A standardized chapter splitting method

Abstract

The present invention discloses a standardized chapter splitting method, which mainly includes the following steps: S1, obtaining standard specification documents; S2, parsing document content, identifying and extracting chapter numbers and title information; S3, constructing a multi-level tree structure through a directed acyclic graph algorithm; S4, matching and correcting chapters by applying template rules; S5, displaying document content through a multi-level tree structure to facilitate user navigation; S6, using a keyword retrieval algorithm combined with hierarchical weights to achieve efficient retrieval; S7, using a graph neural network to associate and bind the clause description with the corresponding chapter to ensure synchronous display of the relevant clause description. The method of the present invention can structuredly process standard specification documents, improve information retrieval efficiency, and enhance user experience.

Description

Technical Field

[0001] The present invention relates to the technical field of document processing and information retrieval, and in particular to a standardized chapter splitting method. Background Art

[0002] In the modern engineering design and construction industry, standard specification documents are indispensable reference materials for design, construction and other work. Standard specification documents are usually issued by authoritative departments and contain detailed technical requirements, operating specifications and quality standards. They are an important basis for ensuring engineering quality and safety. However, the existing technology has some shortcomings in the management and use of standard specification documents, which affects the rapid acquisition and efficient use of information.

[0003] The first is that standard documents are usually published in PDF format. Although PDF format is convenient for reading and printing, it does not support keyword search, making it difficult for users to quickly locate the specific information they need. This limitation requires users to manually flip through a large number of pages when looking for specific terms or regulations, which is time-consuming and easy to miss key content.

[0004] The second is that the existing PDF standard specification document directory is only accurate to the first-level title. Most standard specification documents contain complex hierarchical structures with many chapters and sub-chapters, and only the first-level directory cannot effectively help users quickly locate specific content. The limitations of this directory structure seriously affect the user experience and information acquisition efficiency.

[0005] The third is that the clauses in standard specification documents are usually located in the appendix, which is separated from the main text. When users read the main text, they often need to jump to the appendix frequently to view the clauses, which not only increases the complexity of operation, but also easily leads to incoherent understanding of information by users, reducing the efficiency and accuracy of document use.

[0006] Therefore, how to provide a standardized chapter splitting method is an urgent problem that technicians in this field need to solve. Summary of the invention

[0007] One object of the present invention is to propose a standardized chapter splitting method. The present invention uses a hierarchical relationship parsing algorithm based on a directed acyclic graph and a graph neural network technology to perform structured processing on standard specification documents. Specifically, the method parses the chapter number and title information in the document, constructs a multi-level tree structure, and matches and corrects the document content in combination with predefined template rules. In addition, the present invention also introduces a method based on The hierarchical weighted keyword search algorithm and the association binding technology between the clause description and the text content are combined to achieve efficient content search and information synchronous display. Compared with the traditional PDF document management method, the present invention has the following advantages: significantly improves the efficiency of information search, and users can quickly locate the relevant chapter content; through the synchronous display of the clause description, the jump operation of the user when consulting is reduced, and the convenience of document use and the accuracy of information acquisition are improved.

[0008] A standardized chapter splitting method according to an embodiment of the present invention includes the following steps:

[0009] S1. Obtain standard specification documents in PDF format;

[0010] S2, parse the document content, identify and extract the chapter number and title information in the document;

[0011] S3, using a hierarchical relationship parsing algorithm based on a directed acyclic graph to perform hierarchical parsing on the extracted chapter number and title information to construct a multi-level tree structure of the document;

[0012] S4. Based on the multi-level tree structure, the predefined template rules are applied to match and correct the chapter numbers and title information at different levels;

[0013] S5. Display the document content after hierarchical analysis and correction in the form of a multi-level tree structure, so that users can browse the document intuitively and navigate to specific chapter content;

[0014] S6, combined with the multi-level tree structure, based on Combined with the hierarchical weighted search algorithm, the keyword search function of the document content is realized, so that users can locate the relevant chapter content by entering keywords;

[0015] S7. Use an association binding algorithm based on graph neural network to associate and bind the clause descriptions with the corresponding chapter content, and display the relevant clause descriptions synchronously when the user browses the main text content.

[0016] Optionally, the S3 specifically includes:

[0017] S31, parsing and extracting chapter number and title information from the standard specification document, and organizing the information into a node set N and an edge set E, wherein the node set N represents a unique identifier of each chapter, a title and content corresponding to the chapter number, and the edge set E represents a hierarchical relationship between nodes;

[0018] S32. Construct a directed acyclic graph based on the information of node set N and edge set E. , in the directed acyclic graph G, each node Represents the title and content corresponding to a chapter number. The hierarchical relationship between nodes is represented by each edge in the edge set E;

[0019] S33. In the directed acyclic graph G, define the hierarchical relationship function , used to represent nodes and nodes The priority relationship between them;

[0020] S34, using the numbering format of the chapter number, the style characteristics of the title and the contextual semantic information, applying an adaptive hierarchy adjustment algorithm to sort and layer the nodes in the directed acyclic graph G;

[0021] S35, generating a hierarchical sequence of nodes based on the sorted and layered directed acyclic graph G;

[0022] S36. Construct the generated hierarchical sequence into a multi-level tree structure T. The multi-level tree structure includes all chapters and titles layer by layer starting from the root node. The root node represents the highest level node of the document, and each child node represents a secondary chapter or title in turn.

[0023] Optionally, the S34 specifically includes:

[0024] S341. Obtain the numbering format information of the chapter number from the standard specification document and record the level depth of each chapter and numbering sequence ;

[0025] S342. Extracting style features of the title; defining title saliency function :

[0026] ;

[0027] in, represents the k-th style feature; is the weight coefficient of the corresponding feature, Adjust parameters for noise;

[0028] S343, using a deep semantic analysis model to map chapter content into semantic vectors , the semantic vector captures the theme and key content of the chapter, and defines the semantic relationship measurement function :

[0029] ;

[0030] in, and For Node and The semantic vector of

[0031] S344, the level depth of the chapter , Numbering rules , Title Prominence and semantic relationship measure As input, the adaptive hierarchical adjustment algorithm dynamically calculates the hierarchical relationship index of each node by fusing multi-dimensional features. ;

[0032] ;

[0033] in, For the importance of the chapter, , , , is the balance coefficient, which is used to adjust the contribution of hierarchy depth, title prominence, semantic relationship and chapter importance to the hierarchy relationship index;

[0034] S346, according to the hierarchical relationship index As a result, the order of nodes in the directed acyclic graph G is adjusted, the order of the node set N is reordered, and the hierarchical structure of the document is optimized;

[0035] S347, perform loop-free verification, use an adaptive adjustment mechanism, and adjust the node when a loop is detected. of value and recalculate the node order until the acyclic structure is stable.

[0036] Optionally, the S4 specifically includes:

[0037] S41, establishing a predefined template rule;

[0038] S42, matching the parsed chapter number and title, comparing the extracted information with the predefined template rule, and identifying the numbering pattern of the chapter number;

[0039] S43, correcting the actual style features of the title through the style feature rules, and converting the actual style features of the title into standardized style features;

[0040] S44, determining the depth attribute of each chapter according to the format of the chapter number and the number of separators;

[0041] S45. Adopt an adaptive correction algorithm to optimize the initially identified chapter hierarchy. The adaptive correction algorithm adjusts inconsistent and erroneous hierarchical structures by comparing the relationship between the chapter number format, title style features and chapter content.

[0042] Optionally, the S5 specifically includes:

[0043] S51. Based on the generated multi-level tree structure, the document content is hierarchically organized and a dynamic display framework is created so that each chapter is arranged according to its hierarchical relationship and can be dynamically expanded and collapsed.

[0044] S52. Define interactive navigation nodes , each navigation node corresponds to a chapter node , and contains key information and links to the chapters;

[0045] S53, applying a content focus algorithm to dynamically adjust content display according to the chapter selected by the user. The content focus algorithm combines the structural depth of the chapter, the importance of the content, and the relevance of the sub-chapter, and determines the display priority by calculating the comprehensive significance value;

[0046] S54, implementing adaptive display, automatically adjusting the hierarchical display and content density of chapters by detecting the screen size of the user's device and the user's preference settings;

[0047] Optionally, the S6 specifically includes:

[0048] S61. Build a keyword indexing mechanism in a multi-level tree structure, for each chapter node Extract keywords and generate keyword sets , the initial keyword weight is based on The algorithm calculates and measures the importance of each keyword in the corresponding chapter;

[0049] S62, introducing a hierarchical weight correction model, and adjusting the initial keyword weights through a formula;

[0050] S63, using semantic relevance enhancement model, by calculating the query keyword set Q and chapter nodes The semantic match between them is used to evaluate the relevance and define the comprehensive matching score :

[0051] ;

[0052] in, is the corrected keyword weight, For keywords The semantic vector With chapter nodes The semantic vector The similarity between Adjust parameters for semantic similarity;

[0053] S64. Based on comprehensive matching score Sort chapter nodes and dynamically generate sorting results;

[0054] S65. Based on the user's historical query behavior and click preferences, use a personalized recommendation algorithm to optimize the search results, adjust keyword weights and matching scores, and provide personalized content display.

[0055] Optionally, the S62 specifically includes:

[0056] S621. For each chapter node Keyword collection Perform initial weight calculation using Algorithm determines initial weights , quantitative keywords At the chapter node The importance of

[0057] S622, define layer depth function , calculate each chapter node Hierarchical depth in a multi-level tree structure, hierarchical depth Indicates the position of a node in the document structure;

[0058] S623, introduction of adjustment coefficient , adjustment coefficient It is a constant pre-set according to the characteristics of the document structure, used to balance the influence of keyword weights of nodes at different levels;

[0059] S624: Calculate corrected keyword weights :

[0060] ;

[0061] in, For keywords At the chapter node The initial weight of is the adjustment coefficient, Indicates the hierarchical depth of a node in a multi-level tree structure. It is to modify the initial weight;

[0062] S625. Based on the corrected keyword weight , update chapter nodes Keyword collection The importance weight of each keyword in .

[0063] Optionally, the S7 specifically includes:

[0064] S71. Extract chapter content and clause description Content features, using graph neural network to generate semantic vectors and ;

[0065] S72, constructing a graph structure, where each node represents the content of a chapter or a description of an article, and the edges between the nodes represent reference relationships, and the strength of the reference relationship is quantified by the weight of the edge;

[0066] S73. Use graph neural network to embed features of each node in the graph and generate node feature vectors and ;

[0067] S74, by fusing node feature vectors and citation relationship strength, the comprehensive correlation between chapter content and article description is calculated, and an adaptive adjustment mechanism is introduced, which dynamically adjusts the weight distribution according to the node feature similarity and citation relationship strength;

[0068] S75. According to the comprehensive relevance, the binding relationship between the chapter content and the clause description is determined, and the system automatically identifies and preferentially displays the clause descriptions with relevance;

[0069] S76. When the user browses the chapter content, the system synchronously displays the relevant clause descriptions in real time according to the binding relationship, and dynamically adjusts the displayed chapter content based on the user's real-time operations and interests.

[0070] Optionally, the S74 specifically includes:

[0071] S741, extract chapter content and clause description content features to generate a fusion feature vector , fusion through heterogeneous graph neural networks;

[0072] S742. Introduce a complex reference parsing module to identify and parse multiple reference relationships between chapter content and article descriptions through natural language processing and knowledge graph technology. The parsing module focuses on the contextual dependency of implicit references and the dynamic changes of cross-document reference relationships;

[0073] S743. Adopt a dynamic feature weight optimization mechanism to dynamically adjust feature weights according to the actual complexity of the document content and the relationship between nodes. The dynamic feature weight optimization mechanism combines the semantic diversity of feature vectors and the volatility of citation intensity to generate an optimized weight configuration;

[0074] S744, build a correlation prediction model and use the reinforcement learning algorithm to optimize the model parameters. This model comprehensively considers the feature fusion vector , reference parsing results and optimized feature weights to generate accurate relevance scores;

[0075] S745. Using predicted correlation scores , automatically selects and displays the most relevant clause descriptions, dynamically adjusts the display order of the clause descriptions based on the results output by the prediction model, and updates the displayed content in real time based on user behavior.

[0076] The beneficial effects of the present invention are:

[0077] (1) The present invention uses a hierarchical relationship parsing algorithm based on a directed acyclic graph to accurately parse and extract chapter numbers and title information in standard specification documents and construct a multi-level tree structure. This method can clearly display the hierarchical structure of the document, allowing users to intuitively browse and quickly navigate to specific chapter content, significantly improving the readability of the document and the efficiency of information acquisition.

[0078] (2) The present invention matches and corrects chapter numbers and title information by applying predefined template rules to ensure the accuracy and consistency of document content. The predefined template rules can automatically identify and correct chapter hierarchical relationships, avoid errors in manual operations, and improve the automation level and efficiency of document management.

[0079] (3) The present invention combines Combined with the hierarchical weighted keyword search algorithm, an efficient search function for document content is realized. Users can quickly locate relevant chapter content by entering keywords, solving the problem that existing PDF documents do not support keyword search, greatly improving the convenience and accuracy of users' information search.

[0080] (4) The present invention uses graph neural network technology to associate and bind the article description with the corresponding chapter content, thereby realizing the synchronous display of the article description and the main text content. When users consult the main text, they do not need to frequently jump to the appendix to view the article description, which reduces the operation steps and improves the user experience and the consistency of information acquisition. BRIEF DESCRIPTION OF THE DRAWINGS

[0081] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0082] Figure 1 A flowchart of a standardized chapter splitting method proposed by the present invention;

[0083] Figure 2 A schematic diagram of a multi-level tree structure in a standardized chapter splitting method proposed by the present invention;

[0084] Figure 3 This is a schematic diagram of a keyword search mechanism in a standardized chapter splitting method proposed in the present invention. DETAILED DESCRIPTION

[0085] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, which only illustrate the basic structure of the present invention in a schematic manner, and therefore only show the components related to the present invention.

[0086] refer to Figure 1-Figure 3 , a standardized chapter splitting method, including the following steps:

[0087] S1. Obtain standard specification documents in PDF format;

[0088] S2, parse the document content, identify and extract the chapter number and title information in the document;

[0089] S3, using a hierarchical relationship parsing algorithm based on a directed acyclic graph to perform hierarchical parsing on the extracted chapter number and title information to construct a multi-level tree structure of the document;

[0090] S4. Based on the multi-level tree structure, the predefined template rules are applied to match and correct the chapter numbers and title information at different levels;

[0091] S5. Display the document content after hierarchical analysis and correction in the form of a multi-level tree structure, so that users can browse the document intuitively and navigate to specific chapter content;

[0092] S6, combined with the multi-level tree structure, based on Combined with the hierarchical weighted search algorithm, the keyword search function of the document content is realized, so that users can locate the relevant chapter content by entering keywords;

[0093] S7. Use an association binding algorithm based on graph neural network to associate and bind the clause descriptions with the corresponding chapter content, and display the relevant clause descriptions synchronously when the user browses the main text content.

[0094] In this implementation, S3 specifically includes:

[0095] S31, parsing and extracting chapter number and title information from the standard specification document, and organizing the information into a node set N and an edge set E, wherein the node set N represents a unique identifier of each chapter, a title and content corresponding to the chapter number, and the edge set E represents a hierarchical relationship between nodes;

[0096] S32. Construct a directed acyclic graph based on the information of node set N and edge set E. , in the directed acyclic graph G, each node Represents the title and content corresponding to a chapter number. The hierarchical relationship between nodes is represented by each edge in the edge set E;

[0097] S33. In the directed acyclic graph G, define the hierarchical relationship function , used to represent nodes and nodes The priority relationship between them;

[0098] S34, using the numbering format of the chapter number, the style characteristics of the title and the contextual semantic information, applying an adaptive hierarchy adjustment algorithm to sort and layer the nodes in the directed acyclic graph G;

[0099] S35, generating a hierarchical sequence of nodes based on the sorted and layered directed acyclic graph G;

[0100] S36. Construct the generated hierarchical sequence into a multi-level tree structure T. The multi-level tree structure includes all chapters and titles layer by layer starting from the root node. The root node represents the highest level node of the document, and each child node represents a secondary chapter or title in turn.

[0101] In this implementation manner, the S34 specifically includes:

[0102] S341. Obtain the numbering format information of the chapter number from the standard specification document and record the level depth of each chapter and numbering sequence ;

[0103] S342. Extracting style features of the title; defining title saliency function :

[0104] ;

[0105] in, represents the k-th style feature; is the weight coefficient of the corresponding feature, Adjust parameters for noise;

[0106] S343, using a deep semantic analysis model to map chapter content into semantic vectors , the semantic vector captures the theme and key content of the chapter, and defines the semantic relationship measurement function :

[0107] ;

[0108] in, and For Node and The semantic vector of

[0109] S344, the level depth of the chapter , Numbering rules , Title Prominence and semantic relationship measure As input, the adaptive hierarchical adjustment algorithm dynamically calculates the hierarchical relationship index of each node by fusing multi-dimensional features. ;

[0110] ;

[0111] in, For the importance of the chapter, , , , is the balance coefficient, which is used to adjust the contribution of hierarchy depth, title prominence, semantic relationship and chapter importance to the hierarchy relationship index;

[0112] S346, according to the hierarchical relationship index As a result, the order of nodes in the directed acyclic graph G is adjusted, the order of the node set N is reordered, and the hierarchical structure of the document is optimized;

[0113] S347, perform loop-free verification, use an adaptive adjustment mechanism, and adjust the node when a loop is detected. of value and recalculate the node order until the acyclic structure is stable.

[0114] In this implementation, S4 specifically includes:

[0115] S41, establishing a predefined template rule;

[0116] S42, matching the parsed chapter number and title, comparing the extracted information with the predefined template rule, and identifying the numbering pattern of the chapter number;

[0117] S43, correcting the actual style features of the title through the style feature rules, and converting the actual style features of the title into standardized style features;

[0118] S44, determining the depth attribute of each chapter according to the format of the chapter number and the number of separators;

[0119] S45. Adopt an adaptive correction algorithm to optimize the initially identified chapter hierarchy. The adaptive correction algorithm adjusts inconsistent and erroneous hierarchical structures by comparing the relationship between the chapter number format, title style features and chapter content.

[0120] In this implementation manner, S5 specifically includes:

[0121] S51. Based on the generated multi-level tree structure, the document content is hierarchically organized and a dynamic display framework is created so that each chapter is arranged according to its hierarchical relationship and can be dynamically expanded and collapsed;

[0122] S52. Define interactive navigation nodes , each navigation node corresponds to a chapter node , and contains key information and links to the chapters;

[0123] S53, applying a content focus algorithm to dynamically adjust content display according to the chapter selected by the user. The content focus algorithm combines the structural depth of the chapter, the importance of the content, and the relevance of the sub-chapter, and determines the display priority by calculating the comprehensive significance value;

[0124] S54, implementing adaptive display, automatically adjusting the hierarchical display and content density of chapters by detecting the screen size of the user's device and the user's preference settings;

[0125] In this implementation manner, S6 specifically includes:

[0126] S61. Build a keyword indexing mechanism in a multi-level tree structure, and index each chapter node Extract keywords and generate keyword sets , the initial keyword weight is based on The algorithm calculates and measures the importance of each keyword in the corresponding chapter;

[0127] S62, introducing a hierarchical weight correction model, and adjusting the initial keyword weights through a formula;

[0128] S63, using the semantic relevance enhancement model, by calculating the query keyword set Q and the chapter node The semantic match between them is used to evaluate the relevance and define the comprehensive matching score :

[0129] ;

[0130] in, is the corrected keyword weight, For keywords The semantic vector With chapter nodes The semantic vector The similarity between Adjust parameters for semantic similarity;

[0131] S64. Based on comprehensive matching score Sort chapter nodes and dynamically generate sorting results;

[0132] S65. Based on the user's historical query behavior and click preferences, use a personalized recommendation algorithm to optimize the search results, adjust keyword weights and matching scores, and provide personalized content display.

[0133] In this implementation manner, the S62 specifically includes:

[0134] S621. For each chapter node Keyword collection Perform initial weight calculation using Algorithm determines initial weights , quantitative keywords At the chapter node The importance of

[0135] S622, define layer depth function , calculate each chapter node Hierarchical depth in a multi-level tree structure, hierarchical depth Indicates the position of a node in the document structure;

[0136] S623, introduction of adjustment coefficient , adjustment coefficient It is a constant pre-set according to the characteristics of the document structure, used to balance the influence of keyword weights of nodes at different levels;

[0137] S624: Calculate corrected keyword weights :

[0138] ;

[0139] in, For keywords At the chapter node The initial weight of is the adjustment coefficient, Indicates the hierarchical depth of a node in a multi-level tree structure. It is to modify the initial weight;

[0140] S625. Based on the corrected keyword weight , update chapter nodes Keyword collection The importance weight of each keyword in .

[0141] In this implementation manner, the S7 specifically includes:

[0142] S71. Extract chapter content and clause description Content features, using graph neural network to generate semantic vectors and ;

[0143] S72, constructing a graph structure, where each node represents the content of a chapter or a description of an article, and the edges between the nodes represent reference relationships, and the strength of the reference relationship is quantified by the weight of the edge;

[0144] S73. Use graph neural network to embed features of each node in the graph and generate node feature vectors and ;

[0145] S74, by fusing node feature vectors and citation relationship strength, the comprehensive correlation between chapter content and article description is calculated, and an adaptive adjustment mechanism is introduced, which dynamically adjusts the weight distribution according to the node feature similarity and citation relationship strength;

[0146] S75. According to the comprehensive relevance, the binding relationship between the chapter content and the clause description is determined, and the system automatically identifies and preferentially displays the clause descriptions with relevance;

[0147] S76. When the user browses the chapter content, the system synchronously displays the relevant clause descriptions in real time according to the binding relationship, and dynamically adjusts the displayed chapter content based on the user's real-time operations and interests.

[0148] In this implementation manner, the S74 specifically includes:

[0149] S741, extract chapter content and clause description content features to generate a fusion feature vector , fusion through heterogeneous graph neural networks;

[0150] S742. Introduce a complex reference parsing module to identify and parse multiple reference relationships between chapter content and article descriptions through natural language processing and knowledge graph technology. The parsing module focuses on the contextual dependency of implicit references and the dynamic changes of cross-document reference relationships;

[0151] S743. Adopt a dynamic feature weight optimization mechanism to dynamically adjust feature weights according to the actual complexity of the document content and the relationship between nodes. The dynamic feature weight optimization mechanism combines the semantic diversity of feature vectors and the volatility of citation intensity to generate an optimized weight configuration;

[0152] S744, build a correlation prediction model and use the reinforcement learning algorithm to optimize the model parameters. This model comprehensively considers the feature fusion vector , reference parsing results and optimized feature weights to generate accurate relevance scores;

[0153] S745. Using predicted correlation scores , automatically selects and displays the most relevant clause descriptions, dynamically adjusts the display order of the clause descriptions based on the results output by the prediction model, and updates the displayed content in real time based on user behavior.

[0154] Embodiment 1:

[0155] Company B, a large architectural design company in Shenzhen, Guangdong Province, China, has applied a standardized chapter splitting method of the present invention. Company B has extensive experience in the architectural design industry and is responsible for the design and construction guidance of multiple large-scale engineering projects. Standard specification documents play an important role in these projects, ensuring the compliance and quality of the projects. However, traditional PDF format standard specification documents are difficult to quickly retrieve and consult, especially when facing large-scale and complex engineering design projects. This problem is particularly prominent.

[0156] In order to solve this problem, Company B began to use the method of the present invention to manage and consult standard specification documents from June 2023 to December 2023. The method significantly improves the efficiency of document use and user experience through a series of steps, including obtaining standard specification documents, parsing content, building a multi-level tree structure, matching and correcting chapter information, displaying document content, keyword retrieval, and associating and binding clause descriptions with chapter content.

[0157] During the specific implementation process, we first used automated tools to obtain standard specification documents in PDF format, parsed the document content, and identified and extracted the chapter numbers and title information in the document. Then, we used a hierarchical relationship parsing algorithm based on a directed acyclic graph to build a multi-level tree structure of the document, clarifying the hierarchical relationship between each chapter and sub-chapter. On this basis, we applied predefined template rules to match and correct chapter numbers and title information at different levels to ensure the accuracy and consistency of the document structure.

[0158] In terms of document content display, the multi-level tree structure allows users to browse documents intuitively and easily navigate to specific chapter contents. At the same time, combined with the keyword search function, users can quickly locate relevant chapters, improving the efficiency of information search. In particular, for some chapter contents involving clause descriptions, the system uses an association binding algorithm based on graph neural networks to associate the clause descriptions with the corresponding chapter contents, and synchronously display the relevant clause descriptions when users browse the main text content.

[0159] In the actual application of Company B, the data comparison before and after implementation is as follows:

[0160] Table 1 Comparison of efficiency and accuracy of standard specification document review

[0161]

[0162] In this project, Company B processed a total of 1,000 standard specification documents, with an average length of 200 pages. Each document contains multiple chapters and sub-chapters. After using the method of the present invention, the efficiency of document consultation is greatly improved, and users can accurately find the required clauses and regulations in a shorter time. Through the precise retrieval of keywords and the synchronous display of clause descriptions, users can quickly understand and apply specification requirements, avoiding design errors caused by misreading or omissions. At the same time, the system automatically matches and corrects the document content to ensure the standardization and consistency of the document structure. During the six-month implementation period, more than 5,000 document retrieval and consultation operations were carried out. User feedback shows that the new system has significantly improved work efficiency, reduced consultation time, and reduced workload.

[0163] The present invention can more accurately parse and organize the chapter contents in standard specification documents by utilizing a hierarchical relationship parsing algorithm based on directed acyclic graphs and graph neural network technology. The method intuitively displays the hierarchical structure of the document through a multi-level tree structure, allowing users to quickly locate the required chapter contents. Compared with the traditional PDF format document management method, the method of the present invention can more effectively handle complex document structures and hierarchical relationships, thereby improving the accuracy and efficiency of information search.

[0164] The present invention significantly reduces the reliance on manual operations and improves the efficiency of document management and access through automated document parsing and display processes. The automated process not only speeds up document parsing and matching, but also eliminates subjective judgment differences in manual operations, ensuring the accuracy and consistency of document structure. Combining a variety of information retrieval and display technologies (such as keyword retrieval, clause description association binding, etc.), the system can fully display and evaluate document content and provide users with accurate and timely information support.

[0165] In summary, through the method of the present invention, Company B has achieved remarkable results in the management and review of standardized documents, and improved the compliance and quality control capabilities of engineering projects. The system not only improves the efficiency and accuracy of document review, but also saves a lot of time and human resources for the company, and enhances the user's work experience and satisfaction. This successful application case proves the wide applicability and great potential of the present invention in the field of engineering design and construction.

[0166] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A standardized chapter splitting method, characterized in that: The following steps are involved: S1. Obtain standard specification documents. The documents are Format; S2, parse the document content, identify and extract the chapter number and title information in the document; S3, using a hierarchical relationship parsing algorithm based on a directed acyclic graph to perform hierarchical parsing on the extracted chapter number and title information to construct a multi-level tree structure of the document; S4. Based on the multi-level tree structure, the predefined template rules are applied to match and correct the chapter numbers and title information at different levels; S5. Display the document content after hierarchical analysis and correction in the form of a multi-level tree structure, so that users can browse the document intuitively and navigate to specific chapter content; S6, combined with the multi-level tree structure, based on Combined with the hierarchical weighted search algorithm, the keyword search function of the document content is realized, so that users can locate the relevant chapter content by entering keywords; S7. Use a graph neural network-based association binding algorithm to associate and bind the clause description with the corresponding chapter content, and display the relevant clause description synchronously when the user browses the main text content; The S3 specifically includes: S31. Parse and extract chapter number and title information from standard specification documents and organize this information into node sets and edge sets , where the node set Indicates the unique identifier of each chapter, the title and content corresponding to the chapter number, and the edge set Represents the hierarchical relationship between nodes; S32, based on node set and edge sets , construct a directed acyclic graph , in the directed acyclic graph G, each node Represents the title and content corresponding to a chapter number. The hierarchical relationship between nodes is represented by each edge in the edge set E; S33. In the directed acyclic graph G, define the hierarchical relationship function , used to represent nodes and nodes The priority relationship between them; S34, using the numbering format of the chapter number, the style characteristics of the title and the contextual semantic information, applying an adaptive hierarchy adjustment algorithm to sort and layer the nodes in the directed acyclic graph G; S35, generating a hierarchical sequence of nodes based on the sorted and layered directed acyclic graph G; S36. Construct the generated hierarchical sequence into a multi-level tree structure T. The multi-level tree structure includes all chapters and titles layer by layer starting from the root node. The root node represents the highest level node of the document, and each child node represents a secondary chapter or title in turn.

2. A standardized chapter splitting method according to claim 1, characterized in that: The S34 specifically includes: S341. Obtain the numbering format information of the chapter number from the standard specification document and record the level depth of each chapter and numbering sequence ; S342. Extracting style features of the title; defining title saliency function : ； in, represents the k-th style feature; is the weight coefficient of the corresponding feature, Adjust parameters for noise; S343, using a deep semantic analysis model to map chapter content into semantic vectors , the semantic vector captures the theme and key content of the chapter, and defines the semantic relationship measurement function : ； in, and For Node and The semantic vector of S344, the level depth of the chapter , Numbering Rules , Title Prominence and semantic relationship measure As input, the adaptive hierarchical adjustment algorithm dynamically calculates the hierarchical relationship index of each node by fusing multi-dimensional features. ; ； in, For the importance of the chapter, , , , is the balance coefficient, which is used to adjust the contribution of hierarchy depth, title prominence, semantic relationship and chapter importance to the hierarchy relationship index; S346, according to the hierarchical relationship index As a result, the order of nodes in the directed acyclic graph G is adjusted, the order of the node set N is reordered, and the hierarchical structure of the document is optimized; S347, perform loop-free verification, use an adaptive adjustment mechanism, and adjust the node when a loop is detected. of value and recalculate the node order until the acyclic structure is stable.

3. A standardized chapter splitting method according to claim 1, characterized in that: The S4 specifically includes: S41, establishing a predefined template rule; S42, matching the parsed chapter number and title, comparing the extracted information with the predefined template rule, and identifying the numbering pattern of the chapter number; S43, correcting the actual style features of the title through the style feature rules, and converting the actual style features of the title into standardized style features; S44, determining the depth attribute of each chapter according to the format of the chapter number and the number of separators; S45. Adopt an adaptive correction algorithm to optimize the initially identified chapter hierarchy. The adaptive correction algorithm adjusts inconsistent and erroneous hierarchical structures by comparing the relationship between the chapter number format, title style features and chapter content.

4. A standardized chapter splitting method according to claim 1, characterized in that: The S5 specifically includes: S51. Based on the generated multi-level tree structure, the document content is hierarchically organized and a dynamic display framework is created so that each chapter is arranged according to its hierarchical relationship and can be dynamically expanded and collapsed; S52. Define interactive navigation nodes , each navigation node corresponds to a chapter node , and contains key information and links to the chapters; S53, applying a content focusing algorithm to dynamically adjust content display according to the chapter selected by the user, the content focusing algorithm combines the structural depth of the chapter, the importance of the content, and the relevance of the sub-chapter, and determines the display priority by calculating the comprehensive significance value; S54, implement adaptive display, and automatically adjust the hierarchical display and content density of chapters by detecting the screen size of the user's device and the user's preference settings.

5. A standardized chapter splitting method according to claim 1, characterized in that: The S6 specifically includes: S61. Build a keyword indexing mechanism in a multi-level tree structure, and index each chapter node Extract keywords and generate keyword sets , the initial keyword weight is based on The algorithm calculates and measures the importance of each keyword in the corresponding chapter; S62, introducing a hierarchical weight correction model, and adjusting the initial keyword weights through a formula; S63, using semantic relevance enhancement model, by calculating the query keyword set Q and chapter nodes The semantic match between them is used to evaluate the relevance and define the comprehensive matching score : ； in, is the corrected keyword weight, For keywords The semantic vector With chapter nodes The semantic vector The similarity between Adjust parameters for semantic similarity; S64. Based on comprehensive matching score Sort chapter nodes and dynamically generate sorting results; S65. Combine the user's historical query behavior and click preferences, use personalized recommendation algorithms to optimize search results, adjust keyword weights and matching scores, and provide personalized content display.

6. A standardized chapter splitting method according to claim 5, characterized in that: The S62 specifically includes: S621. For each chapter node Keyword collection Perform initial weight calculation using Algorithm determines initial weights , quantitative keywords At the chapter node The importance of S622, define layer depth function , calculate each chapter node Hierarchical depth in a multi-level tree structure, hierarchical depth Indicates the position of a node in the document structure; S623, introduction of adjustment coefficient , adjustment coefficient It is a constant pre-set according to the characteristics of the document structure, used to balance the influence of keyword weights of nodes at different levels; S624: Calculate corrected keyword weights : ； in, For keywords At the chapter node The initial weight of is the adjustment coefficient, Indicates the hierarchical depth of a node in a multi-level tree structure. It is to modify the initial weight; S625. Based on the corrected keyword weight , update chapter nodes Keyword collection The importance weight of each keyword in .

7. A standardized chapter splitting method according to claim 1, characterized in that: The S7 specifically includes: S71. Extract chapter content and clause description Content features, using graph neural network to generate semantic vectors and ; S72, constructing a graph structure, where each node represents the content of a chapter or a description of an article, and the edges between the nodes represent reference relationships, and the strength of the reference relationship is quantified by the weight of the edge; S73. Use graph neural network to embed features of each node in the graph and generate node feature vectors and ; S74, by fusing node feature vectors and citation relationship strength, the comprehensive correlation between chapter content and article description is calculated, and an adaptive adjustment mechanism is introduced, which dynamically adjusts the weight distribution according to the node feature similarity and citation relationship strength; S75. According to the comprehensive relevance, the binding relationship between the chapter content and the clause description is determined, and the system automatically identifies and preferentially displays the clause descriptions with relevance; S76. When the user browses the chapter content, the system synchronously displays the relevant clause descriptions in real time according to the binding relationship, and dynamically adjusts the displayed chapter content based on the user's real-time operations and interests.

8. A standardized chapter splitting method according to claim 7, characterized in that: The S74 specifically includes: S741, extract chapter content and clause description content features to generate a fusion feature vector , fusion through heterogeneous graph neural networks; S742. Introduce a complex reference parsing module to identify and parse multiple reference relationships between chapter content and article descriptions through natural language processing and knowledge graph technology. The parsing module focuses on the contextual dependency of implicit references and the dynamic changes of cross-document reference relationships; S743. Adopt a dynamic feature weight optimization mechanism to dynamically adjust feature weights according to the actual complexity of the document content and the relationship between nodes. The dynamic feature weight optimization mechanism combines the semantic diversity of feature vectors and the volatility of citation intensity to generate an optimized weight configuration; S744, build a correlation prediction model, use the reinforcement learning algorithm to optimize the model parameters, the model comprehensively considers the feature fusion vector , referencing the parsing results and optimized feature weights to generate a relevance score; S745. Using predicted correlation scores , automatically selects and displays the most relevant clause descriptions, dynamically adjusts the display order of the clause descriptions based on the results output by the prediction model, and updates the displayed content in real time based on user behavior.

Images