Building standardization knowledge base building method and system based on large language model

Abstract

The invention provides a building standardization knowledge base building method and system based on a large language model, and the method comprises the steps: carrying out the semantic partitioning of a building specification document set, and obtaining knowledge unit blocks and the logic association relation thereof; processing the knowledge unit blocks by using a large language model, generating a complete coding identifier of an independent knowledge unit and a boundary association identifier of an associated knowledge unit, and determining an association type mark; for the associated knowledge units, determining adjacent knowledge unit blocks and candidate knowledge units according to association type marks and logic association relationships, matching target knowledge units through global positioning identifiers, and allocating same integration numbers; counting the number of independent knowledge units and the number of integrated numbers, and calculating the sum to determine the total quantity of knowledge entities. According to the method, accurate association and integration of knowledge units are realized through a hierarchical numbering system and semantic association strength. According to the method, the construction efficiency of the building standardized knowledge base and the semantic consistency of the knowledge entities can be improved.

Description

Technical Field

[0001] The present invention relates to the field of building information technology, and more specifically, to a method and system for building a building standardization knowledge base based on a large language model. Background Art

[0002] In the construction industry, with the continuous updating and increasing complexity of building regulations, the management and application of building standardization knowledge faces enormous challenges. Traditional construction knowledge management relies primarily on manual organization and paper document storage, which is not only inefficient but also difficult to achieve rapid retrieval and accurate association of knowledge. In recent years, with the development of information technology, a number of database-based construction knowledge management systems have begun to emerge. These systems improve the efficiency of knowledge storage and query through structured data storage and simple keyword retrieval functions. However, most of these systems lack the ability to deeply analyze the complex semantic associations between knowledge units in building regulation documents, and cannot meet the construction industry's needs for accurate management and efficient application of standardized knowledge.

[0003] In the process of implementing the embodiments of the present invention, the inventors discovered that there are at least the following problems or defects in the prior art: First, traditional methods cannot effectively handle the semantic segmentation and logical association relationships in building specification documents, resulting in the neglect of the association between knowledge units, affecting the integrity and accuracy of knowledge. Second, the prior art lacks an effective method for calculating semantic coverage when processing cross-document links and cross-system mappings, making it difficult to accurately identify and integrate related knowledge units in different documents or systems. In addition, the existing system lacks a dynamic adjustment mechanism during the knowledge integration process and is unable to deal with possible logical conflicts in knowledge entries, resulting in the accuracy and reliability of the knowledge base being affected. Summary of the Invention

[0004] The present invention provides a method and system for building a building standardization knowledge base based on a large language model.

[0005] In a first aspect of the present invention, a method for building a building standardization knowledge base based on a large language model is provided, comprising:

[0006] Performing semantic block processing on the building specification document set to obtain knowledge unit blocks and logical association relationships between the knowledge unit blocks;

[0007] Processing the knowledge unit block using a large language model to obtain complete coding identifiers of independent knowledge units in the knowledge unit block and boundary association identifiers of associated knowledge units, and determining association type tags of the associated knowledge units according to the types of the boundary association identifiers;

[0008] For each associated knowledge unit, the following operations are performed: for a selected knowledge unit, based on the association type tag, the knowledge unit block to which it belongs, and the logical association relationship, an adjacent knowledge unit block is determined, and based on the association type tag, a candidate knowledge unit is determined from all associated knowledge units contained in the adjacent knowledge unit block; and based on the global positioning identifier of the selected knowledge unit boundary identifier and the global positioning identifier of the candidate knowledge unit boundary identifier, a target knowledge unit is determined from the candidate knowledge units, and the target knowledge unit is assigned the same integration number as the selected knowledge unit; wherein the global positioning identifier is a hierarchical numbering system in the building specification document set, and the target knowledge unit is a knowledge unit that has at least a preset semantic association strength with the selected knowledge unit;

[0009] Counting the number of the independent knowledge units and the number of the integration numbers;

[0010] The sum of the number of independent knowledge units and the number of integrated numbers is calculated, and the total number of knowledge entities in the building standardization knowledge base is determined based on the sum.

[0011] Furthermore, in a case where the association type mark includes a horizontal reference mark or a vertical dependency mark, determining the target knowledge unit according to the global positioning mark of the selected knowledge unit boundary mark and the global positioning marks of each candidate knowledge unit boundary mark includes:

[0012] Calculating the hierarchical overlap depth between the global hierarchical number of the longitudinal boundary of the selected knowledge unit and the global hierarchical number of the longitudinal boundary of each candidate knowledge unit;

[0013] The candidate knowledge units whose hierarchical overlap depth is greater than a preset association threshold are used as the target knowledge units.

[0014] Furthermore, the calculating of the hierarchical overlap depth between the global hierarchical number of the longitudinal boundary of the selected knowledge unit and the global hierarchical number of the longitudinal boundary of the candidate knowledge unit includes:

[0015] Determining the type of affiliation between the selected knowledge unit and a candidate knowledge unit based on the global level number of the vertical boundary of the selected knowledge unit, wherein the type of affiliation is a parent clause-child clause dependency relationship or a same-level clause extension relationship;

[0016] In the case where the subordinate relationship is a parent clause-child clause dependency relationship, the difference between the level number of the end of the parent clause's vertical boundary and the level number of the start of the child clause's vertical boundary is calculated as the level number difference; and in the case where the subordinate relationship is a same-level clause extension relationship, the difference between the level number of the right side of the base clause's vertical boundary and the level number of the left side of the extended clause's vertical boundary is calculated as the level number difference;

[0017] When the global level number increases from top to bottom and the level number difference is positive, the difference is used as the level overlap depth; when the global level number increases from bottom to top and the level number difference is negative, the absolute value of the difference is used as the level overlap depth.

[0018] Furthermore, determining the type of subordination relationship includes:

[0019] Obtaining the center level number of the longitudinal boundary of the selected knowledge unit, and obtaining the center level number of the longitudinal boundary of the candidate knowledge unit;

[0020] The subordinate relationship is determined based on the two center level numbers. When the center level number of the knowledge unit on the left is higher than the center level number of the candidate knowledge unit on the right, it is determined to be a parent-sub-clause dependency relationship; when the center level number of the knowledge unit on the left is equal to the center level number of the candidate knowledge unit on the right, it is determined to be a same-level clause extension relationship.

[0021] Furthermore, determining the target knowledge unit according to the global positioning identifier of the selected knowledge unit boundary identifier and the global positioning identifiers of the boundary identifiers of each candidate knowledge unit further includes:

[0022] Calculate chapter span differences based on the global chapter numbers of the horizontal boundaries of the candidate knowledge units;

[0023] When the chapter span difference is less than a preset chapter tolerance, the calculation operation of the hierarchical overlap depth is performed.

[0024] Furthermore, in the case where the association type identifier includes a cross-document link identifier or a cross-system mapping identifier, determining the target knowledge unit according to the global positioning identifier of the selected knowledge unit boundary identifier and the global positioning identifiers of the candidate knowledge unit boundary identifiers includes:

[0025] Calculating the semantic coverage of the global chapter number of the horizontal boundary of the selected knowledge unit and the global chapter number of the horizontal boundary of each candidate knowledge unit;

[0026] The candidate knowledge unit whose semantic coverage is greater than a preset coverage threshold is used as the target knowledge unit.

[0027] Furthermore, the calculation of semantic coverage includes:

[0028] Determining a system affiliation between the selected knowledge unit and a candidate knowledge unit based on the global chapter number of the horizontal boundary of the selected knowledge unit and the global chapter number of the horizontal boundary of the candidate knowledge unit, wherein the system affiliation is a homologous system or a heterogeneous system;

[0029] Under the homologous system relationship, calculate the continuity difference between the ending chapter number of the horizontal boundary of the left knowledge unit and the starting chapter number of the horizontal boundary of the right candidate knowledge unit; under the heterogeneous system relationship, calculate the weight coefficient of the cross-system mapping node of the knowledge unit;

[0030] When the continuity difference reaches the preset coherence standard or the weight coefficient exceeds the preset mapping threshold, the corresponding range value is taken as the valid semantic coverage range.

[0031] Furthermore, it also includes:

[0032] Count the positioning metadata of all related knowledge units with the same integration number, and extract the minimum chapter number, maximum chapter number, lowest level number, and highest level number;

[0033] constructing a knowledge entity topology structure based on the metadata to generate a knowledge container containing a complete semantic boundary;

[0034] Reorganize the content elements of the associated knowledge units in the knowledge container according to the original logical relationship to form standardized knowledge items;

[0035] Verify the semantic integrity of the reorganized knowledge items and correct incorrect associations.

[0036] Furthermore, the method further comprises:

[0037] When verification finds logical conflicts in knowledge items, trace the source of the related knowledge units and determine the knowledge nodes that are the source of the conflicts.

[0038] Dynamically adjust the integrated numbering strategy based on conflict analysis results: add subnumbers in the case of local conflicts, and create new knowledge entity branches in the case of systemic conflicts;

[0039] The statistical integration number number includes the number of valid entities under the sub-numbering system.

[0040] In a second aspect of the present invention, a system for building a building standardization knowledge base based on a large language model is provided, comprising:

[0041] The system comprises a large language model processing module, an association analysis engine and a knowledge integration platform, and is configured to execute the method for building a building standardization knowledge base based on a large language model as described in any one of the first aspects.

[0042] The above embodiments of the present invention have at least the following beneficial effects:

[0043] 1. The present invention can accurately identify knowledge unit blocks and their logical associations by semantically segmenting the building specification document set, and use a large language model to encode and identify knowledge units and mark their association types, thereby achieving accurate associations between knowledge units. When processing cross-document links and cross-system mappings, the semantic coverage can be accurately calculated, and related knowledge units in different documents or systems can be effectively integrated to enrich the content and association of the knowledge base. At the same time, the present invention can also dynamically adjust the integrated number allocation strategy, promptly respond to logical conflicts in knowledge entries, ensure the accuracy and reliability of the knowledge base, and provide strong support for knowledge management and application in the construction industry.

[0044] 2. The present invention can quickly count the number of independent knowledge units and integrated numbers, calculate the total number of knowledge entities, and provide data support for the management and optimization of knowledge bases. By constructing a knowledge entity topology and generating a knowledge container with complete semantic boundaries, the content elements of related knowledge units are reorganized according to their original logical relationships to form standardized knowledge entries, improving the readability and usability of knowledge. This makes the present invention more advantageous in the construction and application of standardized construction knowledge bases, effectively meeting the construction industry's demand for efficient, accurate, and dynamic knowledge management.

[0045] 3. This invention simplifies the specification review process and improves efficiency. Through semantic segmentation and association analysis, users can more quickly locate the required knowledge units and related content, eliminating the need to search word for word in large and complex documents. Simultaneously, it controls the accuracy of building specification information in large model databases. Through precise semantic segmentation, association analysis, and dynamically adjusted and integrated number allocation strategies, it effectively reduces erroneous associations and inaccurate information, thereby improving the reliability of research results. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] The above and other objects, features and advantages of the exemplary embodiments of the present invention will become readily apparent by reading the following detailed description with reference to the accompanying drawings, in which several embodiments of the present invention are shown by way of example and not limitation, in which:

[0047] Figure 1 A flowchart of a method for building a building standardization knowledge base based on a large language model provided by one embodiment of the present invention;

[0048] Figure 2 A schematic diagram of the structure of a system for building a building standardization knowledge base based on a large language model provided by one embodiment of the present invention. DETAILED DESCRIPTION

[0049] The principles and spirit of the present invention will be described below with reference to several exemplary embodiments. It should be understood that these embodiments are provided solely to enable those skilled in the art to better understand and implement the present invention, and are not intended to limit the scope of the present invention in any way. Rather, these embodiments are provided to make the present invention more thorough and complete, and to fully convey the scope of the present invention to those skilled in the art.

[0050] Those skilled in the art will appreciate that the embodiments of the present invention may be implemented as a system, apparatus, device, method, or computer program product. Therefore, the present invention may be implemented in the following forms: entirely in hardware, entirely in software (including firmware, resident software, microcode, etc.), or in a combination of hardware and software.

[0051] It should be noted that any number of elements in the drawings is for illustration only and not for limitation, and any naming is only for distinction and does not have any limiting meaning.

[0052] Example 1

[0053] Reference below Figure 1 , Figure 1 This is a flow chart of a method for building a building standardization knowledge base based on a large language model according to an embodiment of the present invention. Figure 1 As shown, a method for building a building standardization knowledge base based on a large language model includes:

[0054] S1. Performing semantic block processing on the building specification document set to obtain knowledge unit blocks and logical association relationships between the knowledge unit blocks;

[0055] S2. Processing the knowledge unit block using a large language model to obtain complete coding identifiers of independent knowledge units and boundary association identifiers of associated knowledge units in the knowledge unit block, and determining association type tags of the associated knowledge units based on the types of the boundary association identifiers;

[0056] S3. For each associated knowledge unit, perform the following operations: for a selected knowledge unit, determine an adjacent knowledge unit block based on the association type tag, the knowledge unit block to which it belongs, and the logical association relationship, and determine a candidate knowledge unit among all associated knowledge units contained in the adjacent knowledge unit block based on the association type tag; and determine a target knowledge unit among the candidate knowledge units based on the global positioning identifier of the selected knowledge unit boundary identifier and the global positioning identifier of the candidate knowledge unit boundary identifier, and assign the target knowledge unit the same integration number as the selected knowledge unit; wherein the global positioning identifier is a hierarchical numbering system in the building specification document set, and the target knowledge unit is a knowledge unit that has at least a preset semantic association strength with the selected knowledge unit;

[0057] S4. Counting the number of the independent knowledge units and the number of the integrated numbers;

[0058] S5. Calculate the sum of the number of independent knowledge units and the number of integrated numbers, and determine the total number of knowledge entities in the building standardization knowledge base based on the sum.

[0059] It should be noted that the method for constructing a building standardization knowledge base based on a large language model proposed in this invention semantically segments a set of building specification documents into blocks of knowledge and logical relationships between these blocks. Semantic segmentation refers to dividing building specification documents into blocks of independent meaning, such as clauses and chapters, based on their semantic content. Logical relationships refer to dependencies, references, and other relationships between these blocks of knowledge. In this way, complex building specification documents can be broken down into structured data that is easier to process and understand.

[0060] Specifically, a building specification document set refers to a collection of documents including standards, specifications, and regulations for the construction industry. These documents usually exist in text form and cover the specific requirements of each link such as building design, construction, and acceptance. During the implementation process, each document in the document set can be divided into blocks according to its logical structure (such as chapters, clauses, etc.), and each block is a knowledge unit block. For example, a building specification document may be divided into multiple chapters, and each chapter contains several clauses. These chapters and clauses can be regarded as knowledge unit blocks. Logical association relationships can be determined by analyzing references, subordination, and other relationships in the document. For example, a clause may reference the content of another clause, or a chapter may be an extension of another chapter. These relationships can be identified and extracted through natural language processing technology.

[0061] Preferably, semantic segmentation can be implemented using a text segmentation algorithm in natural language processing. Specifically, a pre-trained large language model (such as a model with a Transformer architecture) can be used to perform semantic analysis on the building specification document. Through the model's contextual understanding ability, the document can be divided into unit blocks with independent semantics. For example, the model can identify the boundaries of chapters and clauses based on features such as the title and paragraph structure in the document, and use these boundaries as the basis for segmentation. At the same time, when determining the logical association relationship, the reference marks in the document can be analyzed (such as see Article X) or the reference relationship between clauses can be determined by calculating semantic similarity. For example, by calculating the semantic similarity between two knowledge unit blocks, if the similarity exceeds a certain threshold (such as 0.8), it can be considered that there is a logical association relationship between them. In addition, when processing knowledge unit blocks, each block can also be coded and identified, for example, using a unique number or string as an identifier, so that these knowledge unit blocks can be quickly located and referenced in the subsequent processing process.

[0062] In some embodiments, when the association type mark includes a horizontal reference mark or a vertical dependency mark, determining the target knowledge unit according to the global positioning mark of the selected knowledge unit boundary mark and the global positioning marks of each candidate knowledge unit boundary mark includes:

[0063] Calculating the hierarchical overlap depth between the global hierarchical number of the longitudinal boundary of the selected knowledge unit and the global hierarchical number of the longitudinal boundary of each candidate knowledge unit;

[0064] The candidate knowledge units whose hierarchical overlap depth is greater than a preset association threshold are used as the target knowledge units.

[0065] It should be noted that the present invention further clarifies how to determine the target knowledge unit when processing the case where the association type is marked as a horizontal reference identifier or a vertical dependency identifier. Specifically, by calculating the hierarchical overlap depth between the selected knowledge unit and the candidate knowledge unit, the candidate knowledge unit with a strong semantic association with the selected knowledge unit is screened out as the target knowledge unit. The hierarchical overlap depth refers to the degree of overlap between the selected knowledge unit and the candidate knowledge unit in the hierarchical structure of the building specification document, which is used to measure the strength of the association between them. Through this method, the knowledge units associated with the selected knowledge unit can be identified more accurately, thereby improving the accuracy and efficiency of knowledge base construction.

[0066] Specifically, a horizontal reference identifier refers to a horizontal reference relationship between knowledge units, such as a clause referencing another clause in the same chapter; while a vertical dependency identifier refers to a vertical dependency relationship between knowledge units, such as a clause being a subclause or extension of another clause. During implementation, the global hierarchical numbering of the vertical boundaries of the selected and candidate knowledge units must first be obtained. The global hierarchical numbering system is used to identify the hierarchical position of knowledge units in building specification documents. For example, a document may be numbered by chapter, clause, subclause, and so on. Each hierarchical number can be a numerical sequence, such as 1.1 for the first clause of Chapter 1, or 1.1.1 for the first subclause of the first clause of Chapter 1. By comparing the global hierarchical numbering of the vertical boundaries of the selected and candidate knowledge units, the depth of hierarchical overlap between them can be calculated. For example, if the global hierarchical numbering of the vertical boundary of the selected and candidate knowledge units is 1.1 and the global hierarchical numbering of the vertical boundary of the candidate knowledge unit is 1.1, their hierarchical overlap depth is 2, indicating that they overlap at the first two levels. The preset association threshold is a parameter used to determine whether a candidate knowledge unit is a target knowledge unit. It can be set according to actual needs. For example, it can be set to 2 or 3, which means that the candidate knowledge unit is considered as a target knowledge unit only when the hierarchical overlap depth is greater than or equal to the threshold.

[0067] Preferably, when calculating the hierarchical overlap depth, first, determine the type of subordination of the global hierarchical number of the vertical boundary of the selected knowledge unit and the candidate knowledge unit. Subordination types include parent clause-child clause dependency and same-level clause extension relationship. For example, if the hierarchical number of the selected knowledge unit is 1.1 and the hierarchical number of the candidate knowledge unit is 1.1.1, then there is a parent clause-child clause dependency relationship between them; if the hierarchical number of the selected knowledge unit is 1.1 and the hierarchical number of the candidate knowledge unit is 1.2, then there is a same-level clause extension relationship between them. After determining the subordination type, calculate the hierarchical number difference according to different relationship types. For the parent clause-child clause dependency relationship, calculate the difference between the hierarchical number of the end of the parent clause vertical boundary and the hierarchical number of the starting of the child clause vertical boundary; for the same-level clause extension relationship, calculate the difference between the hierarchical number on the right side of the base clause vertical boundary and the hierarchical number on the left side of the extended clause vertical boundary. For example, for the parent clause-child clause dependency relationship, if the vertical boundary end level number of the parent clause is 1.1 and the vertical boundary starting level number of the child clause is 1.1.1, the level number difference is 1. For the same-level clause extension relationship, if the vertical boundary right level number of the base clause is 1.1 and the vertical boundary left level number of the extended clause is 1.2, the level number difference is also 1. When the global level number increases from top to bottom and the level number difference is a positive value, the difference is used as the level overlap depth; when the global level number increases from bottom to top and the level number difference is a negative value, the absolute value of the difference is used as the level overlap depth. In this way, the strength of the association between the selected knowledge unit and the candidate knowledge unit can be measured more accurately, thereby more precisely determining the target knowledge unit.

[0068] In some embodiments, the step of calculating the level overlap depth between the global level number of the vertical boundary of the selected knowledge unit and the global level number of the vertical boundary of the candidate knowledge unit includes:

[0069] Determining the type of affiliation between the selected knowledge unit and a candidate knowledge unit based on the global level number of the vertical boundary of the selected knowledge unit, wherein the type of affiliation is a parent clause-child clause dependency relationship or a same-level clause extension relationship;

[0070] In the case where the subordinate relationship is a parent clause-child clause dependency relationship, the difference between the level number of the end of the parent clause's vertical boundary and the level number of the start of the child clause's vertical boundary is calculated as the level number difference; and in the case where the subordinate relationship is a same-level clause extension relationship, the difference between the level number of the right side of the base clause's vertical boundary and the level number of the left side of the extended clause's vertical boundary is calculated as the level number difference;

[0071] When the global level number increases from top to bottom and the level number difference is positive, the difference is used as the level overlap depth; when the global level number increases from bottom to top and the level number difference is negative, the absolute value of the difference is used as the level overlap depth.

[0072] It should be noted that the present invention determines the type of subordination between knowledge units by comparing the global hierarchical numbers of the vertical boundaries of the selected knowledge unit and the candidate knowledge unit, determining the type of subordination between them, and calculating the difference in hierarchical numbers according to different subordination types, thereby obtaining the hierarchical overlap depth. The subordination type here refers to the hierarchical relationship between knowledge units, such as the dependency relationship between a parent clause and a child clause, or the extension relationship between clauses at the same level. In this way, the strength of the association between knowledge units can be measured more accurately, ensuring that the selection of the target knowledge unit is more accurate.

[0073] Specifically, the types of subordinate relationships include parent-clause-child clause dependency and same-level clause extension relationships. During implementation, it is first necessary to obtain the global hierarchical number of the vertical boundaries of the selected knowledge unit and the candidate knowledge unit. The global hierarchical number is a numbering system used to identify the hierarchical position of the knowledge unit in the building specification document. For example, 1.1 represents the first clause of the first chapter, and 1.1.1 represents the first subclause of the first clause of the first chapter. By comparing the global hierarchical numbers of the vertical boundaries of two knowledge units, the type of subordinate relationship between them can be determined. For example, if the hierarchical number of the selected knowledge unit is 1.1 and the hierarchical number of the candidate knowledge unit is 1.1.1, then the relationship between them is a parent-clause-child clause dependency relationship; if the hierarchical number of the selected knowledge unit is 1.1 and the hierarchical number of the candidate knowledge unit is 1.2, then the relationship between them is a same-level clause extension relationship. After determining the type of subordinate relationship, the hierarchical number difference is calculated according to different relationship types. For parent-subclause dependencies, the difference between the end level number of the parent clause's vertical boundary and the start level number of the subclause's vertical boundary is calculated. For same-level clause extension relationships, the difference between the right level number of the base clause's vertical boundary and the left level number of the extended clause's vertical boundary is calculated. For example, for a parent-subclause dependency, if the end level number of the parent clause's vertical boundary is 1.1 and the start level number of the subclause's vertical boundary is 1.1.1, the level number difference is 1. For same-level clause extension relationships, if the right level number of the base clause's vertical boundary is 1.1 and the left level number of the extended clause's vertical boundary is 1.2, the level number difference is also 1. When global level numbering increases from top to bottom and the level number difference is positive, the difference is used as the level overlap depth. When global level numbering increases from bottom to top and the level number difference is negative, the absolute value of the difference is used as the level overlap depth. This method can more accurately measure the strength of associations between knowledge units.

[0074] Preferably, the following steps are included when determining the type of subordination. First, the vertical boundary center level number of the selected knowledge unit and the candidate knowledge unit is obtained. The center level number refers to the middle value of the vertical boundary number of the knowledge unit. For example, for the level number 1.1.1, the center level number can be 1.1. By comparing the two center level numbers, the subordination type can be judged more accurately. For example, when the center level number of the knowledge unit on the left is higher than the center level number of the candidate knowledge unit on the right, it is determined to be a parent clause-child clause dependency relationship; when the center level number of the knowledge unit on the left is equal to the center level number of the candidate knowledge unit on the right, it is determined to be a same-level clause extension relationship. When calculating the level number difference, a preset level number difference threshold can be introduced, for example, set to 1 or 2, to further screen the candidate knowledge units. Only when the level number difference is less than or equal to the threshold, the candidate knowledge unit is considered to have a strong correlation with the selected knowledge unit, and is thus determined as the target knowledge unit. In this way, the association relationship between knowledge units can be more accurately identified, improving the accuracy and reliability of knowledge base construction.

[0075] In some embodiments, determining the type of affiliation includes:

[0076] Obtaining the center level number of the longitudinal boundary of the selected knowledge unit, and obtaining the center level number of the longitudinal boundary of the candidate knowledge unit;

[0077] The subordinate relationship is determined based on the two center level numbers. When the center level number of the knowledge unit on the left is higher than the center level number of the candidate knowledge unit on the right, it is determined to be a parent-sub-clause dependency relationship; when the center level number of the knowledge unit on the left is equal to the center level number of the candidate knowledge unit on the right, it is determined to be a same-level clause extension relationship.

[0078] When determining the type of subordinate relationship between knowledge units, the present invention further proposes a method based on the center level number, including comparing the vertical boundary center level numbers of the selected knowledge unit and the candidate knowledge unit, so as to more accurately determine the subordinate relationship between them. The center level number refers to the middle value of the vertical boundary number of the knowledge unit, which is used to reflect the relative position of the knowledge unit in the hierarchical structure. Through this method, the hierarchical relationship between knowledge units can be more accurately identified, thereby providing a more accurate basis for the subsequent calculation of the hierarchical overlap depth, further improving the accuracy and efficiency of knowledge base construction.

[0079] Specifically, the central level number refers to the middle value of the vertical boundary number of the knowledge unit. For example, for the level number 1.1.1, the central level number can be 1.1. During the implementation process, it is first necessary to obtain the vertical boundary central level numbers of the selected knowledge unit and the candidate knowledge unit. For example, if the level number of the selected knowledge unit is 1.1.1, its central level number is 1.1; if the level number of the candidate knowledge unit is 1.1.2, its central level number is also 1.1. By comparing the two central level numbers, the type of subordinate relationship can be determined. When the central level number of the knowledge unit on the left is higher than the central level number of the candidate knowledge unit on the right, it is determined to be a parent clause-child clause dependency relationship; when the central level number of the knowledge unit on the left is equal to the central level number of the candidate knowledge unit on the right, it is determined to be a same-level clause extension relationship. For example, if the central level number of the selected knowledge unit is 1.1 and the central level number of the candidate knowledge unit is 1.2, then they are in a same-level clause extension relationship; if the central level number of the selected knowledge unit is 1.1 and the central level number of the candidate knowledge unit is 1.1.1, then they are in a parent-child clause dependency relationship. This method can more accurately identify the hierarchical relationship between knowledge units, providing a more accurate basis for subsequent calculation of the hierarchical overlap depth.

[0080] Preferably, the type of subordinate relationship is determined. First, the hierarchical structure information of the knowledge unit is extracted by natural language processing technology. For example, a pre-trained large language model (such as a model with a Transformer architecture) is used to perform semantic analysis on the building specification document to extract the hierarchical number of each knowledge unit. Then, the central hierarchical number of each knowledge unit is calculated. For example, for the hierarchical number 1.1.1, the central hierarchical number can be 1.1. When comparing the central hierarchical numbers, a preset hierarchical number tolerance value can be introduced, for example, set to 0.5, to handle small differences between hierarchical numbers. Only when the difference between two central hierarchical numbers is less than or equal to the tolerance value are they considered to have the same hierarchical relationship. In this way, the hierarchical relationship between knowledge units can be identified more accurately, further improving the accuracy and reliability of knowledge base construction. In addition, in practical applications, the hierarchical number tolerance value can be flexibly adjusted according to the specific building specification document structure and requirements to adapt to different hierarchical structures and association relationships.

[0081] In some embodiments, determining the target knowledge unit according to the global positioning identifier of the selected knowledge unit boundary identifier and the global positioning identifiers of the boundary identifiers of each candidate knowledge unit further includes:

[0082] Calculate chapter span differences based on the global chapter numbers of the horizontal boundaries of the candidate knowledge units;

[0083] When the chapter span difference is less than a preset chapter tolerance, the calculation operation of the hierarchical overlap depth is performed.

[0084] It should be noted that the calculation of the chapter span difference, by calculating the difference in the global chapter number of the horizontal boundary between the selected knowledge unit and the candidate knowledge unit, can more accurately screen candidate knowledge units with strong semantic associations with the selected knowledge unit. The chapter span difference refers to the distance between two knowledge units in the horizontal hierarchical structure and is used to measure their association at the chapter level. By combining the calculation of the chapter span difference and the hierarchical overlap depth, the strength of the association between knowledge units can be more comprehensively assessed, thereby improving the recognition accuracy of the target knowledge unit.

[0085] Specifically, the horizontal boundary global chapter number refers to the horizontal hierarchical position number of the knowledge unit in the building specification document. For example, Chapter 1 can be numbered 1, and Chapter 2 can be numbered 2. During implementation, the horizontal boundary global chapter numbers of the selected knowledge unit and the candidate knowledge unit must first be obtained. For example, if the selected knowledge unit is located in Chapter 1, its horizontal boundary global chapter number is 1; if the candidate knowledge unit is located in Chapter 2, its horizontal boundary global chapter number is 2. The chapter span difference is calculated by calculating the difference between these two numbers. For example, if the horizontal boundary global chapter number of the selected knowledge unit is 1 and the horizontal boundary global chapter number of the candidate knowledge unit is 2, the chapter span difference is 1. The preset chapter tolerance is a parameter used to determine whether a candidate knowledge unit is a target knowledge unit. It can be set according to actual needs. For example, it can be set to 1 or 2, indicating that the candidate knowledge unit is considered a target knowledge unit only if the chapter span difference is less than or equal to the tolerance value. This method can more accurately screen candidate knowledge units that have a strong correlation with the selected knowledge unit at the chapter level.

[0086] Preferably, calculating the chapter span difference also includes: first, extracting the chapter structure information of the knowledge unit using natural language processing techniques, such as using a pre-trained large language model to perform semantic analysis on the building specification document and extracting the chapter number of each knowledge unit. Then, calculating the difference between the global chapter numbers of the horizontal boundaries of the selected knowledge unit and the candidate knowledge unit. For example, if the global chapter number of the horizontal boundary of the selected knowledge unit is 1 and the global chapter number of the horizontal boundary of the candidate knowledge unit is 3, then the chapter span difference is 2. In practical applications, the preset chapter tolerance value can be flexibly adjusted based on the specific structure and requirements of the building specification document. For example, for documents with more complex structures, the preset chapter tolerance value can be appropriately increased to accommodate a wider range of chapter associations. Furthermore, the semantic similarity calculation of the knowledge units can be combined to further verify the association between the candidate knowledge unit and the selected knowledge unit. For example, if the chapter span difference value is less than the preset chapter tolerance value and the semantic similarity exceeds a certain threshold (e.g., 0.7), the candidate knowledge unit can be more accurately identified as the target knowledge unit. In this way, the strength of the association between knowledge units can be more comprehensively assessed, improving the accuracy of identifying the target knowledge unit.

[0087] In some embodiments, when the association type flag includes a cross-document link flag or a cross-system mapping flag, determining the target knowledge unit according to the global positioning flag of the selected knowledge unit boundary flag and the global positioning flags of each candidate knowledge unit boundary flag includes:

[0088] Calculating the semantic coverage of the global chapter number of the horizontal boundary of the selected knowledge unit and the global chapter number of the horizontal boundary of each candidate knowledge unit;

[0089] The candidate knowledge unit whose semantic coverage is greater than a preset coverage threshold is used as the target knowledge unit.

[0090] The present invention determines the target knowledge unit based on the calculation method of semantic coverage when the processing association type is marked as a cross-document link identifier or a cross-system mapping identifier. By calculating the semantic coverage of the selected knowledge unit and the candidate knowledge unit, the candidate knowledge unit with a strong semantic association with the selected knowledge unit is screened out as the target knowledge unit. Semantic coverage refers to the degree of overlap between two knowledge units in semantic content, which is used to measure the strength of the association between them. Through this method, the knowledge units associated with the selected knowledge unit across documents or systems can be more accurately identified, thereby improving the accuracy and efficiency of knowledge base construction.

[0091] Specifically, a cross-document link identifier refers to a reference relationship between knowledge units across different documents, such as a clause in one document referencing a clause in another. A cross-system mapping identifier refers to a mapping relationship between knowledge units across different knowledge systems, such as a clause in one standard system having a semantic correspondence with a clause in another standard system. During implementation, the global section numbers of the horizontal boundaries of the selected and candidate knowledge units must first be obtained. Global section numbers are a numbering system used to identify the horizontal position of knowledge units within building standard documents. For example, Chapter 1 might be numbered 1, and Chapter 2 might be numbered 2. By comparing the global section numbers of the horizontal boundaries of the selected and candidate knowledge units, the semantic coverage between them can be calculated. For example, if the global section number of the horizontal boundary of the selected knowledge unit is 1 and the global section number of the horizontal boundary of the candidate knowledge unit is 2, further calculation of their semantic coverage is required. The preset coverage threshold is a parameter used to determine whether a candidate knowledge unit is a target knowledge unit. It can be set according to actual needs. For example, it can be set to 0.7 or 0.8, indicating that the candidate knowledge unit is considered a target knowledge unit only if its semantic coverage is greater than or equal to the threshold. In this way, candidate knowledge units that have a strong semantic association with the selected knowledge unit can be screened out more accurately.

[0092] Preferably, when calculating semantic coverage, first, natural language processing techniques are used to extract the semantic features of knowledge units. For example, a pre-trained large language model (such as a Transformer-based model) is used to perform semantic analysis on building specification documents, extracting semantic features such as keywords and subject terms for each knowledge unit. Then, the system affiliation between the selected knowledge unit and the candidate knowledge unit is determined based on the global chapter numbers of their horizontal boundaries. System affiliation includes both homologous and heterogeneous systems. A homologous system refers to two knowledge units belonging to the same knowledge system, such as different chapters of the same building specification; a heterogeneous system refers to two knowledge units belonging to different knowledge systems, such as clauses between different building specifications. In a homologous system, the continuity difference between the ending chapter number of the horizontal boundary of the knowledge unit on the left and the starting chapter number of the horizontal boundary of the candidate knowledge unit on the right is calculated; in a heterogeneous system, the weight coefficient of the knowledge unit's cross-system mapping node is calculated. For example, if the selected knowledge unit and the candidate knowledge unit belong to the same system, and the ending chapter number of the horizontal boundary of the selected knowledge unit is 1, and the starting chapter number of the horizontal boundary of the candidate knowledge unit is 2, then the continuity difference is 1. If they belong to heterogeneous systems, it is necessary to calculate the weight coefficients of the cross-system mapping nodes, such as the weight coefficients calculated through semantic similarity. When the continuity difference reaches the preset coherence standard (such as less than or equal to 1) or the weight coefficient exceeds the preset mapping threshold (such as greater than 0.8), the corresponding range value is used as the effective semantic coverage range. In this way, the semantic association strength between knowledge units can be more comprehensively evaluated, improving the recognition accuracy of the target knowledge unit.

[0093] In some embodiments, calculating semantic coverage includes:

[0094] Determining a system affiliation between the selected knowledge unit and a candidate knowledge unit based on the global chapter number of the horizontal boundary of the selected knowledge unit and the global chapter number of the horizontal boundary of the candidate knowledge unit, wherein the system affiliation is a homologous system or a heterogeneous system;

[0095] Under the homologous system relationship, calculate the continuity difference between the ending chapter number of the horizontal boundary of the left knowledge unit and the starting chapter number of the horizontal boundary of the right candidate knowledge unit; under the heterogeneous system relationship, calculate the weight coefficient of the cross-system mapping node of the knowledge unit;

[0096] When the continuity difference reaches the preset coherence standard or the weight coefficient exceeds the preset mapping threshold, the corresponding range value is taken as the valid semantic coverage range.

[0097] By analyzing the system affiliation of the selected knowledge unit and candidate knowledge units, and calculating their continuity differences in the homologous system and weight coefficients in the heterogeneous system, the strength of the semantic association between them can be more accurately assessed. System affiliation here refers to whether two knowledge units belong to the same knowledge system. For example, different chapters of the same building code belong to the homologous system, while clauses of different building codes belong to heterogeneous systems. In this way, knowledge units associated with the selected knowledge unit across documents or systems can be more accurately identified, further improving the accuracy and efficiency of knowledge base construction.

[0098] Specifically, a homologous system refers to two knowledge units belonging to the same knowledge system, such as different chapters of the same building code; a heterogeneous system refers to two knowledge units belonging to different knowledge systems, such as clauses in different building codes. During implementation, the system affiliation between the selected knowledge unit and the candidate knowledge unit must first be determined. This can be achieved by analyzing their source documents or chapters. For homologous system relationships, the continuity difference between the ending chapter number of the left-hand knowledge unit's horizontal boundary and the starting chapter number of the right-hand candidate knowledge unit's horizontal boundary is calculated. For example, if the ending chapter number of the selected knowledge unit's horizontal boundary is 1 and the starting chapter number of the candidate knowledge unit's horizontal boundary is 2, the continuity difference is 1. The preset coherence criterion is a parameter used to determine whether the continuity difference is reasonable and can be set according to actual needs, for example, to be less than or equal to 1. For heterogeneous system relationships, the weight coefficient of the knowledge unit's cross-system mapping node is calculated. The weight coefficient can be calculated based on semantic similarity, such as using methods such as cosine similarity or Jaccard similarity. The preset mapping threshold is a parameter used to determine whether the weight coefficient is sufficiently high and can be set according to actual needs, for example, to be greater than 0.8. When the continuity difference reaches the preset coherence standard or the weight coefficient exceeds the preset mapping threshold, the corresponding range value is taken as the effective semantic coverage range. In this way, the semantic association strength between knowledge units can be more comprehensively evaluated.

[0099] Preferably, when calculating the semantic coverage, the system affiliation of the selected knowledge unit and the candidate knowledge unit is determined based on the global chapter numbers of the horizontal boundaries. For homologous system relationships, their chapter structures can be further analyzed, for example, their relevance can be evaluated by calculating the continuity of the chapter numbers. For heterogeneous system relationships, a more complex semantic similarity calculation method can be used, for example, by calculating the cosine similarity between the vector representations of two knowledge units to evaluate their relevance. For example, if the vector representations of two knowledge units are vector A and vector B respectively, their cosine similarity can be obtained by calculating the dot product of vector A and vector B divided by the product of their module lengths. When the cosine similarity exceeds the preset mapping threshold, it can be considered that the two knowledge units have a strong semantic association. In addition, the relevance between knowledge units can be further verified in combination with contextual information, for example, by analyzing their contextual context in the document to determine whether they belong to the same topic or concept. In this way, the semantic association strength between knowledge units can be more comprehensively evaluated, and the recognition accuracy of target knowledge units can be improved.

[0100] In some embodiments, further comprising:

[0101] Count the positioning metadata of all related knowledge units with the same integration number, and extract the minimum chapter number, maximum chapter number, lowest level number, and highest level number;

[0102] constructing a knowledge entity topology structure based on the metadata to generate a knowledge container containing a complete semantic boundary;

[0103] Reorganize the content elements of the associated knowledge units in the knowledge container according to the original logical relationship to form standardized knowledge items;

[0104] Verify the semantic integrity of the reorganized knowledge items and correct incorrect associations.

[0105] It should be noted that in the process of constructing a building standardization knowledge base, the present invention performs location metadata statistics on all associated knowledge units with the same integration number, and constructs a knowledge entity topology structure based on these metadata. By extracting key information such as the minimum chapter number, the maximum chapter number, the lowest level number, and the highest level number, a knowledge container containing complete semantic boundaries can be generated, and the content elements of the associated knowledge units can be reorganized according to the original logical relationships to form standardized knowledge items. In addition, the semantic integrity of the reorganized knowledge items is verified to correct incorrect association relationships. This process can effectively improve the accuracy and usability of the knowledge base and ensure the integrity and logical consistency of the knowledge items.

[0106] Specifically, location metadata refers to information describing the location of a knowledge unit within a building specification document, including section and level numbers. During implementation, the location metadata for all associated knowledge units with the same integration number must first be counted. For example, for a group of knowledge units with the same integration number, their minimum section number, maximum section number, lowest level number, and highest level number are extracted. This metadata can be used to construct the knowledge entity topology, specifically the hierarchical relationships and connections between knowledge units. Using this metadata, a knowledge container with complete semantic boundaries can be generated. This container can reorganize the content elements of related knowledge units according to their original logical relationships to form standardized knowledge entries. For example, if a group of knowledge units has a minimum section number of 1, a maximum section number of 3, a lowest level number of 1.1, and a highest level number of 3.2, a knowledge container can be constructed to reorganize the content of these knowledge units according to their section and level order. Furthermore, the semantic integrity of the reorganized knowledge entries must be verified to ensure semantic integrity and logical consistency. Any logical conflicts must be corrected.

[0107] Preferably, when constructing the knowledge entity topology, the process can be further refined. First, the semantic features and structural information of the knowledge units are extracted using natural language processing techniques. For example, a pre-trained large language model is used to perform semantic analysis on building specification documents to extract the keywords, subject terms, and hierarchical structure of each knowledge unit. Then, the knowledge entity topology is constructed based on the extracted location metadata. For example, a graph database can be used to store knowledge units and the relationships between them, where each knowledge unit is a node and the connections between nodes represent the logical associations between knowledge units. When generating a knowledge container, the scope of the knowledge container can be determined based on the minimum and maximum chapter numbers, and the hierarchical structure of the knowledge container can be determined based on the minimum and maximum level numbers. For example, if the minimum chapter number is 1, the maximum chapter number is 3, the minimum level number is 1.1, and the maximum level number is 3.2, then a knowledge container can be constructed that covers Chapters 1 to 3, with levels 1.1 to 3.2. When reorganizing knowledge items, the content of the knowledge units can be merged according to the chapter and level order to ensure the logical order of the content. Furthermore, semantic verification algorithms can be introduced to validate the reorganized knowledge entries. For example, by calculating the semantic similarity of each part of a knowledge entry, semantic coherence can be ensured. If a logical conflict is discovered, the root cause can be determined through source analysis. The integrated numbering strategy can be dynamically adjusted based on the conflict type, such as adding a subnumber in the case of a local conflict or creating a new knowledge entity branch in the case of a systemic conflict. This approach effectively improves the accuracy and usability of the knowledge base and ensures the integrity and logical consistency of the knowledge entries.

[0108] In some embodiments, the method further comprises:

[0109] When verification finds logical conflicts in knowledge items, trace the source of the related knowledge units and determine the knowledge nodes that are the source of the conflicts.

[0110] Dynamically adjust the integrated numbering strategy based on conflict analysis results: add subnumbers in the case of local conflicts, and create new knowledge entity branches in the case of systemic conflicts;

[0111] The statistical integration number number includes the number of valid entities under the sub-numbering system.

[0112] It should be noted that in the process of constructing a building standardization knowledge base, the present invention discovers logical conflicts when verifying knowledge items and performs traceability analysis on related knowledge units to determine the knowledge nodes that are the source of the conflicts. Based on the results of the conflict analysis, the integrated number allocation strategy is dynamically adjusted, thereby adding sub-numbers in the case of local conflicts and creating new knowledge entity branches in the case of systemic conflicts. This method can effectively solve the problem of logical conflicts in knowledge items and ensure the accuracy and reliability of the knowledge base. In addition, when counting the number of integrated numbers, the number of valid entities under the sub-numbering system is included, which helps to more comprehensively reflect the structure and content of the knowledge base.

[0113] Specifically, logical conflicts refer to semantic inconsistencies or logical incoherence within knowledge items, such as a clause referencing a nonexistent subclause or a contradiction between two clauses. During implementation, the semantic integrity of the reorganized knowledge items must first be verified. This can be achieved through natural language processing techniques, such as using a pre-trained large language model to perform semantic analysis on knowledge items to detect logical contradictions or semantic incoherence. If a logical conflict is found, source analysis of the associated knowledge units is necessary to identify the knowledge node at the root of the conflict. Source analysis can be performed by analyzing the sources and reference relationships of knowledge units, for example, by constructing a reference graph of knowledge units to trace the source of the conflict. Based on the conflict analysis results, the integration number allocation strategy can be dynamically adjusted. In the case of local conflicts, such as contradictions within a clause, subnumbers can be added to distinguish different content. In the case of systemic conflicts, such as contradictions between knowledge units from different systems, new knowledge entity branches can be created. When counting the number of integration numbers, the number of valid entities within the subnumbering system must be considered. For example, a main number may have multiple subnumbers, each representing an independent knowledge entity.

[0114] Preferably, when dynamically adjusting the integrated number allocation strategy, first, perform semantic analysis on the knowledge items through natural language processing technology, extract key information and construct a knowledge graph. For example, use a pre-trained large language model to analyze the knowledge items and extract the keywords, subject terms and reference relationships of each knowledge unit. Then, perform logical conflict detection based on the knowledge graph. For example, by checking the relationship between nodes and edges in the knowledge graph, detect whether there are contradictions or incoherence. If a logical conflict is found, the source of the conflict can be traced back through the traceability function of the knowledge graph. For example, by analyzing the reference path in the knowledge graph, determine the source knowledge node of the conflict. Dynamically adjust the integrated number allocation strategy according to the type and severity of the conflict. For example, for local conflicts, sub-numbers can be added under the main number, such as adding 1.1.1 and 1.1.2 under 1.1 to distinguish different content. For systemic conflicts, new knowledge entity branches can be created, such as creating a new main number 2 to represent a new knowledge system. When counting the number of integrated numbers, it's necessary to consider the number of valid entities under the sub-numbering system. For example, by traversing the nodes in the knowledge graph, we can count the number of sub-numbers under each main number and include them as the number of valid entities. This approach can more comprehensively reflect the structure and content of the knowledge base, ensuring its accuracy and reliability.

[0115] The above embodiments of the present invention have the following beneficial effects:

[0116] 1. By semantically segmenting the building specification document set, it is possible to accurately identify knowledge unit blocks and their logical associations, and use a large language model to encode and identify knowledge units and mark their association types, thereby achieving precise associations between knowledge units. When processing cross-document links and cross-system mappings, the semantic coverage can be accurately calculated, and related knowledge units in different documents or systems can be effectively integrated to enrich the content and relevance of the knowledge base. At the same time, the present invention can also dynamically adjust the integrated number allocation strategy, promptly respond to logical conflicts in knowledge entries, ensure the accuracy and reliability of the knowledge base, and provide strong support for knowledge management and application in the construction industry.

[0117] 2. By quickly counting the number of independent knowledge units and integrated numbers, the total number of knowledge entities is calculated, providing data support for the management and optimization of the knowledge base. By constructing a knowledge entity topology and generating a knowledge container with complete semantic boundaries, the content elements of related knowledge units are reorganized according to their original logical relationships to form standardized knowledge entries, improving the readability and usability of knowledge. This makes the present invention more advantageous in the construction and application of a standardized knowledge base for construction, effectively meeting the construction industry's demand for efficient, accurate, and dynamic knowledge management.

[0118] 3. Through semantic segmentation and association analysis, readers can more quickly locate the required knowledge units and related content, eliminating the need to search word for word in large and complex documents. At the same time, the accuracy of building code information in large model databases is controlled. Through precise semantic segmentation, association analysis, and dynamically adjusted and integrated number allocation strategies, the generation of erroneous associations and inaccurate information is effectively reduced, simplifying the code review process, improving code review efficiency, and enhancing the reliability of results.

[0119] Example 2

[0120] like Figure 2 As shown, some embodiments provide a system for building a building standardization knowledge base based on a large language model, the system comprising:

[0121] The large language model processing module 201, the association analysis engine 202 and the knowledge integration platform 203, the system is configured to execute the method for building a building standardization knowledge base based on a large language model according to any one of the first aspects.

[0122] It is understandable that the modules and references recorded in the building standardization knowledge base construction system based on the large language model Figure 1 The steps in the method for building a building standardization knowledge base based on a large language model correspond to each other. Therefore, the operations, features, and beneficial effects described above for the method for building a building standardization knowledge base based on a large language model are also applicable to the system for building a building standardization knowledge base based on a large language model and the modules contained therein, and will not be repeated here.

[0123] The above descriptions are merely some preferred embodiments of the present invention and an illustration of the technical principles employed. Those skilled in the art should understand that the scope of the invention involved in the embodiments of the present invention is not limited to the technical solutions formed by a specific combination of the above-mentioned technical features, but should also encompass other technical solutions formed by any combination of the above-mentioned technical features or their equivalents without departing from the above-mentioned inventive concept. For example, a technical solution formed by mutually replacing the above-mentioned features with (but not limited to) technical features having similar functions disclosed in the embodiments of the present invention.

Claims

1. A method for building a building standardization knowledge base based on a large language model, characterized in that: include: Performing semantic block processing on the building specification document set to obtain knowledge unit blocks and logical association relationships between the knowledge unit blocks; Processing the knowledge unit block using a large language model to obtain complete coding identifiers of independent knowledge units in the knowledge unit block and boundary association identifiers of associated knowledge units, and determining association type tags of the associated knowledge units according to the types of the boundary association identifiers; For each associated knowledge unit, executing: for a selected knowledge unit, determining an adjacent knowledge unit block according to the association type tag, the knowledge unit block to which it belongs, and the logical association relationship, and determining a candidate knowledge unit among all associated knowledge units contained in the adjacent knowledge unit block according to the association type tag; and, determining a target knowledge unit among the candidate knowledge units according to the global positioning identifier of the selected knowledge unit boundary identifier and the global positioning identifier of the candidate knowledge unit boundary identifier, and assigning the same integration number as that of the selected knowledge unit to the target knowledge unit; wherein the global positioning identifier is a hierarchical numbering system in the building specification document set, and the target knowledge unit is a knowledge unit having at least a preset semantic association strength with the selected knowledge unit; Counting the number of the independent knowledge units and the number of the integration numbers; The sum of the number of independent knowledge units and the number of integrated numbers is calculated, and based on this sum, the total amount of knowledge entities in the building standardization knowledge base is determined.

2. The method according to claim 1, characterized in that In a case where the association type mark includes a horizontal reference mark or a vertical dependency mark, determining the target knowledge unit according to the global positioning mark of the selected knowledge unit boundary mark and the global positioning marks of each candidate knowledge unit boundary mark includes: Calculating the hierarchical overlap depth between the global hierarchical number of the longitudinal boundary of the selected knowledge unit and the global hierarchical number of the longitudinal boundary of each candidate knowledge unit; The candidate knowledge units whose hierarchical overlap depth is greater than a preset association threshold are used as the target knowledge units.

3. The method according to claim 2, characterized in that The calculating of the hierarchical overlap depth between the global hierarchical number of the longitudinal boundary of the selected knowledge unit and the global hierarchical number of the longitudinal boundary of the candidate knowledge unit includes: Determining the type of affiliation between the selected knowledge unit and a candidate knowledge unit based on the global level number of the vertical boundary of the selected knowledge unit, wherein the type of affiliation is a parent clause-child clause dependency relationship or a same-level clause extension relationship; In the case where the subordinate relationship is a parent clause-child clause dependency relationship, the difference between the level number of the end of the parent clause's vertical boundary and the level number of the start of the child clause's vertical boundary is calculated as the level number difference; and in the case where the subordinate relationship is a same-level clause extension relationship, the difference between the level number of the right side of the base clause's vertical boundary and the level number of the left side of the extended clause's vertical boundary is calculated as the level number difference; When the global level number increases from top to bottom and the level number difference is positive, the difference is used as the level overlap depth; when the global level number increases from bottom to top and the level number difference is negative, the absolute value of the difference is used as the level overlap depth.

4. The method according to claim 3, characterized in that Determining the type of affiliation includes: Obtaining the center level number of the longitudinal boundary of the selected knowledge unit, and obtaining the center level number of the longitudinal boundary of the candidate knowledge unit; The subordinate relationship is determined based on the two center level numbers. When the center level number of the knowledge unit on the left is higher than the center level number of the candidate knowledge unit on the right, it is determined to be a parent-sub-clause dependency relationship; when the center level number of the knowledge unit on the left is equal to the center level number of the candidate knowledge unit on the right, it is determined to be a same-level clause extension relationship.

5. The method according to claim 2, characterized in that Determining the target knowledge unit according to the global positioning identifier of the selected knowledge unit boundary identifier and the global positioning identifiers of the boundary identifiers of each candidate knowledge unit further includes: Calculate chapter span differences based on the global chapter numbers of the horizontal boundaries of the candidate knowledge units; When the chapter span difference is less than a preset chapter tolerance, the calculation operation of the hierarchical overlap depth is performed.

6. The method according to claim 1, wherein In a case where the association type identifier includes a cross-document link identifier or a cross-system mapping identifier, determining the target knowledge unit according to the global positioning identifier of the selected knowledge unit boundary identifier and the global positioning identifiers of the boundary identifiers of each candidate knowledge unit includes: Calculating the semantic coverage of the global chapter number of the horizontal boundary of the selected knowledge unit and the global chapter number of the horizontal boundary of each candidate knowledge unit; The candidate knowledge unit whose semantic coverage is greater than a preset coverage threshold is used as the target knowledge unit.

7. The method according to claim 6, characterized in that The computational semantic coverage includes: Determining a system affiliation between the selected knowledge unit and a candidate knowledge unit based on the global chapter number of the horizontal boundary of the selected knowledge unit and the global chapter number of the horizontal boundary of the candidate knowledge unit, wherein the system affiliation is a homologous system or a heterogeneous system; Under the homologous system relationship, calculate the continuity difference between the ending chapter number of the horizontal boundary of the left knowledge unit and the starting chapter number of the horizontal boundary of the right candidate knowledge unit; under the heterogeneous system relationship, calculate the weight coefficient of the cross-system mapping node of the knowledge unit; When the continuity difference reaches the preset coherence standard or the weight coefficient exceeds the preset mapping threshold, the corresponding range value is taken as the valid semantic coverage range.

8. The method according to any one of claims 1 to 7, characterized in that Also includes: Count the positioning metadata of all related knowledge units with the same integration number, and extract the minimum chapter number, maximum chapter number, lowest level number, and highest level number; constructing a knowledge entity topology structure based on the metadata to generate a knowledge container containing a complete semantic boundary; Reorganize the content elements of the associated knowledge units in the knowledge container according to the original logical relationship to form standardized knowledge items; Verify the semantic integrity of the reorganized knowledge items and correct incorrect associations.

9. The method according to claim 8, characterized in that The method further comprises: When verification finds logical conflicts in knowledge items, trace the source of the related knowledge units and determine the knowledge nodes that are the source of the conflicts. Dynamically adjust the integrated numbering strategy based on conflict analysis results: add sub-numbers in the case of local conflicts, and create new knowledge entity branches in the case of systemic conflicts; The statistical integration number count includes the number of valid entities under the sub-numbering system.

10. A knowledge base construction system, characterized in that: The system comprises a large language model processing module, an association analysis engine and a knowledge integration platform, and is configured to execute the method for building a building standardization knowledge base based on a large language model as described in any one of claims 1 to 9.

Images