based on the A search mode to obtain an A page range result. The main text can be searched based on the B search mode to obtain a B page range result. The A page range result and the B page range result are combined to obtain the valid evaluation page range. Optionally, the union of the A page range result and the B page range result is taken as the valid evaluation page range. Further, the A search mode can be used to search the table of contents to obtain an A1 page range result, the A search mode can be used to search the main text to obtain an A2 page range result, and the B search mode can be used to search the main text to obtain a B page range result. The A1 page range result, the A2 page range result, and the B page range result are combined to obtain the valid evaluation page range. Optionally, the union of the A1 page range result, the A2 page range result, and the B page range result is taken as the valid evaluation page range.Of course, it needs to be explained that different retrieval methods have different retrieval accuracy, and the text retrieval is more time-consuming but more accurate and comprehensive than the catalog retrieval, and the catalog retrieval is more time-saving but may be not comprehensive and accurate. When the retrieval is performed based on the combination of different types of retrieval methods and retrieval positions (catalog and text), the balance between the time length and the accuracy and comprehensiveness of the retrieval needs to be considered to determine the appropriate combination method.

[0057] The method for retrieving the selected retrieval position of the bidding document provided in the embodiments of the present application can have two specific methods, and both the first method and the second method can be used in specific implementation.

[0058] The first method includes the following steps.

[0059] In the first step, a title associated with the target evaluation item in the catalog of the bidding document is determined based on a retrieval method, and then the page number of the text associated with the target evaluation item in the bidding document is determined based on the page number of the title, and the determined page number range is recorded as a preliminary page number range.

[0060] In the second step, the page number range of the text associated with the target evaluation item in the text is determined based on a retrieval method, and the determined page number range is recorded as a secondary page number range. The retrieval method used in the first step is different from or the same as the retrieval method used in the second step.

[0061] The effective page number range is determined based on the coincidence degree of the preliminary page number range and the secondary page number range obtained in the first step.

[0062] The second method includes the following steps.

[0063] In the first step, a title associated with the target evaluation item in the catalog of the bidding document is determined based on a retrieval method, and then the page number of the text associated with the target evaluation item in the bidding document is determined based on the page number of the title, and the determined page number range is recorded as a preliminary page number range.

[0064] In the second step, whether there is text associated with the target evaluation item in the text in the preliminary page number range is determined based on a retrieval method, and if it is determined that there is text associated with the target evaluation item, the page number range of the text associated with the target evaluation item in the text is determined, recorded as a secondary page number range, and it is determined that the preliminary page number range is valid, and the preliminary page number range and the secondary page number range are provided as the matching result. The preliminary page number range represents a rough page number range, and the secondary page number range represents a further narrowed and more accurate page number range. If it is determined that there is no text associated with the target evaluation item, it is determined that the preliminary page number range is invalid. The retrieval method used in the first step is different from or the same as the retrieval method used in the second step.

[0065] In an embodiment, after step A, i.e., based on at least one search method, searching in the table of contents of the bidding document to obtain a table of contents search page code range, the method further comprises the following steps:

[0066] Step C, based on at least one search method, searching in the text of the table of contents search page code range to determine whether there is text associated with the target review item in the text of the table of contents search page code range; if yes, go to step C1; otherwise, go to step C2; wherein the search method used in step A is different from or the same as the search method used in step C;

[0067] Step C1, based on the page code range of the text S actually associated with the target review item in the text of the table of contents search page code range as the accurate page code range, taking the table of contents search page code range and the accurate page code range as the search output content at the same time;

[0068] Step C2, determining that the table of contents search page code range result is invalid.

[0069] The embodiments of the present application provide a method for verifying whether the table of contents search result is valid. It should be noted that one search method has corresponding search input content and corresponding search output result. If the search methods used in steps A and C are the same, the search input contents used in the two steps are consistent. Otherwise, if the search methods used in steps A and C are different, the search input contents used in the two steps are inconsistent.

[0070] In an embodiment, the search method in step 3 above has a first search method, and the first search method comprises the following steps:

[0071] Step 301, based on the overall original content of the target review item, generating a first search input content using a large language model, wherein the content generated by the large language model includes keywords, constraints, and prompts of the target review item;

[0072] Step 302, based on the first search input content, performing semantic association analysis using a large language model at a selected search location of the bidding document to determine a first page code range where the text associated with the first search input content, i.e., the target review item, is located, wherein the selected search location is the table of contents part and / or the text part of the bidding document.

[0073] Specifically, the large language model in the embodiments of the present application can be implemented by conventional technology, for example, it can be implemented based on Ali Qwen2.5 model. In the semantic analysis process in the first search method in the embodiments of the present application, the following steps are included:

[0074] (1) Obtain the review item data in the collected historical bidding document dataset to form a historical review item dataset; extract the entity of the review item based on the historical review item dataset, and classify the entity into a primary key entity and a secondary key entity, wherein the primary key entity represents the keywords of the review item, and the secondary key entity represents the constraint conditions of the review item;

[0075] (2) Search for first supplementary information based on the primary key entity, wherein the first supplementary information is used to represent general constraint information of the primary key entity; search for second supplementary information based on the secondary key entity, wherein the second supplementary information is used to represent implicit constraint information contained in the secondary key entity; the second supplementary information and the secondary key entity information are used as auxiliary supplementary information to assist the search for the first supplementary information in the search process, and the search process is based on a preset knowledge graph or a database of the review item dataset in the historical bidding document dataset; in this application, the search for the first supplementary information is based on the second supplementary information and the secondary key entity as auxiliary information, which can make the model more clearly understand the specific application scenario of the review item, and it can be understood that the rules and constraints of the same review item in different application scenarios or business fields are different; on the other hand, the first supplementary information can more clearly understand that the information to be provided is inconsistent with the secondary key entity information, and the secondary key entity can be used as reference information;

[0076] (3) Perform semantic coding and analysis based on the primary key entity, the first supplementary information, the secondary key entity, and the second supplementary information to form the search input content.

[0077] Specifically, considering the data processing amount in the search process of the first search method, especially for hundreds of pages of bidding documents, the first search method has obvious time-consuming differences for the directory part and the text part, which can be distinguished for different application scenarios. In the embodiments of the present application, the first search method is specifically described as follows:

[0078] Step 3011, based on the overall original content of the target review item, a first search input content is generated by using a large language model, wherein the content generated by the large language model includes keywords, constraint conditions, and prompt words of the target review item;

[0079] Step 3012, based on the first search input content, at least one of the following first method and second method is executed to determine the first page code range where the text associated with the target review item is located; wherein the first page code range is determined based on the union of the execution results of the first method and the second method;

[0080] Step 3013, the first method: based on the first search input content, using a large language model to perform semantic association analysis on the directory part of the bid document, inferring the directory title associated with the first search input content, and determining the page code range of the text associated with the target review item based on the page code of the associated directory title. It should be noted that in this method, the page code range of the text content under the title is determined based on the page code of the associated directory title, i.e., the first page code range. Specifically, it can be understood that based on the page code of the associated directory title, combined with the page code of the next sibling directory title of the associated directory title, the page code range of the text content under the associated directory title, i.e., the first page code range, can be determined.

[0081] Step 3014, the second method: based on the first search input content, using a large language model to perform semantic matching analysis on the text part of the bid document, determining the text content that is semantically matched with the first search input content, and determining the page code range of the text associated with the target review item based on the page code of the semantically matched text content.

[0082] In the embodiments of the present application, the semantic adaptability of the target review item overall content and the text in the bid document is analyzed using a large language model, and the keywords and constraint conditions used for searching are parsed based on the review item overall content, for example, if the review item overall content is "bidders' performance Bidders have completed ground centralized nitrogen system design or mine design performance since January 1, 2022, and each provides 3 points, with a maximum of 15 points", the keywords "ground centralized nitrogen system design" and "mine design performance" can be parsed, and the constraint condition "since January 1, 2022" can be parsed. Based on the analysis result of the target review item overall content by the large language model, combined with the prompt information such as Prompt, the "question problem" data is automatically generated, and the "answer" is searched in the entire bid document using the large language model, so as to obtain the content position, i.e., the page code range, of the entire bid document associated with the target review item. The embodiments of the present application realize deeper semantic adaptation.

[0083] In the first search method of the embodiments of the present application, the concept of semantic adaptation is used, the execution time of the first method is less than that of the second method, and in the first method, through the overall semantic understanding of the review item, the affiliation and association degree of each title with the review item are inferred in all titles of the document directory, and the title with the maximum probability of containing the associated text of the review item is determined. This first method not only improves the effectiveness and comprehensiveness of the page code range determination from the perspective of semantic adaptation, but also reduces the data processing amount of the page code range determination compared with the second method, i.e., reduces the time length of the page code range determination.

[0084] Therefore, in actual application, whether to adopt the first method and whether to adopt the second method are determined according to the time limit of the application scene requirement. In the case of sufficient time, the second method obtains the page range where the text associated with the review item is more comprehensive and accurate. Specifically, the full-text page number of the bid document, the historical page number of the bid document analyzed by the second method and the time consumption data can be considered to evaluate the time consumption of the second method for the full text of the current bid document.

[0085] Specifically, the first method further comprises: recording the page range where the text associated with the target review item determined based on the directory title as the first candidate page range;

[0086] Based on the first retrieval input content, semantic matching analysis is performed on the text content in the first candidate page range of the bid document, the text content that is semantically matched with the first retrieval input content is determined, and the page range where the text actually associated with the target review item in the first candidate page range is determined based on the page where the semantically matched text content is located. This step is equivalent to verifying the directory retrieval page range result. If there is no text content in the directory retrieval page range result that is semantically matched with the first retrieval input content, the directory retrieval page range result is invalid. If there is, it can be determined that the directory retrieval page range result is valid, and the page range where the text associated with the target review item can be further limited.

[0087] In an embodiment, the retrieval method in step 3 above has a second retrieval method, which comprises the following steps:

[0088] Step 311, acquiring and parsing the bidding document, and acquiring the content arrangement requirements for the bid document;

[0089] Step 312, generating an association database of "review item-corresponding board-block" based on the content arrangement requirements;

[0090] Step 313, for the target review item, determining whether there is a corresponding "review item-corresponding board-block-content element" data item based on the association database. If there is, the text content of the data item is used as the second retrieval input content to perform retrieval at the selected retrieval position of the bid document, and the second page range where the text associated with the target review item is obtained. The selected retrieval position is the directory part and / or the text part of the bid document.

[0091] The application embodiment fully utilizes the information in the bidding document, parses the content arrangement requirements and content format regulations of the bidding document on the bidding document, and establishes a "review item-corresponding block-content element" association database for the corresponding determination of the related filling content of the review item which should be placed in which block position in the bidding document. It can be understood that the "block" name generally appears as a title. Based on this, the "corresponding block" name can be directly found in the title, so that based on the page number corresponding to the title, the page number range of the text under the title, i.e. the page number range of the text associated with the review item, can be determined.

[0092] In the application embodiment, the conventional retrieval mode is optimized by using the pre-constructed "review item-corresponding block-content element" association database. Based on the generation of the retrieval input content based on the review item information, the generation of the retrieval input content based on the "corresponding block" and "content element" information is expanded. The corresponding relationship between the "review item" and the "corresponding block" and "content element" is used to improve the retrieval efficiency and expand the retrieval input content.

[0093] It should be noted that the text content based on the data item is used as the second retrieval input content, and the selected retrieval position of the bidding document is retrieved respectively, including:

[0094] Directly based on the name of the "corresponding block" as the second retrieval input content, the selected retrieval position of the bidding document is retrieved, and the retrieval includes a text matching manner; the name of the "corresponding block" is directly matched in the text of the bidding document, which saves the retrieval matching time.

[0095] Specifically, in one embodiment, in the second retrieval mode, the step of constructing the association database includes:

[0096] Step 3101, based on the target review item, extracting the keywords of the review item, denoted as the first keyword;

[0097] Step 3102, parsing the bidding document, obtaining the position of the keyword of the review item in the content layout requirement part of the bidding document, and obtaining the context of the keyword position. It should be noted that the context includes the keyword and the context. The content layout requirement part includes all parts that may make requirements for the content layout of the bidding document, such as "answer file format", "answer person must know" and the like.

[0098] Step 3103, identifying the context, determining the keyword in the context that matches the structure title in the format file of the bidding document, denoted as the second keyword, wherein the structure title represents the title framework structure and the structural explanatory text content such as content description in the format file of the bidding document provided in the bidding document;

[0099] Step 3104, based on the distribution position of the first keyword in the context, the distribution position of the second keyword, and the rest of the text other than the first keyword and the second keyword, the semantics in the context are analyzed, and the position relationship that the first keyword corresponding content and the second keyword corresponding content should meet in the entire bidding document is generated; in this step, the context semantics are analyzed to generate the position relationship, which can be obtained by using a preset neural network or deep learning with a sample set through training;

[0100] Step 3105, based on the expression associated with the target review item in all generated position relationship expressions, the association database of "first keyword-second keyword-first keyword related content arrangement requirement keyword" is obtained, that is, the association database of "review item-corresponding block-content element", and the first keyword related content arrangement requirement keyword includes: the key information representing the content arrangement requirement of the review item identified based on the context analysis and irrelevant to the second keyword and its position relationship with the second keyword, for example, "performance proof materials should include contract front page, amount page, etc.", and the key information representing the content arrangement requirement of the performance proof materials in the content arrangement requirement includes the contract front page and the amount page.

[0101] It can be understood that in the above step 3103, there can be no second keyword, and it can be understood that in some intercepted contexts, the position of the review item is not explained, which can be explained as some format requirements of the review item. For the context without identifying the second keyword, the subsequent context semantic analysis step can still be performed and needs to be performed, and steps 3104 and 3105 are continued to be executed, and in step 3104, the semantics of the context are analyzed, not only the position relationship, but also the key information representing the content arrangement requirement and format requirement that the first keyword should meet, and in step 3105, all analysis results are integrated to obtain all content arrangement requirement information about the review item, including position relationship, position arrangement layout information, format requirement, etc. Of course, for the ambiguous format requirement in the bidding document, industry practices can be combined to supplement and clarify, and effective guidance for page range framing is ensured.

[0102] In an embodiment, after the above step 314, that is, after obtaining the second page range where the text associated with the target review item is located, it further includes:

[0103] Step 315, checking whether the text in the second page code range is associated with the target review item, determining whether the text content associated with the review item in the bidding document meets the content arrangement requirements of the bidding document, and if not, marking the format inconsistency information; the checking method comprises: obtaining the semantic understanding features of the overall original content of the target review item and the text in the second page code range respectively by using a large language model, and determining whether the two are associated, that is, based on the semantic features, analyzing whether the two are semantically adapted and associated, so as to determine whether the text in the second page code range can be used for scoring the target review item.

[0104] The embodiment of the present application provides a way to verify whether the bidding document meets the content layout requirements and format requirements of the bidding document. By using the retrieval method of the association database, if the associated content of the "corresponding board" and the "content element" is found in the bidding document under the condition that the content of the bidding document is arranged according to the content arrangement requirements of the bidding document, it can be determined that the second page code range found contains content associated with the target review item. If the text in the second page code range is associated with the target review item, it is determined that the text content associated with the review item in the bidding document meets the content arrangement requirements of the bidding document, otherwise, if the text in the second page code range is not associated with the target review item, mark the format inconsistency information. In the embodiment of the present application, the arrangement requirements of the bidding document for the bidding document are fully utilized to improve the page code range matching efficiency.

[0105] In one embodiment, the retrieval method in step 3 above has a third retrieval method, which comprises the following steps:

[0106] Step 321, based on the target review item, extracting a third retrieval keyword, taking the third retrieval keyword as a third retrieval input content, and performing retrieval at a selected retrieval position of the bidding document to obtain a third page code range where the text associated with the target review item is located, and the retrieval includes a text matching manner. It can be understood that in the above-mentioned first retrieval method, second retrieval method and third retrieval method provided in the present application, the first retrieval method performs retrieval by semantic understanding, and the overall accuracy of the retrieval result is greater than that of the second retrieval method and the third retrieval method, and the second retrieval method fully utilizes the format requirements of the bidding document for the bidding document. In one embodiment, a PDF document page code range matching method based on calibration is provided, comprising the following steps:

[0107] (101) obtaining a PDF bidding document to be analyzed;

[0108] (102) converting the PDF bidding document into an editable bidding document with the structure of the text content style reserved, and performing page code calibration and directory calibration on the editable bidding document to obtain a standardized bidding document;

[0109] (103) The first method of the first search method is used to search in the table of contents of the standardized bidding document to obtain a text page range D1 associated with the review item;

[0110] The second search method is used to search in the table of contents and the text of the standardized bidding document respectively, to obtain a text page range D2 associated with the review item based on the table of contents search and a text page range D3 associated with the review item based on the text search;

[0111] The third search method is used to search in the table of contents and the text of the standardized bidding document respectively to obtain a text page range D4 associated with the review item;

[0112] (104) The effective review page range of the target review item is obtained based on the combination of the page ranges D1, D2, D3 and D4. Further, the union of the page ranges D1, D2, D3 and D4 can be used as the effective review page range of the target review item.

[0113] In an embodiment, another calibration-based PDF document page range matching method is provided, including the following steps:

[0114] (111) A PDF bidding document to be analyzed is obtained;

[0115] (112) The PDF bidding document is converted into an editable bidding document with the content style structure reserved, and the editable bidding document is subjected to page calibration and table of contents calibration to obtain a standardized bidding document;

[0116] (113) The first method of the first search method is used to search in the table of contents of the standardized bidding document to obtain a text page range D1 associated with the review item; the second method of the first search method is used to search in the text of the standardized bidding document to obtain a text page range D2 associated with the review item;

[0117] The second search method is used to search in the table of contents and the text of the standardized bidding document respectively, to obtain a text page range D3 associated with the review item based on the table of contents search and a text page range D4 associated with the review item based on the text search;

[0118] The third search method is used to search in the table of contents and the text of the standardized bidding document respectively to obtain a text page range D5 associated with the review item;

[0119] (114) Based on the union of the page number ranges D1, D2, D3, D4, and D5, the effective evaluation page number range of the target evaluation item is obtained. Further, the union of the page number ranges D1, D2, D3, D4, and D5 can be used as the effective evaluation page number range of the target evaluation item. In this embodiment, multiple retrieval methods are used, and the evaluation item semantic analysis, the bidding document format requirements, and the document page number directory calibration mechanism are combined. Compared with the simple keyword matching of the prior art, the positioning error rate is significantly reduced. For example, for the "similar performance" evaluation item, the content that does not meet the time and type requirements can be accurately excluded, and only the relevant directory range is returned.

[0120] In one embodiment, in the above step 4, the union of the retrieval output content obtained in the directory and / or the text of the standardized bidding document by at least one retrieval method is used to obtain the effective evaluation page number range of the target evaluation item, which includes: the union of the retrieval output content obtained by at least one retrieval method is determined as the effective evaluation page number range of the target evaluation item. Of course, in the embodiments of the present application, the union of multiple retrieval contents can be based on the retrieval output content based on the mutual verification mechanism under multiple retrieval methods, and the page number range determined by multiple retrieval methods at the same time is determined as the final effective evaluation page number range. It can be understood that this method is similar to the voting mechanism, and the page number range recognized by multiple retrieval methods is considered to have greater credibility. However, in the embodiments of the present application, in order not to miss any page number that may exist in the evaluation item associated content, the union of the multiple retrieval output contents is taken, and based on this method, in the subsequent steps, the text of all page number ranges based on the union can be verified and determined whether it is a real and effective page number range by semantic understanding of the text content of each page.

[0121] Referring to Figure 2 The embodiments of the present application provide a PDF document page number range matching system based on calibration, which comprises:

[0122] A PDF document acquisition unit is configured to acquire a PDF bidding document to be analyzed;

[0123] A calibration unit is configured to convert the PDF bidding document into an editable bidding document with the style structure of the text content reserved, and to perform page number calibration and directory calibration on the editable bidding document to obtain a standardized bidding document;

[0124] The searching unit is configured to obtain a target review item for a bidding document, search in the standardized bidding document based on at least one searching mode, and obtain a page range where a text associated with the target review item is located; wherein the searching mode has a searching input content generation method corresponding to the searching mode, and is configured to generate the searching input content by taking the whole original content of the target review item as a data source, execute the corresponding searching mode by using the searching input content to obtain searching output content under the searching mode; the searching input content is configured to represent key information of the target review item; and the searching output content includes the page range where the text associated with the target review item is located.

[0125] The page range obtaining unit is configured to combine the searching output content obtained in the directory and / or the text of the standardized bidding document under at least one searching mode to obtain an effective review page range of the target review item.

[0126] The specific definition of the PDF document page range matching system can refer to the definition of the PDF document page range matching method in the foregoing, and will not be described here. Each unit in the PDF document page range matching system described above can be realized by software, hardware, and a combination thereof, in whole or in part. Each unit described above can be embedded in or independent of a processor in a computer device in a hardware form, or can be stored in a memory in a computer device in a software form, so as to be called and executed by a processor to perform the operation corresponding to each unit.

[0127] In some embodiments, the PDF document page range matching system provided by the embodiments of the present application can be realized in a combination of software and hardware. For example, the PDF document page range matching system provided by the embodiments of the present application can be directly embodied as a combination of software modules executed by a processor. The software modules can be located in a storage medium, the storage medium is located in a memory, and the processor reads executable instructions included in the software modules in the memory, and combines necessary hardware (for example, including the processor and other components connected to the bus) to complete the PDF document page range matching method provided by the embodiments of the present application.

[0128] The present application is not limited to the above-mentioned specific embodiments, and various modifications made by those skilled in the art without creative labor, based on the above-mentioned concept, all fall within the scope of protection of the present application.

Claims

1. A calibration-based PDF document page range matching method, characterized in that, The method comprises the following steps: obtaining a PDF bidding document to be analyzed; converting the PDF bidding document into an editable bidding document with the style structure of the text content reserved, and performing page number calibration and directory calibration on the editable bidding document to obtain a standardized bidding document; obtaining a target evaluation item for the bidding document, performing retrieval in the standardized bidding document based on at least one retrieval manner, and obtaining a page number range where the text associated with the target evaluation item is located; wherein the retrieval manner has a retrieval input content generation method corresponding to the retrieval manner itself for generating retrieval input content with the overall original content of the target evaluation item as a data source, and the retrieval input content is used to execute the corresponding retrieval manner to obtain retrieval output content under the retrieval manner; the retrieval input content is used to represent the key information of the target evaluation item; the retrieval output content includes the page number range where the text associated with the retrieval input content, i.e., the target evaluation item, is located; combining the retrieval output content obtained in the directory and / or the text of the standardized bidding document under at least one retrieval manner to obtain an effective evaluation page number range of the target evaluation item; the step of performing retrieval in the standardized bidding document based on at least one retrieval manner to obtain the page number range where the text associated with the target evaluation item is located comprises: Step A: performing retrieval in the directory of the standardized bidding document based on at least one retrieval manner to obtain the page number range where the text associated with the target evaluation item is located, which is denoted as the directory retrieval page number range; and / or, Step B: performing retrieval in the text of the standardized bidding document based on at least one retrieval manner to obtain the page number range where the text associated with the target evaluation item is located, which is denoted as the text retrieval page number range; wherein the retrieval manner used in Step A is the same as or different from the retrieval manner used in Step B, and the number of types of retrieval manners used in Step A is different from or the same as the number of types of retrieval manners used in Step B; after the step of performing retrieval in the directory of the standardized bidding document based on at least one retrieval manner to obtain the directory retrieval page number range, the method further comprises: Step C: performing retrieval in the text of the directory retrieval page number range based on at least one retrieval manner to determine whether there is text associated with the target evaluation item in the text of the directory retrieval page number range; if yes, go to Step C1; otherwise, go to Step C2; wherein the retrieval manner used in Step A is different from or the same as the retrieval manner used in Step C; Step C1: taking the page number range where the text S actually associated with the target evaluation item in the text of the directory retrieval page number range as the accurate page number range, and taking both the directory retrieval page number range and the accurate page number range as the retrieval output content; Step C2: determining that the directory retrieval page number range result is invalid.

2. The calibration-based PDF document page range matching method of claim 1, wherein, The first retrieval manner in the retrieval manner comprises the following steps: generating a first retrieval input content based on the overall original content of the target evaluation item using a large language model, wherein the content generated by the large language model includes keywords, constraints, and prompts of the target evaluation item; Based on the first search input content, the semantic association analysis is performed on the selected search location of the standardized bid document by using a large language model, and the first page code range where the text associated with the first search input content, i.e., the target review item, is located is determined. The selected search location is the directory part and / or the text part of the bid document.

3. The calibration-based PDF document page range matching method of claim 2, wherein, The first search method includes the following steps: Based on the overall original content of the target review item, a first search input content is generated by using a large language model. The content generated by the large language model includes keywords, constraints, and prompts of the target review item. Based on the first search input content, at least one of the first method and the second method is executed to determine the first page code range where the text associated with the target review item is located. The first page code range is determined based on the union of the execution results of the first method and the second method. The first method: based on the first search input content, the semantic association analysis is performed on the directory part of the standardized bid document by using a large language model, and the directory title associated with the first search input content is inferred. The page code of the associated directory title is used to determine the page code range where the text associated with the target review item is located. The second method: based on the first search input content, the semantic matching analysis is performed on the text part of the standardized bid document by using a large language model, and the text content semantically matched with the first search input content is determined. The page code where the semantically matched text content is located is used to determine the page code range where the text associated with the target review item is located.

4. The method of claim 1, wherein, The search method has a second search method, which includes the following steps: Obtain and parse the bidding document to obtain the content arrangement requirements for the bid document; Based on the content arrangement requirements, generate an association database of "review item-corresponding board-content element"; For the target review item, based on the association database, determine whether there is a corresponding "review item-corresponding board-content element" data item. If there is, use the text content of the data item as the second search input content to search in the selected search location of the bid document, and obtain the second page code range where the text associated with the target review item is located. The selected search location is the directory part and / or the text part of the bid document.

5. The method of claim 4, wherein, After obtaining the second page code range where the text associated with the target review item is located, the following steps are further included: Verify whether the text in the second page code range is associated with the target review item to determine whether the text content associated with the review item in the bid document meets the content arrangement requirements of the bidding document. If not, mark the format inconsistency information; The verification method includes: using a large language model to obtain the semantic understanding features of the overall original content of the target review item and the text in the second page code range, respectively, and determining whether they are associated.

6. The method of claim 1, wherein, The search method has a third search method, which includes the following steps: Based on the target review item, a third search keyword is extracted, which is used as a third search input content to search in the selected search location of the bid document, and the third page code range where the text associated with the target review item is located is obtained. The search includes a text matching manner.

7. The method of claim 1, wherein, The union of the search output contents obtained in the catalog and / or the text of the standardized bidding document under at least one search mode obtains the effective evaluation page range of the target evaluation item.

8. A calibration-based PDF document page range matching system, characterized by, It comprises: A PDF document acquisition unit is configured to acquire a PDF bidding document to be analyzed; A calibration unit is configured to convert the PDF bidding document into an editable bidding document with the style structure of the text content being reserved, and to perform page calibration and catalog calibration on the editable bidding document to obtain a standardized bidding document; A search unit is configured to acquire a target evaluation item for the bidding document, perform search in the standardized bidding document based on at least one search mode, and acquire a page range where the text associated with the target evaluation item is located; wherein the search mode has a search input content generation method corresponding to the search mode itself for generating search input content with the original content of the target evaluation item as a data source, and the search input content is used to execute the corresponding search mode to obtain the search output content under the search mode; the search input content is used to represent the key information of the target evaluation item; and the search output content comprises the page range where the text associated with the search input content, i.e., the target evaluation item, is located; A page range acquisition unit is configured to combine the search output contents obtained in the catalog and / or the text of the standardized bidding document under at least one search mode to obtain the effective evaluation page range of the target evaluation item; In the search unit, the search in the standardized bidding document based on at least one search mode to acquire the page range where the text associated with the target evaluation item is located comprises: Step A: performing search in the catalog of the standardized bidding document based on at least one search mode to obtain the page range where the text associated with the target evaluation item is located, which is denoted as the catalog search page range; And / or, Step B: performing search in the text of the standardized bidding document based on at least one search mode to obtain the page range where the text associated with the target evaluation item is located, which is denoted as the text search page range; wherein the search modes used in step A and step B are the same or different, and the number of types of the search modes used in step A is different from or the same as the number of types of the search modes used in step B; After the search in the catalog of the standardized bidding document based on at least one search mode to obtain the catalog search page range, the method further comprises: Step C: performing search in the text of the catalog search page range based on at least one search mode to determine whether there is text associated with the target evaluation item in the text of the catalog search page range; if yes, go to step C1; otherwise, go to step C2; wherein the search modes used in step A and step C are different or the same; Step C1: taking the page range where the text S actually associated with the target evaluation item in the text of the catalog search page range as an accurate page range, and taking both the catalog search page range and the accurate page range as the search output content; Step C2, determine that the directory search page code range result is invalid.

Images