Construction technology knowledge management method and system based on knowledge base and large language model

Abstract

The invention relates to the technical field of knowledge management, in particular to a construction technology knowledge management method and system based on a knowledge base and a large language model.The method comprises the steps that original construction technology documents are collected and preprocessed to obtain construction technology documents, then the construction technology documents are stored, and a knowledge base of the construction technology field is constructed; based on a natural language processing technology, identifying entities and semantic relationships among the entities from the construction technology document, and constructing a knowledge graph of the construction technology field; generating a construction technology question and answer model based on a pre-trained language model, a knowledge base and a knowledge graph; on the basis of a search engine, querying a construction technology document most related to query content input by a user from a knowledge base; and receiving query content input by a user through the interactive operation page, and displaying a query result or displaying a construction technology question and answer result. According to the invention, intelligent integration, dynamic updating and accurate service of knowledge can be realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of knowledge management, in particular to a construction technology knowledge management method and system based on a knowledge base and a large language model. BACKGROUND

[0002] In the field of engineering construction, efficient management and application of construction technology knowledge is the key to ensuring smooth implementation of projects and improving engineering quality and safety. Traditional knowledge management systems in the field of construction technology rely on scattered document storage and basic retrieval functions, which have significant limitations. On the one hand, construction technology knowledge such as construction specifications, acceptance standards, and operation instruction books is usually stored in unstructured or semi-structured form, resulting in a serious information island phenomenon and low retrieval efficiency. On the other hand, existing knowledge management systems mostly use general databases or simple classification management, lack dynamic updating mechanisms and semantic understanding capabilities, and are difficult to adapt to rapidly changing industry needs.

[0003] In the context of engineering construction, the management of technical knowledge needs to handle massive specifications, cases, and multi-modal data (such as design drawings and process manuals), and has high requirements for real-time and accuracy. Existing construction technology knowledge management systems lack updating mechanisms, have insufficient intelligence, lack personalized services, and cannot provide timely, effective, and reliable technical guidance during the construction process. SUMMARY

[0004] To address the deficiencies in the prior art, the present application provides a construction technology knowledge management method and system based on a knowledge base and a large language model, which realizes intelligent integration, dynamic updating, and precise service of knowledge by constructing a structured knowledge graph, integrating a large language model and domain knowledge.

[0005] To solve the above technical problems, the present application solves them through the following technical solutions:

[0006] A construction technology knowledge management method based on a knowledge base and a large language model, comprising: collecting original construction technology documents, preprocessing them to obtain construction technology documents, and then storing them and constructing a knowledge base in the field of construction technology; based on natural language processing technology, identifying entities and semantic relationships between entities from the construction technology documents, and constructing a knowledge graph in the field of construction technology; based on a pre-trained language model, a knowledge base, and a knowledge graph, generating a construction technology question and answer model, which is used for intelligent answering and generating text for construction technology questions; based on a search engine, querying the most relevant construction technology documents from the knowledge base based on the user input query content; receiving user input query content, displaying query results, or displaying construction technology question and answer results through an interactive operation page.

[0007] In the present application, by constructing core modules such as construction technology knowledge base, knowledge graph, construction technology question and answer model, a complete closed loop covering knowledge collection, processing, storage, retrieval and application is formed, wherein: by collecting and preprocessing the original documents, the dispersed construction technology knowledge is integrated and stored in the knowledge base, realizing the centralized management and structured organization of knowledge, breaking the information island, improving the usability and manageability of knowledge; based on natural language processing technology, the knowledge graph is constructed, the entities and their semantic relationships in the construction technology documents are identified, the knowledge is converted into semantic network form, which is convenient for computer understanding and processing, and lays a foundation for in-depth knowledge query and intelligent question answering; the question and answer model generated by combining the pre-training language model and the knowledge graph can intelligently answer the construction technology problems and generate professional text, providing efficient and accurate technical support for construction personnel, which helps to improve construction efficiency and quality, reduce errors and delays; through the combination of search engine and knowledge base, the rapid query and accurate positioning of construction technology documents are realized, the knowledge retrieval efficiency is improved, and the construction personnel can obtain the required information more timely to support decision-making and operation.

[0008] As preferred, wherein the original construction technology documents are collected, preprocessed to obtain construction technology documents, then stored, and the knowledge base in the construction technology field is constructed, including: collecting original construction technology documents from multiple source heterogeneous data through an automatic data collection program; adding metadata tags to each data, including creator, timestamp and version number; preprocessing the collected construction technology documents, including data cleaning and deduplication, to obtain processed construction technology documents; storing the construction technology documents based on preset knowledge coding rules and naming formats.

[0009] Through the present application, the automatic data collection tool is used to real-time capture data from enterprise internal document library, standard release platform and other channels, reducing the tediousness and errors of manual collection, improving the speed and coverage of data collection, and ensuring that the knowledge base can obtain the latest technical materials in time; the collected data is classified, cleaned, deduplicated and labeled, removing repeated and low-quality information, ensuring the accuracy and reliability of the data in the knowledge base, and providing a high-quality data basis for subsequent knowledge processing and application; metadata tags such as creator, timestamp and version number are added to each data, increasing the traceability and manageability of the data, realizing the whole life cycle management of the documents, facilitating the version control and historical information tracking of the knowledge, and being helpful to the long-term maintenance and update of the knowledge; a unified knowledge coding rule (such as "professional-subfield-document type-version") is adopted, so that the data in the knowledge base has a unified structure and format, which is conducive to the rapid retrieval, sharing and integration of knowledge, and improves the overall performance and scalability of the knowledge base.

[0010] Preferably, the method of identifying entities and semantic relationships between entities from construction technical documents based on natural language processing technology and constructing a knowledge graph in the field of construction technology includes: identifying entities and semantic relationships between entities from construction technical documents based on natural language processing technology; manually verifying and dynamically correcting the identified entities and semantic relationships between entities through expert review; and constructing a knowledge graph in the field of construction technology based on the semantic relationships between entities.

[0011] Through the present invention, entities (such as "rebar" and "concrete") and semantic relationships (such as "binding" and "acceptance") are automatically identified from construction technical documents, key information and knowledge points in the documents are accurately extracted, an entity-relationship network is constructed, and unstructured text is converted into computable knowledge units; a knowledge graph is constructed based on the identified entities and semantic relationships to form a knowledge graph covering the field of construction technology, clearly presenting the associations between knowledge points (such as "construction technology → required materials → quality standards"), and providing structured data support for intelligent question-answering, search and other functions; an expert review mechanism is introduced to manually verify and dynamically correct the recognition results (such as calibrating the accurate pointing of professional terms), thereby ensuring the accuracy and authority of the knowledge graph, avoiding errors or deviations caused by the limitations of natural language processing technology, and improving the quality and credibility of the knowledge graph.

[0012] Preferably, the identifying entities and semantic relationships between entities from construction technical documents based on natural language processing technology includes:

[0013] Based on the entity recognition algorithm, entities are identified from the construction technical documents. Where P(e|ω) is the probability of entity e given word ω, and P(e) is the prior probability of entity e;

[0014] Based on the relation extraction algorithm, the semantic relationship between entities is identified. Among them, P(r|e1,e2) is the probability of relation r given entities e1 and e2, P(e1,e2|r) is the probability that relation r generates entity pair P(e1,e2), and P(r) is the prior probability of relation r.

[0015] This invention uses entity recognition and relationship extraction algorithms to accurately extract entities and semantic relationships from construction technical documents. The application of these algorithms improves the efficiency and accuracy of knowledge graph construction, enabling rapid identification of key information and key knowledge points within documents. This provides rich material for knowledge graph construction, helps to deeply explore the knowledge content within construction technical documents, and enriches the content and semantic hierarchy of the knowledge graph.

[0016] As preferred, wherein the pre-trained language model, knowledge base and knowledge graph are used to generate a construction technology question and answer model for intelligent answering and generating text for construction technology questions, comprising: fine-tuning the pre-trained language model based on the construction technology documents in the knowledge base to obtain a fine-tuned language model; based on the semantic relationship of the knowledge graph, the fine-tuned language model is used to intelligently answer the input construction technology question and generate a professional answer text.

[0017] Through the present application, the fine-tuning process adapts the language model to the specific needs of the construction technology field, improving the model's understanding of the problem and the accuracy of the answer, making it suitable for the professional terminology and business logic of the construction scene, such as understanding the meaning of parameters such as "concrete slump" and "tension stress", using the semantic relationship of the knowledge graph, such as "bridge type → design specification → construction process", to generate answers containing multi-dimensional information, such as simultaneously associating specification articles with actual cases, improving the comprehensiveness of the answer, and generating professional text related to construction technology, including construction plan summaries and technical problem analysis, to meet the actual application needs of the engineering site.

[0018] As preferred, wherein the pre-trained language model is fine-tuned based on the construction technology documents in the knowledge base to obtain a fine-tuned language model, comprising: extracting construction technology documents from the knowledge base as training samples, and fine-tuning the loss function Optimizing the parameters of the language model to obtain a language model adapted to the construction scene, wherein L(θ) is the loss function, y i is the true label; x i is the input text, and θ is the model parameter.

[0019] Through the present application, the fine-tuning loss function is used to optimize the parameters of the language model, and the semantic understanding of construction terminology is strengthened during the model fine-tuning process, so that the fine-tuned language model can generate answers that meet industry standards for construction technology questions, thereby more accurately providing technical support for construction personnel and meeting actual work needs.

[0020] As preferred, wherein the search engine is used to query the most relevant construction technology documents from the knowledge base based on the user's input query content, comprising: querying the construction technology documents related to the user's input query content from the knowledge base; based on the vector space model, calculating the semantic similarity between the construction technology documents and the user's input query content; based on the knowledge graph, adjusting the sorting priority of the retrieval results according to the weight of the entity relationship.

[0021] Through the present application, the search engine and knowledge graph are combined to achieve efficient knowledge retrieval, wherein: based on the vector space model, the construction technology documents and the user query content are converted into vector representations, where sim(d, q) is the similarity of document d and query q, and are the vector representations of the document and query respectively, the semantic similarity between the two is calculated to measure the relevance between the document and the query, compared with the traditional keyword matching method, the semantic intent of the user can be better understood and responded to, and the accuracy and recall rate of the retrieval are improved;Using the weight information of the entity relationship in the knowledge graph, the priority of the retrieval result is adjusted, the documents that are more closely related in semantics and more related in knowledge structure are preferentially displayed to the user, which further improves the quality and practicality of the retrieval result, and helps the user to find the required knowledge more quickly;The most relevant construction technology document is quickly screened out from the large-scale knowledge base, which reduces the time and effort of the user in searching for useful content in the massive information, improves the efficiency of knowledge retrieval, and enables the construction personnel to obtain knowledge support in time, ensuring the construction progress;Combined with the large language model, the user can be provided with a more intelligent and accurate retrieval experience, at this time the user does not need to accurately input keywords or phrases, only needs to express the query intent of natural language, the system can return the documents with high relevance based on semantic understanding and knowledge association, which reduces the user threshold and improves the usability and user satisfaction of the system.

[0022] As preferred, the method further comprises personalized pushing according to user historical operation behavior data.

[0023] In the present application, the user's interest field, focus and knowledge demand mode can be understood in depth through the analysis of the user's historical query, browsing, downloading and other operation behavior data, which provides data support and decision basis for personalized pushing;According to the potential demand mined from the user behavior data, combined with the knowledge content in the knowledge base and knowledge graph, the user is accurately recommended the construction technology knowledge with high relevance to his work and possible interest, which improves the probability of the user obtaining useful knowledge and reduces the time cost of the user's active search.

[0024] The application also provides a construction technology knowledge management system based on a knowledge base and a large language model, which is used for realizing the construction technology knowledge management method based on the knowledge base and the large language model, and comprises: a knowledge base module, which is used for collecting construction technology documents from multi-source heterogeneous data through an automatic data collection tool, and performing data preprocessing on the construction technology documents, the preprocessing comprising classification, tagging, deduplication and standardized storage, and constructing a knowledge base of the construction technology field; a knowledge graph module, which is used for identifying entities and semantic relationships between the entities from the construction technology documents based on a natural language processing technology, and constructing a knowledge graph of the construction technology field; a large language model module, which is used for generating a construction technology question and answer model based on a pre-trained language model, the knowledge base and the knowledge graph, the construction technology question and answer model being used for intelligently answering construction technology questions and generating texts; a search module, which is used for querying construction technology documents most relevant to input query content of a user from the knowledge base based on a search engine; and a user interface module, which is used for receiving the input query content of the user through an interactive operation page, displaying query results or displaying construction technology question and answer results.

[0025] Through the cooperative work of the various functional modules, the whole process coverage of the construction technology knowledge management is realized, and comprehensive support is provided for the effective utilization of the construction technology knowledge. The knowledge acquisition, preprocessing and storage are completed through the knowledge base module, a centralized knowledge warehouse is formed, the knowledge association network is constructed through the knowledge graph module, the semantic level knowledge organization is realized, the intelligent question and answer and text generation capabilities are provided through the large language model module, the function of intelligently solving professional problems is realized, the multi-dimensional efficient retrieval is supported through the search module, the target knowledge is quickly located, the friendly interactive interface is provided through the user interface module, the convenient input and visual output of the knowledge are realized, and the use experience of the user is improved.

[0026] As a preferred, the system further comprises a recommendation module, which is used for performing personalized pushing according to historical operation behavior data of the user.

[0027] Through the application, the personalized knowledge pushing service is provided for the user according to the historical operation behavior data of the user, the interactivity between the system and the user is increased through the personalized pushing function, the user can continuously receive new knowledge and new information meeting the own needs in the use process, so that the continuous use and in-depth exploration of the system by the user are stimulated, the user stickiness and activity are improved, meanwhile, the efficient propagation of the knowledge from the system to the user is realized, the flow and sharing of the construction technology knowledge in the industry are promoted, the technical level and the innovation ability of the whole construction field are improved, meanwhile, the value and importance of the system in the heart of the user are enhanced.

[0028] The application solves the problems of information island, low retrieval efficiency and insufficient intelligence in construction technology knowledge management by deep integration of knowledge base and knowledge graph, combined with the semantic understanding ability of large language model, and provides an efficient tool for digital transformation of engineering construction: by constructing the knowledge base and knowledge graph of the construction technology field, the structured management and semantic understanding of construction technology knowledge are realized, and the problems of information island and low retrieval efficiency in traditional construction technology knowledge management system are solved; based on the pre-training language model and the knowledge graph, the construction technology question and answer model is generated, which can intelligently answer the construction technology problems and generate professional answer texts, providing timely and accurate technical guidance for construction personnel and improving construction efficiency and quality; through the combination of search engines and knowledge graphs, semantic retrieval and accurate query of construction technology documents are realized, improving the efficiency and accuracy of knowledge acquisition; personalized push is carried out according to user historical operation behavior data, which can provide a more intelligent and personalized knowledge management experience for users. BRIEF DESCRIPTION OF DRAWINGS

[0029] Figure 1 A construction technology knowledge management method based on a knowledge base and a large language model in the embodiment.

[0030] Figure 2 A module contact diagram of a construction technology knowledge management system based on a knowledge base and a large language model in the embodiment.

[0031] Figure 3 A system architecture diagram of a construction technology knowledge management system based on a knowledge base and a large language model in the embodiment. DETAILED DESCRIPTION

[0032] In order to further understand the content of the application, the application will be described in detail in combination with the embodiments. It should be understood that the embodiments are only used to explain the application but not to limit the application.

[0033] In the field of engineering construction, the traditional technical knowledge management system depends on scattered document storage and basic retrieval function, and has problems of information island, update lag and low intelligence level. The unstructured storage of construction technical documents such as specifications, cases and drawings leads to low retrieval efficiency of related construction knowledge in the construction process, and the existing general question and answer model lacks field adaptability and is difficult to meet the real-time and accuracy requirements of construction scene.

[0034] As Figure 1As shown, the embodiment provides a construction technology knowledge management method based on a knowledge base and a large language model, which includes: collecting original construction technology documents, preprocessing them to obtain construction technology documents, and then storing them and constructing a knowledge base in the construction technology field; based on natural language processing technology, identifying entities and semantic relationships between entities from construction technology documents, and constructing a knowledge graph in the construction technology field; based on a pre-trained language model, a knowledge base and a knowledge graph, a construction technology question and answer model is generated, which is used for intelligent answering and generating text for construction technology questions; based on a search engine, the most relevant construction technology documents are queried from the knowledge base based on the user input query content; the user input query content is received through an interactive operation page, the query result is displayed or the construction technology question and answer result is displayed.

[0035] Through this embodiment, by constructing core modules such as construction technology knowledge base, knowledge graph and construction technology question and answer model, a complete closed loop covering knowledge collection, processing, storage, retrieval and application is formed, wherein: by collecting and preprocessing original documents, the dispersed construction technology knowledge is integrated and stored in the knowledge base, realizing centralized management and structured organization of knowledge, breaking the information island, and improving the usability and manageability of knowledge; based on natural language processing technology, a knowledge graph is constructed to identify entities and their semantic relationships in construction technology documents, and knowledge is converted into a semantic network form, which is convenient for computer understanding and processing, and lays a foundation for in-depth knowledge query and intelligent answering; the question and answer model generated by combining the pre-trained language model and the knowledge graph can intelligently answer construction technology questions and generate professional text, providing efficient and accurate technical support for construction personnel, which helps to improve construction efficiency and quality, reduce errors and delays; through the combination of search engine and knowledge base, fast query and accurate positioning of construction technology documents are realized, the knowledge retrieval efficiency is improved, and the construction personnel can obtain the required information more timely to support decision-making and operation.

[0036] In this embodiment, original construction technology documents are collected, preprocessed to obtain construction technology documents, and then stored and a knowledge base in the construction technology field is constructed, including: collecting original construction technology documents from multiple heterogeneous data sources through an automatic data collection program; adding metadata tags to each piece of data, including creator, timestamp and version number; preprocessing the collected construction technology documents, including data cleaning and deduplication, to obtain processed construction technology documents; storing the construction technology documents based on pre-set knowledge coding rules and naming formats.

[0037] By using Web crawler technology, through writing automated data collection script program, automatically from internal document management system, external database and even the Internet to collect and engineering project closely related literature, regularly catch the latest published regulations and policies and technical reports. For the collection of multi-source heterogeneous data, that is, from different sources, various storage forms of data, through data cleaning and deduplication, eliminate duplicate data, and add detailed metadata tags, such as creator, timestamp, version number, etc., to facilitate tracking historical change track, and accordingly sorting and filtering, then establish a unified knowledge coding rules, covering naming conventions, classification system, label system, etc. For data storage, ensure that each record can be accurately described and identified. Through a series of processing and storage means, the multi-source heterogeneous data is effectively fused, the data consistency is maintained, and the content of the knowledge base is enriched.

[0038] Through this embodiment, the data is collected from the enterprise internal document library, standard release platform and other channels in real time by using the automated data collection tool, which reduces the tediousness and errors of manual collection, improves the speed and coverage of data collection, and ensures that the knowledge base can obtain the latest technical materials in time; The collected data is classified, cleaned, deduplicated and labeled, which removes duplicate and low-quality information, ensures the accuracy and reliability of the data in the knowledge base, and provides a high-quality data foundation for subsequent knowledge processing and application; Each data is attached with metadata tags such as creator, timestamp and version number, which increases the traceability and manageability of the data, realizes the whole life cycle management of the document, facilitates the version control and historical information tracking of the knowledge, and helps the long-term maintenance and update of the knowledge; A unified knowledge coding rule (such as “professional-subfield-document type-version”) is adopted, so that the data in the knowledge base has a unified structure and format, which is beneficial to the quick retrieval, sharing and integration of the knowledge, and improves the overall performance and scalability of the knowledge base.

[0039] In this embodiment, based on natural language processing technology, entities and semantic relationships between entities in construction technology documents are identified, and a knowledge graph of the construction technology field is constructed, including: based on natural language processing technology, entities and semantic relationships between entities in construction technology documents are identified; Through expert review, the identified entities and semantic relationships between entities are manually verified and dynamically corrected; Based on the semantic relationships between entities, a knowledge graph of the construction technology field is constructed.

[0040] In the process of building the knowledge graph, advanced natural language processing (NLP) algorithms are used to automatically identify entities and their relationships in the text of the construction technical document, and then map them to a pre-defined ontology framework to form a comprehensive knowledge map covering the construction field. At the same time, in order to ensure the quality of the graph, an expert review mechanism is introduced to ensure that all information is correct and accurate.

[0041] Through this embodiment, entities (such as "steel" and "concrete") and semantic relationships (such as "binding" and "acceptance") are automatically identified from the construction technical document, and key information and knowledge points in the document are accurately extracted to construct an entity-relationship network and convert unstructured text into computable knowledge units. Based on the identified entities and semantic relationships, a knowledge graph is constructed to form a knowledge graph covering the construction technical field, clearly presenting the associations between knowledge points (such as "construction process → required materials → quality standards"), providing structured data support for intelligent question answering, search and other functions; an expert review mechanism is introduced to manually verify and dynamically correct the recognition results (such as correcting the accurate direction of professional terms), ensuring the accuracy and authority of the knowledge graph, avoiding errors or biases caused by the limitations of natural language processing technology, and improving the quality and credibility of the knowledge graph.

[0042] In this embodiment, based on natural language processing technology, entities and semantic relationships between entities are identified from the construction technical document, including:

[0043] Based on the entity recognition algorithm, entities are identified from the construction technical document, where P(e|ω) is the probability of entity e given the word ω, and P(e) is the prior probability of entity e;

[0044] Based on the relationship extraction algorithm, the semantic relationships between entities are identified, where P(r|e1,e2) is the probability of relationship r given entities e1 and e2, P(e1,e2|r) is the probability of generating entity pair P(e1,e2) for relationship r, and P(r) is the prior probability of relationship r.

[0045] Through this embodiment, entities and semantic relationships are accurately extracted from the construction technical document through the entity recognition algorithm and the relationship extraction algorithm. The application of the algorithm improves the efficiency and accuracy of the construction of the knowledge graph, which can quickly identify key information and knowledge points in the document, provide rich materials for the construction of the knowledge graph, and help to deeply mine the knowledge connotation in the construction technical document, enrich the content and semantic level of the knowledge graph.

[0046] In this embodiment, based on the pre-trained language model, knowledge base and knowledge graph, a construction technology question and answer model is generated, which is used for intelligent answering and generating text for construction technology problems, including: fine-tuning the pre-trained language model based on the construction technology document of the knowledge base to obtain the fine-tuned language model; based on the semantic relationship of the knowledge graph, the input construction technology problem is intelligently answered and professional answer text is generated through the fine-tuned language model.

[0047] Firstly, a large pre-trained language model suitable for a specific industry is selected as the basis, and then the model is familiarized with the specific business scenario through fine-tuning. For example, a pre-trained language model suitable for bridge construction is selected, and construction technology documents related to bridge construction are used as training samples to fine-tune the pre-trained language model to obtain a fine-tuned language model. When answering questions related to bridge construction, the fine-tuned language model can refer to the latest bridge design specifications and safety guidelines to provide professional intelligent answers to the questions.

[0048] Through this embodiment, the fine-tuning process makes the language model adapt to the specific needs of the construction technology field, improves the understanding ability and answer accuracy of the model, and makes it adapt to the professional terms and business logic of the construction scene, such as understanding the meanings of parameters such as "concrete slump" and "tension stress", using the semantic relationship of the knowledge graph such as "bridge type → design specification → construction technology", generating answers containing multi-dimensional information, such as simultaneously associating specification articles and actual cases, improving the comprehensiveness of the answers, and generating professional text related to construction technology, including construction plan abstracts, technical problem analysis, etc., to meet the actual application needs of the engineering site.

[0049] In this embodiment, the pre-trained language model is fine-tuned based on the construction technology document of the knowledge base to obtain a fine-tuned language model, including: extracting construction technology documents from the knowledge base as training samples, and fine-tuning the loss function Optimize the parameters of the language model to obtain a language model adapted to the construction scene, wherein L(θ) is the loss function, y i is the true label; x i is the input text, and θ is the model parameter.

[0050] Through this embodiment, the fine-tuning loss function is used to optimize the parameters of the language model, and the semantic understanding of construction terms is strengthened during the model fine-tuning process, so that the fine-tuned language model can generate answers that meet industry standards for construction technology problems, thereby more accurately providing technical support for construction personnel and meeting actual work needs

[0051] In this embodiment, based on the search engine, the construction technology document most relevant to the user input query content is queried from the knowledge base, including: querying the construction technology document related to the user input query content from the knowledge base; based on the vector space model, the semantic similarity of the construction technology document and the user input query content is calculated; based on the knowledge graph, the sorting priority of the retrieval result is adjusted according to the weight of the entity relationship.

[0052] For example, after the user inputs the query content "bridge concrete maintenance period", the key entities in the query content are extracted through NLP algorithm word segmentation and entity recognition, and the recognized entities are "bridge", "concrete" and "maintenance period"; based on the vector space model, the cosine similarity of the document and the query content is calculated, that is, Based on the associated path of "concrete-maintenance period-temperature control" in the knowledge graph, the sorting priority of the relevant document is adjusted according to the relevance in turn.

[0053] Through this embodiment, the search engine and the knowledge graph are combined to realize efficient knowledge retrieval, wherein: based on the vector space model, the construction technology document and the user query content are converted into vector representation, wherein sim(d,q) is the similarity of the document d and the query q, and respectively, the semantic similarity between the two is calculated to measure the relevance between the document and the query. Compared with the traditional keyword matching method, the semantic intention of the user can be better understood and responded, and the accuracy and recall rate of the retrieval are improved; the weight information of the entity relationship in the knowledge graph is used to adjust the sorting priority of the retrieval result, and the document that is more closely related in semantics and more associated in knowledge structure is preferentially displayed to the user, which further improves the quality and practicality of the retrieval result, helping the user to find the required knowledge faster; the construction technology document most relevant to the user query is quickly filtered from the large-scale knowledge base, reducing the time and effort of the user in searching for useful content in the vast amount of information, improving the efficiency of knowledge retrieval, enabling the construction personnel to obtain knowledge support in time and ensuring the construction progress; combined with the large language model, the user can be provided with a more intelligent and accurate retrieval experience, at this time the user only needs to express the natural language query intention, and the system can return the document with high relevance based on semantic understanding and knowledge association, reducing the user threshold and improving the ease of use and user satisfaction of the system.

[0054] In this embodiment, personalized push is also included according to user historical operation behavior data.

[0055] For example, by recording the user's historical queries, it is detected that the user frequently retrieves "tunnel waterproofing process", and similar cases are recommended through a collaborative filtering algorithm. In addition, cross-domain technical documents are pushed in combination with the semantic path of the knowledge graph, such as "waterproofing process → material selection → construction case".

[0056] Through this embodiment, the user's interest field, focus, and knowledge demand pattern can be understood in depth through analysis of user historical query, browsing, download, and other operation behavior data, providing data support and decision basis for personalized pushing. According to the potential demand mined from user behavior data, in combination with the knowledge content in the knowledge base and the knowledge graph, the user is accurately recommended construction technical knowledge that is highly relevant to his work and may be of interest, improving the probability of the user obtaining useful knowledge and reducing the time cost of the user's active search.

[0057] As shown in Figure 2 The embodiment also provides a construction technical knowledge management system based on a knowledge base and a large language model, for implementing the construction technical knowledge management method based on a knowledge base and a large language model described above, which comprises: a knowledge base module, configured to collect construction technical documents from multiple source heterogeneous data through an automatic data collection tool, and perform data preprocessing on the construction technical documents, the preprocessing comprising classification, tagging, deduplication, and standardized storage, and constructing a knowledge base of the construction technical field; a knowledge graph module, configured to identify entities and semantic relationships between the entities from the construction technical documents based on natural language processing technology, and construct a knowledge graph of the construction technical field; a large language model module, configured to generate a construction technical question and answer model based on a pre-trained language model, the knowledge base, and the knowledge graph, the construction technical question and answer model being used for intelligent answering and text generation of construction technical problems; a search module, configured to query the most relevant construction technical documents from the knowledge base based on a search engine according to the query content input by a user; and a user interface module, configured to receive the query content input by the user, display the query result, or display the construction technical question and answer result through an interactive operation page.

[0058] Through this embodiment, the full-process coverage of construction technical knowledge management is realized through the collaborative work of various functional modules, providing comprehensive support for the effective utilization of construction technical knowledge. The knowledge acquisition, preprocessing, and storage are completed through the knowledge base module, forming a centralized knowledge warehouse; the knowledge association network is constructed through the knowledge graph module, realizing semantic-level knowledge organization; the intelligent question and answer and text generation capabilities are provided through the large language model module, realizing the function of intelligently solving professional problems; the multi-dimensional efficient retrieval is supported through the search module, assisting in quickly locating the target knowledge; the friendly interactive interface is provided through the user interface module, realizing the convenient input and visual output of knowledge, and improving the user experience.

[0059] In this embodiment, the system further includes a recommendation module for personalized pushing based on user historical operation behavior data.

[0060] Through this embodiment, personalized knowledge pushing services are provided for users based on their historical operation behavior data, and the interactivity between the system and the users is increased through the personalized pushing function, so that users can continuously receive new knowledge and information that meet their own needs during use, thereby stimulating users to continuously use and deeply explore the system, improving user stickiness and activity, while achieving efficient dissemination of knowledge from the system to the users, promoting the flow and sharing of construction technology knowledge in the industry, helping to improve the technical level and innovation ability of the entire construction field, and also enhancing the value and importance of the system in the minds of users.

[0061] As Figure 3 described, the construction technology knowledge management system based on the knowledge base and the large language model in this embodiment adopts a hierarchical architecture design, mainly including a user interface layer, a business logic layer, a knowledge base management layer, a service interface layer, a natural language processing layer, a large language model layer, and a data management layer, which are connected through RESTful API, realizing a loose coupling design concept for convenient later maintenance and extension:

[0062] The business logic layer is used to process user requests and call services provided by the service interface layer to realize intelligent question answering, search, and other functions.

[0063] The knowledge base management layer is responsible for the construction, maintenance, and management of the knowledge graph, including entity recognition, relationship extraction, knowledge updating, etc.

[0064] The service interface layer is used to provide RESTful API for the business logic layer to call, realizing interaction with the knowledge base management layer and the large language model layer.

[0065] The natural language processing layer is responsible for parsing user input, understanding user intent, and retrieving relevant information from the knowledge graph.

[0066] The large language model layer is responsible for generating natural language answers, conducting intelligent question answering, and text generation.

[0067] The data management layer is responsible for storing and managing knowledge data, including the knowledge graph, text data, etc.

[0068] The construction technical knowledge management system not only limits to internal resources of the enterprise, but also includes a large amount of external public information, forms a more complete knowledge ecological system; at the same time, with the powerful analysis of the large language model, more personalized services and supports can be provided for the user; in addition, attention is paid to the detail design, and an operation environment which is not only beautiful but also practical is created, and the learning cost of the user is reduced. Through the centralized storage and intelligent management of the system, the knowledge is more easily discovered, spread and applied, so that the possibility of error is reduced, the work efficiency is improved, the informatization process of the whole industry is promoted, and the market competitiveness of the enterprise is enhanced.

[0069] In the embodiment, through the deep integration of the knowledge base and the knowledge graph, combined with the semantic understanding ability of the large language model, the problems of information island, low efficiency of retrieval and insufficient intelligence in the construction technical knowledge management are solved, and an efficient tool is provided for the digital transformation of engineering construction: by constructing the knowledge base and the knowledge graph in the field of construction technology, the structured management and semantic understanding of the construction technical knowledge are realized, and the problems of information island and low retrieval efficiency in the traditional construction technical knowledge management system are solved; based on the pre-training language model and the knowledge graph, the construction technical question and answer model is generated, which can intelligently answer the construction technical problems and generate professional answer texts, providing timely and accurate technical guidance for construction personnel, improving the construction efficiency and quality; through the combination of the search engine and the knowledge graph, semantic retrieval and accurate query of the construction technical documents are realized, improving the efficiency and accuracy of knowledge acquisition; according to the historical operation behavior data of the user, personalized push is carried out, which can provide a more intelligent and personalized knowledge management experience for the user.

[0070] It is easy to understand that, based on one or more embodiments provided in the present application, other embodiments can be obtained by combining, splitting, recombining, etc. of the embodiments of the present application, and these embodiments do not exceed the protection scope of the present application.

[0071] The above describes the present application and its embodiments in a schematic manner, which is not limited, and the embodiments shown are only part of the embodiments of the present application, and the actual structure is not limited thereto. Therefore, if a person skilled in the art is inspired thereby, without departing from the spirit of the present application, similar structural modes and embodiments can be designed without creativity, which should belong to the protection scope of the present application.

Claims

1. A construction technology knowledge management method based on a knowledge base and a large language model, comprising: Collect original construction technical documents, pre-process them, obtain construction technical documents, store them, and build a knowledge base in the field of construction technology; Based on natural language processing technology, we can identify entities and the semantic relationships between entities from construction technical documents and build a knowledge graph in the field of construction technology. Generate a construction technology question-answering model based on the pre-trained language model, knowledge base, and knowledge graph. The construction technology question-answering model is used to intelligently answer construction technology questions and generate text. Based on the search engine, the most relevant construction technical documents to the query input by the user are retrieved from the knowledge base; Receive query content entered by users through interactive operation pages, display query results or display construction technology question and answer results.

2. A construction technology knowledge management method based on a knowledge base and a large language model according to claim 1, wherein: The process of collecting original construction technical documents, pre-processing them, obtaining construction technical documents, storing them, and building a knowledge base in the field of construction technology includes: Collect original construction technical documents from multi-source heterogeneous data through automated data collection procedures; Attach metadata tags to each piece of data, including creator, timestamp, and version number; Pre-process the collected construction technical documents, including data cleaning and deduplication, to obtain processed construction technical documents; Construction technical documents are stored based on preset knowledge coding rules and naming formats.

3. A construction technology knowledge management system based on a knowledge base and a large language model according to claim 1, wherein: The natural language processing technology is used to identify entities and semantic relationships between entities from construction technical documents and to construct a knowledge graph in the field of construction technology, including: Based on natural language processing technology, identify entities and semantic relationships between entities from construction technical documents; Through expert review, the identified entities and the semantic relationships between them are manually verified and dynamically corrected; Based on the semantic relationship between entities, a knowledge graph in the field of construction technology is constructed.

4. A construction technology knowledge management system based on a knowledge base and a large language model according to claim 3, wherein: The identification of entities and semantic relationships between entities from construction technical documents based on natural language processing technology includes: Based on the entity recognition algorithm, entities are identified from the construction technical documents. Where P(e|ω) is the probability of entity e given word ω, and P(e) is the prior probability of entity e; Based on the relation extraction algorithm, the semantic relationship between entities is identified. Among them, P(r|e1,e2) is the probability of relation r given entities e1 and e2, P(e1,e2|r) is the probability that relation r generates entity pair P(e1,e2), and P(r) is the prior probability of relation r.

5. A construction technology knowledge management system based on a knowledge base and a large language model according to claim 1, wherein: The construction technology question-answering model is generated based on the pre-trained language model, knowledge base and knowledge graph. The construction technology question-answering model is used to intelligently answer construction technology questions and generate text, including: Fine-tune the pre-trained language model based on the construction technical documents in the knowledge base to obtain a fine-tuned language model; Based on the semantic relationship of the knowledge graph, a fine-tuned language model is used to intelligently answer input construction technical questions and generate professional answer texts.

6. A construction technology knowledge management system based on a knowledge base and a large language model according to claim 5, wherein: The pre-trained language model is fine-tuned based on the construction technical documents in the knowledge base to obtain a fine-tuned language model, including: Extract construction technical documents from the knowledge base as training samples and fine-tune the loss function Optimize the parameters of the language model to obtain a language model that is adapted to the construction scenario, where L(θ) is the loss function and y i is the true label; x i is the input text and θ is the model parameter.

7. A construction technology knowledge management system based on a knowledge base and a large language model according to claim 1, wherein: The search engine-based method of searching the knowledge base for the construction technical documents most relevant to the query input by the user includes: Querying the knowledge base for construction technical documents related to the query content input by the user; Based on the vector space model, the semantic similarity between the construction technical documents and the query content entered by the user is calculated; Based on the knowledge graph, the sorting priority of the retrieval results is adjusted according to the weight of the entity relationship.

8. The construction technology knowledge management system based on a knowledge base and a large language model according to claim 1 further comprises: Personalized push notifications are provided based on the user’s historical operational behavior data.

9. A construction technology knowledge management system based on a knowledge base and a large language model, for implementing the construction technology knowledge management method according to any one of claims 1 to 8, comprising: The knowledge base module is used to collect construction technology documents from multi-source heterogeneous data through automated data collection tools and perform data preprocessing, including classification, labeling, deduplication, and standardized storage, to build a knowledge base in the field of construction technology; The knowledge graph module, based on natural language processing technology, identifies entities and the semantic relationships between entities from construction technical documents and constructs a knowledge graph in the field of construction technology; The large language model module is used to generate a construction technology question-answering model based on a pre-trained language model, knowledge base, and knowledge graph. The construction technology question-answering model is used to intelligently answer construction technology questions and generate text; A search module is used to search the knowledge base for the construction technology documents most relevant to the query content input by the user based on the search engine; The user interface module is used to receive query content input by the user through an interactive operation page, display query results or display construction technology question and answer results.

10. The construction technology knowledge management system based on a knowledge base and a large language model according to claim 9, further comprising: The recommendation module is used to provide personalized push notifications based on the user's historical operation behavior data.

Images