Context triple-based building scheme compiling and auditing dialogue state management method

Abstract

The invention relates to a context triple-based building scheme editing and auditing dialogue state management method, which comprises the following steps of: automatically capturing and constructing a context triple when a user browses a building scheme document and initiates a dialogue; performing lightweight structured processing on the context triad to generate a dialogue context snapshot independent of the complete document; establishing a mapping relationship between the dialogue context snapshots and the corresponding dialogue record entries, and storing the mapping relationship; and in response to a selection operation on a historical dialogue record entry, reading a corresponding dialogue context snapshot based on the stored mapping relationship, synchronously driving the document preview pane to restore to an original browsing state, and providing context information in the snapshot to the AI assistant to reconstruct dialogue memory. The method has the beneficial effects that structured packaging and one-key instantaneous recovery of the editing and auditing work site are realized, and the problem of anaphora ambiguity caused by dialogue and visual state isolation is effectively solved.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of building engineering informatization, and particularly relates to a building scheme compilation and review dialogue state management method based on context triplets. BACKGROUND

[0002] In the field of architecture, engineering and construction (AEC), it is prior art to preview scheme documents using computer software and to use AI assistants for natural language interaction. However, when trying to deeply combine the two to build an intelligent review workflow, the architectural defects of the prior art lead to the following insurmountable bottlenecks: 1. Reference ambiguity and inefficient interaction: The running context of the AI assistant and the visual state of the document preview pane are in complete isolation. When the user issues instructions such as "review this paragraph" or "is this data reasonable" based on the current browsed document content (such as a certain clause, a specific chart), the AI cannot perceive the specific visual reference object of the word "this", resulting in the dialogue having to go through multiple tedious clarifications and confirmations, and the interaction efficiency and task accuracy are difficult to guarantee. The essence is the separation of "dialogue instructions" and "visual focus".

[0003] 2. Collaboration site fault and knowledge loss: The building review task is often completed by multiple members in relay. The prior art lacks effective means to encapsulate and share the complete work site (including dialogue records, document visual focus and deep building review business context) created by a member in a specific dialogue in a lightweight and structured manner. This results in the inability of the subsequent participants to restore the previous decision site when the task is switched or taken over by others, and a large amount of time must be spent on manual positioning and background communication, and the collaboration process is severely fragmented. The essence is that the composite knowledge body of "work site" cannot be effectively encapsulated and transmitted.

[0004] 3. AI cognitive overload and context bottleneck: The prior art usually provides the AI with the full amount of long dialogue history to maintain its memory. In long-term, multi-round review tasks, this not only leads to inefficient use of network and computing resources, but more importantly, the excessive amount of pure text context interferes with the core reasoning ability of the AI, making it difficult to accurately focus on the most critical task information from the vast amount of text. At the same time, the pure text form cannot effectively transmit visual information such as charts and drawings, limiting the AI's comprehensive understanding and analysis capabilities in building review. The essence is the failure to build an efficient and low-noise cognitive context for the AI composed of "text instructions" and "visual focus".

[0005] Therefore, there is an urgent need in this field for an innovative technical solution that can break through the mindset of existing architectures and unify and manage the three elements of dialogue instructions, visual focus, and business context through a structured approach, so as to provide such systems with an efficient, lightweight mechanism for accurate saving and instant recovery of the work scene. Summary of the Invention

[0006] The purpose of this invention is to overcome the shortcomings of the prior art and provide a method for managing the state of architectural scheme editing and review dialogue based on context triples.

[0007] Firstly, a method for managing the state of architectural scheme editing and review dialogues based on context triples is provided, applicable to interactive systems including navigation panes, dialog panes, and document preview panes. The method includes: S1. When a user browses architectural design documents and initiates a dialogue, automatically capture and construct a context triplet, which includes the user's dialogue command, the document's visual focus, and the architectural editing and review business context. S2. Perform lightweight structuring processing on the context triples to generate a dialog context snapshot independent of the complete document; S3. Establish and store the mapping relationship between the dialogue context snapshot and the corresponding dialogue record entry; S4. In response to the selection operation of the historical dialogue record entry, the corresponding dialogue context snapshot is read based on the stored mapping relationship, and the document preview pane is synchronously driven to restore the original browsing state. At the same time, the context information in the snapshot is provided to the AI ​​assistant to reconstruct the dialogue memory.

[0008] Preferably, in S1, the document visual focus is used to refer to the specific document area currently focused within the document preview pane that is related to the user's dialogue command; The document's visual focus is characterized by a set of structured positioning data decoupled from the document's original format; When the visual focus of the document is the text content, the structured positioning data includes the document identifier, chapter path, and text object identifier. When the visual focus of the document is graphic content, the structured positioning data includes drawing identifiers, layer information, element coordinates, or globally unique identifiers of building information model components.

[0009] Preferably, the architectural editing and review business context refers to the review task and regulatory basis to which the current dialogue is based; the construction of the architectural editing and review business context includes: Extract task attribute information from the current construction project editing and review tasks; Based on the task attribute information, query the pre-set specification mapping relationship library to retrieve the relevant set of building specification clause numbers; The task attribute information and the set of standard clauses are combined to form the architectural editing and review business context.

[0010] Preferably, S2 includes: S201. Extract key parameters from the context triples and extract the binary content of the complete document; S202. Construct a lightweight snapshot metadata set based on the key parameters; S203. Serialize and encapsulate the metadata set to generate a dialogue context snapshot file.

[0011] Preferably, S3 includes: S301. Assign a globally unique identifier to each dialogue context snapshot; S302. Establish a relation field for storing the globally unique identifier for the dialogue record entry; S303. The mapping query between dialogue record entries and dialogue context snapshots is realized through the relationship field.

[0012] Preferably, in S4, when a user selects a historical conversation record and performs a restore operation, state restoration is performed through an independent context restoration engine, including: S401, Intercept the normal loading process of the document preview pane; S402. Parse the metadata in the dialogue context snapshot; S403. Generate navigation instructions that directly reach the specified visual focus based on the analysis results; S404: The driver document preview pane directly performs positioning and rendering operations.

[0013] Preferably, a context inheritance mechanism across dialogue sessions is also included, specifically including: S501. When a new conversation session is detected and the current browsing state matches an existing snapshot, a prompt interface is provided to the user. S502. Provide the option to inherit the business context from the existing snapshot through the prompt interface; S503: In response to the user's confirmation, set the associated business context as the initial context of the new session and provide it to the AI ​​assistant.

[0014] Preferably, in S3, the dialogue context snapshot is stored in encrypted form in a dedicated snapshot file library and supports access control based on the project level.

[0015] Preferably, in step S4, providing the contextual information from the snapshot to the AI ​​assistant to reconstruct the dialogue memory includes: The business context of architectural editing and review is reconstructed into system prompts to limit the scope of the AI ​​assistant's responses; Resubmit the original user dialogue command as the starting point of the dialogue; Based on the prompts and dialogue instructions, the AI ​​assistant can generate responses without replaying the entire history of dialogues.

[0016] Secondly, an interactive system for compiling and reviewing architectural plans is provided, for implementing the method described in any one of the first aspects, the system comprising: The document preview pane is used to render and interact with architectural design documents; The dialog pane is used to receive user commands and display conversations with the AI ​​assistant; The navigation pane is used to display the conversation history; The context management engine is used to manage the generation, storage, and restoration of conversation context snapshots; The system is configured to avoid transmitting, storing, or loading a complete copy of the architectural design document during the saving and restoring of the dialogue context.

[0017] The beneficial effects of this invention are as follows: By constructing and managing "context triples", this invention realizes the structured encapsulation and one-click instant recovery of the editing and reviewing work site, effectively solves the problem of ambiguity caused by the isolation of dialogue and visual state, overcomes the on-site gaps and knowledge loss in team collaboration, and significantly improves the professional accuracy, task continuity and overall system efficiency of the AI ​​assistant by providing it with strong constraint prompts reconstructed from business context. Attached Figure Description

[0018] Figure 1 This is a flowchart of the dialogue context snapshot generation and state restoration provided by the present invention; Figure 2 This is a schematic diagram of the cross-dialogue session context inheritance mechanism provided by the present invention. Detailed Implementation

[0019] 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.

[0020] Example 1: To address the issues of low interaction efficiency, fragmented collaboration, and AI cognitive overload in existing technologies, Embodiment 1 of this application provides a method for managing the dialogue state of architectural scheme editing and review based on context triples. This method aims to achieve unified association and instantaneous restoration of dialogue instructions, document visual focus, and business context throughout the entire process of construction scheme editing and review (generation, modification, and review). This method is applied to interactive systems that include navigation panes, dialogue panes, and document preview panes, such as... Figure 1 As shown, the method specifically includes: S1. When a user browses architectural design documents and initiates a dialogue, automatically capture and construct a context triplet, which includes the user's dialogue command, the document's visual focus, and the architectural editing and review business context.

[0021] In S1, the document visual focus is used to refer to the specific document area currently focused within the document preview pane that is related to the user's dialogue command; The document's visual focus is characterized by a set of structured positioning data decoupled from the document's original format; When the visual focus of the document is the text content, the structured positioning data includes the document identifier, chapter path, and text object identifier. When the visual focus of the document is graphic content, the structured positioning data includes drawing identifiers, layer information, element coordinates, or globally unique identifiers of building information model components.

[0022] The architectural editing and review business context refers to the review task and regulatory basis to which the current dialogue is based; the construction of the architectural editing and review business context includes: Extract task attribute information from the current construction project editing and review tasks. This information includes at least the task type and task identifier.

[0023] Based on the task attribute information, the system queries a pre-set code mapping database to retrieve a set of relevant building code clause numbers. Specifically, based on the task type, the system queries a pre-set code-task mapping database and automatically retrieves one or more building code clause numbers strongly related to that type of review task, forming a set of code clauses.

[0024] The task attribute information and the set of standard clauses are combined to form the architectural editing and review business context.

[0025] For example, let's assume scenario A: during the solution modification phase.

[0026] The user (the plan drafter) opens the generated draft of the "Special Construction Plan for Foundation Pit Support" in the preview pane. They scroll to the "4.2 Stability Verification" section under "IV. Support Structure Calculation," which contains a calculation formula and result for the "Overturning Safety Factor." The user positions the text cursor on the line containing this formula, or selects the entire paragraph containing the formula using the mouse. They then enter the command in the dialog pane: "Is the basis for this safety factor value sufficient?" and send it.

[0027] The system detects this dialogue initiation behavior, and the context management engine immediately starts, automatically capturing and constructing a context triple: User dialogue command: The system records the original text "Is the basis for this security factor value sufficient?".

[0028] Document visual focus: The system generates a set of lightweight, structured positioning data, including the document identifier (e.g., “SP-Foundation-2024-001”), the path of the currently focused chapter (e.g., “ / Chapter 4 / 4.2”), and the internal identifier of the focused text paragraph or formula within the document structure.

[0029] Architectural editing and review business context: The system extracts the current project task identifier (such as "Task-DesignModification-02") and task stage ("scheme modification"), and queries the business rule base based on the focus content ("support structure calculation"), automatically associating the set of standard clauses, such as ["GB 50330-2013 Clause 5.2.1"].

[0030] For example, let's consider scenario B: during the proposal review stage.

[0031] A reviewer is examining the "Mass Concrete Construction Plan" in the preview pane. He browses to the "Temperature Measurement Point Layout Diagram" under the "V. Temperature Control Monitoring Measures" section (this diagram is embedded in the document as a vector graphic). The user selects the diagram by clicking with the mouse. He then types in the dialog pane: "Can this monitoring point layout effectively capture the area of ​​highest temperature rise?" In this scenario, the capture process of the context triple is as follows: User dialogue command: Record "Can this monitoring point effectively capture the area with the highest temperature rise?".

[0032] Document visual focus: The structured positioning data generated by the system includes: document identifier (such as "SP-Concrete-2024-005"), chapter path (" / Chapter 5"), and element identifier of the embedded diagram, and relative coordinate range within the page (such as [x1, y1, x2, y2]).

[0033] Architectural editing and review business context: The task stage extracted by the system is "formal review", and based on the technical point of "temperature control monitoring", it is associated with the standard ["GB 50496-2018 Article 6.0.8"].

[0034] S2. Perform lightweight structuring processing on the context triples to generate a dialog context snapshot independent of the complete document.

[0035] S2 includes: S201. Data Extraction and Decoupling: Extract and discard the original binary content of the architectural scheme document referenced by the context triple and the full session history, and extract only the key parameters used to relocate and restore the state in future sessions.

[0036] Specifically, the engine strips and discards the original binary content of the complete construction plan document, extracting only key parameters such as user instruction text, visual focus positioning data, task identifiers, and a list of specification clauses.

[0037] S202. Metadata Construction: Based on the extracted key parameters, construct a lightweight snapshot metadata set, which encapsulates the following information: semantic summary of dialogue instructions; structured path information independent of the rendering engine for locating the visual focus of the document; task identifiers and index list of specification clauses representing the business context of architectural editing and review.

[0038] S203, Serialization and Encapsulation: Serialize and encapsulate the metadata set to generate a dialogue context snapshot file.

[0039] Specifically, the snapshot metadata set and a file header used to identify its version and structure are serialized together to generate a final dialog context snapshot file (such as JSON format) that is significantly smaller in size than the complete architectural plan document it is associated with. A globally unique identifier is assigned and mapped to the dialog record entry, and then the snapshot is encrypted and stored in a dedicated system library.

[0040] S3. Establish a mapping relationship between the dialogue context snapshot and the corresponding dialogue record entry and store it.

[0041] S3 includes: S301. In the system database, assign a globally unique identifier to each dialogue context snapshot; S302. In the dialog history list of the navigation pane, create an invisible relationship field for the corresponding dialog record entry, which stores the globally unique identifier; S303. The mapping query between dialogue record entries and dialogue context snapshots is realized through the relationship field.

[0042] The mapping relationship enables the system to instantly locate and load the corresponding complete conversation context snapshot when a user selects any conversation record entry in the graphical list of the navigation pane by querying the relationship field.

[0043] S4. In response to the selection operation of the historical dialogue record entry, the corresponding dialogue context snapshot is read based on the stored mapping relationship, and the document preview pane is synchronously driven to restore the original browsing state. At the same time, the context information in the snapshot is provided to the AI ​​assistant to reconstruct the dialogue memory.

[0044] In S4, when a user selects a historical conversation record and performs a restore operation, state restoration is performed through an independent context restoration engine, including: S401, Intercept the normal loading process of the document preview pane; S402. Parse the metadata in the dialogue context snapshot; S403. Generate navigation instructions that directly reach the specified visual focus based on the analysis results; S404. Upon receiving this instruction, the document preview pane bypasses its standard content initialization sequence and directly performs positioning and rendering operations.

[0045] For example, when a user subsequently clicks on a history conversation entry, the system performs a restore: Snapshot reading and parsing: Locate and decrypt snapshot files through mapping relationships.

[0046] Visual state restoration: The context restoration engine parses location data (such as chapter paths and element identifiers) from the snapshot and generates navigation instructions. The preview pane then jumps directly to the specific chapter in the document and highlights the previously focused content.

[0047] AI Memory Reconstruction: The system reconstructs the business context (task phase, specification clause) in the snapshot into strongly binding system prompts, and submits them to the AI ​​assistant along with the original user instructions, enabling it to generate professional responses in a precise context.

[0048] Example 2: Based on Embodiment 1, Embodiment 2 of this application provides a more specific method for managing the state of architectural scheme editing and reviewing dialogues based on context triples, applied to an interactive system including a navigation pane, a dialog pane, and a document preview pane. The method includes: S1. When a user browses architectural design documents and initiates a dialogue, automatically capture and construct a context triplet, which includes the user's dialogue command, the document's visual focus, and the architectural editing and review business context. S2. Perform lightweight structuring processing on the context triples to generate a dialog context snapshot independent of the complete document.

[0049] S3. Establish a mapping relationship between the dialogue context snapshot and the corresponding dialogue record entry and store it.

[0050] In S3, the dialogue context snapshot is stored in encrypted form in a dedicated snapshot file library and supports access control based on the project level.

[0051] Specifically, the dialogue context snapshot is stored in a dedicated snapshot file library in the form of encrypted data blocks, independent of the dialogue record text it maps to; The snapshot file library is configured to support access control based on the construction project level, ensuring that in a team collaboration environment, only users authorized to participate in a specific project can access and restore snapshots of the dialogue context related to the review process of that project. S4. In response to the selection operation of the historical dialogue record entry, the corresponding dialogue context snapshot is read based on the stored mapping relationship, and the document preview pane is synchronously driven to restore the original browsing state. At the same time, the context information in the snapshot is provided to the AI ​​assistant to reconstruct the dialogue memory.

[0052] In S4, providing the contextual information from the snapshot to the AI ​​assistant to reconstruct the dialogue memory includes: The system reconstructs the architectural editing and review business context into a set of strongly binding system prompts, which clearly define the scope of norms and review perspectives that the AI ​​assistant must follow in its subsequent responses; At the same time, the system resubmits the original user dialogue commands captured in the snapshot as the starting point of the historical dialogue; The combined effect of these two factors allows AI assistants to achieve a predictable and significant improvement in professional relevance and task continuity in their output responses compared to dialogues lacking the aforementioned context, without needing to fully replay the entire historical dialogue.

[0053] S5, cross-dialogue session context inheritance mechanism, such as Figure 2 As shown, it specifically includes: S501. When the system detects that a user has started a new conversation, and the document and visual focus displayed in the current document preview pane highly match the record in a stored conversation context snapshot, the system provides a prompt interface to the user in the conversation pane; the matching conditions include: the document identifier currently being viewed is the same as the document identifier in the snapshot, and the structured positioning data of the current visual focus is consistent with the visual focus positioning data in the snapshot at the semantic level (e.g., belonging to the same chapter path, or pointing to the same graphic element identifier).

[0054] S502. This prompt interface provides the option to inherit the architectural editing and review business context from the stored snapshot to the current new session with one click; S503. In response to the user's confirmation of the option, the system automatically sets the associated building editing and review business context as the initial context of the current new session and provides it to the AI ​​assistant.

[0055] For example, during the review phase, when the system detects that the currently viewed chapter matches a historical snapshot from the revision phase, it can prompt the user to inherit the business context of that time (such as associated specification clauses). After the user confirms, this context is set as the background of the new session, ensuring the continuity of knowledge transfer throughout the entire process.

[0056] 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.

[0057] Example 3: Based on Embodiment 2, Embodiment 3 of this application provides an interactive system for the compilation and review of architectural plans, the system comprising: The document preview pane is used to render and interact with architectural design documents; The dialog pane is used to receive user commands and display conversations with the AI ​​assistant; The navigation pane is used to display the conversation history; The context management engine is used to manage the generation, storage, and restoration of conversation context snapshots; The system is configured to enable lightweight, instantaneous rewind and continuation of work status without transmitting, storing or loading a complete copy of the architectural design document during the saving and restoring of the dialogue context.

[0058] 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.

[0059] In summary, this invention effectively solves problems such as the isolation between AI dialogue and document visual state, the fragmentation of collaboration on-site, and the low efficiency of full data reload by dynamically managing and reusing the aforementioned context triples. It achieves seamless connection and efficient collaboration in the architectural editing and review workflow, significantly improving the collaboration efficiency and work quality of each stage from editing to review.

Claims

1. A method for managing the state of a dialogue for the review of an architectural project based on context triplets, characterized by, Applied to an interactive system comprising a navigation pane, a dialog pane and a document preview pane, the method comprises: S1. When a user browses a building scheme document and initiates a dialog, a context triple is automatically captured and constructed, the triple comprising a user dialog instruction, a document visual focus and a building review business context; S2. The context triple is subjected to lightweight structured processing to generate a dialog context snapshot independent of the complete document; S3. The dialog context snapshot is mapped to a corresponding dialog record entry and stored; S4. In response to a selection operation on a historical dialog record, the corresponding dialog context snapshot is read based on the stored mapping relationship, and the document preview pane is synchronously driven to restore to the original browsing state, while the context information in the snapshot is provided to an AI assistant to reconstruct the dialog memory. 2.The context-triplet-based building scheme review dialogue state management method according to claim 1, wherein, In S1, the document visual focus refers to a specific document area currently focused in the document preview pane and related to the user dialog instruction; The document visual focus is characterized by a set of structured positioning data decoupled from the original format of the document; When the document visual focus is text content, the structured positioning data comprises a document identifier, a chapter path and a text object identifier; When the document visual focus is graphic content, the structured positioning data comprises a drawing identifier, layer information, element coordinates or a globally unique identifier of a building information model component. 3.The context-triplet-based building scheme review dialogue state management method according to claim 2, wherein, The building review business context refers to the review task and the basis of the current dialog; The construction of the building review business context comprises: Extracting task attribute information from the current building project review task; According to the task attribute information, the relevant building code clause number set is retrieved from the pre-stored specification mapping relationship library; The task attribute information and the specification clause set are combined to form the building review business context.

4. The method of claim 3, wherein S2 Comprise: S201. Extract key parameters from the context triple and strip the binary content of the complete document; S202. Construct a lightweight snapshot metadata set based on the key parameters; S203. Serialize and package the metadata set to generate a dialog context snapshot file.

5. The method of claim 4, wherein S3 Comprise: S301. Assign a globally unique identifier to each dialog context snapshot; S302. Establish a relationship field in the dialog record entry to store the globally unique identifier; S303. Map the dialog record entry and the dialog context snapshot through the relationship field. 6.The context-triplet-based building scheme review dialogue state management method according to claim 5, wherein, In S4, when the user selects a historical dialog record and performs a restore operation, a state restoration is performed through an independent context restoration engine, comprising: S401. Intercept the regular loading process of the document preview pane; S402. Parse the metadata in the dialog context snapshot; S403. Generate navigation instructions directly to the specified visual focus based on the parsing result; S404. Drive the document preview pane to directly perform positioning and rendering operations. 7.The context-triplet-based building scheme review dialogue state management method according to claim 6, wherein, Further comprising a context inheritance mechanism across dialog sessions, specifically comprising: S501. When a new dialog session is detected and the current browsing state matches the stored snapshot, a prompt interface is provided to the user; S502, providing an option of inheriting the business context in the existing snapshot through the prompt interface; S503, in response to the user confirmation operation, setting the associated business context as the initial context of the new session and providing it to the AI assistant. 8.The context-triplet-based building scheme review dialogue state management method of claim 7, wherein, In S3, the dialogue context snapshot is stored in an encrypted form in a dedicated snapshot file library, and supports access permission control based on project level. 9.The context-triplet-based building scheme review dialogue state management method of claim 8, wherein, In S4, the context information in the snapshot is provided to the AI assistant to reconstruct the dialogue memory, including: reconstructing the building review business context into system prompt words to limit the answer range of the AI assistant; resubmitting the original user dialogue instruction as the dialogue starting point; based on the prompt words and dialogue instructions, making the AI assistant generate responses without replaying the full historical dialogue.

10. An interactive system for architectural scheme review for implementing the method of any one of claims 1 to 9, characterized in that, The system comprises: a document preview pane for rendering and interacting with the building scheme document; a dialogue pane for receiving user instructions and displaying the dialogue with the AI assistant; a navigation pane for presenting the dialogue history; a context management engine for managing the generation, storage and recovery of dialogue context snapshots; wherein the system is configured to not transmit, store or load the complete copy of the building scheme document during the dialogue context saving and recovery process.

Images