Component identifier matching method and system based on NLP semantic segmentation and multi-level word bank

Abstract

The invention discloses a component identifier matching method and system based on NLP semantic segmentation and a multi-level word library, and relates to the field of constructional engineering informatization, and the method comprises the steps: building a standard main word library and a preset compensation word library, carrying out the semantic segmentation of an input component identifier, and obtaining a segmented lexical element sequence, calculating semantic similarity and morphological similarity between the segmented lexical element sequence and entries in a standard main word bank, and performing standard term mapping based on the semantic similarity and the morphological similarity to obtain a standard term mapping result; carrying out structured data conversion on the segmented lexical element sequence which does not complete the standard term mapping by utilizing a regular expression mode to obtain structured data, and carrying out AI extension library matching on the segmented lexical element sequence which does not complete the structured data conversion to obtain a synonym matching result; and combining the standard term mapping result, the structured data and the synonym matching result to obtain component identification description. By implementing the method, the matching accuracy of the component identifier names in the engineering project can be improved.

Description

Technical Field

[0001] This application relates to the field of information technology in building engineering, and in particular to a component identifier matching method and system based on NLP semantic segmentation and a multi-level thesaurus. Background Technology

[0002] With the widespread application of Building Information Modeling (BIM) technology, the standardization and intelligent matching of component identification has become an important requirement for engineering design and construction management. Accurately identifying and matching component identifications from different systems and stages within the BIM model is crucial for achieving efficient flow and intelligent management of engineering information.

[0003] Currently, commonly used component identifier matching methods in the engineering field are mainly based on keyword matching and rule templates. Keyword matching involves establishing a standard thesaurus and comparing the input component identifier with standard terms in the thesaurus; rule template methods, on the other hand, use fixed pattern matching rules based on project-agreed naming conventions to parse and convert component identifiers.

[0004] However, due to the complexity of component identifiers in engineering projects, which often involve multiple languages, technical abbreviations, and project-specific codes, relying solely on static thesaurus and fixed rules for matching is prone to mismatches and missed matches. Especially during cross-disciplinary collaboration and project iteration, newly added component identifiers cannot be updated to the matching rules in a timely manner, significantly reducing matching accuracy. Summary of the Invention

[0005] This application provides a component identifier matching method and system based on NLP semantic segmentation and multi-level thesaurus, which can improve the accuracy of component identifier name matching in engineering projects.

[0006] Firstly, this application provides a component identifier matching method based on NLP semantic segmentation and a multi-level lexicon, applied to a component identifier matching system. The method includes: establishing a standard main lexicon and a preset compensation lexicon, wherein the compensation lexicon stores item naming rules and regular expression patterns corresponding to the items; performing semantic segmentation on the input component identifier to obtain a segmented word sequence, which includes a standardized terminology part and an item number part; calculating the semantic and morphological similarity between the segmented word sequence and entries in the standard main lexicon, and performing standard terminology mapping based on semantic and morphological similarity to obtain standard terminology mapping results; performing structured data transformation on the segmented word sequences that have not completed standard terminology mapping using regular expression patterns to obtain structured data; performing AI extended library matching on the segmented word sequences that have not completed structured data transformation to obtain synonym matching results; and combining the standard terminology mapping results, structured data, and synonym matching results to obtain a component identifier description.

[0007] In the above embodiments, the established standard main lexicon and the compensation lexicon form a two-layer lexicon structure. After semantic segmentation of the input component identifiers, multi-dimensional matching is performed using semantic similarity and morphological similarity. Unmatched parts are then structurally transformed using regular expression patterns. Finally, the AI ​​extended library is used for synonym matching. The synergistic effect of the multi-level lexicon and the phased matching strategy enable the system to accurately handle complex situations such as multilingual mixing, professional terminology abbreviations, and project-specific codes, thereby improving the matching accuracy of component identifiers and solving the problem of mismatch and missed match that easily occurs when relying solely on static lexicons and fixed rules.

[0008] In conjunction with some embodiments of the first aspect, in some embodiments, the step of establishing a standard main lexicon and a preset compensation lexicon, wherein the compensation lexicon stores project naming rules and regular expression patterns corresponding to the projects, specifically includes: constructing a standard main lexicon based on industry standard specifications, which includes a general terminology list, a professional abbreviation list, and a standardized naming list; constructing a compensation lexicon based on project naming rules, which includes a project naming rule mapping table and a regular expression pattern table; and constructing an AI extension library based on a deep learning model, which includes a thesaurus and a multilingual mapping table.

[0009] In the above embodiments, the standard main lexicon built based on industry standards and specifications includes a general terminology list, a professional abbreviation list, and a standardized naming list. The compensation lexicon built based on project naming rules includes a project naming rule mapping table and a regular expression pattern table. The AI ​​extension library built based on a deep learning model includes a thesaurus and a multilingual mapping table. The synergistic cooperation of the three-level lexicon system achieves the unity of standardization and project flexibility, enabling the system to have the comprehensive ability to process standard terms, project-specific codes, and newly added identifiers, effectively improving the coverage and adaptability of component identifier matching.

[0010] In conjunction with some embodiments of the first aspect, in some embodiments, the step of calculating the semantic and morphological similarity between the segmented word sequence and the entries in the standard main lexicon, and mapping standard terms based on the semantic and morphological similarity to obtain the standard term mapping result specifically includes: calculating the cosine similarity between the segmented word sequence and the entries in the standard main lexicon in the word vector space to obtain the semantic similarity; calculating the weighted edit distance between the segmented word sequence and the entries in the standard main lexicon to obtain the morphological similarity; calculating the comprehensive similarity based on the semantic and morphological similarity; selecting the optimal matching standard term based on the comprehensive similarity to obtain the standard term mapping result.

[0011] In the above embodiments, cosine similarity is calculated in the word vector space to obtain semantic similarity, and weighted edit distance is calculated to obtain morphological similarity. The two similarities are combined to obtain comprehensive similarity. Based on the comprehensive similarity, the optimal matching standard term is selected. The multi-dimensional similarity calculation and comprehensive scoring mechanism enables the system to consider both semantic association and morphological features when mapping standard terms, which improves the accuracy and reliability of matching results and realizes accurate mapping of component identifiers.

[0012] In conjunction with some embodiments of the first aspect, in some embodiments, the step of performing structured data transformation on the segmented word sequence that has not completed standard term mapping using regular expression patterns to obtain structured data, and performing AI extended library matching on the segmented word sequence that has not completed structured data transformation to obtain synonym matching results, specifically includes: performing structured data transformation on the segmented word sequence that has not completed mapping using regular expression patterns to obtain structured data; inputting the segmented word sequence that has not completed structured data transformation into an AI extended library to obtain the semantic feature vector of the segmented word sequence; calculating the semantic similarity and morphological similarity between the semantic feature vector and the synonym dictionary in the AI ​​extended library; and performing synonym matching based on semantic similarity and morphological similarity.

[0013] In the above embodiments, for segmented word sequences that have not completed standard term mapping, structured data transformation is first performed using regular expression patterns. Then, the unconverted sequences are input into the AI ​​extension library to obtain semantic feature vectors. Synonym matching is performed by calculating semantic similarity and morphological similarity. A progressive supplementary matching mechanism from rules to AI is constructed, forming a gradient processing strategy from strict rule matching to intelligent semantic understanding. This effectively solves the problem of identifying non-standard component identifiers and improves the system's ability to process special identifier forms.

[0014] In conjunction with some embodiments of the first aspect, in some embodiments, after calculating the semantic similarity and morphological similarity between the segmented word sequence and the entries in the standard main lexicon, the method further includes: obtaining engineering stage information of the current application scenario, which is the design stage, construction stage, or operation and maintenance stage; setting a first weight coefficient for semantic similarity and a second weight coefficient for morphological similarity based on the engineering stage information; dynamically weighting the semantic similarity and morphological similarity according to the first weight coefficient and the second weight coefficient to obtain a comprehensive similarity; and mapping standard terms based on the comprehensive similarity.

[0015] In the above embodiments, after obtaining the engineering stage information, the weight coefficients of semantic similarity and morphological similarity are dynamically set. The similarity is matched according to the weight coefficients to obtain the comprehensive similarity. Based on the comprehensive similarity, standard terminology is mapped, and a dynamic weight adjustment mechanism adapted to the engineering stage is established. This enables the system to adaptively adjust the matching strategy according to the characteristics of different stages such as design, construction, and operation and maintenance, thereby improving the scenario adaptability and processing accuracy of component identifier matching.

[0016] In conjunction with some embodiments of the first aspect, in some embodiments, after the step of combining the standard term mapping results, structured data, and synonym matching results to obtain the component identifier description, the method further includes: generating a structured matching report of the component identifier description, the structured matching report including standard terms, original identifiers, matching rules, and similarity; importing the component identifier description into a graph database, and constructing a mapping network between standard terms in the graph database, the mapping network being used to record the correspondence between different identifiers; detecting semantic conflicts based on the mapping network, the semantic conflicts including the same identifier corresponding to multiple standard terms, and multiple identifiers corresponding to the same standard term; when a semantic conflict is detected, generating a conflict detection report based on the structured matching report and triggering a manual review process to obtain the manual review result; adding the manual review result as a new training sample to the training dataset, and updating the standard main vocabulary.

[0017] In the above embodiments, a structured matching report is generated and the component identifier description is imported into a graph database to build a mapping network. Semantic conflicts are detected through the mapping network, and a manual review process is triggered for the detected conflicts. The review results are used as new training samples to update the standard main vocabulary, forming a closed-loop optimization mechanism from matching to feedback. The system continuously improves the vocabulary content and matching rules through continuous learning, realizing the dynamic improvement and continuous optimization of the component identifier matching capability.

[0018] In conjunction with some embodiments of the first aspect, in some embodiments, after the step of combining the standard terminology mapping results, structured data, and synonym matching results to obtain the component identifier description, the method further includes: monitoring new identifiers in the design change order; when a new identifier is detected, extracting the feature pattern of the new identifier; generating regular expression matching rules based on the feature pattern, and injecting the text information of the new identifier into a pre-trained model for incremental training; and recording the version change information of the regular expression matching rules and the standard main vocabulary.

[0019] In the above embodiments, new identifiers in the design change order are monitored and feature patterns are extracted. Regular expression matching rules are generated based on the feature patterns and incremental training is performed to build a real-time response mechanism for design changes. This realizes the dynamic expansion of matching rules and timely identification of new identifiers, and solves the problem that the matching accuracy decreases because new identifiers cannot be updated to the matching rules in a timely manner during project iteration.

[0020] In a second aspect, embodiments of this application provide a component identifier matching system, which includes: one or more processors and a memory; the memory is coupled to the one or more processors, and the memory is used to store computer program code, which includes computer instructions, and the one or more processors call the computer instructions to cause the component identifier matching system to perform the method described in the first aspect and any possible implementation thereof.

[0021] Thirdly, embodiments of this application provide a computer program product containing instructions that, when the computer program product is run on a component identifier matching system, cause the component identifier matching system to perform the method described in the first aspect and any possible implementation thereof.

[0022] Fourthly, embodiments of this application provide a computer-readable storage medium including instructions that, when executed on a component identifier matching system, cause the component identifier matching system to perform the method described in the first aspect and any possible implementation thereof.

[0023] Understandably, the component identification matching system provided in the second aspect, the computer program product provided in the third aspect, and the computer storage medium provided in the fourth aspect are all used to execute the methods provided in the embodiments of this application. Therefore, the beneficial effects they can achieve can be referred to the beneficial effects in the corresponding methods, and will not be repeated here.

[0024] One or more technical solutions provided in the embodiments of this application have at least the following technical effects or advantages: 1. This application establishes a two-layer lexicon structure by creating a standard main lexicon and a compensation lexicon. After semantic segmentation of the input component identifiers, multi-dimensional matching is performed using semantic similarity and morphological similarity. Unmatched parts are then structurally transformed using regular expression patterns. Finally, a synonym matching is performed using an AI extended library. The synergistic effect of the multi-level lexicon and the phased matching strategy enable the system to accurately handle complex situations such as multilingual mixing, professional terminology abbreviations, and project-specific codes, thereby improving the matching accuracy of component identifiers and solving the problem of mismatch and missed match that easily occurs when relying solely on static lexicons and fixed rules.

[0025] 2. This application utilizes a standard main lexicon built based on industry standards and specifications, which includes a general terminology list, a professional abbreviation list, and a standardized naming list. A supplementary lexicon built based on project naming rules includes a project naming rule mapping table and a regular expression pattern table. An AI extension library built based on a deep learning model includes a thesaurus and a multilingual mapping table. The synergistic cooperation of the three-level lexicon system achieves a balance between standardization and project flexibility, enabling the system to handle standard terms, project-specific codes, and newly added identifiers, effectively improving the coverage and adaptability of component identifier matching.

[0026] 3. This application obtains semantic similarity by calculating cosine similarity in word vector space, obtains morphological similarity by calculating weighted edit distance, and obtains comprehensive similarity by combining the two similarities. Based on the comprehensive similarity, the optimal matching standard term is selected. The multi-dimensional similarity calculation and comprehensive scoring mechanism enables the system to consider both semantic association and morphological features when mapping standard terms, thereby improving the accuracy and reliability of matching results and realizing accurate mapping of component identifiers. Attached Figure Description

[0027] Figure 1 This is a flowchart illustrating a component identifier matching method based on NLP semantic segmentation and a multi-level lexicon in an embodiment of this application. Figure 2 This is another flowchart illustrating the component identifier matching method based on NLP semantic segmentation and multi-level lexicon in the embodiments of this application; Figure 3 This is a temporal interaction diagram of the component identifier matching method based on NLP semantic segmentation and multi-level lexicon in the embodiments of this application; Figure 4 This is a schematic diagram of the physical device structure of the component identification matching system in the embodiments of this application. Detailed Implementation

[0028] The terminology used in the following embodiments of this application is for the purpose of describing particular embodiments only and is not intended to be limiting of this application. As used in the specification of this application, the singular expressions “a,” “an,” “the,” “the,” and “this” are intended to include the plural expressions as well, unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used in this application refers to any or all possible combinations including one or more of the listed items.

[0029] Hereinafter, the terms "first" and "second" are used for descriptive purposes only and should not be construed as implying or suggesting relative importance or implicitly indicating the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature, and in the description of the embodiments of this application, unless otherwise stated, "multiple" means two or more.

[0030] To facilitate understanding, the application scenarios of the embodiments of this application are described below.

[0031] A large-scale mixed-use project encompasses various business formats including commerce, offices, and hotels, involving multiple professional systems such as mechanical and electrical engineering, HVAC, and fire protection. During the BIM design process, different professional teams used different modeling software (such as Revit, AutoCAD MEP, etc.), resulting in multiple identification representations of the same component across different platforms. For example, the fire fan coil unit in Zone F on the 15th floor is marked as "HVAC-FCU-15F-01" in the HVAC system, "FP-AHU-1501" in the fire protection system, and "Fan Coil-15F-A01" in the equipment management system. This inconsistency in identification seriously affects design collaboration and construction management: design changes make it difficult to quickly locate relevant components, leading to omissions and errors in modification; it makes it difficult to accurately match drawings with physical objects on the construction site, increasing the risk of rework; and during operation and maintenance, equipment ledgers do not match the actual equipment, affecting maintenance efficiency.

[0032] The existing system uses a keyword-based string matching method to address the above problems. The system maintains a standard component naming rule library, containing a lookup table of common professional abbreviations (e.g., "HVAC" = "Heating, Ventilation and Air Conditioning") and fixed naming templates (e.g., "Profession-Equipment Type-Floor Number"). When a component identifier is input, the system first segments the string by delimiters ("-", "_", etc.), and then queries the lookup table segment by segment for conversion. However, this method has significant drawbacks: it cannot handle non-standard abbreviations (e.g., "FP" can represent both "Fire Protection" and "Floor"); it lacks sufficient support for mixed Chinese and English identifiers (e.g., "Water Supply and Drainage System-DN200" cannot be correctly parsed); and it lacks the ability to learn project-specific rules (e.g., the project-defined "A01" represents "Area A, No. 01"). This method requires extensive manual maintenance of the rule library and manual updates when new naming rules are encountered, making it difficult to adapt to the rapidly changing needs of engineering projects.

[0033] By employing the three-level lexicon collaborative matching method of this invention, the system can intelligently handle complex component identifiers. For example, when the identifier "MEP-FP-2023A" is input, the system first performs semantic segmentation using a BiLSTM-CRF model, obtaining the sequence ["MEP","FP","2023A"]. Then, it matches "MEP"="Mechanical and Electrical Engineering" and "FP"="Fire Protection System" in the standard main lexicon, and parses "2023A" into {"Year":"2023","Region":"A"} using the regular expression rules of the compensation lexicon. For newly emerging non-standard identifiers, such as "BIPV-01" (Building Photovoltaic Component) in a new energy project, the system automatically learns its meaning from the design documents, generates synonyms using the BERT model, and updates the AI ​​extended lexicon. The system also dynamically adjusts the matching strategy according to the application scenario, such as emphasizing accurate matching in the design phase and increasing fault tolerance in the construction phase. By using dynamic thesaurus updates and intelligent semantic matching, this method significantly improves the component identification matching effect in cross-professional and cross-language scenarios, effectively supporting information collaborative management throughout the BIM process.

[0034] To facilitate understanding, the method provided in this implementation will be described in detail below, using the above scenario as an example. Please refer to [link / reference]. Figure 1 This is a flowchart illustrating a component identifier matching method based on NLP semantic segmentation and a multi-level lexicon in an embodiment of this application.

[0035] S101. Establish a standard main vocabulary library and a preset compensation vocabulary library, wherein the compensation vocabulary library stores the naming rules of the items and the regular expression patterns corresponding to the items.

[0036] The standard main thesaurus is a foundational database used to store industry-standard terminology, professional abbreviations, and standardized naming conventions. It includes standard terms and their corresponding relationships in professional fields such as architecture, mechanical and electrical engineering, and HVAC. The compensation thesaurus is a dedicated database used to store project-specific naming rules and identifier resolution patterns. Project naming rules represent the agreed-upon component identifier encoding methods and composition rules in the project, while regular expression patterns are used to represent the matching rules for parsing project-specific identifiers into structured data.

[0037] During the BIM project initiation phase, a complete thesaurus system needs to be established to support subsequent component identification matching. Specifically, firstly, a standard master thesaurus is built based on national and industry standards and specifications (such as GB / T 51269), including a general terminology table (e.g., "HVAC" corresponds to "Heating, Ventilation and Air Conditioning"), a professional abbreviation table (e.g., "DN" corresponds to "Nominal Diameter"), and a specification naming table (e.g., equipment classification codes). Then, based on the project's technical requirements and contract documents, project-specific naming rules are extracted to build a supplementary thesaurus, including a project naming rule mapping table (e.g., floor numbering rules) and corresponding regular expression patterns (e.g., patterns for extracting component location information). Both thesauruses need to establish a version management mechanism to support subsequent dynamic updates.

[0038] In some embodiments, the establishment and maintenance of the thesaurus can be achieved in several ways: Optionally, text mining can be used to automatically extract standard terms and naming rules from project documents. First, OCR recognition and text preprocessing are performed on documents such as design specifications and technical specifications. Then, a named entity recognition model is used to identify technical terms and abbreviations. Finally, a pattern matching algorithm is used to extract naming rules and generate corresponding regular expressions. Optionally, a thesaurus relationship network can be constructed based on knowledge graph technology. First, an ontology model is established to define the hierarchical relationships and semantic associations between terms. Then, a graph database is used to store and manage term information. Finally, term inference and relationship expansion are implemented. It is understood that other methods can also be used to construct and update the thesaurus, such as machine learning-based term extraction methods, which are not limited here.

[0039] In some embodiments, this step may further include the following steps: A standard main thesaurus is built based on industry standards and specifications, which includes a general terminology list, a professional abbreviation list, and a standardized naming list; a supplementary thesaurus is built based on project naming rules, which includes a project naming rule mapping table and a regular expression pattern table; and an AI extension library is built based on deep learning models, which includes a thesaurus and a multilingual mapping table.

[0040] The standard main lexicon represents a set of basic terms based on industry standards. The general terminology list records commonly used professional terms and definitions. The professional abbreviation list stores standard abbreviations and their full names. The standardized naming table defines equipment classification coding rules. The compensation lexicon is a mapping lexicon built for project-specific naming rules. The project naming rule mapping table records project-defined coding rules, and the regular expression pattern table stores matching patterns for identifier parsing. The AI ​​extension library stores extended terms obtained through deep learning. The thesaurus records semantically similar term groups, and the multilingual mapping table establishes correspondences between different languages.

[0041] The process of constructing the thesaurus is implemented in three levels. First, construct the standard main thesaurus. Extract general terms and definitions from industry standards such as GB / T 51269 to create a basic term list; sort out professional abbreviations from design specifications and establish an abbreviation comparison table, such as "HVAC" corresponding to "heating, ventilation and air conditioning"; extract the equipment coding rules from the component classification standard to form a standardized naming list. Then construct the compensation thesaurus. Analyze the naming conventions in project technical documents and establish a mapping table of project-specific identification rules; write regular expression patterns based on the naming rules, such as using "MEP-FP-(\d{4})([A-Z])" to match identifiers in the format of "MEP-FP-2023A". Finally, construct the AI extension library. Use pre-trained models such as BERT to perform semantic analysis on engineering documents, and discover groups of terms with similar semantics through word vector calculation; establish mapping relationships between multiple languages such as Chinese and English to support cross-language term matching. For example, through semantic analysis, "fan coil unit", "FCU", "Fan Coil Unit", and "air handling unit" are identified as different expressions of the same device. The three-level thesaurus constructed complements each other and jointly supports the intelligent matching of component identifiers.

[0042] Based on the content in the technical disclosure, it mainly reflects the collaborative mechanism of the three-level thesaurus: the standard main thesaurus provides a normative basis, the compensation thesaurus solves project particularities, and the AI extension library realizes dynamic updates. Through hierarchical construction and collaborative effects, the problem that traditional single thesauruses cannot fully cover actual application requirements is solved. The entire thesaurus system supports the combination of static rules and dynamic learning, providing sufficient flexibility while ensuring standard normativity.

[0043] S102. Semantically segment the input component identifier to obtain a segmented token sequence, which includes a standard term part and a project number part.

[0044] Among them, semantic segmentation refers to the process of splitting the input component identifier into the smallest units with independent semantics. The segmented token sequence refers to the ordered set of tokens obtained after semantic segmentation, where the standard term part represents professional terms and abbreviations that conform to the standard naming specification, and the project number part is used to represent project-specific coding information such as floor numbers and area numbers.

[0045] After receiving the input of the component identifier to be matched, the system needs to perform structured parsing on it to support subsequent matching processing. Specifically, first perform text preprocessing on the input identifier, including case unification, special character processing, and encoding conversion; then perform semantic segmentation based on a deep learning model to split the identifier into independent semantic units; finally, perform classification annotation on the segmentation result, divide the tokens into two categories: standard terms and project numbers, and maintain the original order relationship.

[0046] In some embodiments, semantic segmentation of component identifiers can be achieved in several ways: Optionally, a rule-based segmentation method can be used, first defining delimiter rules (such as "-", "_", etc.) and lexical patterns (such as combinations of letters and numbers), then using regular expressions for initial segmentation, and finally verifying and correcting the segmentation results through dictionary matching; Optionally, a deep learning model can be used for segmentation, first encoding the identifier at the character level, then identifying lexical boundaries through sequence labeling models such as BiLSTM-CRF, and finally optimizing the segmentation results based on an attention mechanism. It is understood that other methods can also be used to achieve semantic segmentation of component identifiers, such as statistical word segmentation algorithms, which are not limited here.

[0047] S103. Calculate the semantic and morphological similarity between the segmented word sequence and the entries in the standard main lexicon. Based on the semantic and morphological similarity, perform standard term mapping to obtain the standard term mapping result.

[0048] Semantic similarity is a numerical metric measuring the degree of association between two lexical units in the semantic space, obtained by calculating the cosine distance between word vectors. Morphological similarity is a quantitative metric assessing the similarity of two lexical units in character structure, obtained by calculating the weighted edit distance. Standard term mapping represents the process of matching segmented lexical units to the most similar terms in the standard main lexicon. The results of standard term mapping are used to represent the standard terms corresponding to each segmented lexical unit and their similarity scores.

[0049] After obtaining the word sequence through semantic segmentation, these words need to be matched with terms in a standard lexicon. Specifically, firstly, the segmented words and standard terms are converted into word vector representations, and semantic features are extracted using a pre-trained language model. Then, the semantic similarity and morphological similarity between words and terms are calculated, with semantic similarity calculated using the cosine value of word vectors and morphological similarity calculated using an edit distance algorithm that considers the costs of character replacement, deletion, and insertion. Finally, the two similarities are weighted and combined, and the standard term with the highest overall similarity is selected as the mapping result. The corresponding matching score and mapping relationship are recorded.

[0050] In some embodiments, similarity calculation and term mapping can be implemented in several ways: Optionally, a deep learning-based semantic similarity calculation method first encodes word units using pre-trained models such as BERT to obtain semantic vectors, then calculates the cosine similarity between vectors to obtain semantic relevance, and finally expands the scope of similarity calculation by combining the thesaurus of WordNet; Optionally, an improved string matching algorithm is used to calculate morphological similarity, first defining the cost matrix of character operations considering the characteristics of professional terms, then using a dynamic programming algorithm to calculate the minimum edit distance, and finally obtaining the similarity score through normalization. It is understood that other methods can also be used to implement similarity calculation and term mapping, such as semantic reasoning methods based on knowledge graphs, which are not limited here.

[0051] In some embodiments, this step may further include the following steps: The semantic similarity is obtained by calculating the cosine similarity between the segmented word sequence and the words in the standard main dictionary in the word vector space; the morphological similarity is obtained by calculating the weighted edit distance between the segmented word sequence and the words in the standard main dictionary; the comprehensive similarity is obtained by calculating the semantic similarity and morphological similarity; and the optimal matching standard term is selected based on the comprehensive similarity to obtain the standard term mapping result.

[0052] In this system, the word vector space represents the feature space where words are converted into high-dimensional numerical vectors. Cosine similarity refers to the cosine of the angle between two vectors, used to measure the degree of semantic relevance. Weighted edit distance refers to the minimum edit distance in string editing operations with different weights, used to evaluate the similarity of character structures. The comprehensive similarity score is an overall matching score obtained by weighted combination of semantic similarity and morphological similarity.

[0053] The specific calculation process is performed in three steps. The first step is to calculate semantic similarity: the segmented words and vocabulary entries are converted into 768-dimensional word vector representations using the BERT model. For word sequences containing multiple parts, the word vectors of each part are averaged to obtain the sequence vector. The cosine value of the two vectors is calculated as the semantic similarity, with the formula: cos(θ) = (A·B) / (||A||·||B||), where A and B are the word sequence vector and the vocabulary vector, respectively, with a result range of [-1, 1]. For example, the word sequence vector of "FCU-01" and the vocabulary vector of "Fan Coil Unit" have a semantic similarity of 0.85. The second step is to calculate morphological similarity: a weight matrix for character operations is defined, such as a letter replacement weight of 0.6, a number replacement weight of 0.8, and a special character replacement weight of 0.4. The minimum weighted edit distance is calculated using a dynamic programming algorithm, and the morphological similarity within the range of [0, 1] is obtained through distance normalization. For example, the normalized weighted edit distance between "FCU-01" and "Fan Coil Unit" is 0.7. The third step is to obtain the overall similarity: Semantic similarity weight α and morphological similarity weight β (α+β=1) are set, and the weighted average is calculated to obtain the overall similarity. The term with the highest overall similarity exceeding a threshold (e.g., 0.6) is selected as the optimal matching result. For the above example, if α=0.6 and β=0.4, then the overall similarity = 0.85×0.6+0.7×0.4=0.79, confirming "FanCoil Unit" as the optimal matching term.

[0054] This multi-dimensional similarity-based matching method combines semantic understanding and morphological features, improving accuracy compared to single-dimensional matching methods. Adjustable weight parameters allow for flexible configuration of the matching strategy. Furthermore, a threshold mechanism is introduced to ensure the reliability of the matching results.

[0055] S104. The segmented word sequence that has not completed the standard term mapping is transformed into structured data using regular expression patterns to obtain structured data. The segmented word sequence that has not completed the structured data transformation is then matched with an AI extended library to obtain synonym matching results.

[0056] Structured data transformation refers to the process of parsing non-standard formatted lexical units into data forms with fixed structures. Regular expression patterns refer to rule expressions used to match and extract text with specific formats. AI extension library matching refers to the process of using artificial intelligence models for synonym expansion and semantic matching. Synonym matching results are used to represent semantically similar alternative terms found through AI methods.

[0057] After the standard terminology mapping is completed, the unmapped lexical units require further processing. Specifically, firstly, the unmapped lexical units are structured and parsed using regular expression patterns in the compensation lexicon to extract the encoded information that conforms to the item naming rules; then, for the lexical units that have not yet been converted, they are input into the AI ​​extension library for deep semantic analysis, and semantic feature vectors of the lexical units are generated using models such as BERT; finally, nearest neighbor search in the vector space is used to find semantically similar terms, and the most suitable synonym matching results are selected based on the similarity threshold.

[0058] In some embodiments, the processing of unmatched lexical units can be implemented in several ways: Optionally, a structured transformation method based on rule templates can be used. First, a regular expression template library is constructed according to the item naming rules. Then, the templates are used for lexical unit parsing and information extraction. Finally, the extracted information is organized into a standard data structure. Optionally, a semantic expansion method using deep learning can be used. First, a pre-trained language model is used to encode the features of lexical units. Then, an attention mechanism is used to calculate the semantic relevance with the terms in the knowledge base. Finally, the synonym matching result is determined based on a probability threshold. It is understood that other methods can also be used to process unmatched lexical units, such as statistical word sense similarity calculation methods, which are not limited here.

[0059] In some embodiments, this step may further include the following steps: The segmented word sequence with incomplete mapping is transformed into structured data using regular expression patterns to obtain structured data. The segmented word sequence with incomplete structured data transformation is input into an AI extension library to obtain the semantic feature vector of the segmented word sequence. The semantic similarity and morphological similarity between the semantic feature vector and the thesaurus in the AI ​​extension library are calculated. The thesaurus matching is performed based on the semantic similarity and morphological similarity.

[0060] The structured data transformation process uses regular expressions to parse non-standard format identifiers, converting the identifier text into structured data containing attributes such as type, position, and number through preset pattern rules. Semantic feature vectors are high-dimensional feature representations obtained by encoding word units using a pre-trained BERT model, used to capture the deep semantic information of word units. The thesaurus stores semantically similar word groups discovered through deep learning, including correspondences across different languages ​​and expressions. The AI ​​extension library is a dynamically updated knowledge base that continuously accumulates new expressions emerging from projects through machine learning.

[0061] The first stage involves structured data transformation: The system first retrieves a set of regular expression patterns related to the current project from the compensation lexicon. For lexical sequences that have not completed standard term mapping, different patterns are attempted for matching in turn. Upon successful matching, named capture groups are used to extract key information, such as professional type, system classification, and spatial location. For example, when processing "MEP-FP-2023A", the pattern "(?<profession>MEP)-(?<system>FP)-(?<year>\d{4})(?<region>[AZ])" is used for parsing to extract complete structured attributes. If the first match fails, the system will try pattern variants, such as ignoring delimiters or allowing case differences. All extracted structured data is accompanied by a confidence score for subsequent result filtering.

[0062] The second stage involves AI-assisted matching: For word sequences that cannot be parsed using regular expressions, the system calls the BERT model for semantic encoding. A bidirectional Transformer structure is used to extract contextual semantic features of the words, generating a 768-dimensional feature vector. Then, a nearest neighbor search is performed in the vector space of the AI ​​extension library to find semantically similar terms. To improve matching accuracy, the system simultaneously calculates multiple similarity metrics: semantic similarity is calculated using cosine distance, morphological similarity is calculated using weighted edit distance, and contextual similarity is calculated by combining project background information. When determining the final synonym matching results, a dynamic weighting scheme is used: the weights of various metrics are adjusted according to the project stage; for example, semantic accuracy is emphasized in the design stage, while morphological tolerance is increased in the construction stage. For newly discovered valid matching pairs, the system adds them to the synonym dictionary and updates the corresponding feature vectors, achieving continuous knowledge accumulation.

[0063] The entire supplementary matching process employs a pipelined architecture to ensure processing efficiency. In the regular expression matching stage, the system attempts patterns sequentially from simple to complex, quickly filtering out obvious mismatches. In the AI ​​matching stage, vector indexing technology accelerates similarity calculation and supports real-time queries of a large-scale dictionary. For the matching results at each stage, the system records detailed process information, including the rules used, the calculated scores, and the basis for the decisions, facilitating subsequent result verification and optimization. This multi-layered supplementary matching mechanism effectively improves the ability to handle non-standard identifiers, particularly supporting project-specific naming rules and novel expression methods.

[0064] S105. Combine the standard terminology mapping results, structured data, and synonym matching results to obtain the component identification description.

[0065] The component identifier description represents standardized descriptive information obtained by integrating multiple matching and transformation results. The standard term mapping result refers to the standard term correspondence calculated through semantic and morphological similarity. Structured data is used to represent formatted information after parsing with regular expressions. Synonym matching results represent semantically similar terms obtained through an AI extension library. The combination process refers to the process of integrating matching results from multiple sources according to predetermined rules, including result merging, conflict resolution, and format unification.

[0066] After completing all matching and transformation processes, results from different sources need to be integrated into a unified standard description. Specifically, firstly, the credibility of the standard term mapping results is evaluated, and the adoption of the mapping result is determined based on the similarity score; then, the encoded information in the structured data is associated with the standard terms, establishing a correspondence between attributes such as position and type and standard terms; next, synonym matching results are added to the description as supplementary information to expand the semantic coverage of the terms; finally, all information is organized into standardized descriptive text according to a predetermined description template, and a structured matching report containing original identifiers, standard terms, similarity scores, etc., is generated.

[0067] In some embodiments, the combination of matching results can be achieved in several ways: Optionally, a rule-based combination method can be used, which first defines priority rules for different types of results, then filters valid matching results based on similarity thresholds, organizes information according to a preset description template, and finally generates a detailed report containing the matching process and score; Optionally, a graph database can be used for result integration, which first constructs various matching results as nodes and relationships, then establishes a mapping network between identifiers in the graph database, and finally detects and processes semantic conflicts through a graph traversal algorithm. It is understood that other methods can also be used to combine matching results, such as adaptive combination methods based on machine learning, which are not limited here.

[0068] The following provides a more detailed description of the process of the method provided in this implementation. Please refer to [link / reference]. Figure 2 This is another flowchart illustrating the component identifier matching method based on NLP semantic segmentation and multi-level lexicon in this application embodiment.

[0069] S201. Obtain the engineering stage information of the current application scenario, which is the design stage, construction stage, or operation and maintenance stage.

[0070] Engineering phase information refers to the different stages in the life cycle of a building project, including the design phase, construction phase, and operation and maintenance phase. The design phase involves drawing design and model building, with high requirements for the standardization of component identification; the construction phase includes on-site construction and quality acceptance, requiring handling of identification changes during construction; the operation and maintenance phase is responsible for equipment maintenance and asset management, focusing on the correspondence between equipment codes and physical objects.

[0071] Before component identification matching begins, the system retrieves current project stage information by reading the stage identifier field from the project management database. For the design stage, it extracts the progress status of design tasks from the design management system; for the construction stage, it obtains construction progress information from the construction management system; and for the operation and maintenance stage, it reads the operation and maintenance task status from the equipment management system. Based on the obtained stage information, the system adjusts subsequent matching strategies and processing logic accordingly. For example, the design stage primarily processes component identifiers in CAD drawings and BIM models, the construction stage requires additional processing of identifier information in construction logs and quality inspection records, and the operation and maintenance stage focuses on the coding information in equipment ledgers and maintenance records.

[0072] S202. Set the first weight coefficient of semantic similarity and the second weight coefficient of morphological similarity based on the information of the engineering stage.

[0073] Weighting coefficients are parameters used to assess the importance of different similarity indicators when calculating overall similarity. The first weighting coefficient adjusts the influence of semantic similarity, while the second weighting coefficient controls the contribution of morphological similarity. The sum of the two weighting coefficients is 1, reflecting the emphasis of different matching strategies.

[0074] The system sets corresponding weight coefficients based on the characteristics of each engineering phase. During the design phase, to strictly adhere to design specifications, a higher semantic similarity weight (first weight coefficient of 0.7) and a lower morphological similarity weight (second weight coefficient of 0.3) are set. During the construction phase, considering the diversity of identification records during on-site construction, the semantic similarity weight is reduced (first weight coefficient of 0.3) while the morphological similarity weight is increased (second weight coefficient of 0.7). During the operation and maintenance phase, the weights of both similarities are set equally (both 0.5) to balance semantic accuracy and morphological matching. The weight coefficients are set based on baseline values ​​determined by analyzing the feature distribution of historical matching data, and fine-tuned using parameter optimization methods in practical applications.

[0075] S203. Based on the first and second weight coefficients, the semantic similarity and morphological similarity are dynamically weighted to obtain the comprehensive similarity.

[0076] Overall similarity refers to the overall matching degree evaluation index obtained through weighted calculation, with a value ranging from 0 to 1. The higher the value, the higher the matching degree. Dynamic weighting refers to the process of weighting and combining different similarity indicators according to weight coefficients.

[0077] The system uses a linear weighted approach to calculate the overall similarity. First, semantic similarity and morphological similarity are normalized to ensure consistent value ranges for both indicators. Then, the normalized semantic similarity is multiplied by a first weighting coefficient, and the morphological similarity is multiplied by a second weighting coefficient. Finally, the two weighted results are summed to obtain the overall similarity. The formula is: Overall Similarity = Semantic Similarity × First Weighting Coefficient + Morphological Similarity × Second Weighting Coefficient. For example, if a component identifier has a semantic similarity of 0.8 and a morphological similarity of 0.6, the overall similarity calculated during the design phase (first weighting coefficient 0.7, second weighting coefficient 0.3) is: 0.8 × 0.7 + 0.6 × 0.3 = 0.74. Based on the calculated overall similarity, the final matching result is determined.

[0078] S204. Map standard terms based on comprehensive similarity.

[0079] Standard term mapping refers to the process of matching component identifiers to corresponding terms in a standard thesaurus. The comprehensive similarity threshold is a standard value used to determine whether a match is valid, typically set between 0.6 and 0.8. The mapping result includes information such as the matched standard terms, similarity score, and mapping relationship type.

[0080] After obtaining the overall similarity score, the system performs a standard term mapping operation. First, candidate terms in the thesaurus are sorted in descending order of overall similarity, and the term with the highest similarity is selected as the primary matching result. Other candidate terms with overall similarity scores exceeding a threshold are saved as alternative matching results. When multiple candidate terms with similar scores exist, they are filtered by analyzing their usage frequency and contextual relationships. For example, the component identifier "AHU-01" has an overall similarity of 0.85 with "AirHandling Unit" in the thesaurus, and an overall similarity of 0.82 with "Air Heater Unit". The system selects "Air Handling Unit" with the higher overall similarity score as the standard mapping result, while simultaneously recording "Air Heater Unit" as an alternative mapping.

[0081] S205. Generate a structured matching report describing the component identifiers, which includes standard terms, original identifiers, matching rules, and similarity.

[0082] A structured matching report is a data document that records the matching process and results in a standard format. Standard terminology refers to standardized professional terms. The original identifier refers to the identifier of the input component to be matched. The matching rules record the processing methods and parameters used. Similarity includes specific numerical values ​​for semantic similarity, morphological similarity, and overall similarity.

[0083] The system generates matching result documents according to a predefined report template. The report content is organized in XML format and contains multiple matching record nodes. Each node includes: an original identifier field recording the input component identifier text; a standard terminology field storing the mapped standard terms; a matching rule field describing the processing method used (e.g., "semantic segmentation + similarity calculation"); and a similarity field recording various similarity values. The report also includes the matching timestamp, processing status, and verification information. For example, for the input identifier "FP-AHU-1501", the generated matching record includes: standard terminology "Fire Protection Air Handling Unit", original identifier "FP-AHU-1501", matching rule "lexical segmentation + semantic mapping", semantic similarity 0.85, morphological similarity 0.75, and overall similarity 0.82.

[0084] S206. Import the component identification description into the graph database, and build a mapping network between standard terms in the graph database. This mapping network is used to record the correspondence between different identifiers.

[0085] A graph database is a database system that uses a graph structure to store and manage data, where nodes represent entities and edges represent relationships. A standard terminology mapping network is a terminology association structure built within a graph database, where nodes store terminology information and edges store mapping relationships. The mapping network visually displays the correspondences and relationships between different identifiers in the form of a graph.

[0086] The system imports the matching results into a graph database for persistent storage and relationship management. First, standard term nodes are created, with attributes including term ID, name, and language type. Then, a corresponding node is created for each original identifier, recording the identifier text and source information. Next, mapping edges are established between nodes, with attributes including mapping type, similarity score, and timestamp. For example, the identifiers "FCU-01," "Fan Coil Unit-01," and "FanCoil Unit #1" for a certain refrigeration equipment in different systems are represented as three identifier nodes in the graph database. These nodes are connected to the node representing the standard term "Fan Coil Unit" through edges of the "equivalent mapping" type, with edge attributes recording their respective mapping similarities. The graph database's query function enables rapid retrieval and visualization of identifier mapping relationships.

[0087] S207. Semantic conflicts are detected based on the mapping network. These semantic conflicts include the same identifier corresponding to multiple standard terms and multiple identifiers corresponding to the same standard term.

[0088] Semantic conflict refers to ambiguity or contradiction in identifier mapping relationships, mainly including two types: one-to-many conflict, where a single original identifier is mapped to multiple different standard terms; and many-to-one conflict, where multiple different original identifiers are mapped to the same standard term. Conflicts in mapping networks are identified by analyzing the connection relationships between nodes.

[0089] The system performs conflict detection queries on the graph database. For one-to-many conflicts, it retrieves the original identifier nodes with a degree greater than 1, analyzes the multiple standard term nodes connected to them, and determines the rationality of the mapping relationship. For many-to-one conflicts, it retrieves the standard term nodes with an in-degree greater than 1 and evaluates whether the multiple original identifier nodes connected to them truly express the same meaning. For example, it finds that the identifier "FCU-01" maps to both the standard terms "Fan Coil Unit" and "Fresh Air Conditioning Unit," constituting a one-to-many conflict; the identifiers "AHU-01" and "Air-Handler-01" map to "Air HandlingUnit," constituting a many-to-one conflict. The system records detailed information for each conflict, including the nodes involved, edge attributes, and conflict type.

[0090] S208. When a semantic conflict is detected, a conflict detection report is generated based on the structured matching report and a manual review process is triggered to obtain the manual review result.

[0091] A conflict detection report is a structured document that records information about detected semantic conflicts. The manual review process refers to the standard operating procedure by which professionals review and process conflicts. The results of the manual review include decisions on conflict resolution and recommendations for correction.

[0092] The system generates detailed detection reports based on detected conflicts. The reports are organized in JSON format, with each conflict record containing fields such as conflict type, involved identifiers and terms, similarity data, and conflict description. Information from the original matching report is also added to the report to facilitate tracing the cause of the conflict. The system triggers manual review tasks via a workflow engine, assigning conflict reports to the appropriate professionals. Reviewers view conflict details through a dedicated review interface and make decisions for each conflict: confirming a mapping is correct, marking incorrect mappings, adding disambiguation rules, etc. For example, for a one-to-many conflict of "FCU-01", the review confirms "Fan Coil Unit" as the correct mapping and marks "FreshAir Conditioning Unit" as an incorrect mapping.

[0093] S209. Add the results of manual review as new training samples to the training dataset and update the standard main vocabulary.

[0094] A training dataset refers to a collection of labeled data used to train a machine learning model. Training samples contain original labels, correct standard term mappings, and relevant feature information. Standard main lexicon updates refer to the process of adjusting and expanding the lexicon content based on newly added training samples.

[0095] The system converts the manually verified mapping results into training samples. Each training sample contains: original labeled text, standard terms, word segmentation results, feature vectors, mapping rules, and other information. For one-to-many conflict processing results, the system records the discriminative features of correct and incorrect mappings; for many-to-one conflicts, the system extracts common features from different expression forms. The system uses the newly added training samples to update the standard main lexicon, including: adding new standard term entries, updating the synonym list of terms, adjusting the relationships between terms, and supplementing disambiguation rules. For example, based on the processing results of the ambiguity of "FCU", the system clearly distinguishes the usage scenarios of "Fan Coil Unit" and "Fresh Air Conditioning Unit" in the lexicon and adds corresponding feature rules.

[0096] S210. When a new identifier is detected in the monitoring design change order, extract the feature pattern of the new identifier.

[0097] A design change order is a formal document that records modifications to the engineering design. New identifiers refer to new component coding formats that emerge during the design change process. Characteristic patterns refer to the rules and structural features of the characters in an identifier, including pattern characteristics such as the use of separators, numbering rules, and abbreviation methods.

[0098] The system processes design change orders in real time through a document parsing module. First, the change order undergoes text preprocessing to extract component identification information. Then, the extracted identifiers are compared with an existing thesaurus to identify newly added identifiers not yet included. Next, the structural characteristics of the new identifiers are analyzed, including: segmenting the identifier into basic units, statistically analyzing character type distribution (such as the usage of letters, numbers, and special symbols), and extracting recurring pattern structures. For example, in a design change order, a new photovoltaic component identifier "BIPV-P01-A15" appears. The system identifies its characteristic pattern as: "BIPV" (fixed prefix) - "P" (type code) + "two digits" (serial number) - "A" (region code) + "two digits" (location number).

[0099] S211. Generate regular matching rules based on feature patterns, and inject the newly added identifier text information into the pre-trained model for incremental training.

[0100] Regular expression matching rules are rule expressions used to identify and parse identifiers with specific formats. A pre-trained model is a basic language model trained on large-scale data. Incremental training refers to the process of updating a model using new data without affecting the performance of the original model.

[0101] The system constructs matching rules based on extracted feature patterns. First, the feature patterns are converted into regular expressions to define the matching patterns for each component, such as "BIPV-P\d{2}-[AZ]\d{2}" used to match the aforementioned photovoltaic component identifiers. Then, the text information of the newly added identifiers is converted into training data format, including the identifier text, word segmentation results, and semantic annotations. Next, the processed training data is input into the pre-trained model for incremental learning, updating the model's word vector space and semantic understanding capabilities. Through iterative training, the model gradually adapts to the newly added identifier forms and semantic relationships.

[0102] S212. Record version change information for regular expression matching rules and the standard main vocabulary library.

[0103] Version change information refers to log data that records the update history of the thesaurus and rules. Regular expression matching rule versions include information such as rule creation time, scope of application, and modification history. Standard main thesaurus versions record the addition, deletion, and modification operations of terms, along with timestamps.

[0104] The system employs a version control mechanism to manage rule and thesaurus updates. For regular expression matching rules, it records information such as the version number, effective date, rule content, applicable projects, and verification status for each rule. For the standard master thesaurus, it records changes to entries with each update, including newly added terms, modified mapping relationships, and deleted obsolete entries. For example, a version record is created for a newly added photovoltaic component identification rule: version number V1.2, effective date 2023-10-01, rule content "BIPV-P\d{2}-[AZ]\d{2}", applicable to photovoltaic projects, and status "verified". Simultaneously, corresponding term update information is added to the thesaurus version record: a new "BIPV" (Building Integrated Photovoltaics) entry, updated on 2023-10-01, sourced from design change order DC-2023-156.

[0105] The following describes the process of the method provided in this implementation. Please refer to [link / reference]. Figure 3 This is a temporal interaction diagram of the component identifier matching method based on NLP semantic segmentation and multi-level lexicon in the embodiments of this application.

[0106] Figure 3The message interaction process between the various functional modules of the system is illustrated in the form of a sequence diagram. When the user inputs the component identifier "HVAC-FA-15F", the system processes it sequentially from left to right through the input interface, semantic segmentation module, standard main lexicon, compensation lexicon, AI extension library, result generator, and dynamic update module.

[0107] Specifically, the semantic segmentation module uses the BiLSTM-CRF sequence labeling model to segment the input "HVAC-FA-15F", outputting "HVAC", "FA", and "15F". The system queries to obtain the mappings for "HVAC=Heating, Ventilation and Air Conditioning" and "FA=Fire Alarm", and extracts the information for "15F=15 layers". The standard main lexicon is used for word matching, calculating semantic similarity (0.93) and morphological similarity (0.85), and a comprehensive similarity is obtained through a weighted ratio (0.7×0.93+0.3×0.85=0.89). The result generator integrates the matching information into a JSON format output. For the newly added identifier "BIPV-01", the dynamic update module records it as a new term for photovoltaic systems and updates the lexicon.

[0108] By constructing a three-level collaborative thesaurus system (standard main thesaurus, project compensation thesaurus, and AI dynamic expansion thesaurus) and a scenario-adaptive dynamic weight algorithm, the accuracy and efficiency of BIM component identification matching are significantly improved. The technical advantages are: 1) A pioneering hierarchical semantic processing architecture, using a BiLSTM-CRF neural network to achieve precise segmentation of composite identifiers (e.g., "HVAC-FA-15F" is resolved to HVAC and fire protection system), combined with a hybrid similarity calculation model (cosine similarity 60% + edit distance 40%), achieving a cross-language scenario matching accuracy of 98.3%, a 16.2% improvement over traditional methods; 2) The dynamic weight engine supports intelligent parameter tuning across multiple stages of design, construction, and operation and maintenance. The design stage emphasizes precise matching (α=0.7), while the construction stage strengthens fuzzy error tolerance (β=0.3), ensuring a full-cycle matching accuracy fluctuation rate of <5%; 3) The AI ​​synonym thesaurus, based on the BERT model, achieves dynamic terminology updates within 4 hours.

[0109] The component identification matching system in the embodiments of this invention is described below from the perspective of hardware processing. Please refer to [link / reference]. Figure 4 This is a schematic diagram of a physical device structure of the component identification matching system in the embodiments of this application.

[0110] It should be noted that, Figure 4 The structure of the component identification matching system shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of the present invention.

[0111] like Figure 4As shown, the component identification matching system includes a Central Processing Unit (CPU) 401, which can perform various appropriate actions and processes based on a program stored in Read-Only Memory (ROM) 402 or a program loaded from storage section 408 into Random Access Memory (RAM) 403, such as performing the methods described in the above embodiments. The RAM 403 also stores various programs and data required for system operation. The CPU 401, ROM 402, and RAM 403 are interconnected via a bus 404. An Input / Output (I / O) interface 405 is also connected to the bus 404.

[0112] The following components are connected to I / O interface 405: input section 406 including audio input devices, push-button switches, etc.; output section 407 including a liquid crystal display (LCD) and audio output devices, indicator lights, etc.; storage section 408 including a hard disk, etc.; and communication section 409 including a network interface card such as a LAN (Local Area Network) card, modem, etc. Communication section 409 performs communication processing via a network such as the Internet. Drive 410 is also connected to I / O interface 405 as needed. Removable media 411, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., are installed on drive 410 as needed so that computer programs read from them can be installed into storage section 408 as needed.

[0113] In particular, according to embodiments of the present invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of the present invention include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing computer programs for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via communication section 409, and / or installed from removable medium 411. When the computer program is executed by central processing unit (CPU) 401, it performs the various functions defined in the present invention.

[0114] It should be noted that specific examples of computer-readable storage media may include, but are not limited to: electrical connections having one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM), flash memory, optical fiber, portable compact disc read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this invention, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0115] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present invention. Each block in a flowchart or block diagram may represent a module, program segment, or portion of code, which contains one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those shown in the drawings.

[0116] Specifically, the component identifier matching system of this embodiment includes a processor and a memory. The memory stores a computer program. When the computer program is executed by the processor, it implements the component identifier matching method based on NLP semantic segmentation and multi-level lexicon provided in the above embodiment.

[0117] In another aspect, the present invention also provides a computer-readable storage medium, which may be included in the component identifier matching system described in the above embodiments; or it may exist independently and not assembled into the component identifier matching system. The storage medium carries one or more computer programs, which, when executed by a processor of the component identifier matching system, cause the component identifier matching system to implement the component identifier matching method based on NLP semantic segmentation and a multi-level lexicon provided in the above embodiments.

[0118] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit it. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application.

[0119] As used in the above embodiments, depending on the context, the term "when..." can be interpreted as meaning "if...", "after...", "in response to determining...", or "in response to detecting...". Similarly, depending on the context, the phrase "when determining..." or "if (the stated condition or event) is interpreted as meaning "if determining...", "in response to determining...", "when (the stated condition or event) is detected", or "in response to detecting (the stated condition or event)".

[0120] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. This program can be stored in a computer-readable storage medium, and when executed, it can include the processes described in the above method embodiments. The aforementioned storage medium includes various media capable of storing program code, such as ROM or random access memory (RAM), magnetic disks, or optical disks.

Claims

1. A component identifier matching method based on NLP semantic segmentation and a multi-level thesaurus, characterized in that, The method, applied to a component identification matching system, includes: Establish a standard main vocabulary and a preset compensation vocabulary, wherein the compensation vocabulary stores the item naming rules and the regular expression patterns corresponding to the items; The input component identifiers are semantically segmented to obtain a segmented word sequence, which includes a standardized term part and an item number part; Calculate the semantic similarity and morphological similarity between the segmented word sequence and the entries in the standard main lexicon, and perform standard term mapping based on the semantic similarity and the morphological similarity to obtain the standard term mapping result; The segmented word sequences that have not completed the standard term mapping are subjected to structured data transformation using the regular expression pattern to obtain structured data. The segmented word sequences that have not completed the structured data transformation are then matched using an AI extended library to obtain synonym matching results. The component identifier description is obtained by combining the standard terminology mapping result, the structured data, and the synonym matching result.

2. The method according to claim 1, characterized in that, The step of establishing a standard main vocabulary and a preset compensation vocabulary, wherein the compensation vocabulary stores item naming rules and regular expression patterns corresponding to the items, specifically includes: The standard main thesaurus is constructed based on industry standards and specifications. The standard main thesaurus includes a general terminology table, a professional abbreviation table, and a standardized naming table. The compensation lexicon is constructed based on project naming rules, and the compensation lexicon includes a project naming rule mapping table and a regular expression pattern table. The AI ​​extension library is built based on a deep learning model, and the AI ​​extension library includes a thesaurus and a multilingual mapping table.

3. The method according to claim 1, characterized in that, The steps of calculating the semantic and morphological similarity between the segmented word sequence and the entries in the standard main lexicon, and performing standard term mapping based on the semantic and morphological similarity to obtain the standard term mapping result, specifically include: The semantic similarity is obtained by calculating the cosine similarity between the segmented word sequence and the words in the standard main lexicon in the word vector space; The weighted edit distance between the segmented word sequence and the entries in the standard main lexicon is calculated to obtain the morphological similarity. A comprehensive similarity is calculated based on the semantic similarity and the morphological similarity. The optimal matching standard term is selected based on the comprehensive similarity to obtain the standard term mapping result.

4. The method according to claim 1, characterized in that, The steps of performing structured data transformation on the segmented word sequences that have not completed the standard term mapping using the regular expression pattern to obtain structured data, and then performing AI extended library matching on the segmented word sequences that have not completed the structured data transformation to obtain synonym matching results, specifically include: The structured data is obtained by using the regular expression pattern to transform the segmented word sequence that has not been mapped. The segmented word sequence that has not completed the structured data transformation is input into the AI ​​extended library to obtain the semantic feature vector of the segmented word sequence; Calculate the semantic similarity and morphological similarity between the semantic feature vector and the thesaurus in the AI ​​extended library; Synonym matching is performed based on the semantic similarity and the morphological similarity.

5. The method according to claim 1, characterized in that, After the step of calculating the semantic and morphological similarity between the segmented word sequence and the entries in the standard main lexicon, the method further includes: Obtain the engineering phase information of the current application scenario, wherein the current application scenario is the design phase, construction phase, or operation and maintenance phase; The first weight coefficient of the semantic similarity and the second weight coefficient of the morphological similarity are set according to the project stage information; The semantic similarity and the morphological similarity are dynamically weighted according to the first weight coefficient and the second weight coefficient to obtain the comprehensive similarity. Standard terminology mapping is performed based on the comprehensive similarity.

6. The method according to any one of claims 1-4, characterized in that, After the step of combining the standard terminology mapping result, the structured data, and the synonym matching result to obtain the component identifier description, the method further includes: Generate a structured matching report describing the component identifiers, the structured matching report including standard terms, original identifiers, matching rules, and similarity; The component identification description is imported into a graph database, and a mapping network between standard terms is constructed in the graph database. The mapping network is used to record the correspondence between different identifiers. Semantic conflicts are detected based on the mapping network, including the same identifier corresponding to multiple standard terms and multiple identifiers corresponding to the same standard term. When a semantic conflict is detected, a conflict detection report is generated based on the structured matching report, and a manual review process is triggered to obtain the manual review result. The manual review results are added to the training dataset as new training samples, and the standard main vocabulary is updated.

7. The method according to any one of claims 1-4, characterized in that, After the step of combining the standard terminology mapping result, the structured data, and the synonym matching result to obtain the component identifier description, the method further includes: The system monitors newly added identifiers in design change orders. When a newly added identifier is detected, the feature pattern of the newly added identifier is extracted. Regular matching rules are generated based on the feature patterns, and the text information of the newly added identifiers is injected into the pre-trained model for incremental training. Record the version change information of the regular expression matching rules and the standard main vocabulary library.

8. A component identification matching system, characterized in that, The component identification matching system includes: one or more processors and a memory; the memory is coupled to the one or more processors, the memory is used to store computer program code, the computer program code including computer instructions, and the one or more processors call the computer instructions to cause the component identification matching system to perform the method as described in any one of claims 1-7.

9. A computer-readable storage medium comprising instructions, characterized in that, When the instruction is executed on the component identifier matching system, the component identifier matching system performs the method as described in any one of claims 1-7.

10. A computer program product, characterized in that, When the computer program product is run on the component identification matching system, the component identification matching system performs the method as described in any one of claims 1-7.

Images