Intelligent question and answer method and system in building construction field

Abstract

The invention provides an intelligent question answering method and system in the field of building construction, and belongs to the technical field of artificial intelligence. Inputting the questioning information into the large question-answering model to obtain reply answers output by the large question-answering model; wherein the question and answer large model is obtained through fine adjustment of the sample question information and a knowledge base containing professional knowledge in the field of building construction, and the knowledge base is used for assisting the question and answer large model to generate reply answers according to the sample question information. The question and answer large model is finely adjusted in combination with the professional knowledge base in the field of building construction, and high-reliability reply answers are generated in an intelligent question and answer scene in the field of building construction.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to an intelligent question-answering method and system in the field of building construction. Background Art

[0002] Existing question-answering systems usually use multimodal inputs, such as images and videos, and perform corresponding retrieval and answers. However, these systems are mainly aimed at general fields, and their retrieval accuracy and depth are often insufficient for retrieval in professional fields, such as the construction field.

[0003] Existing multimodal retrieval technologies usually only perform simple analysis on images or videos, extract text information or keywords, and then perform retrieval and question-answering. This method cannot accurately understand the semantic information in professional fields and it is difficult to give accurate answers to professional questions.

[0004] Therefore, how to generate highly credible responses in intelligent question-answering scenarios in the field of construction has become a technical problem that needs to be solved urgently. Summary of the invention

[0005] The present invention provides an intelligent question-answering method, system, electronic device and storage medium in the field of construction, so as to solve the defects in the prior art and generate highly reliable reply answers in the intelligent question-answering scenario in the field of construction.

[0006] The present invention provides an intelligent question-answering method in the field of building construction, comprising the following steps: Receive question information input by the user; Input the question information into the question-answering model to obtain a reply answer output by the question-answering model; Among them, the question-answering big model is obtained by fine-tuning sample question information and a knowledge base containing professional knowledge in the field of construction. The knowledge base is used to assist the question-answering big model in generating response answers based on the sample question information.

[0007] According to an intelligent question-answering method in the field of building construction provided by the present invention, the method further includes: Get multiple professional knowledge information; Classifying all the professional knowledge information and determining the parent category corresponding to each piece of professional knowledge information; Based on the parent category, each piece of professional knowledge information is further classified to determine a subcategory corresponding to each piece of professional knowledge information; Marking at least one label for each piece of professional knowledge information; According to the classification results and labeling results, all the professional knowledge information is integrated to construct the knowledge base.

[0008] According to an intelligent question-answering method in the field of building construction provided by the present invention, the fine-tuning process of the large question-answering model includes: Get multiple sample question information; Pairing all the sample question information with all the professional knowledge information in the knowledge base to generate a plurality of question-answer pairs; each question-answer pair includes one sample question information and at least one professional knowledge information; All the question-answer pairs are integrated into a general large model to obtain the question-answer large model.

[0009] According to an intelligent question-answering method in the field of construction provided by the present invention, the method of pairing all the sample question information with all the professional knowledge information in the knowledge base to generate multiple question-answer pairs specifically includes: Performing semantic analysis on each of the sample question information, and extracting at least one sample keyword from each of the sample question information; Perform multi-level matching of all sample keywords in each sample question information with parent categories, subcategories, and tags corresponding to all professional knowledge information in the knowledge base to obtain multiple matching results; Based on all the matching results, a plurality of question-answer pairs are constructed.

[0010] According to an intelligent question-answering method in the field of construction provided by the present invention, the question information is input into a large question-answering model to obtain a reply answer output by the large question-answering model, specifically including: Inputting the question information into the question-answering macromodel, and obtaining at least one initial answer associated with the question information through all the question-answer pairs; Determine the output weight corresponding to each of the initial answers; All of the initial answers are sorted according to all of the output weights, and the sorted initial answers are integrated to generate the reply answer.

[0011] According to an intelligent question-answering method in the field of construction provided by the present invention, obtaining at least one initial answer associated with the question information through all the question-answer pairs specifically includes: Performing semantic analysis on each of the question information, and extracting at least one keyword from each of the question information; According to each of the keywords, in all the professional knowledge information contained in all the question-answer pairs, it is determined whether there is matching professional knowledge information; the text similarity between any one of the parent category, subcategory or label corresponding to the matching professional knowledge information and the keyword is greater than a preset similarity threshold; When it is determined that there is matching professional knowledge information, each piece of matching professional knowledge information is used as an initial answer; When it is determined that there is no matching professional knowledge information, at least one relevant professional knowledge information is selected from all the professional knowledge information contained in all the question-answer pairs; the text similarity between any one of the parent categories, subcategories or tags corresponding to the relevant professional knowledge information and the keyword is greater than the text similarity between any one of the parent categories, subcategories or tags corresponding to the remaining professional knowledge information and the keyword; Based on each of the related professional knowledge information, an initial answer is generated.

[0012] According to an intelligent question-answering method in the field of building construction provided by the present invention, the method further includes: Obtain the user's historical question information and interaction record information, and obtain the position information corresponding to the user account; Performing keyword extraction and semantic relevance analysis on the current question information to obtain a plurality of current keywords and current semantic representations, and performing keyword extraction and semantic relevance analysis on the historical question information to obtain a plurality of historical keywords and a plurality of historical semantic representations; Clustering all the historical keywords and all the historical semantic expressions to generate multiple interest topics; each of the historical question information corresponds to one interest topic; Determine the interest weight corresponding to each interest topic according to the historical question time corresponding to each historical question information and according to the time decay strategy; generating long-term interest information according to each of the interest topics and the interest weight corresponding to each of the interest topics; Generate short-term interest information according to the interaction record information, all the current keywords and all the current semantic expressions; Determine job-related knowledge requirement information based on the job information; Generate corresponding recommendation questions according to the long-term interest information, the short-term interest information and the job-related knowledge requirement information.

[0013] The present invention also provides an intelligent question-answering system in the field of building construction, comprising the following modules: A receiving module, used for receiving question information input by a user; A processing module is used to input the question information into the question-answering big model to obtain a reply answer output by the question-answering big model; wherein the question-answering big model is obtained by fine-tuning sample question information and a knowledge base containing professional knowledge in the field of construction, and the knowledge base is used to assist the question-answering big model in generating a reply answer based on the sample question information.

[0014] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, an intelligent question-answering method in the field of construction as described above is implemented.

[0015] The present invention also provides a non-transitory computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, it implements any of the above-mentioned intelligent question-answering methods in the field of construction.

[0016] The present invention also provides a computer program product, comprising a computer program, which, when executed by a processor, implements any of the above-mentioned intelligent question-answering methods in the field of construction.

[0017] In summary, one or more technical solutions provided in the embodiments of the present application have at least the following technical effects or advantages: By receiving the question information input by the user, the specific needs of the user can be obtained, providing basic input for subsequent processing; by inputting the question information into the question-answering big model, the reasoning and generation capabilities of the big model can be used to quickly obtain the reply answer related to the question; among them, the question-answering big model has been fine-tuned with sample question information and a knowledge base containing professional knowledge in the field of construction. Through this solution, the big model not only has general language understanding capabilities, but also can perform well in professional scenarios in the field of construction and provide more targeted answers. By introducing a professional knowledge base to fine-tune the big model, the authoritative information in the knowledge base can be used to assist the big model in generating accurate reply answers, thereby effectively improving the professionalism and accuracy of the answers. Through this solution, when facing complex questions in the field of construction, it is possible to combine the unique knowledge in the field to generate more in-depth and accurate answers, and realize the generation of highly credible reply answers in the intelligent question-answering scenario in the field of construction. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0019] Figure 1 This is one of the flow charts of the intelligent question-answering method in the field of construction provided by the present invention.

[0020] Figure 2 This is the second flow chart of the intelligent question-answering method in the field of construction provided by the present invention.

[0021] Figure 3 This is the third flow chart of the intelligent question-answering method in the field of construction provided by the present invention.

[0022] Figure 4 This is the fourth flow chart of the intelligent question-answering method in the field of construction provided by the present invention.

[0023] Figure 5 This is the fifth flow chart of the intelligent question-answering method in the field of construction provided by the present invention.

[0024] Figure 6 This is the sixth flow chart of the intelligent question-answering method in the field of construction provided by the present invention.

[0025] Figure 7 This is the seventh flow chart of the intelligent question-answering method in the field of construction provided by the present invention.

[0026] Figure 8 It is a structural schematic diagram of the intelligent question-answering system in the field of building construction provided by the present invention.

[0027] Fig. 9 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0028] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. 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.

[0029] It should be noted that, in the description of the present invention, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "include one..." do not exclude the existence of other identical elements in the process, method, article or device including the elements. The orientation or positional relationship indicated by the terms "upper", "lower", etc. is based on the orientation or positional relationship shown in the accompanying drawings, and is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the system or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation on the present invention. For those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood according to the specific circumstances.

[0030] The terms "first", "second", etc. in the present invention are used to distinguish similar objects, and are not used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged under appropriate circumstances, so that the embodiments of the present invention can be implemented in an order other than those illustrated or described herein, and the objects distinguished by "first", "second", etc. are generally of the same type, and the number of objects is not limited. For example, the first object can be one or more. In addition, "and / or" means at least one of the connected objects, and the character " / " generally indicates that the objects associated with each other are in an "or" relationship.

[0031] Combine the following Figure 1-Figure 9 The invention describes an intelligent question-answering method, system, electronic device and storage medium in the field of building construction.

[0032] Figure 1 is one of the flow charts of the intelligent question-answering method in the field of construction provided by the present invention, such as Figure 1 As shown, including but not limited to the following steps: Step 101: receiving question information input by a user.

[0033] In step 101, receiving the question information input by the user is the starting point and key step of the entire multimodal intelligent question-answering process. Its main purpose is to obtain the relevant issues in the field of construction that the user is concerned about, so that accurate model reasoning and response generation can be performed in the subsequent steps. Specifically, during the implementation process, the user inputs the question through the user terminal device (such as a smart phone, tablet or computer), and the input form of the question can be multimodal information such as text, picture, video, audio, etc. The system receives and recognizes these inputs through the multimodal processing module, and after necessary conversion of non-text inputs (for example, performing OCR recognition on the picture, converting the text content in the picture into text information, such as extracting the text information in the construction drawing for subsequent retrieval and question-answering; extracting frames from the video, dividing the video into multiple independent pictures, such as dividing a one-hour construction training video into multiple 5-minute segments for frame-by-frame analysis and processing; audio-to-text conversion, converting the voice content in the audio into text information for subsequent retrieval and question-answering methods), the question content is unified into a processable text form.

[0034] The significance of receiving the question information input by the user is to provide the question-answering model with the content that is most relevant to the user's needs, so that in the subsequent steps, the fine-tuned question-answering model can generate targeted responses. By preprocessing multimodal information in this way, the system can be compatible with the diverse input needs of users. Whether the user takes pictures of the on-site construction, uploads relevant videos or simple text descriptions, the system can abstract, understand and transform the content, so that the subsequent question-answering model can process this input in a unified manner. In addition, the unification of multimodal input also lays the foundation for subsequent semantic analysis, ensuring that the question information is in a state that can be efficiently analyzed before entering the question-answering model.

[0035] Step 102: Input the question information into the question-answering big model to obtain the reply answer output by the question-answering big model; wherein, the question-answering big model is obtained by fine-tuning sample question information and a knowledge base containing professional knowledge in the field of construction, and the knowledge base is used to assist the question-answering big model in generating reply answers based on sample question information.

[0036] In step 102, the question information input by the user is input into the question-answering model, and the reply answer output by the question-answering model is obtained. This is the core part of the intelligent question-answering system, which mainly uses the question-answering model to perform knowledge reasoning and answer generation. The purpose of this is to make full use of the reasoning ability of the pre-fine-tuned question-answering model, so as to provide accurate, professional and in-depth replies to professional questions in the field of construction raised by users.

[0037] In the specific implementation process, the question information input by the user is received and multimodal converted in step 101, and then enters the question-answering model for further processing. The question-answering model has been fine-tuned through a large amount of sample question information and professional knowledge information in the field of construction. The fine-tuning process uses the question-answer pairs generated by pairing the sample question information with the relevant field knowledge in the knowledge base, so that the model can understand the user's questions more specifically and generate accurate answers when facing similar construction scenarios.

[0038] In the subsequent embodiments, the specific method of constructing the knowledge base, the method of fine-tuning the question and answer big model, and the implementation method of inputting question information into the question and answer big model in actual application to obtain the reply answer output by the question and answer big model will be described in detail.

[0039] In one possible implementation, refer to Figure 2 , Figure 2 This is the second flow chart of the intelligent question-answering method in the field of construction provided by the present invention, such as Figure 2 As shown, the method further includes constructing the knowledge base in step 102. The specific method of constructing the knowledge base includes steps 201-205: Step 201: Acquire multiple professional knowledge information.

[0040] During the specific implementation process, professional knowledge information comes from various knowledge resources such as various specifications and standard documents, operation manuals, industry cases, expert experience bases, etc. in the field of construction. These resources include but are not limited to text materials, construction drawings, technical regulations, construction quality acceptance standards, typical construction site videos, etc., which are digitized and converted into structured data that can be used for subsequent processing. The collected content covers not only the basic construction process and standard operating procedures, but also multiple sub-fields such as building materials, quality control, site management, safety protection, etc., to ensure the comprehensiveness and depth of the knowledge base in all aspects.

[0041] Step 202: Classify all professional knowledge information and determine the parent category corresponding to each piece of professional knowledge information.

[0042] In step 202, classifying all professional knowledge information and determining the parent category corresponding to each professional knowledge information is one of the key links in building a knowledge base in the field of construction. The purpose of this is to establish a hierarchical structure of the knowledge base so that the subsequent question-answering model can efficiently retrieve professional knowledge information related to the user's questions, thereby improving the accuracy and efficiency of the system's answers. By reasonably classifying knowledge, the professional knowledge in the knowledge base can be summarized and organized according to different aspects of the field of construction, so that the complex knowledge system has a clear hierarchy and order, which is easy to manage and call.

[0043] In the specific implementation process, we first analyze the content of each piece of professional knowledge information and classify it into the main parent categories in the field of construction, such as construction process, acceptance criteria, quality defects, safety protection, material technology, construction equipment, etc. To achieve this goal, the system uses natural language processing (NLP) technology to perform text analysis on knowledge information, combines pre-trained models to understand the content semantically, and extracts core features and keywords. Based on the extracted features, the system classifies each piece of knowledge information and aggregates similar knowledge under the corresponding parent category. The setting of parent categories usually follows the practical division of labor and standardized classification in the field of construction, ensuring that each type of knowledge has a high degree of homogeneity in a specific semantic context.

[0044] Step 203: Based on the parent category, each piece of professional knowledge information is further refined and classified to determine the subcategory corresponding to each piece of professional knowledge information.

[0045] In the specific implementation process, based on the established parent categories, the system further uses natural language processing and machine learning technology to conduct in-depth semantic analysis on each piece of professional knowledge information, extract its specific content features, and determine the most appropriate sub-category. For example, for the knowledge information under the parent category of "quality defects", the system will further divide it into sub-categories such as "cracks", "hollowing", and "leakage" according to different construction stages. In this process, the system extracts the features of the knowledge content and compares it with the preset classification standards to attribute each piece of knowledge information to the corresponding refined sub-category to form a more specific knowledge hierarchical structure.

[0046] Through this detailed classification process, the system makes each piece of information in the knowledge base not only have a macro-domain affiliation (parent category), but also have specific knowledge attributes at the micro level (sub-category). The effect of this is that when the question-answering model actually answers user questions, it can not only quickly determine the macro-category of the user's question, but also further find the answer that best meets the specific situation and detail requirements through sub-category matching.

[0047] Step 204: label each piece of professional knowledge information with at least one tag.

[0048] In the specific implementation process, after classifying each piece of professional knowledge information into parent categories and subcategories, the system further analyzes the content of the knowledge information through natural language processing technology, and extracts keywords or phrases that can represent its core meaning and annotates them as labels. For example, for a piece of knowledge information about "rebar binding construction technology" in the "structural construction" subcategory under the "construction process" parent category, the system will extract key phrases such as "rebar binding", "construction steps", "quality control", and "safety precautions" as labels for the knowledge. These labels can cover multiple aspects such as the main content, technical characteristics, and applicable conditions of the knowledge information, so that each piece of knowledge information can be described and positioned in different dimensions.

[0049] Through this labeling process, the system adds semantic tags to each piece of professional knowledge in the knowledge base, thereby adding more associated paths in the knowledge retrieval and matching process. The effect of this is that when the user's question information involves multiple concepts or specific keywords, the system can quickly locate the relevant knowledge information through the tags, rather than relying solely on the classification path of the parent category and subcategory.

[0050] Step 205: According to the classification results and labeling results, all professional knowledge information is integrated to build a knowledge base.

[0051] During the specific implementation process, the system integrates all the professional knowledge information that has been classified and labeled by parent categories and subcategories, and organizes them according to different hierarchical relationships. First, the system groups the knowledge information according to parent categories to establish the basic structure of the knowledge base. Then, within each parent category group, it is further refined into subcategories, so that the knowledge information logically presents a hierarchical classification hierarchy from large to small and from broad to specific. Subsequently, the system uses the labels of the knowledge information to construct the association relationship between different categories in the form of a knowledge graph or an associated database.

[0052] During the knowledge integration process, the system also stores knowledge information in a standardized manner to ensure that each piece of knowledge can be effectively queried based on the category path or label. Specifically, the system uses a unified data storage model to assign a unique identifier to each piece of knowledge information, and establishes a query interface in the knowledge base so that the subsequent question-answering model can quickly locate and call relevant knowledge after receiving the question information. In addition, the consistency and redundancy of data are also considered during the knowledge integration process to ensure the stability of the knowledge base structure and reduce the storage of duplicate information.

[0053] Through this integration and construction process, the system has formed a multi-level, multi-dimensional, and highly interconnected knowledge base in the field of construction. The effect of this is that the systematic organization of the knowledge base provides an efficient knowledge call path for the question-answering model, so that each time the search can not only find the target knowledge based on the classification, but also quickly realize cross-category information association through tags. For complex questions from users, the system can find the most relevant knowledge information based on the diverse characteristics of the questions, so that the generated responses are more accurate and comprehensive. At the same time, the integration of the knowledge base makes knowledge updating and expansion more convenient, ensuring that the question-answering system can be updated at any time based on the latest industry information, so that the model has the ability to continuously learn and improve, and further enhances the practicality and professionalism of the question-answering system.

[0054] In one possible implementation, refer to Figure 3 , Figure 3 This is the third flow chart of the intelligent question-answering method in the field of construction provided by the present invention, such as Figure 3 As shown, the fine-tuning process of the question-answering large model specifically includes steps 301-303: Step 301: Obtain multiple sample question information.

[0055] During the specific implementation process, the system collects sample question information in the field of construction through various channels. These samples may include questions from industry experts, summaries of common problems on construction sites, content from industry Q&A communities, and questions sorted out from relevant documents. These questions can be in various forms, including text descriptions, specific construction issues annotated with images or videos, etc. Through these channels, the system obtains sample question information covering construction specifications, quality control, safety measures, material use, and more. After obtaining the sample question information, the system cleans and organizes these questions, removes redundant content, retains representative and targeted questions, and ensures the quality and diversity of the data. In this process, the system will also annotate the sample questions and add corresponding field labels to make subsequent matching with the knowledge base more efficient and accurate.

[0056] Step 302: Pair all sample question information with all professional knowledge information in the knowledge base to generate multiple question-answer pairs; each question-answer pair includes a sample question information and at least one professional knowledge information.

[0057] In step 302, all sample question information is paired with all professional knowledge information in the knowledge base to generate multiple question-answer pairs, which is the core step in the fine-tuning process of the large question-answer model. The purpose of this is to build a fine-tuning dataset for fine-tuning the model so that the model can learn how to generate appropriate answers to questions in the specific field of construction through these question-answer pairs. By pairing sample question information with professional knowledge, the system can effectively establish an input-output mapping relationship for the large question-answer model, thereby enhancing the model's understanding and reasoning ability of professional construction problems.

[0058] In one possible implementation, refer to Figure 4 , Figure 4 This is a fourth flow chart of the intelligent question-answering method in the field of construction provided by the present invention, such as Figure 4 As shown, step 302 specifically includes steps 401-403: Step 401: semantically analyze each sample question information, and extract at least one sample keyword from each sample question information.

[0059] During the specific implementation process, the system first performs text preprocessing on the input sample question information, which includes removing interference elements such as stop words and punctuation marks, and retaining key information. Next, the system uses natural language processing (NLP) technology to deeply analyze each sample question, such as analyzing the sentence structure and semantic relationship of the question through dependency syntax analysis, entity recognition and other methods. Through these methods, the system can extract key entities, verb phrases and important modifiers in the expression, which constitute the core content of the question. For example, for the sample question "How to ensure the strength of concrete in low-temperature construction?" The system can extract keywords such as "concrete", "low-temperature construction" and "strength" through semantic analysis. These keywords represent the main problem dimensions and situational characteristics that users are concerned about in the question.

[0060] The extracted keywords not only reflect the surface semantics of the user's question, but also understand the actual meaning of these keywords in the question through contextual relationship analysis. In this process, the system will also tag the keywords with parts of speech, and use synonyms and semantic expansion to ensure that the extracted keywords can cover the important information and potential meanings in the question to the greatest extent possible, such as associating "low temperature construction" with "winter construction" and "cold environment" to ensure the comprehensiveness of subsequent matching.

[0061] Step 402: Perform multi-level matching on all sample keywords in each sample question information with parent categories, subcategories and tags corresponding to all professional knowledge information in the knowledge base to obtain multiple matching results.

[0062] In the specific implementation process, the system first obtains the sample keywords extracted in step 401, and then matches these keywords with the knowledge in the knowledge base in turn. The matching process adopts a multi-level strategy to gradually match parent categories, subcategories, and tags to ensure that all relevant content in the knowledge base is covered. Specifically, it can be achieved through the following process: Match all sample keywords in each sample question information with the parent categories corresponding to all professional knowledge information in the knowledge base to determine multiple first matching results; and construct multiple first sub-question-answer pairs based on all first matching results.

[0063] Match all sample keywords in each sample question information with subcategories corresponding to all professional knowledge information in the knowledge base to determine multiple second matching results; and construct multiple second sub-question-answer pairs based on all second matching results.

[0064] Match all sample keywords in each sample question information with labels corresponding to all professional knowledge information in the knowledge base to determine multiple third matching results; and construct multiple third sub-question-answer pairs based on all third matching results.

[0065] The plurality of first question-answer sub-pairs, the plurality of second question-answer sub-pairs, and the plurality of third question-answer sub-pairs together constitute a plurality of matching results.

[0066] In the parent category matching stage, the system first performs a preliminary search for each keyword to determine its corresponding macro field. For example, "concrete" may be classified under the parent category of "material technology", which can quickly narrow the matching scope. After determining the parent category, the system further enters the subcategory matching stage and uses a more detailed classification structure to conduct in-depth matching of keywords. For example, "concrete pouring" can be classified under the subcategory of "concrete construction technology" to further improve the accuracy and pertinence of the matching.

[0067] After completing the matching of parent categories and subcategories, the system will also use the labels annotated for each knowledge information in the knowledge base to perform a multi-dimensional semantic comparison of keywords. Through these labels, you can further find knowledge items that may be semantically related to the keywords in addition to the parent categories and subcategories. For example, if a user's question involves "concrete strength in low-temperature construction", the system will match the knowledge corresponding to relevant labels such as "temperature control" and "antifreeze use" in the knowledge base based on the keyword "low-temperature construction", thereby establishing a more comprehensive association. This process uses text similarity calculations (such as word embedding similarity) and semantic expansion technology to ensure that each keyword can match relevant knowledge to the greatest extent.

[0068] Through this multi-level matching process, the system can find the professional knowledge information that best fits the semantics and domain background of each sample question. The effect of this is that the system can not only quickly locate the macro-classification of sample questions, but also go into detail to find the knowledge items that are most likely to answer the questions, ensuring the quality and representativeness of the question-answer pairs. This multi-level matching method effectively improves the correlation between sample questions and knowledge base information, and then in the process of fine-tuning the large question-answer model, the model can learn how to answer questions at different levels and different semantic backgrounds through these high-quality question-answer pairs. Ultimately, this matching process lays a solid foundation for the professionalism and pertinence of the model, enabling the model to better handle complex problems in the field of construction in practical applications and generate accurate and reference-worthy answers.

[0069] Step 403: Construct multiple question-answer pairs based on all matching results.

[0070] In the specific implementation process, the system first collects all professional knowledge information related to each sample question information according to the matching results in step 402. Each sample question may match knowledge items in multiple knowledge bases, including information under different parent categories, subcategories or tags. These matching items represent knowledge sources that can effectively answer the question. The system selects the most relevant knowledge information from these matching results, taking into account factors such as the similarity score of the knowledge items, contextual relevance, and the authority of the knowledge, to ensure that the final selected answer can provide the most valuable answer to the sample question.

[0071] Next, the system pairs each sample question with the knowledge item it matches to form a question-answer pair. For example, if the sample question is "How to ensure the construction quality of concrete in a low-temperature environment?" The system may match multiple knowledge items such as "Technical standards for low-temperature concrete construction", "How to use antifreeze", and "Quality control measures for low-temperature construction". These items will form a question-answer pair together with the question. When constructing a question-answer pair, the system will also format the answer part to ensure that the answer has an easy-to-understand structure and sufficient details, while retaining relevant field terminology and standard specifications to maintain the professionalism and practicality of the answer.

[0072] Furthermore, the diversity and coverage of the constructed question-answer data can be improved in the following three ways so that it can better match the business needs of the construction industry.

[0073] 1. Use artificial, rule-based and generative pre-trained language models to rewrite the constructed question samples in a variety of ways, such as adjusting the way the question is expressed, expanding the context of the question, etc., so that the training data can better meet the actual needs of the construction field and cover a wider range; 2. Introduce real user conversation data. The real conversation data comes from real user questions and answers collected from professional construction websites, forums, and technical exchange platforms. After screening and cleaning, and de-identifying user information, high-quality real conversation data is obtained. It is further enhanced by the processing method mentioned in 1 to generate question-answer data that is more in line with actual construction business scenarios; 3. Architectural scenario simulation question and answer data, architectural experts design scenario dialogue data. For example, in the construction plan preparation scenario, simulate the multi-round dialogue between the construction engineer and the intelligent question and answer system when the human-machine collaborative preparation of the construction plan. The scenario simulation data comes from: (1) domain expert design, experts sort out key issues and decision logic according to the expert preparation process, form preliminary question and answer pairs, and then design multi-round interaction methods and model output standards according to the plan preparation process. Finally, the generated multi-round question and answer data is verified by experts; (2) model-assisted generation, organize the experience of expert preparation into a standard process, use a large language model combined with expert process data, convert each link into multi-round dialogue content, and perform data enhancement and optimization.

[0074] The above three data construction methods will be used in conjunction in the actual generation process to further enhance the authenticity of the data and make the question-and-answer effect of the model more in line with the actual needs of the construction industry.

[0075] Step 303: Integrate all question-answer pairs into a general large model to obtain a large question-answer model.

[0076] In the specific implementation process, the system first organizes all the question-answer pairs constructed in step 403 into a fine-tuning dataset. These question-answer pairs have been matched and constructed at multiple levels to ensure that they cover a wide range of scenarios and detailed issues in the field of construction, and have high quality and representativeness. The system performs data preprocessing on these question-answer pairs, such as standardizing questions and answers to ensure that their formats meet the input and output requirements of the large model, and screening and weighting according to the corpus quality of the question-answer pairs so that the model can prioritize learning of key knowledge items.

[0077] Next, the system inputs these processed question-answer pairs into the general large model for fine-tuning. The general large model is usually a pre-trained natural language understanding model that has general language understanding and reasoning capabilities, but lacks professional knowledge in the field of construction. The fine-tuning process uses supervised learning to optimize the parameters of the model through question-answer pairs as training data, so that when it receives questions in the field of construction, it can use professional knowledge to generate accurate answers. During the fine-tuning process, the model continuously adjusts its internal weights through back propagation, so that when faced with similar construction problems, it can output professional answers that are consistent with the content of the knowledge base. In addition, the model will combine the semantic features of the question-answer pairs to learn common terms, technical standards, and construction scenarios in the field of construction, and gradually enhance its knowledge mastery and problem-solving capabilities in specific fields.

[0078] Through this fine-tuning process, the system finally generated a large question-answering model suitable for the field of construction. The effect of this is that the fine-tuned large question-answering model not only retains the powerful ability of the original general large model in natural language understanding, but also has the professionalism to deal with specific problems in the field of construction. When users ask questions related to construction, the model can effectively understand its semantics and generate authoritative and targeted answers based on the learned domain knowledge. Through this fine-tuning process, the model has the actual ability to be applied in construction scenarios, and can provide users with professional advice in many aspects from construction quality control to material selection, safety protection, etc., effectively improving the practicality and user experience of the question-answering system.

[0079] Based on the above steps, in a possible implementation manner, referring to Figure 5 , Figure 5 FIG. 5 is a flowchart of the intelligent question-answering method in the field of construction provided by the present invention. Figure 5 As shown, in actual application, step 102 specifically includes steps 501-503: Step 501: Input the question information into the question-answering model, and obtain at least one initial answer associated with the question information through all question-answer pairs.

[0080] In step 501, the question information is input into the question-answering model, and at least one initial answer associated with the question information is obtained through all question-answer pairs. The purpose of this process is to generate multiple possible answers to the user's question by using the fine-tuned question-answering model. This helps the model to make a diversified understanding of the question, which facilitates the subsequent screening of the most accurate final answer that meets the user's needs.

[0081] In one possible implementation, refer to Figure 6 , Figure 6 This is the sixth flow chart of the intelligent question-answering method in the field of construction provided by the present invention, such as Figure 6 As shown, step 501 specifically includes steps 601-605: Step 601: perform semantic analysis on each question information and extract at least one keyword from each question information.

[0082] The specific implementation method of step 601 is consistent with the specific implementation method of step 401, so it will not be described in detail here.

[0083] Step 602: Based on each keyword, determine whether there is matching professional knowledge information in all professional knowledge information contained in all question-answer pairs; the text similarity between any one of the parent categories, subcategories or tags corresponding to the matching professional knowledge information and the keyword is greater than a preset similarity threshold.

[0084] In the specific implementation process, the system first uses the keywords extracted in step 601 and performs multi-level matching of each keyword with all the professional knowledge information in the knowledge base, including different dimensions such as parent categories, subcategories, and tags. Specifically, the system performs the following matching on each keyword in turn: Parent category matching: The system first matches the keyword with its parent category. The purpose of this matching process is to quickly narrow the scope of the match and ensure that the answer is only found in the general category of knowledge that is semantically similar to the keyword. For example, for the keyword "low temperature construction", the system first matches it to the general category of "construction environment management" or "concrete construction" in the knowledge base, so as to focus on the knowledge area that is most relevant to the question.

[0085] Subcategory matching: After the parent category is successfully matched, the system will further match the keywords with the subcategories in the knowledge base to further refine the knowledge scope. For example, "low temperature construction" can match the subcategory "low temperature concrete construction technology" under the parent category, thereby further limiting the search scope to more specific knowledge nodes.

[0086] Tag matching: Finally, the system will use the tags annotated for each professional knowledge information in the knowledge base to match the keywords. This level of matching helps to find knowledge items that are closely related to the keywords but not necessarily identical. For example, the keyword "low temperature construction" may also match tags such as "temperature control" and "antifreeze use", making knowledge retrieval more flexible and diverse.

[0087] During the matching process, the system uses text similarity calculations, such as cosine similarity or semantic similarity based on word embedding, to evaluate the similarity between keywords and professional knowledge information to determine whether they match. If the parent category, subcategory or label corresponding to the keyword and the knowledge item in the knowledge base reaches a preset similarity threshold in the similarity calculation, it is determined that there is matching professional knowledge information.

[0088] Step 603: When it is determined that there is matching professional knowledge information, each matching professional knowledge information is used as an initial answer.

[0089] In the specific implementation process, in step 602, the system has found professional knowledge information related to the user's question through keyword matching, including the matched parent category, subcategory and tag content. When it is confirmed that the keyword and certain knowledge items in the knowledge base reach the preset similarity threshold, the system will directly extract these matching knowledge items and use them as initial answers. Each initial answer represents a possible solution direction and provides corresponding domain knowledge based on the content that has been labeled and classified in the knowledge base.

[0090] In this way, the system can generate one or more initial answers for each user's question. These answers are directly based on the matched professional knowledge information and have high domain relevance and knowledge depth. The effect of this is that the initial answer construction process provides a high-quality foundation for the screening and optimization of subsequent answers, so that when the system faces professional problems in the field of construction, it can rely on authoritatively verified knowledge information and directly provide answers with reference value.

[0091] Step 604: When it is determined that there is no matching professional knowledge information, at least one relevant professional knowledge information is screened from all the professional knowledge information contained in all question-answer pairs; the text similarity between any one of the parent categories, subcategories or tags corresponding to the relevant professional knowledge information and the keyword is greater than the text similarity between any one of the parent categories, subcategories or tags corresponding to the remaining professional knowledge information and the keyword.

[0092] In the specific implementation process, when the system fails to find professional knowledge information with a similarity greater than a preset threshold with the question keyword in step 602, the system will adopt a second-best strategy and select at least one professional knowledge item closest to the keyword from the knowledge base based on the principle of the highest relative similarity. The system uses a text similarity calculation method, such as cosine similarity based on a word vector model, to calculate the similarity between each keyword and each professional knowledge item in the knowledge base to find the most semantically consistent content. In this process, the system will evaluate the similarity between the parent category, subcategory or label corresponding to the keyword and the knowledge item one by one, and sort them according to the size of the similarity.

[0093] In the specific screening, the system first performs a preliminary similarity screening based on the parent category, and includes knowledge items with a certain degree of similarity to the keywords into the candidate set. Then it further performs a refined screening based on the similarity of subcategories and tags, and finally selects the items with the highest similarity as the relevant professional knowledge information. For example, if a user asks "How to maintain the strength of concrete in a high humidity environment?", if there is no completely matching knowledge information in the knowledge base, the system may find related items such as "concrete humidity control" or "construction environment management" and provide these contents as alternative answers.

[0094] Step 605: Generate an initial answer based on each relevant professional knowledge information.

[0095] During the specific implementation process, the system first extracts and organizes the most relevant professional knowledge information screened out in step 604 to ensure that this information has sufficient reference value and can form the basis for the answer. The system processes the relevant knowledge information, including streamlining the text content, removing redundant and irrelevant parts, and reorganizing the key content to ensure that the structure of the answer is clear and easy to understand. In this process, the system will also associate and emphasize the key words mentioned in the question and the key points in the knowledge information based on the characteristics of the user's question to improve the pertinence of the answer. For example, if the user's question involves "concrete construction in a high humidity environment", and the relevant knowledge information screened out is mainly about "the impact of humidity on building materials", the system will focus the answer on how to control the impact of humidity on concrete performance, so as to fit the user's question intention as much as possible.

[0096] When generating the initial answer, the system will structure the answer based on the domain terminology and existing construction standards to make it conform to the answer specifications in the construction field. For example, the system may divide the answer into several parts such as "influencing factors", "treatment measures", and "recommended practices" so that users can quickly obtain key information. The content of the answer strives to be concise and professional, and highlights the most relevant content to the question to improve the practicality of the answer.

[0097] Through this generation process, the system can generate at least one piece of information based on existing relevant knowledge as an initial answer to the user's question. The effect of this is that even when the question-answering system does not find a completely matching knowledge entry, it can still provide the user with a reasonable answer, avoiding the situation of missing answers to the greatest extent, and ensuring the continuity and practicality of the system. This method of generating initial answers enables the system to respond to diverse questions in the field of construction, and even in the face of scenarios that are not fully covered by the knowledge base, it can provide users with answers with reference value, thereby effectively improving the user experience and the application value of the system.

[0098] Step 502: Determine the output weight corresponding to each initial answer.

[0099] In the specific implementation process, the system first classifies all the initial answers generated in step 501 and step 605, and assigns corresponding weights according to the professional knowledge categories and tags involved in the initial answers. For example, the initial answers may involve different parent categories, such as "construction quality management", "safety protection", "material selection", etc., as well as further subcategories and detailed tags, such as "safety hazards at construction sites", "concrete strength control", etc. The system adjusts the weight of each answer based on these categories and tags, thereby quantifying their importance in answering user questions.

[0100] In the weight distribution process, the system mainly considers the following two aspects: Category weight adjustment: Based on the actual needs of the construction field, different parent categories and subcategories may have different importance for answering questions. For example, if a user asks a question about safety issues at a construction site, the system will assign a higher weight to the initial answer in the "safety protection" category to highlight the priority of safety-related content. For example, when the system recognizes that there are keywords related to "site safety" or "hidden dangers" in the question, the answer from the "safety protection" category will be given a higher weight.

[0101] Label weight adjustment: In addition to adjusting the categories, the system will further evaluate the importance of the answers based on the labels of each knowledge item in the knowledge base. Labels can identify knowledge information in a more granular manner. For example, an initial answer may have labels such as "quality inspection", "temperature control", and "construction standards". The system will adjust the weights of relevant labels based on the specific context and needs of the user's question to ensure that the content most closely related to the user's question receives a higher weight. For example, when a user mentions "winter construction" in a question, the system will increase the weight of answers with the label "temperature control" to ensure that the answer can effectively address the impact of low temperature environments on construction quality.

[0102] By adjusting the comprehensive weights of categories and labels, the system calculates a final output weight for each initial answer to quantify the importance of the answer in the entire question answer. This weight distribution can reflect the relevance of each initial answer to the user's question and the priority of the answer, ensuring that these initial answers can be reasonably sorted in subsequent steps to select the best answer.

[0103] Step 503: sort all initial answers according to all output weights, and integrate the sorted initial answers to generate reply answers.

[0104] In the specific implementation process, the system first sorts all the initial answers according to the output weight calculated for each initial answer in step 502. The sorting is based on the output weight of each initial answer. The higher the weight, the more relevant the answer is to the question and the more professional it is. In the sorting process, the system will give priority to answers with higher weights so that they occupy a core position in the final reply.

[0105] After the sorting is completed, the system will integrate all the initial answers to generate the final response answer. During the integration process, the system will combine the content of different initial answers, remove duplicate information, and sort out the logic of the answers to ensure that the content of the final response is complete, coherent and easy to understand. For example, for a user's question "How to ensure the construction quality of concrete in winter?", the system may get multiple initial answers, involving "using antifreeze", "appropriately extending the curing time", "covering insulation materials" and other aspects. When integrating these answers, the system will remove content redundancy and organize different measures in a logically reasonable order, and finally generate a comprehensive response answer covering a variety of effective measures.

[0106] Through this sorting and integration process, the system can generate a highly targeted and reference-worthy response to the user's question. The effect of this is that by weighting and integrating the various initial answers, the system can not only ensure the professionalism and accuracy of the answers, but also make the final answer structure more complete and logically clear, making it easier for users to understand and adopt.

[0107] In one possible implementation, refer to Figure 7 , Figure 7 FIG. 7 is a flow chart of the intelligent question-answering method in the field of construction provided by the present invention. Figure 7 As shown, the method also includes the following steps: Step 701: Obtain the user's historical question information and interaction record information, and obtain the position information corresponding to the user account.

[0108] Step 702: perform keyword extraction and semantic relevance analysis on the current question information to obtain multiple current keywords and current semantic representations, and perform keyword extraction and semantic relevance analysis on the historical question information to obtain multiple historical keywords and multiple historical semantic representations.

[0109] Step 703: cluster all historical keywords and all historical semantic expressions to generate multiple interest topics; each historical question information corresponds to one interest topic.

[0110] Step 704: Determine the interest weight corresponding to each interest topic according to the historical question time corresponding to each historical question information and the time decay strategy.

[0111] Step 705: Generate long-term interest information according to each interest topic and the interest weight corresponding to each interest topic.

[0112] Step 706: Generate short-term interest information based on the interaction record information, all current keywords and all current semantic expressions.

[0113] Step 707: Determine job-related knowledge requirement information based on job information.

[0114] Step 708: Generate corresponding recommendation questions based on the long-term interest information, short-term interest information and job-related knowledge requirement information.

[0115] In steps 701-708, by obtaining the user's multi-dimensional information, analyzing the question content and behavior records, and combining the job requirements in the construction field and dynamic interest modeling, the system can generate recommended questions related to the user's questions, providing users with a more personalized and professional knowledge exploration path. This process aims to comprehensively improve the pertinence and practicality of recommended questions from the dual dimensions of long-term and short-term interests and the professional perspective of job requirements.

[0116] In step 701, the system first obtains the user's historical question information and interaction record information, and combines it with the job information corresponding to the user account. The purpose of this step is to provide multi-dimensional basic data for subsequent interest modeling and recommendation question generation, where historical questions reflect the knowledge points that users have been paying attention to for a long time, and interaction records reflect the user's recent behavior characteristics. For example, the user gave positive feedback (likes) to the answer to the question "concrete maintenance". Job information is directly related to professional needs in the field of construction (such as construction managers, safety officers, materials engineers, etc.). Through the collection of this comprehensive information, the system can comprehensively portray user portraits.

[0117] In step 702, the system performs keyword extraction and semantic relevance analysis on the current question information and the historical question information, respectively, to generate multiple current keywords and current semantic representations, as well as multiple historical keywords and historical semantic representations. Through natural language processing technology, the system can identify the core content of the user's question and construct a deep representation of the user's question in combination with the contextual semantics. For example, the keywords of the user's current question "How to prevent concrete cracks?" may be "concrete", "cracks", and "prevention", and its semantic representation reflects the user's focus on construction quality management.

[0118] In step 703, the system performs topic clustering on all historical keywords and historical semantic representations, generates multiple interest topics, and maps each historical question information to an interest topic. Through cluster analysis, the system can classify user questions into specific knowledge areas, such as "quality management" and "construction material selection", reflecting the long-term interest distribution of users in different topics in the field of construction.

[0119] Then, in step 704, the system dynamically adjusts the interest weight of each interest topic based on the question time of each historical question information and the time decay strategy. Newer questions or frequently asked topics are given higher weights, while older questions gradually decay their weights. Through this dynamic adjustment mechanism, the system can highlight topics that users have paid more attention to recently, thereby improving the timeliness and pertinence of interest modeling.

[0120] In step 705, the system generates the user's long-term interest information by combining the interest topics and their weights. This step presents the user's long-term accumulated interest points in a structured manner, highlighting the user's core knowledge needs. For example, the long-term interest information may include the user's accumulated attention preferences in the fields of "construction safety" and "construction quality management".

[0121] Then, in step 706, the system generates short-term interest information based on the interaction record information, current keywords and current semantic representation. Short-term interest information reflects the association between the user's current question content and its real-time behavior characteristics. For example, when the user asks "concrete insulation in low-temperature construction", the short-term interest dynamically generated by the system may focus on the topic of "winter construction specifications".

[0122] In step 707, the system uses the user's job information to determine job-related knowledge demand information, combining the user's interests with their professional needs in the field of construction. For example, the job information of a project manager may be associated with "construction progress control" and "resource optimization", while the job information of a safety officer may be associated with "safety hazard investigation" and "emergency measures".

[0123] Finally, in step 708, the system integrates the long-term interest information, short-term interest information and job-related knowledge requirement information to generate recommendation questions related to the question information. The generation of recommendation questions is achieved through a multi-level semantic matching strategy, combining the matching degree of keywords and semantic representations with the tags, categories and upper and lower relationships in the architectural knowledge base, and at the same time assigning different weights to long-term interests, short-term interests and job requirements to expand the scope of topics that users are concerned about.

[0124] For example, for the user's current question "What issues should be paid attention to when pouring concrete?", the system obtains historical question information including: "Under what conditions is it necessary to use high-performance concrete?"; "What is the best way to maintain concrete?"; "What are the precautions for the treatment of construction joints?". The user interaction record obtained is: positive feedback (likes) was given to the answer to the question "Concrete maintenance". Then the system may generate the following recommended questions: 1. "How to choose a suitable concrete mix ratio?" (related to the material selection requirements of the user's position); 2. "After pouring concrete, how to ensure that the strength meets the standard?" (an extension question of short-term interest); 3. "How to avoid air inclusion during pouring?" (a question directly related to the current question); 4. "What construction precautions are there when using different types of concrete?" (an extension question of long-term interest).

[0125] Through the above steps, the system can generate multi-dimensional recommended questions based on the knowledge areas that users have long been concerned about, the real-time needs of current questions, and information related to job responsibilities. The effect of this is that it not only expands the depth and breadth of user knowledge acquisition, but also ensures the personalization and specialization of recommended content, significantly improving the user experience and the practical application value of the question-answering system.

[0126] Reference Figure 8 , Figure 8 : is a schematic diagram of the structure of the intelligent question-answering system in the field of construction provided by the present invention, the system comprises: A receiving module, used for receiving question information input by a user; The processing module is used to input the question information into the question-answering big model and obtain the reply answer output by the question-answering big model; wherein the question-answering big model is obtained by fine-tuning the sample question information and the knowledge base containing the professional knowledge in the field of construction, and the knowledge base is used to assist the question-answering big model to generate the reply answer according to the sample question information.

[0127] In a possible implementation, the system further includes a knowledge base construction module, which is used to: Get multiple professional knowledge information; Classify all professional knowledge information and determine the parent category corresponding to each professional knowledge information; On the basis of the parent category, each professional knowledge information is further refined and classified to determine the subcategory corresponding to each professional knowledge information; Label each piece of professional knowledge information with at least one tag; According to the classification results and labeling results, all professional knowledge information is integrated to build a knowledge base.

[0128] In a possible implementation, the system further includes a training module, which is used to: Get multiple sample question information; Pair all sample question information with all professional knowledge information in the knowledge base to generate multiple question-answer pairs; each question-answer pair includes a sample question information and at least one professional knowledge information; Integrate all question-answer pairs into a general large model to obtain a question-answer large model.

[0129] In a possible implementation, the training module is further used to: Perform semantic analysis on each sample question information, and extract at least one sample keyword from each sample question information; Perform multi-level matching of all sample keywords in each sample question information with the parent categories, subcategories, and labels corresponding to all professional knowledge information in the knowledge base to obtain multiple matching results; Based on all matching results, multiple question-answer pairs are constructed.

[0130] In a possible implementation manner, the processing module is further configured to: Input the question information into the question-answering model, and obtain at least one initial answer associated with the question information through all question-answer pairs; Determine the output weight corresponding to each initial answer; All initial answers are sorted according to all output weights, and the sorted initial answers are integrated to generate a reply answer.

[0131] In a possible implementation manner, the processing module is further configured to: Perform semantic analysis on each question information and extract at least one keyword from each question information; Based on each keyword, determine whether there is matching professional knowledge information in all professional knowledge information contained in all question-answer pairs; the text similarity between any one of the parent categories, subcategories or tags corresponding to the matching professional knowledge information and the keyword is greater than a preset similarity threshold; When it is determined that there is matching professional knowledge information, each matching professional knowledge information is used as an initial answer; When it is determined that there is no matching professional knowledge information, at least one relevant professional knowledge information is selected from all professional knowledge information contained in all question-answer pairs; the text similarity between any one of the parent categories, subcategories or tags corresponding to the relevant professional knowledge information and the keyword is greater than the text similarity between any one of the parent categories, subcategories or tags corresponding to the remaining professional knowledge information and the keyword; Generate an initial answer based on each relevant expertise information.

[0132] In a possible implementation manner, the acquisition module is further used to acquire the user's historical question information and interaction record information, and acquire the position information corresponding to the user account; The processing module is also used to: Perform keyword extraction and semantic relevance analysis on the current question information to obtain multiple current keywords and current semantic representations, and perform keyword extraction and semantic relevance analysis on the historical question information to obtain multiple historical keywords and multiple historical semantic representations; Cluster all historical keywords and all historical semantic representations to generate multiple interest topics; each historical question information corresponds to an interest topic; According to the historical question time corresponding to each historical question information, the interest weight corresponding to each interest topic is determined according to the time decay strategy; Generate long-term interest information according to each interest topic and the interest weight corresponding to each interest topic; Generate short-term interest information based on the interaction record information, all current keywords and all current semantic representations; Determine the job-related knowledge requirement information based on job information; Generate corresponding recommendation questions based on long-term interest information, short-term interest information and job-related knowledge requirement information.

[0133] It should be noted that the intelligent question and answer system in the field of construction provided by the present invention can execute the intelligent question and answer method in the field of construction of any of the above-mentioned embodiments during specific operation, which will not be elaborated in this embodiment.

[0134] Fig. 9 is a schematic diagram of the structure of the electronic device provided by the present invention, such as Fig. 9 As shown, the electronic device may include: a processor 910 (processor), a communication interface 920 (CommunicationsInterface), a memory 930 (memory) and a communication bus 940, wherein the processor 910, the communication interface 920, and the memory 930 communicate with each other through the communication bus 940. The processor 910 may call the logic instructions in the memory 930 to execute the intelligent question-answering method in the field of construction, the method comprising: receiving the question information input by the user; inputting the question information into the question-answering big model, and obtaining the reply answer output by the question-answering big model; wherein the question-answering big model is obtained by fine-tuning the sample question information and the knowledge base containing the professional knowledge in the field of construction, and the knowledge base is used to assist the question-answering big model in generating the reply answer according to the sample question information.

[0135] In addition, the logic instructions in the above-mentioned memory 930 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on this understanding, the technical solution of the present invention can be essentially or partly embodied in the form of a software product that contributes to the prior art. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods of various embodiments of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc. Various media that can store program codes.

[0136] On the other hand, the present invention also provides a computer program product, which includes a computer program stored on a non-transitory computer-readable storage medium, and the computer program includes program instructions. When the program instructions are executed by a computer, the computer can execute the intelligent question and answer method in the field of construction provided by the above-mentioned embodiments, and the method includes: receiving question information input by a user; inputting the question information into a question and answer big model, and obtaining a reply answer output by the question and answer big model; wherein the question and answer big model is obtained by fine-tuning sample question information and a knowledge base containing professional knowledge in the field of construction, and the knowledge base is used to assist the question and answer big model in generating a reply answer based on the sample question information.

[0137] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to execute the intelligent question and answer method in the field of construction provided by the above-mentioned embodiments, the method comprising: receiving question information input by a user; inputting the question information into a question and answer big model, and obtaining a reply answer output by the question and answer big model; wherein the question and answer big model is obtained by fine-tuning sample question information and a knowledge base containing professional knowledge in the field of construction, and the knowledge base is used to assist the question and answer big model in generating a reply answer based on the sample question information.

[0138] The system embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, i.e., they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Those of ordinary skill in the art may understand and implement it without creative effort.

[0139] Through the description of the above implementation modes, those skilled in the art can clearly understand that each implementation mode can be implemented by means of software plus a necessary general hardware platform, or of course by hardware. Based on such an understanding, the above technical solution can essentially or in other words be embodied in the form of a software product that contributes to the prior art. The computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a magnetic disk, an optical disk, etc., and includes a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods of each embodiment or some parts of the embodiment.

[0140] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. An intelligent question-answering method in the field of building construction, characterized in that: include: Receive question information input by the user; Input the question information into the question-answering model to obtain a reply answer output by the question-answering model; Among them, the question-answering big model is obtained by fine-tuning sample question information and a knowledge base containing professional knowledge in the field of construction. The knowledge base is used to assist the question-answering big model in generating response answers based on the sample question information.

2. The intelligent question-answering method in the field of construction according to claim 1, characterized in that: The method further comprises: Get multiple professional knowledge information; Classifying all the professional knowledge information and determining the parent category corresponding to each piece of professional knowledge information; Based on the parent category, each piece of professional knowledge information is further classified to determine a subcategory corresponding to each piece of professional knowledge information; Marking at least one label for each piece of professional knowledge information; According to the classification results and labeling results, all the professional knowledge information is integrated to construct the knowledge base.

3. The intelligent question-answering method in the field of construction according to claim 2, characterized in that: The fine-tuning process of the large question-answering model includes: Get multiple sample question information; Pairing all the sample question information with all the professional knowledge information in the knowledge base to generate a plurality of question-answer pairs; each question-answer pair includes one sample question information and at least one professional knowledge information; All the question-answer pairs are integrated into a general large model to obtain the question-answer large model.

4. The intelligent question-answering method in the field of construction according to claim 3, characterized in that: The step of pairing all the sample question information with all the professional knowledge information in the knowledge base to generate a plurality of question-answer pairs specifically includes: Performing semantic analysis on each of the sample question information, and extracting at least one sample keyword from each of the sample question information; Perform multi-level matching of all sample keywords in each sample question information with parent categories, subcategories, and tags corresponding to all professional knowledge information in the knowledge base to obtain multiple matching results; Based on all the matching results, a plurality of question-answer pairs are constructed.

5. The intelligent question-answering method in the field of construction according to claim 3, characterized in that: Inputting the question information into the question-answering model to obtain a reply answer output by the question-answering model specifically includes: Inputting the question information into the question-answering macromodel, and obtaining at least one initial answer associated with the question information through all the question-answer pairs; Determine the output weight corresponding to each of the initial answers; All of the initial answers are sorted according to all of the output weights, and the sorted initial answers are integrated to generate the reply answer.

6. The intelligent question-answering method in the field of construction according to claim 5, characterized in that: The obtaining of at least one initial answer associated with the question information through all the question-answer pairs specifically includes: Performing semantic analysis on each of the question information, and extracting at least one keyword from each of the question information; According to each of the keywords, in all the professional knowledge information contained in all the question-answer pairs, it is determined whether there is matching professional knowledge information; the text similarity between any one of the parent category, subcategory or label corresponding to the matching professional knowledge information and the keyword is greater than a preset similarity threshold; When it is determined that there is matching professional knowledge information, each piece of matching professional knowledge information is used as an initial answer; When it is determined that there is no matching professional knowledge information, at least one relevant professional knowledge information is selected from all the professional knowledge information contained in all the question-answer pairs; the text similarity between any one of the parent categories, subcategories or tags corresponding to the relevant professional knowledge information and the keyword is greater than the text similarity between any one of the parent categories, subcategories or tags corresponding to the remaining professional knowledge information and the keyword; Based on each of the related professional knowledge information, an initial answer is generated.

7. The intelligent question-answering method in the field of construction according to claim 1, characterized in that: The method further comprises: Obtain the user's historical question information and interaction record information, and obtain the position information corresponding to the user account; Performing keyword extraction and semantic relevance analysis on the current question information to obtain a plurality of current keywords and current semantic representations, and performing keyword extraction and semantic relevance analysis on the historical question information to obtain a plurality of historical keywords and a plurality of historical semantic representations; Clustering all the historical keywords and all the historical semantic expressions to generate multiple interest topics; each of the historical question information corresponds to one interest topic; Determine the interest weight corresponding to each interest topic according to the historical question time corresponding to each historical question information and according to the time decay strategy; generating long-term interest information according to each of the interest topics and the interest weight corresponding to each of the interest topics; Generate short-term interest information according to the interaction record information, all the current keywords and all the current semantic expressions; Determine job-related knowledge requirement information based on the job information; Generate corresponding recommendation questions according to the long-term interest information, the short-term interest information and the job-related knowledge requirement information.

8. An intelligent question-answering system in the field of construction, characterized in that: include: A receiving module, used for receiving question information input by a user; A processing module is used to input the question information into the question-answering big model to obtain a reply answer output by the question-answering big model; wherein the question-answering big model is obtained by fine-tuning sample question information and a knowledge base containing professional knowledge in the field of construction, and the knowledge base is used to assist the question-answering big model in generating a reply answer based on the sample question information.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that: When the processor executes the computer program, the intelligent question-answering method in the field of construction as described in any one of claims 1 to 7 is implemented.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the intelligent question-answering method in the field of construction is implemented as described in any one of claims 1 to 7.

Images