Intelligent answering method and system for building construction process information

Abstract

The invention relates to the field of intelligent construction, in particular to an intelligent answering method and system for building construction process information. The invention provides an intelligent answering method and system for building construction process information, and the key point is that a building construction process vector database is constructed through a large language model; analyzing, reconstructing and vectorizing the problem, performing retrieval and rearrangement secondary retrieval on the problem in a building construction process vector database to obtain a retrieval result, generating a reply for the input problem based on the generation capability of a large language model in combination with a prompt template, and finally performing weighted score evaluation on the reply. The invention aims to realize efficient query and management of construction process information and promote the transformation of the building industry to intelligence and digitization. The illusion problem of intelligent answering of a common large language model and the problem that knowledge base content is not updated in time are solved, questions and answers in the general field are converted into questions and answers in the vertical field, complex user queries can be processed and responded more accurately, and therefore the overall search efficiency and accuracy are improved.

Description

Technical Field

[0001] The present invention relates to the field of intelligent construction, and in particular to a method and system for intelligently answering construction process information. Background Art

[0002] At present, artificial intelligence is developing rapidly. NLP natural language processing technology represented by large language models has demonstrated powerful capabilities, and various intelligent question-answering system technologies have been given development opportunities. Traditional large language model intelligent question-answering is limited by the training corpus, cannot perceive the latest content, and is prone to hallucinations and incorrect content. At the same time, it is difficult to provide accurate answers for specific fields. Retrieval-enhanced generation can optimize the output effect of large language models by integrating external knowledge bases.

[0003] In the construction process in the building industry, there are problems such as unclear specific information on a certain construction process and fragmented process details, resulting in a lack of efficient query and management methods for construction process information.

[0004] In summary, there is an urgent need for a method to apply retrieval enhancement generation technology to the construction field, to achieve efficient query and management of construction process information, to improve the utilization rate of construction process knowledge, and to promote the transformation of the construction industry to intelligent digitalization. It can also improve the problem that existing technologies are not accurate and convenient enough, and to improve the problem that knowledge in the construction field is extensive and fragmented. Summary of the invention

[0005] The purpose of the present invention is to apply retrieval enhancement generation technology to the method of the construction field, to achieve efficient query and management of construction process information, to improve the utilization rate of construction process knowledge, to promote the transformation of the construction industry to intelligent digitalization, to improve the common hallucination problem of large language model intelligent answering and the problem of untimely updating of knowledge base content, to shift from general field question and answer to vertical field, to be able to more accurately process and respond to complex user queries, thereby improving the overall search efficiency and accuracy.

[0006] In a first aspect, the present invention provides a method for intelligently answering construction process information, comprising the following steps: Through the large language model, the structured and unstructured knowledge documents of different types of construction processes are parsed, cut, vectorized and stored as a construction process vector database; After analyzing, reconstructing and vectorizing the input questions, they are searched and rearranged in the construction process vector database for secondary search to obtain construction process knowledge; Generate a response to the input question based on the prompt template and the building construction process knowledge based on the generation capability of the large language model; The input questions and generated responses are stored as historical conversation information in a construction process vector database; The weighted score is used to evaluate whether the response generated by the large language model meets the standard. If it meets the standard, the response is output; if it does not meet the standard, the question is input again until a response that meets the standard is obtained.

[0007] The intelligent answering method for construction process information provided by the present invention has accurate and convenient technical effects.

[0008] Optionally, the analysis is to convert the construction process document into structured text information using text recognition, computer graphics and intelligent image processing engine processing methods, and the formats of the construction process document include PDF, EXCEL, WORD, PPT, and PNG.

[0009] Optionally, the cutting is to cut the parsed construction process document into blocks, taking into account the construction personnel's queries and the nature of the indexed construction process document when dividing the blocks, using a recursive character text segmentation method to segment the text according to the position of the separator, and specifying a text block of 100 tokens as the size of the final block.

[0010] Optionally, the analyzing and reconstructing the input question includes rewriting the input question according to historical information combined with contextual semantics, splitting complex construction process problems into multiple sub-queries, and optimizing the problem when the description of the proposed question is not professional enough.

[0011] Optionally, the retrieval uses a hybrid retrieval method to retrieve the construction technology related knowledge corresponding to the question vector, and the rearranged secondary retrieval uses the input question and the retrieved construction technology related knowledge as input content for secondary retrieval, sorts the results according to the degree of match with the context, and selects the construction technology related knowledge with the highest degree of match as output.

[0012] Optionally, the use of weighted scores to evaluate whether the responses generated by the large language model meet the standards uses context relevance, fidelity, and answer relevance as evaluation indicators.

[0013] In a second aspect, the present invention provides an intelligent answering system for construction process information, comprising: The building construction process information storage module is used to store different types of building construction process knowledge content and historical dialogue information, and supports the building construction process information retrieval module to sort out and retrieve questions; The building process information retrieval module is used to sort out the construction problems raised and retrieve the corresponding building construction process-related knowledge based on the construction problems; The human-computer interaction module calls the large language model to generate humanized responses based on the relevant knowledge of the construction process and realizes human-computer interaction through the front-end page; The evaluation and optimization module is used to evaluate whether the generated response meets the standards. If it does not meet the standards, retrieval enhancement generation will be performed again to optimize the response content.

[0014] The intelligent answering system for building construction process information provided by the present invention has clear and visual technical effects.

[0015] Optionally, the building construction process information storage module includes: The construction process document parsing submodule is used to parse the different types of construction process document information that need to be stored into structured text information; The construction document cutting and segmenting submodule is used to cut the parsed construction process documents; Vectorized construction document submodule, used for vectorizing the construction process document blocks obtained by cutting; The building construction process vector database submodule is used to store the vector data of the building construction process document blocks; Establishing an index submodule, which is used to give index attributes to the cut building construction process document blocks, and mapping the vector data of the building construction process document blocks with the building construction process document blocks; The historical information storage submodule is used to store the dialogue information generated during human-computer interaction.

[0016] Optionally, the construction process information retrieval module includes: The query reconstruction submodule is used to analyze and reconstruct the input question and vectorize the sub-questions obtained after the analysis and reconstruction; The building construction process knowledge retrieval submodule is used to retrieve vector data from the building construction process vector database and index the building construction process document block corresponding to the input question through the vector data; The re-ranking knowledge sub-module is used to perform secondary retrieval using the input question and the construction process document blocks obtained by the preliminary retrieval as input, sort the results according to the degree of context matching, and select the construction process document block with the highest matching degree as output.

[0017] Optionally, the human-computer interaction module includes: The front-end page model submodule is used to create the front-end page to realize the visualization of human-computer interaction; Integrate various LLM sub-modules to integrate online large models and locally deployed large models through the distributed multi-model service system. According to the type of construction process problem queried, call various large models integrated in the front-end page to generate responses; The API interface submodule is used to open the API interface to add, delete and improve the front-end page functions. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 It is a flow chart of a method for intelligent answering of construction process information provided by the present invention; Figure 2 It is a flow chart of constructing a vector database in a method for intelligent answering of building construction process information provided by the present invention; Figure 3 This is an architecture diagram of an intelligent answering system for building construction process information provided by the present invention. DETAILED DESCRIPTION

[0019] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention. Unless otherwise defined, the technical terms or scientific terms used herein should be understood by people with general skills in the field to which the present invention belongs. "Including" and similar words used in this article mean that the elements or objects appearing before the word include the elements or objects listed after the word and their equivalents, without excluding other elements or objects.

[0020] The embodiment of the present invention provides a method for intelligently answering construction process information, comprising the following steps: S1. Parse, cut, vectorize and store structured and unstructured knowledge documents of different types of construction processes as a construction process vector database through a large language model; S2, analyzing, reconstructing and vectorizing the input question, searching and rearranging it in the building construction process vector database for secondary retrieval to obtain building construction process knowledge; S3, generating a response to the input question based on the prompt template and the building construction process knowledge based on the generation capability of the large language model; S4, storing the input questions and generated responses as historical dialogue information in a construction process vector database; S5. Use the weighted score to evaluate whether the response generated by the large language model meets the standard. If it meets the standard, the response is output. If it does not meet the standard, the question is input again until a response that meets the standard is obtained.

[0021] The flowchart of a method for intelligent answering of construction process information provided by the present invention is as follows: Figure 1 shown.

[0022] When executing step S1, the analysis is to convert the construction process document into structured text information by using text recognition, computer graphics and intelligent image processing engine processing methods. The formats of the construction process document include PDF, EXCEL, WORD, PPT, and PNG.

[0023] This embodiment uses masonry wall construction process PDF document data as building construction process knowledge.

[0024] Use the PyPDFLoader command under Document Loade to parse and get the text format as follows: "Page 0: page_content='Masonry wall construction technology

[0025] 1. Construction preparation

[0026] (1) Be familiar with the construction drawings and understand the design requirements.

[0027] (2) Prepare qualified materials: bricks, blocks, mortar, tie bars, etc.

[0028] (3) Prepare construction tools: brick cutter, level, plumb line hammer, mortar mixer, etc.

[0029] (4) Clean up the construction site to ensure a clean and tidy construction environment.

[0030] ' metadata={'source': 'F:\\Masonry Wall Construction Technology.pdf', 'page': 0}"

[0031] The cutting is to cut the parsed construction process document into blocks, taking into account the construction personnel's query and the nature of the indexed construction process document when dividing the blocks, using a recursive character text segmentation method to segment the text according to the position of the separator, and specifying a text block of 100 tokens as the size of the final block.

[0032] Use the RecursiveCharacterTextSplitte command to split the document. The resulting document is as follows: “Document 1: Split 1: ('id', None)

[0033] Split 2:('metadata', {'source': 'F:\\Masonry Wall Construction Technology.pdf', 'page':0})

[0034] Split 3: ('page_content', 'Masonry wall construction technology\n1. Construction preparation\n(1) Familiarize yourself with the construction drawings and clarify the design requirements.\n(2) Prepare qualified materials: bricks, blocks, mortar, tie bars, etc.\n(3) Prepare construction tools: brick cutter, level ruler, vertical line hammer, mortar mixer, etc.')

[0035] Split 4: ('type', 'Document')"

[0036] The cut construction process documents are divided into blocks and converted into vector data. After conversion into vector data, the degree of association can be determined by calculating the similarity between the vector data.

[0037] The vector data obtained by converting the PDF document after cutting the masonry wall construction process in this embodiment is as follows: "Embedding 1: [0.045131747, 0.03196441, -0.06771896]

[0038] Embedding 2: [0.051496707, 0.04866433, -0.060791768]"

[0039] Create a construction process vector database to store vector data converted from different types of construction process documents.

[0040] The converted construction process vector data and construction process documents are divided into blocks and mapped to construct vector indexes, providing a calculation basis for subsequent vector retrieval and other retrieval operations.

[0041] The resulting index ID format is as follows: “Added documents with IDs: ['1d1b4a5a-bf3f-4457-bbcc-cf501f648180', 'd80f8880-939d-4add-af12-c0d65dd44434']"

[0042] The flowchart of constructing a vector database in a method for intelligent answering of construction process information provided by the present invention is as follows: Figure 2 shown.

[0043] When executing step S2, the analysis and reconstruction of the input question includes rewriting the input question based on historical information combined with contextual semantics, splitting complex construction process problems into multiple sub-queries, and optimizing the problem when the description of the proposed problem is not professional enough.

[0044] In this embodiment, the question input by the construction personnel is: how to deal with the door and window openings encountered during masonry, and where to leave the construction joints. At this time, the question is divided into two sub-questions: how to deal with the door and window openings encountered during masonry, and where to leave the construction joints during masonry.

[0045] The reconstructed construction process problem is vectorized, and then the reconstructed problem vector is averaged to obtain a new vector Z; , Among them, F is the vector conversion encoding operation, It is a problem of original construction technology. represents the obtained sub-construction process problem, and X represents the number of new problems transformed by reconstruction.

[0046] In this embodiment, the two sub-problems obtained by splitting are vectorized to obtain two sub-problem vectors. At this time, the new vector is calculated as = .

[0047] The retrieval uses a hybrid retrieval method to retrieve the construction technology-related knowledge corresponding to the question vector. The rearranged secondary retrieval uses the input question and the retrieved construction technology-related knowledge as input content for secondary retrieval, sorts the results according to the degree of match with the context, and selects the construction technology-related knowledge with the highest degree of match as output.

[0048] Among them, the hybrid search method is used to retrieve the construction process-related knowledge corresponding to the construction problem by combining keyword matching and semantic understanding. The hybrid search method refers to the fusion of the search results of sparse search and dense search, as follows: The scores of the two methods are normalized and combined, and the parameter α is used for weighting to determine the weight of each algorithm and affect the re-ranking of the results. The hybrid scoring formula is: hybrid_score=(1-α)*sparse_score+α*dense_score

[0049] Among them, hybrid_score represents the hybrid score, parse_score represents the sparse score, and dense_score represents the dense score. The value of α is usually between 0 and 1. When α=1, it means that the dense search method is used purely and only vector search is used. α=0 means that the sparse search method is used purely and only keywords are searched.

[0050] In this embodiment, sparse search uses the BM25 search method, and dense search uses the BERT vector similarity calculation method. At the same time, α is set to 0.3, and the scoring formula is set to: hybrid_score=0.7*sparse_score+0.3*dense_score

[0051] The DLM reranking method is used for reranking, and the bge-reranker-base reranker is used for sorting. The top-ranked and most relevant documents are selected as the output responses. At this time, the highest-ranked document content is: 5. Door and window opening treatment

[0052] (1) Reserve door and window openings to ensure the accuracy of door and window frame installation.

[0053] (2) The masonry on both sides of the opening should be built symmetrically to prevent deformation of the opening.

[0054] 6. Construction joint treatment

[0055] (1) Construction joints should be left at locations where the structure is subjected to less stress.

[0056] (2) The joints should be cleaned and mortar should be laid to ensure a firm connection.

[0057] When executing step S3, when generating a response to the input question based on the prompt template and the generation capability of the large language model based on the construction process knowledge, first create a front-end page to realize the visualization of human-computer interaction with front-end and back-end separation, wherein the front-end page has three models, including a user model, a dialogue model and an information data model, and the model data is stored in a table.

[0058] The user model is used to express user permission management and user password. One user can create multiple conversations.

[0059] The dialog model is used to create a dialog with the large model.

[0060] The Message data model is used to create and model multiple messages in a dialog window.

[0061] In this embodiment, the front end uses Vue to build the front end page.

[0062] Run a distributed multi-model service system to integrate the online big model and the locally deployed big model, including running the conrtoller service, starting the model worker local model on the back end, and starting the server service for calling (the online big model is called through the API Key, and the locally deployed big model is called through the model worker registered in the controller). After the front-end page is integrated, the construction personnel can call the big model integrated on the page according to the category of the construction process problem.

[0063] In this embodiment, a multimodal distribution system is run based on Fastchat, and the GLM4 model is selected to answer the question.

[0064] Open API interfaces to add, delete and improve front-end page functions, as well as call construction field tools to improve the overall performance of the intelligent answering system for construction process information.

[0065] The construction process knowledge is transmitted to the large language model integrated and called by the front-end page, and a response is generated according to the prompt template based on the generation capability of the large language model. The prompt template is as follows: PromptTemplate = { "Input_variables": ["question"], "Template": """ You are a construction professional with extensive experience in the construction field.

[0066] Please provide professional and detailed answers to the following questions: {question} Looking forward to your professional answer.

[0067] """}

[0068] The construction process knowledge obtained after reordering is transferred to the selected GLM4 large language model, and the PromptTemplate is used to generate a response using the generation ability of the large language model: "When encountering door and window openings during the masonry process, door and window openings should be reserved to ensure accurate installation of door and window frames. The masonry on both sides of the opening should be built symmetrically to prevent deformation of the opening, and tie bars should be set according to design requirements to enhance the integrity and seismic resistance of the structure. Construction joints should be left in parts of the structure with less stress, and ensure that the joints are cleaned and mortar is laid to ensure a firm connection. (Refer to 5. Door and window opening treatment and 6. Construction joint treatment in context)"

[0069] When executing step S5, the weighted score is used to evaluate whether the response generated by the large language model meets the standards, with context relevance, fidelity and answer relevance as evaluation indicators.

[0070] Context relevance: The degree of match between the question and the retrieved query documents is calculated as: CR= , Faithfulness measures the relevance of the process knowledge answers generated by the model to the context, and calculates the score based on the manually set correct answers and the retrieved context. The higher the score between 0 and 1, the more relevant the answer. The calculation formula is: F= , Answer relevancy (answer relevancy) does not consider the correctness of the answer, but only compares the similarity between the answer and the question. The calculation formula is: AR= ), Among them, S represents the similarity comparison, p represents the original question, Indicates potential problems generated; The scores of the three evaluation indicators in this embodiment are: "Context_relevancy_score:0.76584558555224 Faithfulness_score:1.0 Answer_relevancy_score:0.709024777916293”

[0071] The weighted score is used to evaluate whether the response generated by the large language model meets the standard. If it meets the standard, the response is output; if it does not meet the standard, the question is input again until a response that meets the standard is obtained.

[0072] Score=Context_relevancy_score*30+Faithfulness_score*50+Answer_relevancy_score*20

[0073] In this embodiment: Score=Context_relevancy_score*30+Faithfulness_score*50+Answer_relevancy_score*20=87.155, indicating that this reply meets the standard, and the reply is output.

[0074] The embodiment of the present invention also provides a building construction process information intelligent answering system, including: The building construction process information storage module is used to store different types of building construction process knowledge content and historical dialogue information, and supports the building construction process information retrieval module to sort out and retrieve questions; The building process information retrieval module is used to sort out the construction problems raised and retrieve the corresponding building construction process-related knowledge based on the construction problems; The human-computer interaction module calls the large language model to generate humanized responses based on the relevant knowledge of the construction process and realizes human-computer interaction through the front-end page; The evaluation and optimization module is used to evaluate whether the generated response meets the standards. If it does not meet the standards, retrieval enhancement generation will be performed again to optimize the response content.

[0075] The architecture diagram of the intelligent answering system for construction process information provided by the present invention is as follows: Figure 3 shown.

[0076] Wherein, the building construction process information storage module includes: The construction process document parsing submodule is used to parse the different types of construction process document information that need to be stored into structured text information; The construction document cutting and segmenting submodule is used to cut the parsed construction process documents; Vectorized construction document submodule, used for vectorizing the construction process document blocks obtained by cutting; The building construction process vector database submodule is used to store the vector data of the building construction process document blocks, and uses the chroma vector database to store the obtained vector data; Establishing an index submodule, which is used to give index attributes to the cut building construction process document blocks, and mapping the vector data of the building construction process document blocks with the building construction process document blocks; The historical information storage submodule is used to store the dialogue information generated during human-computer interaction.

[0077] This embodiment uses Langchain, GLM4, Loade, PyPDF, embedding-2, chroma, and RecursiveCharacterTextSplitte to process construction process documents.

[0078] The construction process information retrieval module includes: The query reconstruction submodule is used to analyze and reconstruct the input question and vectorize the sub-questions obtained after the analysis and reconstruction; The building construction process knowledge retrieval submodule is used to retrieve vector data from the building construction process vector database and index the building construction process document block corresponding to the input question through the vector data;

[0079] The re-ranking knowledge sub-module is used to perform secondary retrieval using the input question and the construction process document blocks obtained by the preliminary retrieval as input, sort the results according to the degree of context matching, and select the construction process document block with the highest matching degree as output.

[0080] This embodiment uses keyword retrieval technology, BERT model, and bge-reranker-base model to realize the retrieval of construction process documents and the reranking of knowledge.

[0081] The human-computer interaction module comprises: The front-end page model submodule is used to create the front-end page to realize the visualization of human-computer interaction with the separation of front-end and back-end; Integrate various LLM sub-modules to integrate online large models and locally deployed large models through the distributed multi-model service system. According to the type of construction process problem queried, call various large models integrated in the front-end page to generate responses; The API interface submodule is used to open the API interface to add, delete and improve the front-end page functions.

[0082] The evaluation optimization module selects three evaluation indicators: context relevance, answer relevance and fidelity. This embodiment uses Pychram as the evaluation test platform.

[0083] Although the embodiments of the present invention are described in detail above, it is obvious to those skilled in the art that various modifications and variations can be made to these embodiments. However, it should be understood that such modifications and variations are within the scope and spirit of the present invention as described in the claims. Moreover, the present invention described herein may have other embodiments and may be implemented or realized in a variety of ways.

Claims

1. A method for intelligently answering construction process information, characterized in that: The following steps are involved: Through the large language model, the structured and unstructured knowledge documents of different types of construction processes are parsed, cut, vectorized and stored as a construction process vector database; After analyzing, reconstructing and vectorizing the input questions, they are searched and rearranged in the construction process vector database for secondary search to obtain construction process knowledge; Generate a response to the input question based on the prompt template and the building construction process knowledge based on the generation capability of the large language model; The input questions and generated responses are stored as historical conversation information in a construction process vector database; The weighted score is used to evaluate whether the response generated by the large language model meets the standard. If it meets the standard, the response is output; if it does not meet the standard, the question is input again until a response that meets the standard is obtained.

2. The intelligent answering method for construction process information according to claim 1 is characterized in that: The analysis is to convert the construction process document into structured text information using text recognition, computer graphics and intelligent image processing engine processing methods. The formats of the knowledge document include PDF, EXCEL, WORD, PPT, and PNG.

3. The intelligent answering method for construction process information according to claim 1 is characterized in that: The cutting is to cut the parsed construction process document into blocks, taking into account the construction personnel's query and the nature of the indexed construction process document when dividing the blocks, using a recursive character text segmentation method to segment the text according to the position of the separator, and specifying a text block of 100 tokens as the size of the final block.

4. The intelligent answering method for construction process information according to claim 1 is characterized in that: The analysis and reconstruction of the input question includes rewriting the input question according to historical information combined with contextual semantics, splitting complex construction process problems into multiple sub-queries, and optimizing the problem when the description of the proposed question is not professional enough.

5. The intelligent answering method for construction process information according to claim 1 is characterized in that: The retrieval uses a hybrid retrieval method to retrieve the construction technology-related knowledge corresponding to the question vector. The rearranged secondary retrieval uses the input question and the retrieved construction technology-related knowledge as input content for secondary retrieval, sorts the results according to the degree of match with the context, and selects the construction technology-related knowledge with the highest degree of match as output.

6. The intelligent answering method for construction process information according to claim 1 is characterized in that: The described use of weighted scores to evaluate whether the responses generated by the large language model meet the standards uses context relevance, fidelity, and answer relevance as evaluation indicators.

7. An intelligent answering system for construction process information, characterized in that: include: The building construction process information storage module is used to store different types of building construction process knowledge content and historical dialogue information, and supports the building construction process information retrieval module to sort out and retrieve questions; The building process information retrieval module is used to sort out the construction problems raised and retrieve the corresponding building construction process-related knowledge based on the construction problems; The human-computer interaction module calls the large language model to generate humanized responses based on the relevant knowledge of the construction process and realizes human-computer interaction through the front-end page; The evaluation and optimization module is used to evaluate whether the generated response meets the standards. If it does not meet the standards, retrieval enhancement generation will be performed again to optimize the response content.

8. The intelligent answering system for construction process information according to claim 7 is characterized in that: The building construction process information storage module comprises: The construction process document parsing submodule is used to parse the different types of construction process document information that need to be stored into structured text information; The construction document cutting and segmenting submodule is used to cut the parsed construction process documents; Vectorized construction document submodule, used for vectorizing the construction process document blocks obtained by cutting; The building construction process vector database submodule is used to store the vector data of the building construction process document blocks; Establishing an index submodule, which is used to give index attributes to the cut building construction process document blocks, and mapping the vector data of the building construction process document blocks with the building construction process document blocks; The historical information storage submodule is used to store the dialogue information generated during human-computer interaction.

9. The intelligent answering system for construction process information according to claim 8 is characterized in that: The construction process information retrieval module comprises: The query reconstruction submodule is used to analyze and reconstruct the input question and vectorize the sub-questions obtained after the analysis and reconstruction; The building construction process knowledge retrieval submodule is used to retrieve vector data from the building construction process vector database and index the building construction process document block corresponding to the input question through the vector data; The re-ranking knowledge sub-module is used to perform secondary retrieval using the input question and the construction process document blocks obtained by the preliminary retrieval as input, sort the results according to the degree of context matching, and select the construction process document block with the highest matching degree as output.

10. The intelligent answering system for construction process information according to claim 9 is characterized in that: The human-computer interaction module comprises: The front-end page model submodule is used to create the front-end page to realize the visualization of human-computer interaction; Integrate various LLM sub-modules to integrate online large models and locally deployed large models through the distributed multi-model service system. According to the type of construction process problem queried, call various large models integrated in the front-end page to generate responses; The API interface submodule is used to open the API interface to add, delete and improve the front-end page functions.

Images