Intelligent data question and answer method and system fusing field large language model

Abstract

The invention provides an intelligent data question and answer method and system fusing a field large language model, and relates to the technical field of large language models.The method comprises the steps that action data and view angle data of a user in a pre-constructed three-dimensional virtual scene corresponding to a target field are obtained, and question and answer data of the target field are collected; performing association analysis on the action data and the view angle data to determine a user query intention; performing associated coding processing on the query intention of the user and the scene position of the three-dimensional virtual scene to construct an intelligent question and answer library; obtaining a resource allocation scheme by using a software defined network technology; searching target question and answer data corresponding to the query request data from an intelligent question and answer library through a domain large language model, and generating a target question and answer result and a visual instruction; based on the resource allocation scheme and the visualization instruction, intelligent data question answering is completed, and high-immersion, low-delay and accurate-intention intelligent data question answering and visualization presentation oriented to the specific field are achieved.

Description

Technical Field

[0001] This application relates to the field of large language model technology, and in particular to an intelligent data question answering method and system that integrates domain-specific large language models. Background Technology

[0002] Users have an increasing demand for complex data queries in 3D virtual environments, creating a need for a technical solution that integrates spatial interactive behavior, high-fidelity visualization, and low-latency intelligent response to achieve a natural, accurate, and immersive question-and-answer experience.

[0003] To address the aforementioned needs, existing mainstream solutions have attempted to combine general-purpose large language models with virtual reality platforms. By collecting user gaze direction and gesture input through head-mounted devices, and using preset spatial anchor points, the question-and-answer results are projected into the virtual scene as floating text or two-dimensional cards. Network resources are statically partitioned to support basic interactive question-and-answer functions.

[0004] However, existing solutions have significant shortcomings in multi-dimensional collaboration: the recognition of user interaction intent does not fully integrate action trajectories and scene context, resulting in ambiguous query positioning. The visualization output does not consider the physical characteristics of optical display devices, making it prone to offset or distortion within the overlay. Furthermore, the rigid network resource allocation strategy makes it difficult to maintain stable data transmission and real-time rendering performance in high-concurrency or dynamic scenarios, severely impacting user experience and question-answering accuracy. Summary of the Invention

[0005] The purpose of this application is to provide an intelligent data question answering method and system that integrates a large domain language model to solve problems such as ambiguous query positioning in the prior art.

[0006] Firstly, this application provides an intelligent data question-answering method that integrates a domain-specific large language model, including:

[0007] Acquire user action and viewpoint data in a pre-constructed 3D virtual scene corresponding to the target domain, and collect question-and-answer data from the target domain;

[0008] Perform correlation analysis on motion data and viewpoint data to obtain correlation information, and determine the user's query intent based on the correlation information;

[0009] The system performs association encoding on the user's query intent and the scene location in the 3D virtual scene to generate query request data. The question and answer data is then associated and bound with scene elements in the 3D virtual scene to build an intelligent question and answer library.

[0010] Software-defined networking technology is used to allocate network transmission resources for query request data and to allocate computing resources for the corresponding data processing flow to obtain a resource allocation scheme.

[0011] The target question and answer data corresponding to the query request data is retrieved from the intelligent question and answer database using a domain-specific large language model. The target question and answer data is then processed with the attribute information and association information in the 3D virtual scene to generate the target question and answer results and visualization instructions.

[0012] Based on resource allocation schemes and visualization instructions, optical waveguide technology is used to fuse and render the target question-and-answer results with the target scene position in the 3D virtual scene in order to complete intelligent data question-and-answer.

[0013] Optionally, target question-and-answer data corresponding to the query request data is retrieved from the intelligent question-and-answer database using a domain-specific large language model. This target question-and-answer data is then processed along with attribute and association information from the 3D virtual scene to generate target question-and-answer results and visualization instructions, including:

[0014] The query request data is parsed using a domain-specific large language model to obtain search conditions, which include query keywords and coordinate information of target scene elements.

[0015] Based on the search criteria, target question and answer data associated with the target scene elements are retrieved from the intelligent question and answer database, and attribute information and scene association information of the target scene elements are obtained.

[0016] The target question-and-answer data is matched with attribute information to obtain the first matching result. The target question-and-answer data is matched with scene association information to obtain the second matching result. The first matching result and the second matching result are integrated to obtain the target question-and-answer result.

[0017] The coordinate information of the target question-and-answer results and the target scene elements is transformed and integrated to obtain visualization instructions.

[0018] Optionally, the coordinate information of the target question-and-answer results and the target scene elements is transformed and integrated to obtain visualization instructions, including:

[0019] The target question-and-answer results are converted to a standard format to obtain text conversion results, and the coordinate information of the target scene elements is converted to three-dimensional coordinates to obtain coordinate conversion results.

[0020] Based on the presentation specifications of the target domain and the visual characteristics of the 3D virtual scene, the rendering format is determined, including display style, display color, and display transparency.

[0021] Based on a preset fixed field order, the text conversion result, coordinate conversion result, and rendering format are combined to obtain combined data;

[0022] Based on the command parsing capability of optical waveguide technology and the data transmission requirements of three-dimensional virtual scenes, encoding rules adapted to optical waveguide technology are determined.

[0023] According to the encoding rules, the combined data is encoded to obtain complete encoded data. The complete encoded data is then format-validated to obtain the format validation result. If the format validation result is that the validation fails, the combined data is re-encoded until the validation passes. If the format validation result is that the validation passes, the validated complete encoded data is encapsulated to obtain visual instructions.

[0024] Optionally, correlation analysis can be performed on motion data and viewpoint data to obtain correlation information, including:

[0025] Extract motion trajectory features from motion data and calculate the angle difference between adjacent time points in viewpoint data. Integrate target viewpoint data with angle differences exceeding a preset threshold to form angle change features.

[0026] The motion trajectory features are matched with scene elements in the 3D virtual scene to obtain the motion matching results, and the angle change features are associated with scene regions in the 3D virtual scene to obtain the viewpoint association results.

[0027] The scene elements corresponding to the action matching results and the scene areas corresponding to the viewpoint association results are linked and integrated to obtain the association information.

[0028] Optionally, the user's query intent and the scene location of the 3D virtual scene are correlated and encoded to generate query request data. The question-and-answer data is then associated and bound to scene elements in the 3D virtual scene to construct an intelligent question-and-answer database, including:

[0029] Semantic segmentation is performed on the user's query intent to extract query keywords from the user's query intent;

[0030] Obtain the coordinate information of the scene location in the 3D virtual scene that corresponds to the user's query intent, and encode and combine the query keywords and coordinate information according to the preset data format to obtain the query request data;

[0031] Based on the information categories of the target domain, the question-and-answer data is classified to obtain multiple subsets of question-and-answer data;

[0032] The binding result is obtained by associating and binding scene element identifiers in a 3D virtual scene with question and answer data subset identifiers in a question and answer data subset, in order to construct an intelligent question and answer library.

[0033] Optionally, software-defined networking technology is used to allocate network transmission resources for query request data and to allocate computing resources for the corresponding data processing flow to obtain a resource allocation scheme, including:

[0034] Calculate the transmission rate based on the size of the query request data, and calculate the computational complexity parameter based on the query complexity of the query request data.

[0035] Based on the transmission rate and computational load parameters, determine the first priority for transmitting query request data and the second priority for the data processing flow;

[0036] Based on the first priority, software-defined networking technology is used to allocate corresponding network bandwidth for the transmission of query request data, and based on the second priority, software-defined networking technology is used to allocate corresponding computing resources for the data processing flow.

[0037] By integrating network bandwidth and computing resources, a resource allocation scheme is obtained.

[0038] Optionally, based on resource allocation schemes and visualization instructions, optical waveguide technology is used to fuse the target question-and-answer results with the target scene location in the 3D virtual scene for rendering processing, in order to complete intelligent data question-and-answer, including:

[0039] According to the resource allocation scheme, the corresponding rendering resources are called from the preset rendering resource pool using optical waveguide technology. The rendering resources include bandwidth resources and computing resources.

[0040] Based on the rendering position in the visualization instructions, determine the target scene position in the 3D virtual scene that corresponds to the target question-and-answer result;

[0041] Spatially calibrate the displayed content of the target question-and-answer results with the target scene location to obtain the spatial calibration results;

[0042] Based on the display color and transparency in the rendering format of the visualization instructions, the visual attributes of the spatial calibration results are adjusted to obtain the visual adaptation results;

[0043] The visual adaptation results and the 3D virtual scene are pixel-wise fused to obtain the fused rendering result, thus completing the question-and-answer data.

[0044] Secondly, this application provides an intelligent data question-answering system that integrates a domain-specific large language model, including:

[0045] The acquisition module is used to acquire the user's action data and perspective data in a pre-built 3D virtual scene corresponding to the target domain, and to collect question and answer data in the target domain;

[0046] The association module is used to perform correlation analysis on action data and viewpoint data to obtain association information, and to determine the user's query intent based on the association information;

[0047] The association module is also used to perform association encoding processing on the user's query intent and the scene location of the 3D virtual scene, generate query request data, and associate and bind the question and answer data with the scene elements in the 3D virtual scene to build an intelligent question and answer library.

[0048] The allocation module is used to allocate network transmission resources for query request data using software-defined networking technology, and to allocate computing resources for the corresponding data processing flow to obtain a resource allocation scheme.

[0049] The retrieval module is used to retrieve target question and answer data corresponding to the query request data from the intelligent question and answer database through the domain large language model, and process the target question and answer data with the attribute information and association information in the 3D virtual scene to generate target question and answer results and visualization instructions;

[0050] The fusion module is used to fuse the target question-and-answer results with the target scene position in the 3D virtual scene based on the resource allocation scheme and visualization instructions, using optical waveguide technology to complete intelligent data question-and-answer.

[0051] Thirdly, this application provides an electronic device, comprising:

[0052] Memory, used to store computer programs;

[0053] A processor, used to execute computer programs, implements the steps of an intelligent data question-answering method that integrates a large language model of a fusion domain, as described in the first aspect above.

[0054] Fourthly, this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, can implement the steps of the intelligent data question-answering method that integrates a domain-wide large language model as described in the first aspect above.

[0055] This application presents an intelligent data question-answering method that integrates a domain-specific large language model. By acquiring user action and perspective data in a pre-constructed 3D virtual scene corresponding to the target domain, and collecting question-answering data from the target domain, it provides a high-fidelity, context-rich input foundation for subsequent intent recognition and contextualized question answering, supporting accurate understanding of complex query tasks. It achieves deep mining of users' implicit query intent, overcoming the intent ambiguity problem caused by relying on a single input channel, improving the scene adaptability and semantic relevance of the answer, and realizing high-fidelity, low-distortion virtual-real fusion presentation, enhancing immersion and information readability.

[0056] Furthermore, the query request data is parsed using a domain-specific large language model to obtain search criteria including query keywords and target scene element coordinates. Based on these criteria, target question-and-answer data associated with the target scene elements is retrieved from the intelligent question-and-answer database, and its attribute information and scene association information are obtained. Subsequently, the target question-and-answer data is matched with both the attribute information and the scene association information, and the matching results are integrated to form the target question-and-answer results. Finally, the target question-and-answer results are transformed and integrated using the coordinate information of the target scene elements to generate visual instructions.

[0057] It solves the problems of question-and-answer content drift, misalignment, or semantic disconnect caused by the lack of fine-grained scene binding in existing solutions, and improves the accuracy, consistency, and visual credibility of immersive intelligent question answering. Attached Figure Description

[0058] To more clearly illustrate the technical solutions in the embodiments of this application 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 some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0059] Figure 1 A flowchart illustrating an intelligent data question-answering method that integrates a domain-specific large language model, provided as an embodiment of this application;

[0060] Figure 2 A flowchart illustrating the generation of target question-and-answer results and visualization instructions is provided for an embodiment of this application.

[0061] Figure 3 This is a schematic diagram of the structure of an intelligent data question-answering system that integrates a large domain language model, as provided in an embodiment of this application. Detailed Implementation

[0062] Existing solutions suffer from several problems in 3D virtual environments, including difficulty in accurately identifying user query intent based on complex actions and perspectives, lack of semantic-spatial depth binding between question and answer content and scene elements, failure to adapt visualization overlays to the physical characteristics of optical displays, and rendering stuttering under high concurrency due to static allocation of network resources.

[0063] This application collects user action and perspective data in a 3D virtual scene within a target domain, and combines this with correlation analysis to achieve contextual awareness and understanding of query intent. Then, it co-encodes the intent and scene location to construct an intelligent question-and-answer library that closely links question-and-answer data with scene elements. Next, relying on a domain-specific large language model, it retrieves and integrates scene attributes and related information from the intelligent question-and-answer library to generate target question-and-answer results and visualization instructions that are both semantically accurate and spatially based. Finally, it utilizes optical waveguide technology to precisely fuse the question-and-answer results to the target scene location for rendering, thereby systematically solving the collaborative bottlenecks of existing technologies in areas such as ambiguous intent recognition, distortion from virtual-real overlay, and rigid resource scheduling, achieving a natural, accurate, and immersive intelligent data question-and-answer experience.

[0064] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions 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, 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.

[0065] The core of this application is to provide an intelligent data question-answering method that integrates a large domain language model. A flowchart illustrating one specific implementation is shown below. Figure 1 As shown, the method includes:

[0066] Step 101: Obtain the user's action data and viewpoint data in the pre-constructed 3D virtual scene corresponding to the target domain, and collect question and answer data of the target domain.

[0067] In this step, the 3D virtual scene refers to a pre-constructed digital 3D simulation scene adapted to a specific professional field. The target field refers to the specific professional field that intelligent data question answering addresses, such as mechanical manufacturing, biomedicine, aerospace, etc.

[0068] In this embodiment, the user's motion and perspective data generated in a pre-constructed 3D virtual scene corresponding to the target domain are first collected, along with question-and-answer data within the target domain, completing the initial data acquisition. For example, a 3D virtual scene corresponding to the biomedical field is first constructed. This scene may include various virtual cells and virtual organelles, as well as microscopic biochemical reaction scenarios. Then, motion capture devices are used to collect the user's limb movements in this 3D virtual scene, such as pointing at a specific organelle or simulating a dissection operation. Visual acquisition devices are used to collect the user's perspective data while observing the scene. Simultaneously, question-and-answer data, such as cell structure queries and gene function analyses within the biomedical field, are collected.

[0069] Step 102: Perform correlation analysis on motion data and viewpoint data to obtain correlation information, and determine the user's query intent based on the correlation information.

[0070] In this embodiment, the acquired action data and viewpoint data are first analyzed for correlation. Correlation information is obtained by analyzing their correspondence, including scene elements corresponding to the action matching results and scene regions corresponding to the viewpoint correlation results. Next, a pre-defined intent mapping library is retrieved, which stores the correspondence between different scene elements and scene regions within the 3D virtual scene and various user query intents. Then, the scene elements and scene regions in the correlation information are matched one by one with the correspondences in the intent mapping library, filtering out the intent entries with the highest matching degree. Finally, the user's query intent is determined based on the specific content of the intent entry and the feature details of the correlation information.

[0071] Step 103: Perform association encoding processing on the user's query intent and the scene location of the 3D virtual scene to generate query request data, and associate and bind the question and answer data with the scene elements in the 3D virtual scene to build an intelligent question and answer library.

[0072] In this step, scene location refers to the three-dimensional spatial coordinates of each scene element within the 3D virtual scene. Scene elements refer to the various entity objects included in the 3D virtual scene.

[0073] Step 104: Use software-defined networking technology to allocate network transmission resources for query request data and allocate computing resources for the corresponding data processing flow to obtain a resource allocation scheme.

[0074] In this step, Software Defined Networking (SDN) refers to a technical architecture that achieves centralized network management by separating the control plane and data plane of network devices. It separates the control plane and data plane of network devices, enabling flexible scheduling of network resources through software programming. Network transmission resources refer to various resources used for data transmission in the network, including bandwidth and transmission channels.

[0075] A data processing flow refers to a series of processes involving parsing, retrieving, and matching query request data. Computing resources refer to the computing power resources used to support the data processing flow, including CPU processing power and memory space.

[0076] Step 105: Retrieve the target question and answer data corresponding to the query request data from the intelligent question and answer database using the domain-specific large language model, and process the target question and answer data with the attribute information and association information in the 3D virtual scene to generate the target question and answer results and visualization instructions.

[0077] In this step, the domain-specific large language model refers to a large language model trained and optimized for a specific target domain, capable of handling specialized data and problems within that domain. Attribute information refers to the feature information of scene elements within a 3D virtual scene, such as material, size, and performance parameters.

[0078] Step 106: Based on the resource allocation scheme and visualization instructions, use optical waveguide technology to fuse and render the target question-and-answer results with the target scene position in the 3D virtual scene to complete intelligent data question-and-answer.

[0079] In this step, optical waveguide technology refers to the technology that uses optical waveguide media to transmit and image optical signals, enabling the fusion and display of virtual images with real-world scenes. The target scene location refers to the specific position of the scene element within the 3D virtual scene corresponding to the target question-and-answer result.

[0080] This application's embodiments acquire user action and perspective data in a pre-constructed 3D virtual scene corresponding to the target domain, and collect question-and-answer data from the target domain. This provides a high-fidelity, context-rich input foundation for subsequent intent recognition and contextualized question answering, supporting accurate understanding of complex query tasks. It enables deep mining of users' implicit query intent, overcoming the intent ambiguity problem caused by relying on only a single input channel, and improving the accuracy of intent recognition. It avoids transmission delays or rendering stutters caused by resource contention, ensuring the smoothness and timeliness of the immersive question-and-answer process. It improves the scene adaptability and semantic relevance of the answers. It achieves high-fidelity, low-distortion virtual-real fusion presentation, enhancing immersion and information readability.

[0081] This application provides a specific embodiment. Step 102 involves performing correlation analysis on motion data and viewpoint data to obtain correlation information, specifically including the following steps:

[0082] Step 201: Extract motion trajectory features from motion data and calculate the angle difference between adjacent time points in the viewpoint data. Integrate the target viewpoint data whose angle difference exceeds a preset threshold to form angle change features.

[0083] In this step, motion trajectory features refer to the feature information extracted from motion data that reflects the user's limb movement path and trend. Scene area refers to a specific region in a 3D virtual scene divided according to spatial range or functional attributes.

[0084] In this embodiment, the motion trajectory features that reflect the user's movement path can first be extracted from the motion data. Then, the angle difference is calculated by selecting the angle values ​​of the viewpoint data at adjacent time points. Target viewpoint data with angle differences exceeding a preset threshold are then selected and integrated to finally form angle change features that reflect the user's viewpoint change pattern.

[0085] Step 202: Match the motion trajectory features with scene elements in the 3D virtual scene to obtain the motion matching result, and associate the angle change features with the scene area in the 3D virtual scene to obtain the viewpoint association result.

[0086] In this embodiment of the application, the extracted motion trajectory features can be compared and matched with various scene elements in the three-dimensional virtual scene to determine the scene elements corresponding to the motion trajectory features and obtain the motion matching result. Then, the formed angle change features can be associated with different scene areas in the three-dimensional virtual scene to determine the scene area corresponding to the angle change features and obtain the viewpoint association result.

[0087] For example, the extracted user action trajectory features pointing to the virtual cell nucleus can be compared one by one with scene elements such as virtual cell nuclei, virtual mitochondria, and virtual ribosomes in the 3D virtual scene to determine that the action trajectory features correspond to the virtual cell nucleus, thus obtaining the action matching result as the virtual cell nucleus. At the same time, the angle change features reflecting the user's downward view of the cell nucleus surface can be associated with scene areas such as the cell nucleus surface area, the cell nucleus interior area, and the nuclear pore area in the 3D virtual scene to determine that the angle change features correspond to the cell nucleus surface area, thus obtaining the viewpoint association result as the cell nucleus surface area.

[0088] Step 203: Link and integrate the scene elements corresponding to the action matching results and the scene areas corresponding to the viewpoint association results to obtain association information.

[0089] In this embodiment, the specific scene element corresponding to the action matching result can be identified first, and then the specific scene area corresponding to the viewpoint association result can be identified. These two are then linked and integrated to clarify and merge the correspondence between scene elements and scene areas, ultimately obtaining association information that reflects the relationship between user actions and viewpoints. For example, after identifying that the scene element corresponding to the action matching result is a virtual cell nucleus and the scene area corresponding to the viewpoint association result is the cell nucleus surface area, these two can be linked and integrated to clarify and merge the correspondence between the virtual cell nucleus and the cell nucleus surface area, thereby obtaining association information where the user action points to the virtual cell nucleus and the viewpoint is focused on the cell nucleus surface area.

[0090] The embodiments of this application realize refined correlation analysis of user action data and perspective data, which solves the problem that the interpretation of user behavior is one-sided and the basis for intent judgment is singular due to the analysis of action data or perspective data alone in the prior art, and improves the correlation and comprehensiveness of user behavior analysis.

[0091] This application provides a specific embodiment. Step 103 involves associating the user's query intent with the scene location of the 3D virtual scene using encoding to generate query request data. The question-and-answer data is then associated and bound with scene elements in the 3D virtual scene to construct an intelligent question-and-answer database. This specifically includes the following steps:

[0092] Step 301: Perform semantic segmentation on the user's query intent to extract query keywords from the user's query intent.

[0093] In this embodiment of the application, semantic segmentation processing is first performed on the determined user query intent. The statement of the user query intent is split into semantic units, and then words that can reflect the core content of the query are selected from the split semantic units. Finally, the query keywords in the user query intent are extracted.

[0094] Step 302: Obtain the coordinate information of the scene location in the 3D virtual scene corresponding to the user's query intent. According to the preset data format, encode and combine the query keywords and coordinate information to obtain the query request data.

[0095] In this step, the preset data format refers to the pre-defined structured data format used to integrate query keywords and coordinate information, which can ensure the standardization and parsability of the data encoding combination.

[0096] In this embodiment, the scene location in the three-dimensional virtual scene that matches the user's query intent is first located, and then the three-dimensional spatial coordinate information corresponding to the scene location is obtained. Then, according to the field requirements of the preset data format, the extracted query keywords and the obtained coordinate information are sequentially filled into the corresponding fields. Finally, the filled content is encoded and combined to generate structured query request data.

[0097] For example, the location of the virtual cell nucleus in a 3D virtual scene, corresponding to the intention of understanding the function of the virtual cell nucleus, is located. The 3D coordinate information of the scene location (X-axis, Y-axis, Z-axis) is obtained through a spatial coordinate detection tool. Assuming that a preset data format of keywords plus coordinates is set as the field, the extracted query keywords such as virtual cell, cell nucleus, and function, along with the coordinate information, can be filled into the corresponding fields according to this format. Then, the filled content is encoded and combined to finally generate query request data including keywords and coordinates.

[0098] Step 303: Based on the information categories of the target domain, classify the question-and-answer data to obtain multiple subsets of question-and-answer data.

[0099] In this step, information category refers to the classification standard of question and answer data in the target domain according to knowledge attributes or application scenarios, which can realize the orderly classification of question and answer data.

[0100] In this embodiment of the application, the information categories of the target domain are first sorted out, and then the collected question and answer data of the target domain are matched one by one according to the sorted information categories. Question and answer data belonging to the same information category are grouped together, and finally multiple subsets of question and answer data corresponding to different information categories are obtained.

[0101] Step 304: Associate and bind the scene element identifiers in the 3D virtual scene with the question and answer data subset identifiers in the question and answer data subset to obtain the binding result, so as to construct an intelligent question and answer library.

[0102] In this step, the scene element identifier refers to a unique identifier assigned to each scene element in the 3D virtual scene, used to distinguish different scene elements. The question-and-answer data subset identifier refers to a unique identifier assigned to each question-and-answer data subset, used to distinguish different question-and-answer data subsets.

[0103] In this embodiment, the scene element identifier of each scene element in the three-dimensional virtual scene is first retrieved, then the question and answer data subset identifier of each question and answer data subset is retrieved, then the question and answer data subset identifier related to the scene element and the scene element identifier of the scene element are associated and bound together to obtain the binding result of the scene element and the question and answer data subset. Finally, all binding results are stored and a structured intelligent question and answer library is constructed based on the results.

[0104] For example, the unique scene element identifiers of scene elements such as virtual cells, virtual cell nuclei, and virtual mitochondria in a 3D virtual scene are retrieved, assuming they are numbered C001, C002, and C003 respectively. Then, the unique question-and-answer data subset identifiers of question-and-answer data subsets of cell structure, drug mechanism, and experimental operation are retrieved, assuming they are numbered D001, D002, and D003 respectively. The cell structure question-and-answer data subset identifier number D001 related to the virtual cell nucleus can be associated and bound with the scene element identifier number C002 of the virtual cell nucleus, resulting in the binding result of number C002 and number D001. The binding of all relevant scene element identifiers and question-and-answer data subset identifiers is completed in this way. After storing all the binding results, an intelligent question-and-answer database in the biomedical field is constructed based on these binding results.

[0105] This application's embodiments achieve precise association between user query intent and virtual scene location, as well as orderly binding of question-and-answer data and scene elements. It solves the problems of non-standard generation of query request data, disordered management of domain question-and-answer data, and loose association between data and virtual scene elements in the prior art, thereby improving the effectiveness of query request data and the scene adaptability of question-and-answer data.

[0106] This application provides a specific embodiment. Step 104 involves using software-defined networking technology to allocate network transmission resources for query request data and allocating computing resources for the corresponding data processing flow to obtain a resource allocation scheme. This specifically includes the following steps:

[0107] Step 401: Calculate the transmission rate based on the size of the query request data, and calculate the computational complexity parameter based on the query complexity of the query request data.

[0108] In this embodiment, the size of the query request data is first obtained, and the transmission rate is calculated in combination with the preset transmission time requirement. The transmission rate is equal to the data size divided by the transmission time. Then, the query complexity of the query request data is analyzed, and the computational quantity parameter is calculated in combination with the basic computational quantity. The computational quantity parameter is equal to the query complexity coefficient multiplied by the basic computational quantity. The query complexity coefficient is a coefficient determined according to the retrieval dimension and matching level of the query request data, and the basic computational quantity is the computational quantity required to process a single-dimensional basic query.

[0109] For example, first, the size of the query request data corresponding to the function of the cell nucleus of the virtual cell is obtained, assuming it is 2KB, and the preset transmission time is 0.004 seconds. Based on this, the transmission rate is calculated, and the transmission rate value is calculated. Then, the query complexity of the query request data is analyzed. Assuming that it needs to retrieve cell structure data and perform single-dimensional matching, the query complexity coefficient is determined to be 1.2, and the basic computational cost is 200 MIPS. Based on this, the computational cost parameter is calculated.

[0110] Step 402: Determine the first priority of transmitting the query request data and the second priority of the data processing flow based on the transmission rate and computational load parameters.

[0111] In this embodiment, firstly, a priority threshold range for transmission rate is defined, the calculated transmission rate is compared with the threshold range, and the first priority for transmitting query request data is determined based on the comparison result. Then, a priority threshold range for computational quantity parameters is defined, the calculated computational quantity parameters are compared with the threshold range, and the second priority for data processing flow is determined based on the comparison result, thus completing the determination of the two priorities.

[0112] Step 403: Based on the first priority, allocate the corresponding network bandwidth for the transmission of query request data using software-defined networking technology; based on the second priority, allocate the corresponding computing resources for the data processing flow using software-defined networking technology.

[0113] In this step, network bandwidth refers to the channel capacity available for data transmission in the network. It is an important indicator for measuring network transmission capacity and determines the amount of data that the network can transmit per unit of time.

[0114] In this embodiment of the application, firstly, according to the determined first priority, the corresponding network bandwidth is selected from the total network bandwidth using software-defined networking technology and allocated to the transmission process of query request data. Then, according to the determined second priority, the corresponding computing resources are selected from the total computing resources using software-defined networking technology and allocated to the data processing flow corresponding to the query request data, thus completing the allocation operation of the two types of resources.

[0115] Step 404: Integrate network bandwidth and computing resources to obtain a resource allocation scheme.

[0116] In this embodiment, key information such as the value of the allocated network bandwidth and the allocation object is sorted out first, and then key information such as the value of the allocated computing resources and the allocation object is sorted out. The information of the two types of resources is integrated according to a preset resource sorting format, and the allocation information of different types of resources is summarized into the same structured carrier, and finally a resource allocation scheme including network bandwidth and computing resource allocation details is formed.

[0117] For example, the final step involves sorting out the values ​​of the allocated 5Mbps network bandwidth and the data transmission of the virtual cell nucleus query request, as well as the values ​​of the allocated 250MIPS computing resources and the data processing flow of the cell nucleus function analysis. This information is then integrated according to a preset format, and the allocation information of network bandwidth and computing resources is summarized into a structured document, ultimately forming a clear resource allocation scheme. The scheme clearly records the allocation of various resources.

[0118] The embodiments of this application realize the precise scheduling of network transmission resources and computing resources, solve the problems of lack of data support for resource allocation and imbalance in the scheduling of transmission and processing resources in the prior art, and improve the rationality and effectiveness of resource allocation.

[0119] This application provides a specific embodiment, such as Figure 2 As shown, step 105 involves retrieving target question-and-answer data corresponding to the query request data from the intelligent question-and-answer database using a domain-specific large language model, and then processing the target question-and-answer data with attribute and association information in the 3D virtual scene to generate target question-and-answer results and visualization instructions. Specifically, this includes the following steps:

[0120] Step 501: Parse the query request data using the domain-wide language model to obtain the search conditions, which include query keywords and coordinate information of target scene elements.

[0121] In this step, query keywords refer to words extracted from the user's query intent that can reflect the core content of the query and are the core basis for retrieving target question and answer data; target scene elements refer to specific scene entity objects in the three-dimensional virtual scene that correspond to the user's query intent and are the scene anchors for the retrieved data.

[0122] In this embodiment, the query request data is first input into the domain language model, which then decomposes and analyzes the text structure and data content of the query request data. The model then extracts query keywords that point to the core of the query from the decomposed content and identifies the coordinate information of the target scene elements included in the query request data. Finally, the query keywords and the coordinate information of the target scene elements are integrated to form the retrieval conditions for retrieving data.

[0123] For example, the query request data corresponding to understanding the function of the cell nucleus of a virtual cell is first input into a large language model in the biomedical field. This model decomposes and analyzes the text and data content of the query request data, extracting query keywords such as virtual cell, cell nucleus, and function from the decomposed content. Simultaneously, it identifies the three-dimensional coordinate information of the virtual cell nucleus included in the query request data. Then, these query keywords and the coordinate information of the virtual cell nucleus are integrated to form search conditions containing query keywords of virtual cell, cell nucleus, and function, as well as the coordinate information of target scene elements.

[0124] Step 502: Based on the search criteria, retrieve the target question and answer data associated with the target scene element from the intelligent question and answer database, and obtain the attribute information and scene association information of the target scene element.

[0125] In this embodiment of the application, the formed search conditions are input into the search module of the intelligent question and answer database. The search module matches the corresponding question and answer data categories in the intelligent question and answer database based on the query keywords, and then locates the question and answer data related to the target scene element under the category by combining the coordinate information of the target scene element. The content with the highest matching degree is then selected as the target question and answer data.

[0126] Subsequently, the characteristic information of the target scene element itself is retrieved from the scene element information library of the 3D virtual scene as attribute information, and the relationship information between the target scene element and other scene elements is retrieved as scene association information.

[0127] For example, the generated search criteria are input into the search module of an intelligent question-and-answer database in the biomedical field. The search module matches the query keywords such as "virtual cell," "cell nucleus," and "function" to the cell structure category of the intelligent question-and-answer database. Then, combining the coordinate information of the virtual cell nucleus, it locates the question-and-answer data related to the virtual cell nucleus within that category, and selects the content with the highest matching degree—that the cell nucleus is the control center of the cell and is responsible for storing genetic material and regulating cellular life activities—as the target question-and-answer data.

[0128] Then, the attribute information of the virtual cell nucleus is retrieved from the scene element information library of the 3D virtual scene, which may include the number of nuclear membrane layers, the number of nucleoli, etc. At the same time, the scene association information of the virtual cell nucleus with the virtual cytoplasm and virtual chromosomes is retrieved, which may include the material exchange mode and the genetic information transmission path, etc.

[0129] Step 503: Match the target question and answer data with the attribute information to obtain the first matching result; match the target question and answer data with the scene association information to obtain the second matching result; integrate the first matching result and the second matching result to obtain the target question and answer result.

[0130] In this embodiment, the target question-and-answer data is first compared dimension-by-dimensionally with the acquired attribute information to verify the suitability of the target question-and-answer data with the attribute information. Based on the suitability, a first matching result including matching dimensions and suitability conclusions is generated. Next, the target question-and-answer data is compared dimension-by-dimensionally with scene-related information to verify the fit between the target question-and-answer data and the scene-related information. Based on the fit, a second matching result including matching dimensions and fit conclusions is generated. Finally, the contents of the first and second matching results are sorted, duplicate information is removed, and valid content is integrated to form a complete target question-and-answer result.

[0131] For example, the target question-and-answer data—that the cell nucleus is the control center of the cell, responsible for storing genetic material and regulating cellular life activities—is compared dimension-by-dimensionally with the attribute information of the virtual cell nucleus, such as the number of nuclear membrane layers and the number of nucleoli. This verifies the compatibility between the question-and-answer data and the attribute information, generating a first matching result: the target question-and-answer data matches the structural attributes of the virtual cell nucleus, and the characteristics of the nuclear membrane and nucleoli support the functions of storing and regulating genetic material. Then, the target question-and-answer data is compared dimension-by-dimensionally with the scene association information of the virtual cell nucleus, virtual cytoplasm, and virtual chromosomes, verifying the fit between the question-and-answer data and the scene association information. This generates a second matching result: the target question-and-answer data matches the association relationship with the virtual cell nucleus, and the genetic information transmission path conforms to the logic of regulating cellular life activities.

[0132] Finally, the contents of the two matching results were sorted out, duplicate descriptions were removed, and effective information was integrated to form the target question-and-answer result that the cell nucleus is the control center of the cell and relies on the double nuclear membrane and nucleolus to store genetic material and to regulate cell life activities through the exchange of substances with the cytoplasm and the transmission of genetic information with the chromosomes.

[0133] Step 504: Transform and integrate the coordinate information of the target question-and-answer results and the target scene elements to obtain visualization instructions.

[0134] In this embodiment, the generated target question and answer results are first converted into a standardized text format to obtain regular text data. Then, the coordinate information of the target scene elements is converted into a coordinate parameter format that can be recognized by the 3D virtual scene rendering module to obtain standardized coordinate data. Then, according to the preset instruction generation rules, the converted text data and coordinate data are combined, and after adding rendering-related configuration information, a visualization instruction that can guide the fusion rendering is finally generated.

[0135] For example, the question-and-answer result regarding the cell nucleus as the cell's control center, storing genetic material through a double nuclear membrane and nucleolus, and regulating cellular life activities through material exchange with the cytoplasm and the transmission of genetic information with chromosomes, is converted into a standardized string text format, resulting in well-organized text data. The coordinate information of the virtual cell nucleus is then converted into a 3D coordinate parameter format recognizable by the 3D virtual scene rendering module, yielding standardized coordinate data. Following preset instruction generation rules, this text data and coordinate data are combined, and rendering configuration information is added to the target scene element positions as floating text, ultimately generating visualization instructions including display content and display position.

[0136] The embodiments of this application realize accurate retrieval and contextualized processing of target question-and-answer data, solving the problems in the prior art where the retrieval results of domain-wide language models are disconnected from the three-dimensional virtual scene, the question-and-answer results lack contextual adaptability, and the visualization instructions are generated without basis, thereby improving the contextual relevance and instructiveness of intelligent data question-and-answer.

[0137] This application provides a specific embodiment. Step 504 involves transforming and integrating the coordinate information of the target question-and-answer result and the target scene element to obtain visualization instructions. This specifically includes the following steps:

[0138] Step 511: Convert the target question and answer results to a standard format to obtain a text conversion result, and convert the coordinate information of the target scene elements to three-dimensional coordinates to obtain a coordinate conversion result.

[0139] In this embodiment, firstly, a text format commonly used in the target domain is selected as the standard format. The original text of the target question and answer result is adjusted according to the character arrangement and content structure requirements of the standard format. After completing the standard format conversion operation, the text conversion result is obtained. Then, a coordinate format recognizable by the 3D virtual scene rendering module is selected as the conversion basis. The coordinate information of the target scene elements is adjusted according to the parameter expression and dimension identification requirements of the coordinate format. After completing the 3D coordinate conversion operation, the coordinate conversion result is obtained.

[0140] Step 512: Determine the rendering format based on the presentation specifications of the target domain and the visual characteristics of the 3D virtual scene. The rendering format includes display style, display color, and display transparency.

[0141] In this step, presentation specifications refer to the standards and guidelines for information presentation within the target domain, used to ensure the professionalism and consistency of information display. Visual features refer to the visual characteristics of the 3D virtual scene, including color scheme, scene brightness, and element layout. Display style refers to the specific form in which the target question-and-answer results are displayed in the 3D virtual scene. Display color refers to the colors used to display the target question-and-answer results. Display transparency refers to the degree of transparency of the target question-and-answer results.

[0142] In this embodiment, the presentation specifications of the target domain are first sorted out to clarify the requirements for the style, color and other aspects of information display within the domain. Then, the visual characteristics of the three-dimensional virtual scene are analyzed to determine the color tone, brightness and other characteristics of the scene. Next, the display style, display color and display transparency of the target question and answer results are determined by combining the requirements of the presentation specifications and the characteristics of the visual features. The rendering form is obtained by integrating these three types of parameters.

[0143] Step 513: Combine the text conversion result, coordinate conversion result, and rendering format according to the preset fixed field order to obtain combined data.

[0144] In this step, the preset fixed field order refers to the pre-defined order of fields used to combine different data, which can ensure the structure and parsability of the combined data.

[0145] In this embodiment, a pre-set field arrangement order is first retrieved, which determines the order in which the text conversion result, coordinate conversion result, and rendering format are combined. Then, according to the pre-set fixed field order, the text conversion result, coordinate conversion result, and rendering format are sequentially filled into the pre-set data frame. After the data combination operation is completed, the combined data is obtained.

[0146] Step 514: Based on the instruction parsing capability of optical waveguide technology and the data transmission requirements of the three-dimensional virtual scene, determine the encoding rules that are compatible with optical waveguide technology.

[0147] In this step, command parsing capability refers to the ability of optical waveguide technology to identify, interpret, and process input command data, which determines the format adaptation requirements of the command data; data transmission requirements refer to the requirements of the 3D virtual scene on data format, transmission rate, data integrity, etc. during data transmission.

[0148] In this application embodiment, the instruction parsing capability of optical waveguide technology is first analyzed to clarify the instruction encoding format and data identification rules that the technology can recognize. Then, the data transmission requirements of the three-dimensional virtual scene are analyzed to clarify the scene's requirements for the simplicity of data encoding and the efficiency of transmission. Then, based on the analysis results of these two aspects, encoding rules that are compatible with both optical waveguide technology parsing and three-dimensional virtual scene transmission are formulated.

[0149] Step 515: According to the encoding rules, encode the combined data to obtain complete encoded data, perform format verification on the complete encoded data to obtain the format verification result. If the format verification result is that the verification fails, re-encode the combined data until the verification passes. If the format verification result is that the verification passes, encapsulate the verified complete encoded data to obtain the visualization instructions.

[0150] In this embodiment, each field of the combined data is first encoded segment by segment according to the established encoding rules. The encoded segments are then concatenated to obtain complete encoded data. A preset format verification standard is then selected to verify the encoding format, data length, and identifier integrity of the complete encoded data, and a format verification result is obtained. If the format verification result is that the verification fails, the encoding rules are readjusted and the combined data is encoded again. This operation is repeated until the verification passes. If the format verification result is that the verification passes, an instruction encapsulation format recognizable by optical waveguide technology is selected, and the verified complete encoded data is encapsulated to finally obtain a visual instruction.

[0151] For example, according to the established encoding rules, the text conversion result, coordinate conversion result, and rendering format field of the combined data are encoded in hexadecimal segment by segment. The encoded segments containing cell nucleus interpretation and coordinates are then concatenated to obtain the complete encoded data. A preset format verification standard is then selected to check whether the encoding format of the complete encoded data is hexadecimal and whether the data identifier length is 8 bits. The format verification result is obtained. If the verification finds that the data identifier length is inconsistent, the encoding rules are readjusted to change the identifier length to 8 bits and the encoding is repeated until the verification passes. If the verification passes, the complete encoded data that has passed the verification is encapsulated according to the instruction encapsulation format that can be recognized by optical waveguide technology, and finally the visualization instruction is obtained.

[0152] This application embodiment realizes the standardization and scene-based generation of visualization instructions, which solves the problems of lack of unified standards for visualization instruction generation and poor adaptability to optical waveguide technology and 3D virtual scenes in the prior art. It improves the parsability and scene adaptability of visualization instructions, and provides accurate instruction support for subsequent fusion rendering.

[0153] This application provides a specific embodiment. Step 106 involves fusing and rendering the target question-and-answer result with the target scene position in the 3D virtual scene using optical waveguide technology, based on a resource allocation scheme and visualization instructions, to complete intelligent data question-and-answer. The specific steps include:

[0154] Step 601: According to the resource allocation scheme, use optical waveguide technology to call the corresponding rendering resources from the preset rendering resource pool. The rendering resources include bandwidth resources and computing resources.

[0155] In this step, the pre-built rendering resource pool refers to a collection of various resources pre-built for storing and managing the fusion rendering, including rendering-related resources such as bandwidth resources and computing resources. Bandwidth resources refer to the network channel capacity used to support the data transmission of fusion rendering.

[0156] In this embodiment, the specific values ​​of bandwidth resources and computing resources defined in the resource allocation scheme are first read. Then, relying on the resource call interface of optical waveguide technology, bandwidth resources and computing resources that match the values ​​are selected from the preset rendering resource pool. Next, the resource call operation is performed to allocate the selected rendering resources to the fusion rendering process, thus completing the rendering resource call work.

[0157] Step 602: Determine the target scene location in the 3D virtual scene corresponding to the target question-and-answer result based on the rendering location in the visualization instructions.

[0158] In this embodiment, the content of the visualization instruction is first parsed to extract the rendering position parameter used to indicate the display position. Then, the rendering position parameter is matched with the spatial coordinate system of the three-dimensional virtual scene to locate the specific spatial position in the three-dimensional virtual scene corresponding to the parameter. This position is then determined as the target scene position corresponding to the target question and answer result.

[0159] Step 603: Spatially calibrate the displayed content of the target question-and-answer result with the target scene location to obtain the spatial calibration result.

[0160] In this embodiment, the geometric center coordinates of the displayed content of the target question and answer result are first extracted, and then the spatial coordinates of the target scene position are obtained. The offset between the two coordinates is calculated. The offset is equal to the geometric center coordinates of the displayed content minus the spatial coordinates of the target scene position. Then, the spatial position of the displayed content is adjusted according to the offset so that the geometric center of the displayed content coincides with the spatial coordinates of the target scene position. After the spatial calibration operation is completed, the spatial calibration result is obtained.

[0161] Step 604: Adjust the visual attributes of the spatial calibration results according to the display color and display transparency in the rendering form of the visualization instructions to obtain the visual adaptation results.

[0162] In this step, visual attributes refer to the visual characteristics of the content displayed in the target question-and-answer results, including adjustable visual feature parameters such as display color and display transparency.

[0163] In this embodiment, the specific parameters of display color and display transparency are first extracted from the rendering form of the visualization instruction. Then, the original visual attribute parameters of the spatial calibration result are retrieved. The extracted display color parameters replace the color parameters in the original visual attributes. At the same time, the transparency parameters in the original visual attributes are adjusted according to the display transparency parameters. After the visual attribute adjustment operation is completed, the visual adaptation result is obtained.

[0164] Step 605: Perform pixel fusion of the visual adaptation results and the 3D virtual scene to obtain the fused rendering result, thereby completing the question-and-answer data.

[0165] In this embodiment, the pixel matrix of the visual adaptation result and the pixel matrix of the three-dimensional virtual scene are first obtained. The fusion weight of the two pixel matrices is calculated. The fusion weight is equal to the pixel ratio of the visual adaptation result multiplied by the pixel ratio of the scene. Then, according to the fusion weight, the corresponding pixels of the two pixel matrices are superimposed for color and brightness calculation to obtain the fused pixel matrix. The pixel matrix is ​​used as the fused rendering result to complete the entire process of question answering data.

[0166] For example, the pixel matrix of the visual adaptation result and the pixel matrix of the 3D virtual scene are obtained. Assuming that the pixel ratio of the visual adaptation result is set to 0.7 and the pixel ratio of the scene is 0.3, the fusion weight is calculated. Based on the weight, the corresponding pixels of the two pixel matrices are superimposed on the color and brightness to obtain the fused pixel matrix. This matrix can present the effect of the dark blue semi-transparent target question and answer result display content superimposed on the virtual cell nucleus position of the 3D virtual scene. This is used as the fusion rendering result to complete the entire process of question and answer data question and answer.

[0167] This application embodiment achieves accurate fusion rendering of target question-and-answer results and three-dimensional virtual scenes, solving the problems of disordered scheduling of fusion rendering resources, deviation in matching target question-and-answer results with scene positions, and incoordination between visual presentation and scene in the prior art, thereby improving the visual presentation effect and scene integration of intelligent data question-and-answer.

[0168] Figure 3 This is a schematic diagram illustrating a specific implementation of an intelligent data question-answering system that integrates a large domain language model, as provided in this application embodiment. (Refer to...) Figure 3 The system may include:

[0169] The acquisition module 21 is used to acquire the user's action data and viewpoint data in a pre-constructed 3D virtual scene corresponding to the target domain, and to collect question and answer data in the target domain;

[0170] The association module 22 is used to perform association analysis on action data and viewpoint data to obtain association information, and determine the user's query intent based on the association information;

[0171] The association module 22 is also used to perform association encoding processing on the user's query intent and the scene location of the 3D virtual scene, generate query request data, and associate and bind the question and answer data with the scene elements in the 3D virtual scene to build an intelligent question and answer library.

[0172] The allocation module 23 is used to allocate network transmission resources for query request data using software-defined networking technology, and to allocate computing resources for the corresponding data processing flow to obtain a resource allocation scheme.

[0173] The retrieval module 24 is used to retrieve the target question and answer data corresponding to the query request data from the intelligent question and answer library through the domain large language model, and process the target question and answer data with the attribute information and association information in the three-dimensional virtual scene to generate the target question and answer results and visualization instructions;

[0174] The fusion module 25 is used to fuse the target question-and-answer results and the target scene position in the 3D virtual scene based on the resource allocation scheme and visualization instructions, using optical waveguide technology to complete intelligent data question-and-answer.

[0175] This application provides an intelligent data question answering system that integrates a large domain language model to implement the aforementioned intelligent data question answering method that integrates a large domain language model. Therefore, the specific implementation of the intelligent data question answering system that integrates a large domain language model can be found in the embodiment section of the intelligent data question answering method that integrates a large domain language model mentioned above. The specific implementation can be referred to the description of the corresponding embodiments, which will not be repeated here.

[0176] This application also provides an electronic device, comprising: a memory for storing a computer program; and a processor for executing the computer program to implement the steps of an intelligent data question-answering method that integrates a large language model of a fusion domain, as described above.

[0177] This application also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of any of the above-described intelligent data question-answering methods that integrate a domain-wide large language model.

[0178] In one exemplary embodiment, the aforementioned computer-readable storage medium may include, but is not limited to, various media capable of storing computer programs, such as USB flash drives, read-only memory, random access memory, portable hard drives, magnetic disks, or optical disks.

[0179] Embodiments of the present invention also provide a computer program product, which includes a computer program that, when executed by a processor, implements the steps in any of the embodiments of the intelligent data question answering method for integrating a large language model in a fusion domain.

[0180] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0181] The above provides a detailed description of the intelligent data question-answering method and system that integrates a large domain language model, as provided in this application. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the embodiments above are merely for the purpose of helping to understand the method and its core ideas. It should be noted that those skilled in the art can make various improvements and modifications to this application without departing from its principles, and these improvements and modifications also fall within the protection scope of this application.

Claims

1. An intelligent data question answering method integrating a domain-specific large language model, characterized in that, include: Acquire user action and viewpoint data in a pre-constructed 3D virtual scene corresponding to the target domain, and collect question-and-answer data from the target domain; The motion data and viewpoint data are correlated to obtain correlation information, and the user's query intent is determined based on the correlation information. The user's query intent and the scene location of the three-dimensional virtual scene are associated and encoded to generate query request data. The question and answer data are then associated and bound with scene elements in the three-dimensional virtual scene to construct an intelligent question and answer library. Software-defined networking technology is used to allocate network transmission resources for the query request data and to allocate computing resources for the corresponding data processing flow to obtain a resource allocation scheme; The target question and answer data corresponding to the query request data is retrieved from the intelligent question and answer database using a domain-specific large language model. The target question and answer data is then processed with the attribute information and association information in the three-dimensional virtual scene to generate target question and answer results and visualization instructions. Based on the resource allocation scheme and the visualization instructions, the target question-and-answer result and the target scene position in the three-dimensional virtual scene are fused and rendered using optical waveguide technology to complete intelligent data question-and-answer.

2. The method according to claim 1, characterized in that, The system retrieves target question-and-answer data corresponding to the query request data from the intelligent question-and-answer database using a domain-specific large language model, and processes the target question-and-answer data with attribute information and association information in the 3D virtual scene to generate target question-and-answer results and visualization instructions, including: The query request data is parsed using a domain-specific large language model to obtain search conditions, which include query keywords and coordinate information of target scene elements. Based on the search criteria, target question and answer data associated with the target scene element are retrieved from the intelligent question and answer database, and attribute information and scene association information of the target scene element are obtained. The target question-and-answer data is matched with the attribute information to obtain a first matching result; the target question-and-answer data is matched with the scene association information to obtain a second matching result; the first matching result and the second matching result are integrated to obtain the target question-and-answer result. The target question-and-answer results and the coordinate information of the target scene elements are transformed and integrated to obtain visualization instructions.

3. The method according to claim 2, characterized in that, The target question-and-answer results and the coordinate information of the target scene elements are transformed and integrated to obtain visualization instructions, including: The target question-and-answer results are converted to a standard format to obtain a text conversion result, and the coordinate information of the target scene elements is converted to three-dimensional coordinates to obtain a coordinate conversion result. Based on the presentation specifications of the target domain and the visual characteristics of the three-dimensional virtual scene, the rendering format is determined, which includes display style, display color, and display transparency. According to a preset fixed field order, the text conversion result, the coordinate conversion result, and the rendering format are combined to obtain combined data; Based on the instruction parsing capability of the optical waveguide technology and the data transmission requirements of the three-dimensional virtual scene, an encoding rule adapted to the optical waveguide technology is determined. According to the encoding rules, the combined data is encoded to obtain complete encoded data. The complete encoded data is then format-validated to obtain a format validation result. If the format validation result is a failure, the combined data is re-encoded until the validation passes. If the format validation result is a pass, the validated complete encoded data is encapsulated to obtain visualization instructions.

4. The method according to claim 1, characterized in that, The motion data and viewpoint data are correlated to obtain correlation information, including: Extract motion trajectory features from the motion data and calculate the angle difference between adjacent time points in the viewpoint data. Integrate target viewpoint data whose angle difference exceeds a preset threshold to form angle change features. The motion trajectory features are matched with scene elements in the 3D virtual scene to obtain motion matching results, and the angle change features are associated with scene regions in the 3D virtual scene to obtain viewpoint association results. The scene elements corresponding to the action matching result and the scene regions corresponding to the viewpoint association result are associated and integrated to obtain association information.

5. The method according to claim 1, characterized in that, The user's query intent and the scene location of the 3D virtual scene are correlated and encoded to generate query request data. The question-and-answer data is then associated and bound to scene elements in the 3D virtual scene to construct an intelligent question-and-answer database, including: The user's query intent is semantically segmented to extract query keywords from the user's query intent; Obtain the coordinate information of the scene location in the three-dimensional virtual scene that corresponds to the user's query intent, and encode and combine the query keywords and the coordinate information according to a preset data format to obtain query request data; Based on the information categories of the target domain, the question-and-answer data is classified to obtain multiple subsets of question-and-answer data; The scene element identifiers in the 3D virtual scene are associated and bound with the question and answer data subset identifiers in the question and answer data subset to obtain the binding result, so as to construct an intelligent question and answer library.

6. The method according to claim 1, characterized in that, Software-defined networking (SDN) technology is used to allocate network transmission resources for the query request data and to allocate computing resources for the corresponding data processing flow to obtain a resource allocation scheme, including: The transmission rate is calculated based on the data size of the query request data, and the computational complexity parameter is calculated based on the query complexity of the query request data. Based on the transmission rate and the computational load parameters, determine the first priority of transmitting the query request data and the second priority of the data processing flow; Based on the first priority, software-defined networking technology is used to allocate corresponding network bandwidth for the transmission of the query request data, and based on the second priority, software-defined networking technology is used to allocate corresponding computing resources for the data processing flow. By integrating the network bandwidth and the computing resources, a resource allocation scheme is obtained.

7. The method according to claim 1, characterized in that, Based on the resource allocation scheme and the visualization instructions, optical waveguide technology is used to fuse and render the target question-and-answer result and the target scene position in the 3D virtual scene to complete intelligent data question-and-answer, including: According to the resource allocation scheme, optical waveguide technology is used to call corresponding rendering resources from a preset rendering resource pool, the rendering resources including bandwidth resources and computing resources; Based on the rendering position in the visualization instructions, determine the target scene position in the 3D virtual scene that corresponds to the target question-and-answer result; The displayed content of the target question-and-answer result is spatially calibrated with the target scene location to obtain a spatial calibration result; Based on the display color and display transparency in the rendering format of the visualization instructions, the visual attributes of the spatial calibration results are adjusted to obtain the visual adaptation results; The visual adaptation result and the three-dimensional virtual scene are pixel-wise fused to obtain a fused rendering result, thereby completing the question-and-answer data.

8. An intelligent data question-answering system integrating a domain-specific large language model, characterized in that, include: The acquisition module is used to acquire the user's action data and perspective data in a pre-built 3D virtual scene corresponding to the target domain, and to collect question and answer data in the target domain; The association module is used to perform association analysis on the action data and viewpoint data to obtain association information, and determine the user's query intent based on the association information; The association module is also used to perform association encoding processing on the user's query intent and the scene location of the three-dimensional virtual scene, generate query request data, and associate and bind the question and answer data with scene elements in the three-dimensional virtual scene to build an intelligent question and answer library. The allocation module is used to allocate network transmission resources for the query request data using software-defined networking technology, and to allocate computing resources for the corresponding data processing flow to obtain a resource allocation scheme. The retrieval module is used to retrieve target question and answer data corresponding to the query request data from the intelligent question and answer database through the domain large language model, and process the target question and answer data with the attribute information and association information in the three-dimensional virtual scene to generate target question and answer results and visualization instructions; The fusion module is used to perform fusion rendering processing on the target question-and-answer result and the target scene position in the three-dimensional virtual scene based on the resource allocation scheme and the visualization instructions, using optical waveguide technology, so as to complete intelligent data question-and-answer.

9. An electronic device, characterized in that, include: Memory, used to store computer programs; A processor, configured to implement the steps of an intelligent data question-answering method for integrating a domain-wide large language model as described in any one of claims 1 to 7 when executing the computer program.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, enables an intelligent data question-answering method that integrates a large domain language model as described in any one of claims 1 to 7.

Images