Building scheme comprehension visualization guiding method based on structured semantic memory

Abstract

The application relates to a building scheme understanding degree visualization guiding method based on structured semantic memory, which comprises the following steps: through the construction of a dynamic task planning and a project semantic graph dual-core driving mechanism, a building scheme compilation and review process is converted into an interactive, guidable and measurable structured cognitive cycle. The application has the beneficial effects that the application effectively solves the problems of semantic understanding fragmentation, low human-computer interaction efficiency and rigid task process under the traditional mode, realizes the deep fusion of transparent presentation of AI understanding degree and accurate user guidance, and significantly improves the intelligent level and cooperation efficiency of building scheme compilation and review.

Description

Technical Field

[0001] This invention belongs to the field of information technology in building engineering, and in particular relates to a visualization-guided method for understanding building schemes based on structured semantic memory. Background Technology

[0002] The preparation and review of architectural plans is a crucial preliminary stage in the field of architectural engineering, involving multidisciplinary collaboration, the reference of a vast number of code provisions, and complex spatial logic deductions. Traditional architectural plan preparation and review processes rely heavily on human experience, typically employing computer-aided design (CAD / BIM) tools based on two-dimensional drawings or three-dimensional models. While these tools provide support in geometric representation and basic data management, they have significant shortcomings in the deeper levels of "semantic understanding" and "collaborative cognition" of the plans.

[0003] Specifically, the existing technology has the following main limitations: 1. Fragmented and Unstructured Information Understanding: Current CAD / BIM systems can store and manage geometric information and some attribute information, but lack the ability to systematically and structurally model the semantic knowledge behind the design scheme, such as design intent, functional logic, and regulatory constraints. Key elements in the design scheme (such as spatial functional flow lines, structural system logic, and material performance requirements) and their interrelationships are often implicit in drawings, documents, or the designer's mind, failing to be explicitly captured and constructed into structured knowledge that machines can understand and reason about. This makes it difficult for AI-assisted tools to deeply understand the scheme's connotation, and their assistance often remains at a superficial level of retrieval or simple rule checking.

[0004] 2. Low Human-Computer Interaction Efficiency and Cognitive Gap: During the editing and review process, designers and reviewers need to communicate frequently to clarify intentions and confirm understanding. Existing interaction modes are mostly based on linear dialogues or isolated information annotations, lacking a shared view that can dynamically and intuitively display the AI's current understanding of the solution (including the degree of understanding and uncertainty). Users find it difficult to quickly grasp the AI's cognitive progress and existing confusion, and cannot directly and efficiently intervene in and guide the AI's "structured memory." The interaction process often involves multiple repetitions and detours, resulting in high cognitive costs.

[0005] 3. Rigid Task Implementation and Lack of Adaptive Adjustment: Existing workflow management systems or task list tools typically employ static or predefined task planning, failing to dynamically adjust based on the evolving semantic understanding of the solution during the editing and review process. When users modify or supplement information about the solution, the system often cannot intelligently reassess the necessity and priority of subsequent review steps, potentially leading to redundant operations or omissions of key checks, resulting in poor process flexibility.

[0006] 4. Disconnect between visualization and semantics: Although some studies attempt to combine knowledge graphs with visualization, the visualization presentation is often disconnected from the dynamic task execution progress and fails to effectively map the key indicator of "understanding" (such as confidence level) to visual variables. Users cannot intuitively perceive through the interface which parts have been fully understood and which are still questionable, which limits the actual effectiveness of visualization for collaborative cognition and decision support.

[0007] Therefore, there is an urgent need in this field for a new method that can progressively and structurally memorize the semantics of architectural schemes, and on this basis, realize visualized guidance for comprehension and dynamic task planning and adjustment, in order to overcome the above-mentioned defects and significantly improve the intelligence level and collaborative efficiency of architectural scheme compilation and review. Summary of the Invention

[0008] The purpose of this invention is to overcome the shortcomings of the prior art and provide a visualization-guided method for understanding architectural schemes based on structured semantic memory.

[0009] Firstly, it provides a visual guidance method for understanding architectural schemes based on structured semantic memory, including: S100, Dynamic Task Planning: Based on user-initiated architectural scheme editing and review tasks and initial information, a dynamic task plan containing multiple execution stages is generated and maintained by a large model. S200, Progressive Structuring: Execute the current step of the dynamic task planning, guide the user to input information, parse and structure the input content to form or update the project semantic graph, and assign understanding confidence to the graph elements; wherein, the project semantic graph is composed of multiple semantic gene units and their relationships, and can dynamically extract local structures related to the current execution stage according to the task context to form a project semantic graph subgraph; S300, Integrated Visualization of Understanding: The progress of the dynamic task planning, the current active steps, and the associated semantic graph subgraphs are integrated and visualized in the interactive interface. The visual style of the graph elements is configured differently based on their understanding confidence. S400, Direct Editing and Guidance of Structured Memory: Responding to user editing operations on visual graph elements; S500, Semantic Incremental Reasoning and Planning Adjustment: The editing operation is converted into a semantic incremental instruction, which triggers the large model to re-reason based on the updated semantic graph, update the understanding confidence and dynamically adjust the subsequent dynamic task planning. S600, Loop and Output: Loop through S200 to S500 until the dynamic task planning is completed, and output the editing and review results based on the final semantic graph.

[0010] As a preferred embodiment, S200 includes: S210. The large model actively initiates a request based on the objective of the current step. S220. Receive a response from the user to the request, the response including at least one of text input, file upload, and parameter selection; S230. Call the semantic parsing model in the architectural field to parse the response and identify predefined semantic gene units and their relationships; S240. The identified semantic gene units and their associations are added or updated to the project semantic graph in an incremental manner, and an initial understanding confidence is calculated for each newly added or changed graph element.

[0011] As a preferred embodiment, S300 includes: S310. Render a unified visual workspace in the interactive interface, which includes at least a task planning view area and a semantic memory view area. S320. In the task planning view area, the dynamic task plan is dynamically displayed in the form of a flowchart or Gantt chart, and different visual states are used to distinguish the completed stage, the currently active stage, and the stage to be executed. S330. In the semantic memory view area, the project semantic graph subgraph associated with the currently active step is dynamically displayed in an interactive graph structure. S340. Configure visual mapping rules for the graph nodes and edges in the semantic memory view area: wherein the fill color or size of the node maps to its understanding confidence, and the solid or void style or color of the edge maps to the degree of certainty of the relationship it represents.

[0012] Preferably, in S400, the editing operation includes at least one of the following: Content modification operation: Directly modify the semantic content or parameter values ​​of graph nodes; Relationship editing operations: Create, delete, or change semantic relationship edges between nodes; Confidence intervention: Manually adjust the understanding confidence value of a specified node; Annotation anchoring operation: Add annotations, references to standard provisions, or file links to nodes.

[0013] As a preferred embodiment, S500 includes: S510. The user's editing operation is solidified into an update instruction for the project semantic graph, triggering the graph version evolution, and the update is synchronized to the dialogue context of the large model. S520, the large model, based on the updated graph context, re-evaluates the preconditions and execution necessity of each step to be executed in the incomplete dynamic task planning. S530. Based on the results of the reassessment, perform at least one of the following operations on the dynamic task planning: skipping redundant steps, adding necessary steps, or adjusting the order of steps.

[0014] As a preferred option, during the content correction operation of S400, when the node parameter modified by the user belongs to the predefined set of key engineering parameters, the engineering rule verification process is automatically triggered: the chain effect caused by the parameter modification is simulated and evaluated, and the evaluation results are displayed in the corresponding node in the form of warnings or suggestions.

[0015] As a preferred option, it also includes: Global view switching: In response to user operation, switch from the view showing the details of the current step to the integrated overview view; in the integrated overview view, the dynamic task planning and the complete project semantic graph are overlaid and presented, wherein each stage node of the task planning is visually connected to the core semantic nodes generated in the graph at that stage, and the graph nodes are clustered according to the task stage to which they belong.

[0016] As a preferred option, it also includes: Context-aware information prefetching and filling: In the semantic memory view area of ​​the currently active step, if a graph node is detected as a required parameter but the content is missing or the confidence level is too low, the system will automatically retrieve the semantic graph of the architectural domain knowledge base or historical projects and suggest the obtained candidate values ​​near the node for the user to adopt and fill.

[0017] In a second aspect, a system for implementing any of the methods described in the first aspect is provided, comprising: The task scheduler is used to execute the logic for generating, maintaining, and advancing dynamic task plans. Semantic memory is used to store and manage the project's semantic graph in the form of a graph database, recording the change history of all graph elements and the changes in understanding confidence. An interactive rendering engine is used to connect the task scheduler and the semantic memory, and it includes an interactive instruction abstraction layer and a visualization presentation layer. The interactive instruction abstraction layer is used to normalize raw user operation events from different input devices into semantic graph editing instructions or planning control instructions; The visualization layer is used to generate and maintain a unified visualization workspace on the user end based on the state of the semantic memory and the progress of the task scheduler, and to respond to instructions transmitted by the interaction instruction abstraction layer.

[0018] Thirdly, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the computer program, implements the method as described in any of the first aspects.

[0019] The beneficial effects of this invention are as follows: By constructing a dual-core driving mechanism of "dynamic task planning" and "project semantic graph," this invention transforms the architectural design review process into an interactive, guided, and measurable structured cognitive loop. This method effectively solves the problems of fragmented semantic understanding, low human-computer interaction efficiency, and rigid task processes in traditional methods, achieving a deep integration of transparent presentation of AI understanding and precise user guidance, significantly improving the intelligence level and collaborative efficiency of architectural design review. Attached Figure Description

[0020] Figure 1 This is an overall flowchart of the visual guidance method for architectural scheme comprehension based on structured semantic memory provided by the present invention; Figure 2 This is a flowchart of the dual-ring coupling update mechanism provided by the present invention; Figure 3 This is a schematic diagram of the architecture of the building scheme understanding visualization guidance system provided by the present invention. Detailed Implementation

[0021] The present invention will be further described below with reference to embodiments. The description of the embodiments below is only for the purpose of helping to understand the present invention. It should be noted that those skilled in the art can make several modifications to the present invention without departing from the principle of the present invention, and these improvements and modifications also fall within the protection scope of the claims of the present invention.

[0022] Example 1: To address the problems of existing technologies, Embodiment 1 of this application provides a visualization guidance method for understanding architectural schemes based on structured semantic memory, which is suitable for intelligent review of architectural schemes in the initial design stage.

[0023] Specifically, the method includes: S100, Dynamic Task Planning: Based on user-initiated architectural design editing and review tasks and initial information, a large model generates and maintains dynamic task planning that includes multiple execution stages.

[0024] For example, a user starts the system, creates a new task titled "Architectural Design Scheme for a Municipal General Hospital," and uploads preliminary design specifications, site plan, and partial architectural floor plans. The system calls a large model (such as a domain fine-tuning model based on the GPT-4 architecture) to parse the task and generate a hierarchical dynamic task plan. The plan includes the following stages: Phase 1: Overall Layout Understanding (including steps: identification of functional zones in the master plan, circulation analysis, and assessment of green space and parking). Phase 2: Functional Analysis of Individual Buildings (including steps: functional layout of outpatient building, analysis of inpatient nursing units, and examination of the correlation between medical technology departments); Phase 3: Preliminary assessment of compliance with regulations (including steps: fire evacuation simulation pre-check, solar radiation simulation input extraction, and accessibility facility identification).

[0025] The plan is initially presented in the system in the form of a flowchart, and the current active stage is "Overall Layout Understanding".

[0026] S200, Progressive Structuring: Execute the current step of the dynamic task planning, guide the user to input information, parse and structure the input content to form or update the project semantic graph, and assign understanding confidence to the graph elements; wherein, the project semantic graph is composed of multiple semantic gene units and their relationships, and can dynamically extract local structures related to the current execution stage according to the task context to form a project semantic graph subgraph.

[0027] S200 includes: S210, The large model actively initiates a request based on the goal of the current step.

[0028] These requests include requests for information or data collection.

[0029] S220. Receive a response from the user to the request, the response including at least one of text input, file upload, and parameter selection.

[0030] S230. Call the semantic parsing model in the architectural field to parse the response and identify the predefined semantic gene units and their relationships.

[0031] Among them, the semantic gene unit is a predefined, minimal knowledge component with a clear engineering meaning.

[0032] S240. The identified semantic gene units and their associations are added or updated incrementally to the project semantic graph as the structured output of the current step. Based on the completeness of the parsing and the consistency of the domain, an initial understanding confidence is calculated for each newly added or changed graph element.

[0033] For example, when the system executes the current step "Site Plan Functional Zoning Identification", the large model proactively asks the user: "Please confirm the functional type of each building in the site plan, or upload a functional zoning description file." The user uploads a functional zoning table (Excel format).

[0034] The system invokes a built-in semantic parsing model for the architectural domain to identify semantic gene units such as "outpatient building," "inpatient building," "medical technology building," and "administrative and logistics building," and adds them as nodes to the project semantic graph. Edge relationships include "adjacent to," "serving," and "connected to." The system assigns an initial understanding confidence score to each node (e.g., "outpatient building" has a confidence score of 0.9, and "medical technology building" has a confidence score of 0.6 due to unclear labeling).

[0035] S300, Integrated Visualization of Understanding: The interactive interface integrates and visualizes the progress of the dynamic task planning, the current active steps, and the associated semantic graph subgraphs. The visual style of the graph elements is configured differently based on their understanding confidence.

[0036] The S300 includes: S310. Render a unified visual workspace in the interactive interface, which includes at least a task planning view area and a semantic memory view area. S320. In the task planning view area, the dynamic task plan is dynamically displayed in the form of a flowchart or Gantt chart, and different visual states are used to distinguish the completed stage, the currently active stage, and the stage to be executed. S330. In the semantic memory view area, the project semantic graph subgraph associated with the currently active step is dynamically displayed in an interactive graph structure. S340. Configure visual mapping rules for the graph nodes and edges in the semantic memory view area: wherein the fill color or size of the node maps to its understanding confidence, and the solid or void style or color of the edge maps to the degree of certainty of the relationship it represents.

[0037] Furthermore, when a user hovers the cursor over any graph element, detailed information about that element floats on the interface, including its semantic content, source, confidence calculation basis, and version history.

[0038] For example, the task planning view area on the right side of the interactive interface displays three stages and their progress in the form of a Gantt chart, with the current "Overall Layout Understanding" stage highlighted. The semantic memory view area on the left displays the semantic subgraphs related to the current step in a graph structure: the node colors gradually change from dark green (high confidence) to light yellow (low confidence), and the solid or dashed borders indicate the degree of certainty of the relationship. When the user hovers over the "Medical Technology Building" node, a floating window displays that its source is an uploaded form, the confidence calculation basis is "field missing," and indicates that it is editable.

[0039] S400, Direct Editing and Guidance of Structured Memory: Responds to user editing operations on visual graph elements.

[0040] For example, a user found that the description of the "Medical Technology Building" node was incomplete. Clicking on the node allowed the user to correct the content, adding "Includes Radiology Department, Laboratory Department, and Surgical Center." The user also added an annotation to the "Inpatient Building" node, citing Article 4.2.1 of the "Code for Design of General Hospital Buildings" GB 51039-2014, noting that "Nursing units should be relatively independent."

[0041] S500, Semantic Incremental Reasoning and Planning Adjustment: The editing operation is converted into a semantic incremental instruction, which triggers the large model to re-reason based on the updated semantic graph, update the understanding confidence, and dynamically adjust the subsequent dynamic task planning.

[0042] For example, the system translates user edits into semantic incremental instructions, triggering a re-inference of the large model. Based on the supplementary information, the model increases the confidence level of "Medical Technology Building" to 0.85 and identifies that "Surgical Center" needs to establish an "emergency access connection" with "Inpatient Building," automatically adding this edge. Simultaneously, because the Inpatient Building is already associated with regulatory provisions, the "accessibility facility identification" step in the planning assessment cycle "Phase 3" can be performed in advance to the current phase, with the system dynamically adjusting the task order.

[0043] S600, Loop and Output: Loop through S200 to S500 until the dynamic task planning is completed, and output the editing and review results based on the final semantic graph.

[0044] Finally, a stable project semantic graph is generated, which includes all building entities, functional attributes, specification associations, and review conclusions, and is output as a structured JSON file and a visual review report.

[0045] Example 2: Building upon Example 1, Example 2 of this application provides a more specific method for visually guiding architectural scheme comprehension based on semantic increment and structured memory, such as... Figure 1 As shown, it includes: S100, Dynamic Task Planning: Based on the user-initiated architectural scheme editing and review task and initial information, a dynamic task plan of a certain level is generated and maintained by a large model. The plan includes multiple sequential or parallel execution stages, and each stage includes at least one semantic understanding step.

[0046] S200, Progressive Structuring: In the current step of the dynamic task planning, the user is guided to input information or upload materials. The input content is parsed and structured into semantic entities, attributes and relationships using knowledge of the architectural field, forming or updating the corresponding nodes and edges in the project semantic graph as the structured memory of AI, and assigning a degree of understanding confidence to each element.

[0047] It should be noted that the system's built-in semantic parsing model for the architectural domain is based on a pre-trained BERT architecture and has been fine-tuned on a large number of building codes, design specifications, and drawing annotation texts. The model segments user input (text or document) into sentences or fields and identifies entities that match a predefined "semantic gene unit library".

[0048] For example, when a user inputs "The emergency entrance is located on the first floor of the inpatient building, connected to the outpatient building via a corridor," the model identifies: Semantic gene units: “Inpatient building” (type: building unit), “Emergency entrance” (type: building component), “Outpatient building” (type: building unit), “Connecting corridor” (type: connecting component).

[0049] Relationship: "Inpatient Building - Includes -> Emergency Entrance" "Inpatient Building - Connects to -> Outpatient Building (via connecting corridor)".

[0050] The confidence calculation module assigns initial values ​​based on the following rules: Source clarity: From structured tables or annotated drawings → +0.2; Completeness of representation: Fields are complete and without missing data → +0.15; Domain consistency: Matches the attribute range of similar entities in the existing graph → +0.1; Contextual support: Appears multiple times in the same sentence or paragraph or has a clear relationship with other entities → +0.05.

[0051] Initial confidence level = base value 0.5 + various bonuses, with a maximum of 0.95 and a minimum of 0.3.

[0052] The identification results are updated incrementally to the project semantic graph database (such as Neo4j), and each node stores attributes: ID, type, description, confidence, source, and update time.

[0053] S300, Integrated presentation of comprehension visualization: On the interactive interface, the dynamic task planning is presented in an interactive graph format, integrating the overall progress, current active steps, and project semantic graph subgraphs associated with each step. The visual style of the graph elements is configured differently based on their comprehension confidence.

[0054] S400, Direct Editing and Guidance of Structured Memory: Receives user editing operations on any element in the visualization graph, including modifying the semantic content of graph nodes or edges, adding annotation constraints to nodes, and confirming or questioning the AI's understanding confidence.

[0055] In S400, the editing operation includes at least one of the following: Content modification operation: Directly modify the semantic content or parameter values ​​of a graph node. Specifically, the user selects a specific graph node in the semantic memory view area and directly modifies the semantic description or specific parameter values ​​of the node in its associated text input box.

[0056] During content correction, when the node parameters modified by the user belong to a predefined set of key engineering parameters, the system automatically calls the built-in engineering rule validator after completing the graph update to quickly simulate and evaluate the chain effects caused by the modified parameters. The evaluation results of the validator are displayed in the semantic memory view area in association with the relevant nodes in the form of early warning prompts or adjustment suggestions.

[0057] For example, the predefined set of key engineering parameters includes: structural column cross-sectional dimensions, floor slab thickness, fire compartment area, evacuation staircase width, floor clear height, window-to-floor ratio, etc.

[0058] Scenario: The user changes the "window-to-floor ratio" from "0.25" to "0.18" in the semantic memory view.

[0059] The system recognizes that this parameter belongs to the set of key parameters and triggers the verification process.

[0060] Invoke the built-in project rule validator to retrieve associated rules: Article 4.0.2 of the "Standard for Daylighting Design of Buildings" GB 50033-2013 states that the window-to-floor ratio in hospital wards should not be less than 0.25.

[0061] The impact of the calibrator simulation is as follows: if the room is a ward, the modification will not meet the standards; if it is a warehouse, it may be compliant.

[0062] The system displays a warning message next to the "Window-to-Land Ratio" node: "The modified window-to-floor ratio of 0.18 may be lower than the minimum requirement of 0.25 for wards. Related room: Room 302, Inpatient Building. It is recommended to verify the room's function or adjust the parameters." Users can click on the prompt to view the detailed specifications or further modify the room's functional attributes.

[0063] Relationship editing operations: Create, delete, or change semantic relationship edges between nodes. Specifically, users can create new semantic relationship edges between existing graph nodes, or delete or change the type of existing associated edges, by dragging and dropping connection lines or selecting from the menu.

[0064] Confidence Intervention: Manually adjust the confidence value of a specified node. Specifically, users can manually increase or decrease the confidence value of nodes in the graph that are displayed as ambiguous or questionable through interactive controls. This operation acts as a strong feedback signal that directly affects subsequent inference.

[0065] Annotation anchoring operation: Add annotations, references to standard provisions, or links to documents to nodes. Specifically, users can add free text annotations, references to standard provisions, or links to external documents to any graph node. These annotations will be treated as strong constraint attributes of that node and will be given priority in subsequent steps.

[0066] S500, Semantic Incremental Reasoning and Planning Adjustment: Transforms user editing operations into semantic incremental instructions, triggering the large model to perform local re-reasoning based on the updated structured memory, updating the relevant understanding confidence, and dynamically adjusting subsequent dynamic task planning accordingly.

[0067] The implementation of S500 relies on a dual-loop coupled update mechanism. Specifically, S500 includes: S510, First Update Ring – Memory Synchronization Ring: Solidifies user editing operations into update instructions for the project semantic graph, triggers graph version evolution, and synchronizes the update to the dialogue context of the large model.

[0068] The following example illustrates the dual-loop coupling update mechanism in S500. Scenario: In the "Fire Evacuation Simulation Pre-Check" step, the user modifies the "Staircase Width" node in the graph from the original value "1.2m" to "1.5m".

[0069] For example, in S510, the system generates a map update command: JSON: { "op": "update_node", "node_id": "stair_001", "attribute": "width", "old_value": "1.2", "new_value": "1.5", "user": "designer_01 } The command is submitted to the graph database, node attributes are updated, and the version number is incremented by 1. Simultaneously, this update is synchronized as a context message to the dialogue memory of the larger model.

[0070] S520, Second Update Loop – Planning Evaluation Loop: Based on the updated graph context, the large model re-evaluates the preconditions and execution necessity of each step in the incomplete dynamic task planning.

[0071] For example, the large model re-evaluates incomplete steps based on the updated atlas (increased stair width): In the "evacuation distance verification" step, a width of 1.5m may meet a more lenient distance requirement, and this step may be skipped. The "staircase evacuation capacity verification" step must be performed due to parameter changes; The "Emergency Lighting Configuration Check" remains unaffected and its order remains unchanged.

[0072] S530. Based on the results of the reassessment, perform at least one of the following operations on the dynamic task planning: skipping steps that have become redundant due to the premise being met, adding steps that are necessary due to changes in context, or adjusting the logical order and dependencies between steps.

[0073] For example, the system automatically: In the task planning, mark "Evacuation Distance Verification" as "Conditions Met, Skip Recommended" and provide a skip option; Move the "Staircase Evacuation Capacity Verification" step to be executed immediately after the currently active step; Keep the other steps in the same order.

[0074] Users can view adjustment suggestions in the task planning view and confirm the execution with one click.

[0075] S600, Loop and Output: Loop through steps S200 to S500 until all stages of dynamic task planning are completed, and output the edited and reviewed digital data of the architectural scheme based on the final stable project semantic graph.

[0076] In addition, the visualization workspace provides a global perspective switching control, and the above methods also include: Global view switching: In response to the user's operation of the global view switching control, the view switches from the view showing the details of the current step to the integrated overview view; in the integrated overview view, the dynamic task planning and the complete project semantic graph are overlaid and presented, wherein each stage node of the task planning is visually connected to the core semantic nodes generated in the graph at that stage, and the graph nodes are clustered according to the task stage to which they belong.

[0077] For example, the user clicks the "Overview Mode" control in the upper right corner of the interface. The interface switches from a detailed view of the current step to an integrated overview view: the left side displays the task planning Gantt chart, and the right side displays the complete project semantic graph. Graph nodes are clustered according to their respective task stages (e.g., all "Stage 1" nodes are clustered on the left, in light blue; "Stage 2" nodes are centered, in green).

[0078] There are visual connections from the stage nodes to the core entity nodes generated in that stage (e.g., the connection from "Stage 1" to the "Outpatient Building" and "Inpatient Building" nodes).

[0079] In addition, it also includes: Context-aware information prefetching and filling: In the semantic memory view area of ​​the currently active step, if a graph node is detected as having a required parameter but lacking content or having low confidence, the system automatically retrieves semantic graphs from the architectural knowledge base or historical projects, and displays the obtained candidate values ​​near the node for the user to adopt and fill. For example, relevant candidate values ​​or typical ranges retrieved can be displayed near the node in a non-intrusive manner, allowing the user to adopt the candidate value and fill it into the node with a single click.

[0080] For example, in the "Fire Evacuation Simulation Pre-Check" step, the system detected that the confidence level of the "Number of People per Floor" node was only 0.4 (because the user did not provide this information). The system automatically: Search the knowledge base for the typical range of "average number of people per outpatient floor" (e.g., 200-300 people) for similar hospital buildings. Search for the preset number of people in rooms with the same functions in the semantic graph of similar past projects; Display a prompt next to the "Number of People per Floor" node: "Reference values: Outpatient floors typically see 200-300 people. Click to fill in: ① 250 people (median) ② 280 people (maximum) ③ Custom…" When a user clicks on "250 people", the node value is automatically filled, and the confidence level is increased to 0.75.

[0081] It should be noted that the parts in this embodiment that are the same as or similar to those in Embodiment 1 can be referred to each other, and will not be repeated in this application.

[0082] Example 3: Based on Embodiment 2, Embodiment 3 of this application provides a visual guidance system for architectural scheme comprehension based on semantic increment and structured memory, including: The task scheduler is used to execute the logic for generating, maintaining, and advancing dynamic task plans.

[0083] Specifically, the task scheduler is implemented based on a rule engine and a reinforcement learning model. It has a built-in planning utility evaluation module that simulates 3-5 paths before each planning adjustment, calculates the interaction cost (estimated number of user operations) and completeness score (percentage of review points covered), and recommends the optimal path.

[0084] Semantic memory is used to store and manage the project's semantic graph in the form of a graph database, recording the change history of all graph elements and the changes in understanding confidence.

[0085] Specifically, the semantic memory uses the Neo4j graph database to store the project's semantic graph. Each node has a version chain, supporting backtracking to any historical state.

[0086] An interactive rendering engine, used to connect the task scheduler and the semantic memory, includes an interactive instruction abstraction layer and a visualization rendering layer.

[0087] The interactive instruction abstraction layer is used to normalize raw user operation events from different input devices into semantic graph editing instructions or planning control instructions.

[0088] Specifically, the interaction instruction abstraction layer receives events from the web front-end, mobile touch input, and voice input, and normalizes them into semantic instructions such as "add node", "modify attribute", and "adjust confidence".

[0089] The visualization layer is used to generate and maintain a unified visualization workspace on the user end based on the state of the semantic memory and the progress of the task scheduler, and to respond to instructions transmitted by the interaction instruction abstraction layer.

[0090] Specifically, the visualization layer is based on D3.js and Three.js to render two-dimensional / three-dimensional maps, supporting drag-and-drop, zooming, selection by bounding box, and hovering over details.

[0091] The task scheduler includes a built-in planning utility evaluation module. Before each adjustment to the dynamic task planning, this module simulates different planning paths before and after the adjustment based on the current state of the project semantic graph. The module calculates the expected interaction cost and scheme completeness score for each simulated path, and provides the planning adjustment scheme corresponding to the path with the highest score as the preferred suggestion to the large model as a decision reference.

[0092] Furthermore, the simulation calculation of the planning utility evaluation module introduces a user-personalized collaboration model trained based on historical interaction logs; this personalized collaboration model can predict the possible feedback efficiency and preference choices of a specific user when faced with different types of information inquiries or decision branches; the calculation of the expected interaction cost will be weighted and corrected by combining the predicted output of the personalized collaboration model.

[0093] Furthermore, the interactive instruction abstraction layer also includes a device adaptation submodule; the device adaptation submodule is used for: a) Detect the device type and input capabilities of the current operating environment; b) Based on the detection results, output layout strategy parameters to the visualization presentation layer, and dynamically adjust the specific rule set that maps the original user operation events to semantic instructions; c) Wherein, the device type includes at least desktop and mobile touch terminals.

[0094] It should be noted that the system provided in this embodiment is the corresponding system of the method provided in embodiment 2. Therefore, the parts that are the same as or similar to those in embodiment 2 in this embodiment can be referred to each other, and will not be described again in this application.

Claims

1. A method for visually guiding architectural scheme comprehension based on structured semantic memory, characterized in that: include: S100, Dynamic Task Planning: Based on user-initiated architectural scheme editing and review tasks and initial information, a dynamic task plan containing multiple execution stages is generated and maintained by a large model. S200, Progressive Structuring: Execute the current step of the dynamic task planning, guide the user to input information, parse and structure the input content to form or update the project semantic graph, and assign understanding confidence to the graph elements; wherein, the project semantic graph is composed of multiple semantic gene units and their relationships, and can dynamically extract local structures related to the current execution stage according to the task context to form a project semantic graph subgraph; S300, Integrated Visualization of Understanding: The progress of the dynamic task planning, the current active steps, and the associated semantic graph subgraphs are integrated and visualized in the interactive interface. The visual style of the graph elements is configured differently based on their understanding confidence. S400, Direct Editing and Guidance of Structured Memory: Responding to user editing operations on visual graph elements; S500, Semantic Incremental Reasoning and Planning Adjustment: The editing operation is converted into a semantic incremental instruction, which triggers the large model to re-reason based on the updated semantic graph, update the understanding confidence and dynamically adjust the subsequent dynamic task planning. S600, Loop and Output: Loop through S200 to S500 until the dynamic task planning is completed, and output the editing and review results based on the final semantic graph.

2. The method for visually guiding architectural scheme comprehension according to claim 1, characterized in that, S200 includes: S210. The large model actively initiates a request based on the objective of the current step. S220. Receive a response from the user to the request, the response including at least one of text input, file upload, and parameter selection; S230. Call the semantic parsing model in the architectural field to parse the response and identify predefined semantic gene units and their relationships; S240. The identified semantic gene units and their associations are added or updated to the project semantic graph in an incremental manner, and an initial understanding confidence is calculated for each newly added or changed graph element.

3. The method for visually guiding architectural scheme comprehension according to claim 2, characterized in that, The S300 includes: S310. Render a unified visual workspace in the interactive interface, which includes at least a task planning view area and a semantic memory view area. S320. In the task planning view area, the dynamic task plan is dynamically displayed in the form of a flowchart or Gantt chart, and different visual states are used to distinguish the completed stage, the currently active stage, and the stage to be executed. S330. In the semantic memory view area, the project semantic graph subgraph associated with the currently active step is dynamically displayed in an interactive graph structure. S340. Configure visual mapping rules for the graph nodes and edges in the semantic memory view area: wherein the fill color or size of the node maps to its understanding confidence, and the solid or void style or color of the edge maps to the degree of certainty of the relationship it represents.

4. The method for visually guiding architectural scheme comprehension according to claim 3, characterized in that, In S400, the editing operation includes at least one of the following: Content modification operation: Directly modify the semantic content or parameter values ​​of graph nodes; Relationship editing operations: Create, delete, or change semantic relationship edges between nodes; Confidence intervention: Manually adjust the understanding confidence value of a specified node; Annotation anchoring operation: Add annotations, references to standard provisions, or file links to nodes.

5. The method for visually guiding architectural scheme comprehension according to claim 4, characterized in that, The S500 includes: S510. The user's editing operation is solidified into an update instruction for the project semantic graph, triggering the graph version evolution, and the update is synchronized to the dialogue context of the large model. S520, the large model, based on the updated graph context, re-evaluates the preconditions and execution necessity of each step to be executed in the incomplete dynamic task planning. S530. Based on the results of the reassessment, perform at least one of the following operations on the dynamic task planning: skipping redundant steps, adding necessary steps, or adjusting the order of steps.

6. The method for visually guiding architectural scheme comprehension according to claim 5, characterized in that, In the S400 content correction operation, when the node parameter modified by the user belongs to the predefined set of key engineering parameters, the engineering rule verification process is automatically triggered: the chain effect caused by the parameter modification is simulated and evaluated, and the evaluation results are displayed in the form of warnings or suggestions on the corresponding nodes.

7. The method for visually guiding architectural scheme comprehension according to claim 6, characterized in that, Also includes: Global view switching: In response to user actions, switch from a view showing details of the current step to an integrated overview view; In the integrated overview view, the dynamic task planning and the complete project semantic graph are overlaid. The nodes of each stage of the task planning are visually connected to the core semantic nodes generated in the graph at that stage, and the graph nodes are clustered according to the task stage to which they belong.

8. The method for visually guiding architectural scheme comprehension according to claim 7, characterized in that, Also includes: Context-aware information prefetching and filling: In the semantic memory view area of ​​the currently active step, if a graph node is detected as a required parameter but the content is missing or the confidence level is too low, the system will automatically retrieve the semantic graph of the architectural domain knowledge base or historical projects and suggest the obtained candidate values ​​near the node for the user to adopt and fill.

9. A system for implementing the method of any one of claims 1 to 8, characterized in that, include: The task scheduler is used to execute the logic for generating, maintaining, and advancing dynamic task planning steps. Semantic memory is used to store and manage the project's semantic graph in the form of a graph database, recording the change history of all graph elements and the changes in understanding confidence. An interactive rendering engine is used to connect the task scheduler and the semantic memory, and it includes an interactive instruction abstraction layer and a visualization presentation layer. The interactive instruction abstraction layer is used to normalize raw user operation events from different input devices into semantic graph editing instructions or planning control instructions; The visualization layer is used to generate and maintain a unified visualization workspace on the user end based on the state of the semantic memory and the progress of the task scheduler, and to respond to instructions transmitted by the interaction instruction abstraction layer.

10. A computer device, characterized in that, The method includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the computer program, implements the method as claimed in any one of claims 1 to 8.

Images