A method for querying building design schemes and related equipment
By constructing prompt text for a large language model and performing query processing, the accuracy problem caused by the large number of language model parameters is solved, and more accurate building design scheme output is achieved.
Patent Information
- Application Number
- CN202410776970.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-06-17
- Publication Date
- 2026-11-17
- Estimated Expiration
- 2044-06-17
AI Technical Summary
Existing language models have a large number of parameters, making it difficult to accurately output text that meets user needs.
By acquiring consultation dialogue data and building design scheme databases, prompt text is constructed for a large language model, and then input into the large language model for query processing to output the target design scheme.
It solves the accuracy problem caused by the large number of language model parameters, and can output building design solutions that meet user needs more accurately.
Smart Images

Figure CN118779418B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of artificial intelligence technology, and in particular to a method and related equipment for querying building design schemes. Background Technology
[0002] Large language models are a type of artificial intelligence model designed to understand and generate human language. They typically involve iteratively training the model by collecting user feedback on their responses, aiming to output more accurate and user-relevant replies. Currently, many businesses are starting to use language models. For example, search engines are incorporating language models to analyze users' true search needs; voice assistants are using language models to provide more intelligent responses to user questions; and intelligent customer service uses language models to analyze user questions and provide more personalized answers. However, currently, due to the large number of parameters in language models, it is difficult to accurately output text that meets user needs. Summary of the Invention
[0003] This invention provides a method for querying building design schemes, aiming to solve the problem in existing methods where the language model has a large number of parameters, making it difficult to accurately output text that meets user needs. By constructing prompt text for a large language model based on consultation dialogue data and a building design scheme database, and inputting the prompt text into the large language model for query processing, the target design scheme output by the large language model is obtained and returned to the customer. This method effectively solves the problem of existing methods having a large number of language model parameters, making it difficult to accurately output text that meets user needs.
[0004] In a first aspect, embodiments of the present invention provide a method for querying building design schemes, the method comprising the following steps:
[0005] Acquire consultation dialogue data, which includes the customer's first dialogue data and the customer service representative's second dialogue data;
[0006] Based on the consultation dialogue data and the building design scheme database, a prompt text for a large language model is constructed. The prompt text includes a design scheme result output request. The building design scheme database includes multiple building design units that constitute a building design scheme.
[0007] The prompt text is input into the large language model for query processing to obtain the target design scheme output by the large language model.
[0008] The target design solution is then returned to the client.
[0009] Optionally, the consultation dialogue data is multimodal data. The first dialogue data includes at least one modality of text data, visual data, and speech data. The second dialogue data includes at least one modality of text data, visual data, and speech data. The step of constructing prompt text for a large language model based on the consultation dialogue data and the building design scheme database includes:
[0010] If the first dialogue data includes the visual data and / or the voice data, then the visual data and / or the voice data are prompted to obtain the first data to be searched;
[0011] If the second dialogue data includes the visual data and / or the voice data, then the context data of the visual data and / or the voice data is retrieved from the first dialogue data, and based on the context data, it is determined whether the visual data and / or the voice data is the second data to be searched.
[0012] Perform semantic search on the first data to be searched and / or the second data to be searched to obtain multiple semantically related reference data;
[0013] Based on multiple reference data, consultation dialogue data, and building design scheme database, a prompt text for a large language model is constructed.
[0014] Optionally, the step of determining whether the visual data and / or the voice data is the second search data based on the context data includes:
[0015] Semantic extraction is performed on the context data to obtain the semantics of the context data;
[0016] Determine whether the customer approves of the video data and / or the voice data based on the semantics of the context data;
[0017] If it is determined that the customer approves of the video data and / or the voice data, then the video data and / or the voice data are determined as the second data to be searched;
[0018] If it is determined that the customer does not approve of the video data and / or the voice data, then it is determined that the video data and / or the voice data are not the second data to be searched.
[0019] Optionally, the step of performing a semantic search on the first search data and / or the second search data to obtain semantically related parameter data includes:
[0020] The first search data and / or the second search data are semantically searched using the large language model to obtain multiple candidate data. Each candidate data corresponds to a confidence score output by the large language model and a semantic relevance score to the first search data and / or the second search data.
[0021] Based on the confidence level and the semantic relevance, multiple reference data are determined from the multiple candidate data, wherein the number of reference data is less than or equal to the number of candidate data.
[0022] Optionally, the step of constructing the prompt text of the large language model based on multiple reference data, the consultation dialogue data, and the building design scheme database includes:
[0023] Obtain a prompt template, which includes a reference data input area, a consultation dialogue input area, a database input area, and a preset design scheme result output request;
[0024] The multiple reference data are sorted according to the confidence level and the semantic relevance, and then sequentially filled into the reference data filling area; and
[0025] Fill in the consultation dialogue data in the consultation dialogue input area; and
[0026] Enter the access method for the building design scheme database in the database input area;
[0027] Once completed, you will receive the prompt text from the large language model.
[0028] Optionally, the step of inputting the prompt text into the large language model for query processing to obtain the target design scheme output by the large language model includes:
[0029] The prompt text is input into the large language model, and the semantics and intent of the prompt text are extracted through the large language model;
[0030] The large language model accesses the building design scheme database based on the semantics and intent of the prompt text to find the target building design unit set, which includes multiple target building design units.
[0031] The target building design units in the target building design unit set are combined into at least one target design scheme using the large language model, and at least one target design scheme is output.
[0032] Optionally, before the step of constructing the prompt text for the eye model based on the consultation dialogue data and the building design scheme database, the method further includes:
[0033] Obtain multiple sample building design schemes;
[0034] Multiple sample building design schemes are divided into units to obtain multiple building design units;
[0035] A building design scheme database is constructed based on multiple building design units.
[0036] Secondly, embodiments of the present invention also provide a building design scheme query device, the building design scheme query device comprising:
[0037] The first acquisition module is used to acquire consultation dialogue data, which includes the customer's first dialogue data and the customer service's second dialogue data.
[0038] The first construction module is used to construct prompt text for a large language model based on the consultation dialogue data and the building design scheme database. The prompt text includes a design scheme result output request. The building design scheme database includes multiple building design units that constitute a building design scheme.
[0039] The query processing module is used to input the prompt text into the large language model for query processing, and obtain the target design scheme output by the large language model.
[0040] The sending module is used to return the target design scheme to the customer.
[0041] Thirdly, embodiments of the present invention provide an electronic device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps in the building design scheme query method provided in embodiments of the present invention.
[0042] Fourthly, embodiments of the present invention provide a computer-readable storage medium storing a computer program, wherein the computer program, when executed by a processor, implements the steps in the building design scheme query method provided in embodiments of the present invention.
[0043] In this embodiment of the invention, consultation dialogue data is acquired, including first dialogue data from the customer and second dialogue data from the customer service representative. Based on the consultation dialogue data and a building design scheme database, a prompt text for a large language model is constructed. The prompt text includes a request for the output of a design scheme result. The building design scheme database includes multiple building design units constituting a building design scheme. The prompt text is input into the large language model for query processing to obtain the target design scheme output by the large language model. The target design scheme is then returned to the customer. By constructing a prompt text for a large language model based on consultation dialogue data and a building design scheme database, and inputting the prompt text into the large language model for query processing to obtain the target design scheme output by the large language model, and then returning the target design scheme to the customer, the problem of existing methods having a large number of language model parameters, making it difficult to accurately output text that meets user needs, can be solved. Attached Figure Description
[0044] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0045] Figure 1 This is a flowchart of a building design scheme query method provided by an embodiment of the present invention;
[0046] Figure 2 This is a schematic diagram of the structure of a building design scheme query device provided in an embodiment of the present invention;
[0047] Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation
[0048] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0049] like Figure 1 As shown, Figure 1 This is a flowchart illustrating a building design scheme query method provided in an embodiment of the present invention. The building design scheme query method includes the following steps:
[0050] 101. Obtain consultation dialogue data.
[0051] In this embodiment of the invention, the consultation dialogue data includes the customer's first dialogue data and the customer service's second dialogue data.
[0052] The aforementioned first dialogue data can be understood as the questions or needs raised by the customer during the consultation process, which can be text data, visual data, and voice data.
[0053] The aforementioned second dialogue data can be understood as the answers, suggestions, or solutions given by customer service to customers' questions or needs, which can be text data, visual data, and voice data.
[0054] The above consultation dialogue data can be all the communication content generated between the customer and customer service during the consultation process.
[0055] It should be noted that consultation dialogue data can be used as a text dataset for natural language processing models.
[0056] 102. Based on consultation dialogue data and building design scheme database, construct prompt text for a large language model.
[0057] In this embodiment of the invention, the above-mentioned prompt text includes a design scheme result output request, and the above-mentioned building design scheme database includes multiple building design units that constitute a building design scheme.
[0058] The aforementioned large language models can be machine learning models capable of processing massive amounts of data and performing various complex tasks, such as natural language processing, computer vision, and speech recognition. Large language models learn complex patterns and structures in language through training, thereby generating coherent and natural text. In consultation dialogue scenarios, large language models can help the system understand the user's questions or needs and provide relevant suggestions and answers.
[0059] Prompt text for a large language model can be constructed based on consultation dialogue data and building design scheme database.
[0060] Data preprocessing can be performed on consultation dialogue data and building design scheme databases to construct prompt text for a large language model, so that the model can respond with corresponding text, voice, and visual data based on the text, voice, and visual data proposed by the user during use.
[0061] The above preprocessing can be methods such as text extraction, semantic extraction, and visual extraction.
[0062] 103. Input the prompt text into the large language model for query processing to obtain the target design scheme output by the large language model.
[0063] In this embodiment of the invention, the prompt text can be input into a large language model for query processing. The semantics and intent of the prompt text can be extracted through the large language model. Based on the semantics and intent of the prompt text, the building design scheme database can be accessed to find the target building design unit set. Then, the target building design unit set can be retrieved through the large language model, and at least one target design scheme can be output.
[0064] The aforementioned set of target building design units includes multiple target building design units, and the combination of target building design units constitutes at least one target design scheme.
[0065] The semantics of the above prompt text can be understood as the meaning and implications expressed by the prompt text, and the intent of the above prompt text can be understood as the goal and purpose that the user hopes to achieve through the prompt text.
[0066] 104. Return the target design solution to the client.
[0067] In this embodiment of the invention, after the large language model generates the target design scheme, the scheme is returned or transmitted to the client. The client can obtain the desired results or suggestions by viewing, evaluating, and selecting the target design scheme.
[0068] In this embodiment of the invention, the large language model can be customized according to the user's needs and goals to meet the user's expectations and requirements.
[0069] In this embodiment of the invention, consultation dialogue data is acquired, including first dialogue data from the customer and second dialogue data from the customer service representative. Based on the consultation dialogue data and a building design scheme database, a prompt text for a large language model is constructed. The prompt text includes a request for the output of the design scheme result, and the building design scheme database includes multiple building design units constituting a building design scheme. The prompt text is input into the large language model for query processing to obtain the target design scheme output by the large language model. The target design scheme is then returned to the customer. This invention can solve the problem in existing methods where the language model has a large number of parameters, making it difficult to accurately output text that meets user needs.
[0070] Optionally, the consultation dialogue data is multimodal data. The first dialogue data includes at least one modality of text data, visual data, and speech data. The second dialogue data includes at least one modality of text data, visual data, and speech data. In the step of constructing prompt text for a large language model based on the consultation dialogue data and the building design scheme database, if the first dialogue data includes visual data and / or speech data, the visual data and / or speech data are extracted to obtain the first search data. If the second dialogue data includes visual data and / or speech data, the context data of the visual data and / or speech data is found in the first dialogue data. Based on the context data, it is determined whether the visual data and / or speech data is the second search data. Semantic search is performed on the first search data and / or the second search data to obtain multiple semantically related reference data. Based on the multiple reference data, the consultation dialogue data, and the building design scheme database, the prompt text for a large language model is constructed.
[0071] In this embodiment of the invention, if the first dialogue data includes visual data and speech data, then the visual data and speech data are extracted and used as the first search data for the large language model. If the first dialogue data consists of visual data, then the visual data is extracted and used as the first search data for the large language model. If the first dialogue data consists of speech data, then the speech data is extracted and used as the first search data for the large language model.
[0072] Furthermore, if the second dialogue data includes both visual and speech data, then the context data for the visual and speech data is retrieved from the first dialogue, and based on the context data, it is determined whether the visual and speech data are the second data to be searched. If the second dialogue data consists of visual data, then the context data for the visual data is retrieved from the first dialogue, and based on the context data, it is determined whether the visual data is the second data to be searched. If the second dialogue data consists of speech data, then the context data for the speech data is retrieved from the first dialogue, and based on the context data, it is determined whether the speech data is the second data to be searched.
[0073] The semantic search described above can be understood as a search process that involves understanding the context and meaning of the search query and the search content.
[0074] In this embodiment of the invention, prompt text for a large language model can be constructed based on multiple reference data, consultation dialogue data, and a building design scheme database.
[0075] Optionally, in the step of determining whether visual data and / or voice data is the second searchable data based on context data, semantic extraction can be performed on the context data to obtain the semantics of the context data; the semantics of the context data can be used to determine whether the customer approves of the video data and / or voice data; if the customer approves of the video data and / or voice data, then the video data and / or voice data are determined to be the second searchable data; if the customer does not approve of the video data and / or voice data, then the video data and / or voice data are determined not to be the second searchable data.
[0076] In this embodiment of the invention, the above semantic extraction can be understood as the process of extracting relevant data by analyzing the semantic information in the context data.
[0077] Semantic extraction can be performed on the contextual data of visual and audio data to obtain the semantics of the contextual data. Based on the contextual semantics, it can be determined whether the customer approves of the video and audio data. If the customer approves of the video and audio data, then the video and audio data are designated as the second search data. Semantic extraction can also be performed on the contextual data of visual and audio data to obtain the semantics of the contextual data. Based on the contextual semantics, it can be determined whether the customer approves of the video data. If the customer approves of the video data, then the video data is designated as the second search data. Similarly, semantic extraction can be performed on the contextual data of audio data to obtain the semantics of the contextual data. Based on the contextual semantics, it can be determined whether the customer approves of the audio data. If the customer approves of the audio data, then the audio data is designated as the second search data.
[0078] If it is determined that the customer does not approve of the video and audio data, then the video and audio data are not the second searchable data. If it is determined that the customer does not approve of the video data, then the video data is not the second searchable data. If it is determined that the customer does not approve of the audio data, then the audio data is not the second searchable data.
[0079] Optionally, in the step of performing a semantic search on the first search data and / or the second search data to obtain multiple semantically related reference data, a large language model can be used to perform a semantic search on the first search data and / or the second search data to obtain multiple candidate data. Each candidate data corresponds to a confidence score output by a large language model and a semantic relevance score corresponding to the first search data and / or the second search data. Based on the confidence score and semantic relevance score, multiple reference data are determined from the multiple candidate data, and the number of reference data is less than or equal to the number of candidate data.
[0080] In this embodiment of the invention, the aforementioned confidence level can be understood as the credibility of the relevance between the candidate data and the first and / or second search data when the large language model performs semantic search on the candidate data. The higher the confidence level, the stronger the relevance between the candidate data and the target data.
[0081] A semantic search can be performed on the first and second search data using a language model to obtain multiple candidate data. Each candidate data corresponds to a confidence score output by a large language model and a semantic relevance score between the first and second search data. The same process is repeated for the second search data.
[0082] Furthermore, multiple reference data can be determined from multiple candidate data based on confidence level and semantic relevance. The number of reference data is less than or equal to the number of candidate data. The K candidate data with the highest confidence level or semantic relevance can be used as reference data, or the K candidate data with the highest weighted sum of confidence level and relevance can be used as reference data.
[0083] Optionally, in the step of constructing the prompt text for the large language model based on multiple reference data, consultation dialogue data, and building design scheme database, a prompt template can be obtained. The prompt template includes a reference data input area, a consultation dialogue input area, a database input area, and a preset design scheme result output request. Multiple reference data are sorted according to confidence level and semantic relevance and filled in the reference data input area in sequence. Consultation dialogue data is filled in the consultation dialogue input area. The access method of the building design scheme database is filled in the database input area. After filling in, the prompt text for the large language model is obtained.
[0084] In this embodiment of the invention, the reference data field is used to fill in reference data. The consultation dialogue field is used to fill in consultation dialogue data. The database field is used to fill in the access method for the building design scheme database.
[0085] The prompt text of the aforementioned large language model includes reference data, consultation dialogue data, and database access methods, aiming to guide the large language model to better understand user needs and provide accurate and useful results.
[0086] Optionally, in the step of inputting the prompt text into a large language model for query processing to obtain the target design scheme output by the large language model, the prompt text can be input into the large language model, and the semantics and intent of the prompt text can be extracted by the large language model; the large language model can access the building design scheme database according to the semantics and intent of the prompt text to find the target building design unit set, which includes multiple target building design units; the large language model can combine the target building design units in the target building design unit set into at least one target design scheme and output at least one target design scheme.
[0087] In this embodiment of the invention, the semantics of the above-mentioned prompt text can be understood as the meaning and significance expressed by the prompt text, and the intent of the above-mentioned prompt text can be understood as the goal and purpose that the user hopes to achieve through the prompt text.
[0088] By inputting the prompt text into the large language model, the large language model can extract the semantics and intent of the prompt text; and then access the building design scheme database based on the extracted semantics and intent to find the target building design unit set; and finally combine the target building design units in the target building design unit set into at least one target design scheme and output at least one target design scheme.
[0089] The aforementioned set of target building design units includes multiple target building design units, each of which is a design scheme related to user needs.
[0090] Furthermore, multiple target building design units are combined to output at least one target design scheme to meet the user's needs.
[0091] Optionally, before constructing the prompt text of the large language model based on consultation dialogue data and building design scheme database, multiple sample building design schemes can be obtained; the multiple sample building design schemes can be divided into units to obtain multiple building design units; and a building design scheme database can be constructed based on the multiple building design units.
[0092] In this embodiment of the invention, the above-mentioned modular decomposition can be understood as breaking down a complex system or problem into several relatively simple units or modules for better understanding and processing. Modular decomposition divides the entire design scheme into several specific design elements or components, such as architectural style, functional layout, structural form, etc.
[0093] The aforementioned building design scheme database includes multiple building design units that constitute a building design scheme.
[0094] In this embodiment of the invention, various building design schemes can be collected and organized, and the collected building design schemes can be divided according to certain standards or rules to obtain multiple building design units. A building design scheme database is then established based on these multiple building design units. The building design scheme database is used to store and manage various building design schemes for reference by those seeking information.
[0095] like Figure 2 As shown, this embodiment of the invention provides a building design scheme query device. The building design scheme query device includes:
[0096] The first acquisition module 201 is used to acquire consultation dialogue data, which includes the customer's first dialogue data and the customer service's second dialogue data.
[0097] The first construction module 202 is used to construct prompt text for a large language model based on the consultation dialogue data and the building design scheme database. The prompt text includes a design scheme result output request, and the building design scheme database includes multiple building design units that constitute a building design scheme.
[0098] The query processing module 203 is used to input the prompt text into the large language model for query processing, and obtain the target design scheme output by the large language model;
[0099] The return module 204 is used to return the target design scheme to the customer.
[0100] Optionally, the building module 202 includes:
[0101] The prompting submodule is used to prompt the visual data and / or the voice data if the first dialogue data includes the visual data and / or the voice data, so as to obtain the first data to be searched;
[0102] The determination submodule is used to, if the second dialogue data includes the visual data and / or the voice data, find the context data of the visual data and / or the voice data in the first dialogue data, and determine whether the visual data and / or the voice data is the second data to be searched based on the context data;
[0103] The semantic search submodule is used to perform semantic search on the first data to be searched and / or the second data to be searched to obtain multiple semantically related reference data.
[0104] A submodule is constructed to build prompt text for a large language model based on multiple reference data, consultation dialogue data, and building design scheme database.
[0105] Optionally, the determining submodule includes:
[0106] A semantic extraction unit is used to extract semantics from the context data to obtain the semantics of the context data;
[0107] The first determining unit is configured to determine, based on the semantics of the context data, whether the customer approves of the video data and / or the voice data;
[0108] The second determining unit is configured to determine the video data and / or the voice data as the second search data if it is determined that the customer approves of the video data and / or the voice data.
[0109] The third determining unit is configured to determine that the video data and / or the voice data are not the second search data if it is determined that the customer does not approve of the video data and / or the voice data.
[0110] Optionally, the semantic search submodule includes:
[0111] The semantic search unit is used to perform semantic search on the first search data and / or the second search data through the large language model to obtain multiple candidate data. Each candidate data corresponds to a confidence score output by the large language model and a semantic relevance score corresponding to the first search data and / or the second search data.
[0112] The fourth determining unit is used to determine multiple reference data from the multiple candidate data based on the confidence level and the semantic relevance, wherein the number of reference data is less than or equal to the number of candidate data.
[0113] Optionally, the construction submodule includes:
[0114] The acquisition unit is used to acquire a prompt template, which includes a reference data input area, a consultation dialogue input area, a database input area, and a preset design scheme result output request.
[0115] A sorting unit is used to sort multiple reference data according to the confidence level and the semantic relevance, and then fill them sequentially in the reference data filling area; and
[0116] The first filling unit is used to fill in the consultation dialogue data in the consultation dialogue filling area; and
[0117] The second filling unit is used to fill in the access method of the building design scheme database into the database filling area;
[0118] The processing unit is used to complete the input and obtain the prompt text for the large language model.
[0119] Optionally, the query processing module 203 includes:
[0120] An extraction submodule is used to input the prompt text into the large language model and extract the semantics and intent of the prompt text through the large language model;
[0121] The access submodule is used to access the building design scheme database through the large language model based on the semantics and intent of the prompt text, and to find the target building design unit set, which includes multiple target building design units;
[0122] The output submodule is used to combine the target building design units in the target building design unit set into at least one target design scheme through the large language model, and output at least one of the target design schemes.
[0123] Optionally, the device further includes:
[0124] The second acquisition module is used to acquire multiple sample building design schemes;
[0125] The splitting module is used to break down multiple sample building design schemes into units, resulting in multiple building design units;
[0126] The second construction module is used to construct a building design scheme database based on multiple building design units.
[0127] It should be noted that the building design scheme query device provided in this embodiment of the invention can be applied to devices such as smartphones, computers, and servers that can perform building design scheme query methods.
[0128] The building design scheme query device provided in this embodiment of the invention can implement all the processes implemented by the building design scheme query method in the above-described method embodiments, and can achieve the same beneficial effects. To avoid repetition, it will not be described again here.
[0129] See Figure 3 , Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention, such as... Figure 3 As shown, it includes: a memory 302, a processor 301, and a computer program for querying building design schemes stored in the memory 302 and executable on the processor 301, wherein:
[0130] The processor 301 is used to call the computer program stored in the memory 302 and perform the following steps:
[0131] Acquire consultation dialogue data, which includes the customer's first dialogue data and the customer service representative's second dialogue data;
[0132] Based on the consultation dialogue data and the building design scheme database, a prompt text for a large language model is constructed. The prompt text includes a design scheme result output request. The building design scheme database includes multiple building design units that constitute a building design scheme.
[0133] The prompt text is input into the large language model for query processing to obtain the target design scheme output by the large language model.
[0134] The target design solution is then returned to the client.
[0135] Optionally, the consultation dialogue data is multimodal data. The first dialogue data includes at least one modality of text data, visual data, and speech data. The second dialogue data includes at least one modality of text data, visual data, and speech data. The step of constructing prompt text of a large language model based on the consultation dialogue data and the building design scheme database, executed by the processor 301, includes:
[0136] If the first dialogue data includes the visual data and / or the voice data, then the visual data and / or the voice data are prompted to obtain the first data to be searched;
[0137] If the second dialogue data includes the visual data and / or the voice data, then the context data of the visual data and / or the voice data is retrieved from the first dialogue data, and based on the context data, it is determined whether the visual data and / or the voice data is the second data to be searched.
[0138] Perform semantic search on the first data to be searched and / or the second data to be searched to obtain multiple semantically related reference data;
[0139] Based on multiple reference data, consultation dialogue data, and building design scheme database, a prompt text for a large language model is constructed.
[0140] Optionally, the step of determining whether the visual data and / or the voice data is the second search data based on the context data, performed by the processor 301, includes:
[0141] Semantic extraction is performed on the context data to obtain the semantics of the context data;
[0142] Determine whether the customer approves of the video data and / or the voice data based on the semantics of the context data;
[0143] If it is determined that the customer approves of the video data and / or the voice data, then the video data and / or the voice data are determined as the second data to be searched;
[0144] If it is determined that the customer does not approve of the video data and / or the voice data, then it is determined that the video data and / or the voice data are not the second data to be searched.
[0145] Optionally, the step of performing a semantic search on the first search data and / or the second search data to obtain semantically related parameter data, executed by the processor 301, includes:
[0146] The first search data and / or the second search data are semantically searched using the large language model to obtain multiple candidate data. Each candidate data corresponds to a confidence score output by the large language model and a semantic relevance score to the first search data and / or the second search data.
[0147] Based on the confidence level and the semantic relevance, multiple reference data are determined from the multiple candidate data, wherein the number of reference data is less than or equal to the number of candidate data.
[0148] Optionally, the step of processor 301 constructing prompt text based on multiple reference data, the consultation dialogue data, and the building design scheme database includes:
[0149] Obtain a prompt template, which includes a reference data input area, a consultation dialogue input area, a database input area, and a preset design scheme result output request;
[0150] The multiple reference data are sorted according to the confidence level and the semantic relevance, and then sequentially filled into the reference data filling area; and
[0151] Fill in the consultation dialogue data in the consultation dialogue input area; and
[0152] Enter the access method for the building design scheme database in the database input area;
[0153] Once completed, you will receive the prompt text from the large language model.
[0154] Optionally, the step of processor 301 inputting the prompt text into the large language model for query processing to obtain the target design scheme output by the large language model includes:
[0155] The prompt text is input into the large language model, and the semantics and intent of the prompt text are extracted through the large language model;
[0156] The large language model accesses the building design scheme database based on the semantics and intent of the prompt text to find the target building design unit set, which includes multiple target building design units.
[0157] The target building design units in the target building design unit set are combined into at least one target design scheme using the large language model, and at least one target design scheme is output.
[0158] Optionally, before the step of constructing the prompt text for the eye model based on the consultation dialogue data and the building design scheme database, the method executed by the processor 301 further includes:
[0159] Obtain multiple sample building design schemes;
[0160] Multiple sample building design schemes are divided into units to obtain multiple building design units;
[0161] A building design scheme database is constructed based on multiple building design units.
[0162] It should be noted that the electronic device provided in the embodiments of the present invention can be applied to devices such as smartphones, computers, and servers that can perform building design scheme query methods.
[0163] The electronic device provided in this embodiment of the invention can implement all the processes of the building design scheme query method in the above-described method embodiments, and can achieve the same beneficial effects. To avoid repetition, it will not be described again here.
[0164] The above description discloses only preferred embodiments of the present invention and should not be construed as limiting the scope of the present invention. Therefore, equivalent variations made in accordance with the claims of the present invention are still within the scope of the present invention.
Claims
1. A method for querying building design schemes, characterized in that, The method for querying building design schemes includes the following steps: Acquire consultation dialogue data, which includes the customer's first dialogue data and the customer service representative's second dialogue data; Based on the consultation dialogue data and building design scheme database, a prompt text for a large language model is constructed. Specifically, the consultation dialogue data is multimodal data. The first dialogue data includes at least one modality of text data, visual data, and speech data. The second dialogue data includes at least one modality of text data, visual data, and speech data. If the first dialogue data includes the visual data and / or the speech data, the visual data and / or the speech data are extracted to obtain the first search data. If the second dialogue data includes the visual data and / or the speech data, the context data of the visual data and / or the speech data is found in the first dialogue data, and semantic extraction is performed on the context data to obtain the semantics of the context data. Based on the... The semantics of the context data determine whether the customer approves of the visual data and / or the voice data; if the customer approves of the visual data and / or the voice data, then the visual data and / or the voice data are determined to be the second search data; if the customer does not approve of the visual data and / or the voice data, then the visual data and / or the voice data are not determined to be the second search data; a semantic search is performed on the first search data and / or the second search data to obtain multiple semantically related reference data; based on the multiple reference data, the consultation dialogue data, and the building design scheme database, a prompt text for a large language model is constructed; the prompt text includes a design scheme result output request, and the building design scheme database includes multiple building design units constituting a building design scheme; The prompt text is input into the large language model for query processing to obtain the target design scheme output by the large language model. The target design solution is then returned to the client.
2. The building design scheme query method as described in claim 1, characterized in that, The step of performing a semantic search on the first search data and / or the second search data to obtain multiple semantically related reference data includes: The first search data and / or the second search data are semantically searched using the large language model to obtain multiple candidate data. Each candidate data corresponds to a confidence score output by the large language model and a semantic relevance score to the first search data and / or the second search data. Based on the confidence level and the semantic relevance, multiple reference data are determined from the multiple candidate data, wherein the number of reference data is less than or equal to the number of candidate data.
3. The building design scheme query method as described in claim 2, characterized in that, The steps for constructing prompt text for a large language model based on multiple reference data, consultation dialogue data, and a building design scheme database include: Obtain a prompt template, which includes a reference data input area, a consultation dialogue input area, a database input area, and a preset design scheme result output request; The multiple reference data are sorted according to the confidence level and the semantic relevance, and then sequentially filled into the reference data filling area; and Fill in the consultation dialogue data in the consultation dialogue input area; and Enter the access method for the building design scheme database in the database input area; Once completed, you will receive the prompt text from the large language model.
4. The building design scheme query method as described in any one of claims 1 to 3, characterized in that, The step of inputting the prompt text into the large language model for query processing to obtain the target design scheme output by the large language model includes: The prompt text is input into the large language model, and the semantics and intent of the prompt text are extracted through the large language model; The large language model accesses the building design scheme database based on the semantics and intent of the prompt text to find the target building design unit set, which includes multiple target building design units. The target building design units in the target building design unit set are combined into at least one target design scheme using the large language model, and at least one target design scheme is output.
5. The building design scheme query method as described in any one of claims 1 to 3, characterized in that, Before the step of constructing prompt text for a large language model based on the consultation dialogue data and the building design scheme database, the method further includes: Obtain multiple sample building design schemes; Multiple sample building design schemes are divided into units to obtain multiple building design units; A building design scheme database is constructed based on multiple building design units.
6. A building design scheme query device, characterized in that, The building design scheme query device includes: The first acquisition module is used to acquire consultation dialogue data, which includes the customer's first dialogue data and the customer service's second dialogue data. The first construction module is used to construct prompt text for a large language model based on the consultation dialogue data and the building design scheme database. Specifically, the consultation dialogue data is multimodal data. The first dialogue data includes at least one modality of text data, visual data, and speech data. The second dialogue data includes at least one modality of text data, visual data, and speech data. If the first dialogue data includes the visual data and / or the speech data, the visual data and / or the speech data are extracted to obtain the first search data. If the second dialogue data includes the visual data and / or the speech data, the context data of the visual data and / or the speech data is found in the first dialogue data, and semantic extraction is performed on the context data to obtain the semantics of the context data. Based on the semantics of the context data, determine whether the customer approves of the visual data and / or the voice data; if the customer approves of the visual data and / or the voice data, then determine the visual data and / or the voice data as the second search data; if the customer does not approve of the visual data and / or the voice data, then determine that the visual data and / or the voice data are not the second search data; perform a semantic search on the first search data and / or the second search data to obtain multiple semantically related reference data; based on the multiple reference data, the consultation dialogue data, and the building design scheme database, construct a prompt text for a large language model; the prompt text includes a design scheme result output request, and the building design scheme database includes multiple building design units constituting a building design scheme; The query processing module is used to input the prompt text into the large language model for query processing, and obtain the target design scheme output by the large language model. The return module is used to return the target design scheme to the customer.
7. An electronic device, characterized in that, include: The memory, the processor, and the computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the building design scheme query method as described in any one of claims 1 to 5.
8. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the building design scheme query method as described in any one of claims 1 to 5.
Citation Information
Patent Citations
Generative large language model training method and model-based man-machine voice interaction method
CN116127045A
Text generation method and device, electronic equipment and storage medium
CN117332046A
