Collaborative intelligent agent-oriented high-fidelity style reference relationship cognitive enhancement method

Abstract

The application discloses a kind of high fidelity style reference relationship cognitive enhancement methods for collaborative intelligent agent and related devices, wherein the method comprises: generating design context vector according to fine-grained interactive behavior flow and the structure metadata of current design document;According to the design context vector, a dynamically evolving design cognitive ontology is constructed, and an initial design cognitive graph is constructed according to the design cognitive ontology and cross-document reference link;The initial design cognitive graph is input into the comparative graph neural network and causal reasoning network model to identify implicit style reference and design intent, and output high-order reference relationship graph with intent label and confidence;According to the high-order reference relationship graph, the origin node and evolution tree of the design element, the use context and team consensus intensity score are dynamically presented.The application realizes the systematic representation and organization of design style knowledge, improves the accuracy of design style reference relationship identification, dynamically analyzes and predicts the evolution trend of design style, and provides support for design decision.

Description

Technical Field

[0001] This invention relates to the field of software design technology, specifically to a high-fidelity style reference relationship recognition enhancement method, apparatus, and computing device for collaborative intelligent agents. Background Technology

[0002] In the field of modern design, the reference and evolution relationships of design styles are complex and diverse. Traditional design style analysis methods mainly rely on human experience, resulting in inefficiency, strong subjectivity, and difficulty in handling large-scale data. With the development of artificial intelligence and big data technologies, knowledge graphs, as a semantic network structure, offer new possibilities for the intelligent analysis of design style relationships. However, current knowledge graph-based design style analysis systems still face the following challenges: the representation and organization of design style knowledge are not systematic enough; the reference relationships between design styles are difficult to accurately identify and quantify; there is a lack of dynamic analysis capabilities for the evolution process of design styles; and the interpretability and visualization of analysis results are insufficient.

[0003] To address the aforementioned issues, this invention proposes a high-fidelity style reference relationship recognition enhancement method for collaborative intelligent agents, which improves the accuracy of design style reference relationship identification, dynamically analyzes and predicts the evolution trend of design styles, and provides support for design decisions. Summary of the Invention

[0004] In view of the above problems, the present invention provides a high-fidelity style reference relationship recognition enhancement method, apparatus, and computing device for collaborative intelligent agents.

[0005] According to one aspect of the present invention, a high-fidelity style reference relationship recognition enhancement method for cooperative intelligent agents is provided, comprising:

[0006] The system collects fine-grained interaction behavior flows from designers during canvas operations in real time, and generates a design context vector based on these fine-grained interaction behavior flows and the structural metadata of the current design document. The fine-grained interaction behavior flows include component instantiation, style overriding, main component synchronization, variant derivation, and cross-project template reuse. The structural metadata includes layer topology, style tokens, version snapshots, and collaborator identities.

[0007] A dynamically evolving design cognitive ontology is constructed based on the design context vector. An initial design cognitive graph is constructed based on the design cognitive ontology and cross-document reference links. The nodes of the design cognitive ontology include atomic styles, composite specifications, and organizational-level design assets. The edge relationships of the design cognitive ontology include explicit inheritance, implicit imitation, conflict coverage, and team adoption.

[0008] The initial design cognitive map is input into a contrastive graph neural network and causal reasoning network model to identify implicit style references and design intentions, and outputs a high-order reference relationship graph with intention labels and confidence scores; wherein, the contrastive graph neural network and causal reasoning network model includes a semantic encoder composed of Tree-LSTM network layers, an instance visual encoder composed of CNN and GNN, a causal intervention decoupler composed of counterfactual intervention layers and causal attention layers, and a multi-view contrast alignment head;

[0009] When a designer selects any design element, the origin node and evolution tree of the design element, usage context, and team consensus strength score are dynamically presented according to the high-order reference relationship graph. When the main component or design token of the design element changes, the impact propagation simulation is performed according to the high-order reference relationship graph to predict the visual consistency risk of downstream projects and generate a multi-level remediation strategy package.

[0010] In one alternative approach, generating the design context vector based on the fine-grained interaction behavior flow and the structural metadata of the current design document further includes:

[0011] Feature extraction is performed on the component instantiation operations in the fine-grained interactive behavior flow to identify the main component ID it references and the initial attribute snapshot at the time of instantiation;

[0012] Perform a difference analysis on the style overlay to calculate the style difference vector between the attribute values ​​of the current component instance and the referenced main component or atomic style;

[0013] Version tracking is performed on the main component, and style token change logs are recorded before and after synchronization.

[0014] Perform topology analysis on the variant derivation to construct a variant inheritance chain between the derived component and the source component;

[0015] The cross-project template reuse is matched with the organizational asset library to identify the reused template ID and its adaptation modification records in the target project;

[0016] The extracted main component ID, initial attribute snapshot, style difference vector, style token change log, variant inheritance chain, template ID, adaptation modification record, and structural metadata are fused to generate the design context vector of the current design state and historical operation sequence.

[0017] In one alternative approach, constructing a dynamically evolving design cognitive ontology based on the design context vector further includes:

[0018] The design context vector is subjected to temporal clustering to identify design patterns that frequently co-occur on the time axis and are semantically stable, and these patterns are extracted into atomic style nodes; multiple atomic style nodes with hierarchical dependencies or compositional relationships are aggregated into composite canonical nodes;

[0019] Elevate frequently reused atomic patterns or composite specification nodes that have been confirmed by the team to organizational-level design asset nodes; identify the edge relationships between nodes by analyzing the operation sequences contained in the design context vector;

[0020] Specifically, if a component is detected referencing a parent style via the main component ID during instantiation, an explicit inheritance edge is established; if two nodes with no direct reference relationship are detected to be highly similar in visual attributes via style difference vectors, an implicit imitation edge is established; if a style manually modified by the designer is detected to conflict with the inherited parent style, a conflict overriding edge is established; if an atomic style or composite specification is detected to be repeatedly used by multiple collaborators in different projects, a team adoption edge is established, thereby generating a design cognitive ontology containing nodes, edges, and their evolution sequence.

[0021] In one alternative approach, constructing an initial design cognitive graph based on the design cognitive ontology and cross-document reference links further includes:

[0022] The first type of node is defined as atomic patterns, composite specifications, and organizational-level design assets in the design cognitive ontology.

[0023] Traverse the current project and its referenced external project libraries to extract cross-document reference links between all design elements. These cross-document reference links include cross-file main component reference relationships and cross-project style library link relationships.

[0024] The extracted cross-document reference links are added to the design cognition ontology as second-type nodes and edges to form the initial design cognition graph; wherein, the node attributes of the initial design cognition graph include node type, unique identifier, visual attribute vector and creation timestamp, and the edge attributes include relation type, initial confidence value and last synchronization timestamp.

[0025] In an alternative approach, the method further includes:

[0026] When identifying implicit style references, the contrastive graph neural network and causal reasoning network model uses the canonical semantics of the first type of cognitive nodes and the visual features of the second type of instance nodes to perform comparative learning, thereby obtaining the potential implicit reference relationships between instances and canonical references or between instances.

[0027] When conducting impact propagation simulations, the contrastive graph neural network and causal reasoning network model starts with the first type of cognitive node corresponding to the main component or design token that has been changed, and propagates along the instantiation reference edge to all second type of instance nodes. It predicts the visual impact of the change on the design instance, thereby providing downstream projects with visual consistency risk assessment and multi-level remediation strategy packages.

[0028] In one alternative approach, dynamically presenting the origin node and evolution tree of the design element, usage context, and team consensus strength score based on the higher-order reference graph further includes:

[0029] Using the design element as an index, query the corresponding target node in the higher-order reference graph;

[0030] By traversing back through the explicit inheritance edges and implicit reference edges with confidence exceeding a preset threshold pointing to the target node in the higher-order reference graph, all upstream nodes are traced to construct the origin node and evolution tree of the design element, so as to show its evolution path from the basic style to the current form.

[0031] By forward traversing all outgoing edges from the target node, all direct and indirect downstream nodes are aggregated, and a list of usage contexts for the design element is generated by grouping and statistically analyzing them by project, page, or team.

[0032] The team consensus strength score is calculated based on the number of team adoption edges converged to the target node, the diversity of their sources, and the authority of the relevant collaborator identities.

[0033] In one alternative approach, performing impact propagation simulation based on the higher-order reference graph to predict visual consistency risks in downstream projects and generate a multi-level remediation strategy package further includes:

[0034] When a designer initiates a change operation on the main component or design token of the source design element, the source design element is locked. Starting from the node of the source design element in the higher-order reference graph, a breadth-first traversal is performed along all outgoing edges to identify all downstream elements that are directly or indirectly affected by the change and obtain the influence propagation graph.

[0035] Based on the project to which the downstream element belongs, the type of relationship with the source design element, and the conflict coverage of its edges, assess the visual consistency risk of each affected project to generate a risk report, wherein the risks include style conflicts, component deviations, and brand specification violations.

[0036] Based on the visual consistency risk of each downstream element in the impact propagation diagram, a multi-level remediation strategy package is automatically generated. The multi-level remediation strategy package includes an automatic batch synchronization strategy for downstream elements without conflict coverage, a manual review and recommendation strategy for downstream elements with conflict coverage, and a version rollback or new variant creation strategy for cross-project references.

[0037] In an alternative approach, after generating the multi-level remediation strategy package, the method further includes:

[0038] Automatically traverse all downstream elements that meet the preset conditions, synchronize the changed attributes of the source design elements to the downstream elements, and record this batch synchronization operation in the version history;

[0039] In response to the designer clicking the manual review recommendation strategy button, highlight all downstream elements with conflicting coverage and display their current style difference vector and the latest value of the parent style.

[0040] In response to designers clicking the version rollback or new variant strategy button, an isolated branch containing the current stable version is created for the affected downstream projects or component libraries, or the affected downstream elements are automatically derived into new component variants, breaking their inheritance relationship with the source design elements to maintain the visual stability of the original project.

[0041] According to another aspect of the present invention, a high-fidelity style reference relationship recognition enhancement device for cooperative intelligent agents is provided, comprising:

[0042] The design context vector generation module is used to collect fine-grained interaction behavior flows of designers during canvas page operations in real time, and generate design context vectors based on the fine-grained interaction behavior flows and the structural metadata of the current design document; wherein, the fine-grained interaction behavior flows include component instantiation, style overriding, main component synchronization, variant derivation, and cross-project template reuse; the structural metadata includes layer topology, style token, version snapshot, and collaborator identity;

[0043] The initial design cognitive graph construction module is used to construct a dynamically evolving design cognitive ontology based on the design context vector, and to construct an initial design cognitive graph based on the design cognitive ontology and cross-document reference links. The nodes of the design cognitive ontology include atomic styles, composite specifications and organizational-level design assets, and the edge relationships of the design cognitive ontology include explicit inheritance, implicit imitation, conflict coverage and team adoption.

[0044] The high-order citation graph generation module is used to input the initial design cognitive graph into a comparative graph neural network and causal reasoning network model to identify implicit style citations and design intentions, and output a high-order citation graph with intention labels and confidence scores.

[0045] The design element recognition and presentation module is used to dynamically present the origin node and evolution tree of any design element, its usage context, and the team consensus strength score based on the high-order reference relationship graph when a designer selects any design element. When the main component or design token of the design element changes, the module simulates the impact propagation based on the high-order reference relationship graph, predicts the visual consistency risk of downstream projects, and generates a multi-level remediation strategy package.

[0046] According to another aspect of the present invention, a computing device is provided, comprising: a processor, a memory, a communication interface, and a communication bus, wherein the processor, the memory, and the communication interface communicate with each other via the communication bus;

[0047] The memory is used to store at least one executable instruction, which causes the processor to perform the operation corresponding to the high-fidelity style reference relationship recognition enhancement method for cooperative agents described above.

[0048] According to the solution provided by this invention, fine-grained interactive behavior flow of designers during canvas page operations is collected in real time. A design context vector is generated based on the fine-grained interactive behavior flow and the structural metadata of the current design document. The fine-grained interactive behavior flow includes component instantiation, style overriding, main component synchronization, variant derivation, and cross-project template reuse. The structural metadata includes layer topology, style tokens, version snapshots, and collaborator identities. A dynamically evolving design cognitive ontology is constructed based on the design context vector. An initial design cognitive graph is constructed based on the design cognitive ontology and cross-document reference links. The nodes of the design cognitive ontology include atomic styles, composite specifications, and organizational-level design assets. The edge relationships of the design cognitive ontology include explicit inheritance, implicit imitation, conflict overriding, and team adoption. The initial design cognitive graph is then... The input is fed into a contrastive graph neural network and causal reasoning network model to identify implicit style references and design intent, and outputs a high-order reference graph with intent labels and confidence levels. The contrastive graph neural network and causal reasoning network model includes a semantic encoder composed of Tree-LSTM network layers, an instance visual encoder composed of CNN and GNN, a causal intervention decoupler composed of counterfactual intervention layers and causal attention layers, and a multi-view contrast alignment head. When a designer selects any design element, the origin node and evolution tree of that design element, its usage context, and team consensus strength score are dynamically presented based on the high-order reference graph. When the main component or design token of that design element changes, the impact propagation simulation is performed based on the high-order reference graph to predict the visual consistency risk of downstream projects and generate a multi-level remediation strategy package. This invention achieves a systematic representation and organization of design style knowledge, significantly improving the accuracy and comprehensiveness of design style reference relationship identification, and enabling dynamic analysis and prediction of design style evolution trends, providing support for design decisions. Furthermore, by utilizing knowledge graphs and causal reasoning networks, it achieves implicit reference relationship mining across projects and documents and intelligent reasoning for global style conflicts.

[0049] The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it in accordance with the contents of the specification, and in order to make the above and other objects, features and advantages of the present invention more apparent and understandable, specific embodiments of the present invention are described below. Attached Figure Description

[0050] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings:

[0051] Figure 1A flowchart illustrating the high-fidelity style reference relationship recognition enhancement method for collaborative intelligent agents according to an embodiment of the present invention is shown.

[0052] Figure 2 A schematic diagram illustrating the repair strategy and cognitive enhancement process of an embodiment of the present invention is shown;

[0053] Figure 3 A schematic diagram illustrating the process of generating cross-document reference links according to an embodiment of the present invention is shown;

[0054] Figure 4 This diagram illustrates the framework of a high-fidelity style reference relationship recognition enhancement device for collaborative intelligent agents according to an embodiment of the present invention.

[0055] Figure 5 A schematic diagram of the structure of a computing device according to an embodiment of the present invention is shown. Detailed Implementation

[0056] Exemplary embodiments of the invention will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the invention are shown in the drawings, it should be understood that the invention may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this invention will be thorough and complete, and will fully convey the scope of the invention to those skilled in the art.

[0057] Figure 1 A flowchart illustrating a high-fidelity style reference relationship recognition enhancement method for cooperative intelligent agents according to an embodiment of the present invention is shown. Specifically, as... Figure 1 As shown, it includes the following steps:

[0058] Step S101: Real-time acquisition of fine-grained interactive behavior flow during the designer's operation on the canvas page, and generation of design context vector based on the fine-grained interactive behavior flow and the structural metadata of the current design document; wherein, the fine-grained interactive behavior flow includes component instantiation, style overriding, main component synchronization, variant derivation and cross-project template reuse; the structural metadata includes layer topology, style token, version snapshot and collaborator identity.

[0059] In this embodiment, by capturing fine-grained operations such as component instantiation and style overriding, the designer's true intentions during the creative process can be accurately restored, avoiding semantic gaps caused by relying solely on static document structures. Integrating interactive behavior flows with structural metadata such as layer topology and version snapshots allows for dynamic perception of the complete lifecycle of design assets from creation, modification to reuse. Collaborator identities and cross-project template reuse behaviors not only understand the internal logic of individual design files but also identify the propagation paths of team collaboration patterns and organizational design guidelines. For example, when designing a login page, a designer drags a main component named "PrimaryButton" from the organization's component library (component instantiation); manually changes the button's background color from brand blue to green (style overriding); later discovers an update to the brand guidelines and clicks "Sync Main Component" to restore it to blue (main component synchronization); creates a "DangerButton" variant based on this button, modifying only the color and border (variant derivation); and finally reuses the entire form template from another project (cross-project template reuse). Simultaneously, the layer topology of the current canvas is retrieved (the button is located under the "Form / Submit" path); the style token used by the button (color.primary); the document version snapshot (v2.3); and the identity of the operator / collaborator (Alice, UI team, senior designer). This behavioral flow is fused with metadata to generate a design context vector, which not only records that the button has been modified and synchronized, but also implicitly includes semantics such as the button belonging to organizational assets, having variants, and being reused across projects. Although the designer manually changed the color, the team's specifications were ultimately adopted, and the component has high reusability. This allows for subsequent prompts to other collaborators that the button is a team-consensus asset, or for accurately predicting the scope of impact when the main component changes.

[0060] In one alternative approach, generating the design context vector based on the fine-grained interaction behavior flow and the structural metadata of the current design document further includes:

[0061] Feature extraction is performed on the component instantiation operations in the fine-grained interactive behavior flow to identify the main component ID it references and the initial attribute snapshot at the time of instantiation;

[0062] Perform a difference analysis on the style overlay to calculate the style difference vector between the attribute values ​​of the current component instance and the referenced main component or atomic style;

[0063] Version tracking is performed on the main component, and style token change logs are recorded before and after synchronization.

[0064] Perform topology analysis on the variant derivation to construct a variant inheritance chain between the derived component and the source component;

[0065] The cross-project template reuse is matched with the organizational asset library to identify the reused template ID and its adaptation modification records in the target project;

[0066] The extracted main component ID, initial attribute snapshot, style difference vector, style token change log, variant inheritance chain, template ID, adaptation modification record, and structural metadata are fused to generate the design context vector of the current design state and historical operation sequence.

[0067] In this embodiment, by extracting deep features (such as main component ID, difference vectors, and change logs) from operations like component instantiation, style overriding, and main component synchronization, it is possible not only to record "what was done" but also to understand "why it was done" and "how it evolved," thereby constructing a high-fidelity design semantic representation. Style difference vectors and variant inheritance chains provide quantitative evidence for identifying implicit imitation, conflict overriding, and other non-explicit reference relationships, breaking through the cognitive limitations of relying solely on explicit links. By matching cross-project template reuse with an organizational-level asset library, the propagation path and adaptation modifications of design assets across different projects can be accurately tracked.

[0068] Step S102: Construct a dynamically evolving design cognitive ontology based on the design context vector, and construct an initial design cognitive graph based on the design cognitive ontology and cross-document reference links. The nodes of the design cognitive ontology include atomic styles, composite specifications, and organizational-level design assets, and the edge relationships of the design cognitive ontology include explicit inheritance, implicit imitation, conflict coverage, and team adoption.

[0069] In this embodiment, simply recording the file structure cannot capture the implicit intentions of designers during operations (such as why to override styles or why to imitate a certain component). By constructing a dynamically evolving cognitive ontology, the designer's unconscious interactive behaviors (such as repeatedly adjusting rounded corners) are transformed into quantifiable design knowledge nodes (such as the "rounded corners 8px" atomic style). The design cognitive ontology is "dynamically evolving," continuously updating the state of nodes and the weight of edges with the operation sequence on the timeline, reflecting the entire process of the birth, evolution, and elimination of design specifications.

[0070] For example, a company's design team uses the same button system in multiple projects. Designer A created a blue rounded corner button in the "Payment App" using the color token color.primary, rounded corner radius.m, and inner margin space.md. Designer B did not reference this component in the "Customer Service Backend" but manually created a nearly identical button (color #007AFF ≈ color.primary, rounded corner 8px ≈ radius.m). Designer C referenced A's main component in the "Marketing Campaign Page" but changed the color to red to highlight "Claim Now". At the same time, all three projects are linked to the company's unified Design Tokens library. Based on the design context vector analysis, {color.primary, radius.m, space.md} is extracted as an atomic style node S1; "Main button = S1 + text style + interaction state" is aggregated into a composite specification node C1; because C1 is used by 3 designers in 3 projects, it is promoted to an organizational-level design asset node A1; edge relationships are established: A's button → A1 (explicit inheritance); B's button → A1 (implicit imitation, due to high visual similarity); C's button → A1 (conflict overriding, due to manual color modification); A1 ← Team adoption edge (from A, B, C);

[0071] Meanwhile, all three projects referenced the same Tokens library, adding it as a cross-document reference link to the graph. The resulting initial design cognitive graph not only showed the canonical form of the "main button" (A1) but also revealed three typical relationships in actual use: compliance with specifications, unconscious imitation, and intentional deviation. When brand colors change in the future, this can be used to accurately distinguish: button B should be automatically suggested for synchronization (implicit imitation needs to be corrected), button C requires manual confirmation (conflict coverage preserves intent), while button A can be safely updated in batches, significantly improving the consistency and flexibility of collaborative design.

[0072] In one alternative approach, constructing a dynamically evolving design cognitive ontology based on the design context vector further includes:

[0073] The design context vector is subjected to temporal clustering to identify design patterns that frequently co-occur on the time axis and are semantically stable, and these patterns are extracted into atomic style nodes; multiple atomic style nodes with hierarchical dependencies or compositional relationships are aggregated into composite canonical nodes;

[0074] Elevate frequently reused atomic patterns or composite specification nodes that have been confirmed by the team to organizational-level design asset nodes; identify the edge relationships between nodes by analyzing the operation sequences contained in the design context vector;

[0075] Specifically, if a component is detected referencing a parent style via the main component ID during instantiation, an explicit inheritance edge is established; if two nodes with no direct reference relationship are detected to be highly similar in visual attributes via style difference vectors, an implicit imitation edge is established; if a style manually modified by the designer is detected to conflict with the inherited parent style, a conflict overriding edge is established; if an atomic style or composite specification is detected to be repeatedly used by multiple collaborators in different projects, a team adoption edge is established, thereby generating a design cognitive ontology containing nodes, edges, and their evolution sequence.

[0076] In this embodiment, by performing temporal clustering on the design context vector, it automatically identifies visually stable combination patterns that recur repeatedly over time (such as "rounded corners + main color fill + inner margin"), abstracting them into reusable atomic style nodes, avoiding reliance on manual annotation or predefined specifications. Atomic styles are aggregated into composite specification nodes according to hierarchical dependencies or combination logic (such as "button = background color + font + hover state"), and frequently used specifications across projects are elevated to organizational-level design asset nodes, forming a "atomic → composite → organizational" knowledge system. It not only records explicit component references (explicit inheritance edges), but also discovers designs with no links but high visual similarity (implicit imitation edges) through style difference vectors, identifies specification deviations caused by manual modifications (conflict overriding edges), and consensus formed by multiple people and multiple projects (team adoption edges), thereby constructing a more realistic collaborative design relationship network. The cognitive ontology is continuously updated with the designer's operations, tracking the evolution path of the design language and supporting functions such as version rollback and specification migration. For example, multiple designers design notification banner components in multiple product lines. Designer A (payment app) created a banner with a background color of color.warning (#FFC107), 12px padding, and an icon + text layout. Designer B (email client) did not use any components but manually drew an almost identical banner (color #FFC000 ≈ color.warning, 12px padding). Designer C (admin backend) used Designer A's main component, changed the icon to an exclamation mark, and slightly adjusted the text color. All three designers came from different teams but used the company's unified DesignTokens. High-frequency combinations {color: ~#FFC100, padding: 12px} were clustered from the context vectors and abstracted into the atomic style node S_warn_bg. S_warn_bg frequently co-occurs with "left-aligned icon + 14px body text," aggregating into the composite specification node C_alert_banner. Because C_alert_banner was used by three designers in three independent projects without major conflicts, it was promoted to an organizational asset A_alert_v1. By establishing edge relationships, banner A → A_alert_v1 (explicit inheritance); banner B → A_alert_v1 (implicit imitation, due to high visual similarity but no reference); banner C → A_alert_v1 (conflict overriding, due to icon and text color modification); A_alert_v1 ← Team adoption edge (from A, B, and C, including identity and timestamp). When a future brand upgrade requires changing the warning color to orange-red, banner B should be suggested for synchronization (implicit imitation needs correction), banner C requires manual confirmation, and banner A can be automatically updated, significantly improving the consistency and flexibility of large-scale collaborative design.

[0077] In one alternative approach, constructing an initial design cognitive graph based on the design cognitive ontology and cross-document reference links further includes:

[0078] The first type of node is defined as atomic patterns, composite specifications, and organizational-level design assets in the design cognitive ontology.

[0079] Traverse the current project and its referenced external project libraries to extract cross-document reference links between all design elements. These cross-document reference links include cross-file main component reference relationships and cross-project style library link relationships.

[0080] The extracted cross-document reference links are added to the design cognition ontology as second-type nodes and edges to form the initial design cognition graph; wherein, the node attributes of the initial design cognition graph include node type, unique identifier, visual attribute vector and creation timestamp, and the edge attributes include relation type, initial confidence value and last synchronization timestamp.

[0081] In this embodiment, cross-file main component references and cross-project style library links are explicitly extracted, enabling any modification to core assets within the organization to be accurately tracked to all affected downstream scenarios. By distinguishing between specification definitions and instance references, it is possible to identify which projects are still using older versions of assets and which references have not been synchronized for a long time, thereby accurately assessing the scope of risks during brand upgrades or specification iterations. Timestamps and confidence levels give the graph evolutionary memory capabilities, significantly improving the transparency of system recommendations. For example, a company has a unified Design System library (project DS), which defines atomic styles: color.primary = #007AFF (node ​​ID: style_color_001); composite specification: PrimaryButton (node ​​ID: comp_btn_pri_v2), composed of atomic styles such as color, rounded corners, and font; and organizational-level assets: the entire CoreComponents library (node ​​ID: asset_core_v5). Three business projects reference this library. Project A (Payment App): The login page references comp_btn_pri_v2 and it is always synchronized, with the most recent synchronization time being 2026-02-28. Project B (Customer Service Backend) references the same button, but it has not been synchronized since December 2025 and its color has been manually changed to green. Project C (Marketing Campaign) directly links to asset_core_v5 and reuses multiple components. Add style_color_001, comp_btn_pri_v2, and asset_core_v5 as first-class nodes to the graph; extract cross-document references: A → comp_btn_pri_v2 (main component reference); B → comp_btn_pri_v2 (main component reference, but not synchronized for a long time); C → asset_core_v5 (library-level link); create second-class nodes (e.g., inst_A_login_btn) for specific button instances in A, B, and C; establish reference edges and set attributes: A's edge: relation = "instantiation", confidence = 1.0, last synchronized = 2026-02-28; B's edge: relation = "instantiation", confidence = 0.4 (due to style overriding and not synchronized for a long time), last synchronized = 2025-12-10; C's edge: relation = "library dependency", confidence = 0.9, last synchronized = 2026-02-20. The resulting initial design awareness map shows that the specification asset comp_btn_pri_v2 is used by three projects; Project A is up-to-date, Project B has a high-risk deviation, and Project C has stable overall dependencies. When the Design System team updates color.primary to #0056CC, the system can immediately and automatically synchronize Project A based on this map; issue a "high-risk conflict" warning to Project B, recommending manual review; and check whether all derived components in Project C are affected.

[0082] In an alternative approach, the method further includes:

[0083] When identifying implicit style references, the contrastive graph neural network and causal reasoning network model uses the canonical semantics of the first type of cognitive nodes and the visual features of the second type of instance nodes to perform comparative learning, thereby obtaining the potential implicit reference relationships between instances and canonical references or between instances.

[0084] When conducting impact propagation simulations, the contrastive graph neural network and causal reasoning network model starts with the first type of cognitive node corresponding to the main component or design token that has been changed, and propagates along the instantiation reference edge to all second type of instance nodes. It predicts the visual impact of the change on the design instance, thereby providing downstream projects with visual consistency risk assessment and multi-level remediation strategy packages.

[0085] In this embodiment, as Figure 3 As shown, comparative learning can not only capture explicit component references (such as main component ID references), but also identify implicit imitation relationships between visually similar design elements that are not directly linked. This helps to discover undocumented patterns and trends in the design system. Causal inference networks can more accurately infer designers' design intentions in specific contexts, such as why a particular color, shape, or layout was chosen (even if these decisions are not explicitly stated, they are important for maintaining brand consistency). When a designer modifies a core component or style token, impact propagation simulations can automatically track all affected design instances along the instantiated reference edges, thus providing early warnings of potential visual consistency risks (such as style conflicts, component distortion, etc.) and generating targeted remediation strategies. Based on the specific circumstances of each downstream instance (such as whether there is conflict coverage, whether there are cross-project references, etc.), differentiated remediation suggestions are automatically generated. These suggestions include strategies such as automatic batch synchronization, manual review and recommendation, and version rollback or creating new variations, ensuring optimal handling for each scenario.

[0086] Step S103: Input the initial design cognitive map into the contrastive graph neural network and causal reasoning network model to identify implicit style references and design intentions, and output a high-order reference relationship graph with intention labels and confidence scores; wherein, the contrastive graph neural network and causal reasoning network model includes a semantic encoder composed of Tree-LSTM network layers, an instance visual encoder composed of CNN and GNN, a causal intervention decoupler composed of counterfactual intervention layers and causal attention layers, and a multi-view contrast alignment head.

[0087] In this embodiment, two types of nodes in the initial design cognitive graph are encoded separately. The first type of nodes (atomic styles, composite specifications, etc.) are fed into the Tree-LSTM semantic encoder, which uses its tree structure modeling capabilities (such as nested combinations of style tokens) to generate semantic vectors. The second type of nodes (design instances) are fed into the CNN+GNN visual encoder. The CNN extracts local visual features (such as color distribution and shape contours), and the GNN aggregates their contextual neighbor information in the canvas (such as their relative positions with icons and text). Before fusing semantic and visual features, a counterfactual intervention layer simulates hypothetical scenarios such as "what would happen if a certain interfering factor (such as the project's main color) were removed." A causal attention layer weights each feature dimension, suppressing non-causal related terms (such as color coincidences caused by the project theme) and strengthening commonalities truly driven by design specifications. Positive and negative sample pairs are constructed. Positive samples include explicit inheritance pairs and high-frequency co-occurrence pairs; negative samples are random unrelated node pairs. A multi-view comparison alignment head calculates the similarity between the semantic view and the visual view, bringing different expressions under the same design intent (such as main buttons in different projects) closer together in the embedding space. Calculate the implicit reference score for all nodes in the graph. If the score exceeds the threshold and the causal confidence is high, add an edge with an intent label (such as "implicit imitation - brand alignment") and a confidence value (0~1). At the same time, retain the original explicit edges and add intent enhancement labels (such as "explicit inheritance - forced synchronization") to them, and output a higher-order reference graph with a richer structure and clearer semantics.

[0088] Step S104: When a designer selects any design element, the origin node and evolution tree, usage context, and team consensus strength score of the design element are dynamically presented according to the higher-order reference relationship graph. When the main component or design token of the design element changes, the impact propagation simulation is performed according to the higher-order reference relationship graph to predict the visual consistency risk of downstream projects and generate a multi-level repair strategy package.

[0089] In this embodiment, by traversing the explicit inheritance edges and high-confidence implicit reference edges in the higher-order graph in reverse, the evolution path of a design element from atomic style → composite specification → organizational asset → current instance is fully reconstructed, enabling designers to understand "where the element comes from and why it looks the way it does." Downstream usage nodes are aggregated in a forward manner and grouped by project / page / team to generate a usage context list, helping designers quickly determine the reusability and collaborative dependencies of the element, avoiding the risk of "changing one thing and causing problems everywhere." Based on the number of adopted edges in the team, the diversity of collaborators, and their authority (e.g., senior designer vs. intern), a team consensus strength score is calculated, providing an objective basis for whether to elevate a style to an organizational-level asset. Figure 2As shown, impact propagation simulations are initiated before changes occur, predicting potential downstream risks such as style conflicts, component deviations, and brand inconsistencies, and automatically generating differentiated multi-level remediation strategy packages (automatic synchronization / manual review / new variant creation), shifting consistency governance from post-event correction to pre-event prevention. This is particularly suitable for large design organizations spanning multiple teams and product lines, significantly reducing issues such as duplicated work, style drift, and brand dilution caused by information asymmetry.

[0090] In one alternative approach, dynamically presenting the origin node and evolution tree of the design element, usage context, and team consensus strength score based on the higher-order reference graph further includes:

[0091] Using the design element as an index, query the corresponding target node in the higher-order reference graph;

[0092] By traversing back through the explicit inheritance edges and implicit reference edges with confidence exceeding a preset threshold pointing to the target node in the higher-order reference graph, all upstream nodes are traced to construct the origin node and evolution tree of the design element, so as to show its evolution path from the basic style to the current form.

[0093] By forward traversing all outgoing edges from the target node, all direct and indirect downstream nodes are aggregated, and a list of usage contexts for the design element is generated by grouping and statistically analyzing them by project, page, or team.

[0094] The team consensus strength score is calculated based on the number of team adoption edges converged to the target node, the diversity of their sources, and the authority of the relevant collaborator identities.

[0095] In this embodiment, all downstream user nodes are aggregated in a forward manner and presented in a structured way according to dimensions such as project, page, and team. This allows designers to clearly grasp the scope of influence of an element at a glance, which is especially suitable for cross-functional and cross-product collaborative scenarios in large organizations. The evolution tree and usage context are presented in a graphical and interactive way, making the inference results highly interpretable, thus making people more willing to adopt intelligent assistance.

[0096] In one alternative approach, performing impact propagation simulation based on the higher-order reference graph to predict visual consistency risks in downstream projects and generate a multi-level remediation strategy package further includes:

[0097] When a designer initiates a change operation on the main component or design token of the source design element, the source design element is locked. Starting from the node of the source design element in the higher-order reference graph, a breadth-first traversal is performed along all outgoing edges to identify all downstream elements that are directly or indirectly affected by the change and obtain the influence propagation graph.

[0098] Based on the project to which the downstream element belongs, the type of relationship with the source design element, and the conflict coverage of its edges, assess the visual consistency risk of each affected project to generate a risk report, wherein the risks include style conflicts, component deviations, and brand specification violations.

[0099] Based on the visual consistency risk of each downstream element in the impact propagation diagram, a multi-level remediation strategy package is automatically generated. The multi-level remediation strategy package includes an automatic batch synchronization strategy for downstream elements without conflict coverage, a manual review and recommendation strategy for downstream elements with conflict coverage, and a version rollback or new variant creation strategy for cross-project references.

[0100] In this embodiment, breadth-first search (BFS) is used to identify all potentially affected design elements starting from the source of the change, ensuring that no potential impact is overlooked. Customized remediation solutions are provided based on the specific circumstances of downstream elements, including automatic synchronization, manual review suggestions, and version management strategies. This makes remediation measures both efficient and flexible, adaptable to different project needs and complexities, reducing time wasted due to manual checks and poor communication, and improving team collaboration efficiency. By assessing and handling specific risk types such as style conflicts, component distortion, and brand specification violations, consistency in product appearance and brand recognition are ensured even during frequent design iterations. For example, an e-commerce website's design team decides to adjust the color of its button component (from blue to green). The main component or design token representing the button color is locked, and all places where that button color is used (such as login pages, product detail pages, and shopping cart pages) are automatically searched and recorded. Some pages have already customized button colors for specific scenarios, which may lead to style conflicts; at the same time, for pages that strictly adhere to global design guidelines, brand specification violations may occur. Specific solutions are proposed for different situations. For pages without customized colors, it is recommended to automatically synchronize the new colors. For pages with personalized settings, designers are advised to review and confirm the update one by one. If multiple sub-brands' independent sites are involved, new button variations can be created or a rollback to a stable version can be implemented to ensure that the consistency and uniqueness of the visual style of each sub-brand are not affected. This not only improves the efficiency of design changes but also ensures the consistency of user experience and brand coherence.

[0101] In an alternative approach, after generating the multi-level remediation strategy package, the method further includes:

[0102] Automatically traverse all downstream elements that meet the preset conditions, synchronize the changed attributes of the source design elements to the downstream elements, and record this batch synchronization operation in the version history;

[0103] In response to the designer clicking the manual review recommendation strategy button, highlight all downstream elements with conflicting coverage and display their current style difference vector and the latest value of the parent style.

[0104] In response to designers clicking the version rollback or new variant strategy button, an isolated branch containing the current stable version is created for the affected downstream projects or component libraries, or the affected downstream elements are automatically derived into new component variants, breaking their inheritance relationship with the source design elements to maintain the visual stability of the original project.

[0105] In this embodiment, analysis results are directly transformed into actionable steps, improving collaborative efficiency. Different risk scenarios are handled in layers: low-risk instances (no conflicts) are automatically batch synchronized; medium- and high-risk instances (with coverage) are highlighted and their differences visualized; key cross-project references are isolated into branches / new variants to ensure the stability of existing projects. All automatic synchronization operations are recorded in the version history, ensuring auditable and rollbackable changes; simultaneously, floating difference vectors reduce the risk of accidental operations. Some projects are allowed to proactively deviate from the main specification (by creating new variants), maintaining core brand consistency while preserving space for local innovation. In enterprise-level design systems, a single basic style change (such as primary color or rounded corners) can affect hundreds of pages; automation significantly improves the feasibility of specification iteration.

[0106] According to the solution provided by this invention, fine-grained interactive behavior flow of designers during canvas page operations is collected in real time. A design context vector is generated based on the fine-grained interactive behavior flow and the structural metadata of the current design document. The fine-grained interactive behavior flow includes component instantiation, style overriding, main component synchronization, variant derivation, and cross-project template reuse. The structural metadata includes layer topology, style tokens, version snapshots, and collaborator identities. A dynamically evolving design cognitive ontology is constructed based on the design context vector. An initial design cognitive graph is constructed based on the design cognitive ontology and cross-document reference links. The nodes of the design cognitive ontology include atomic styles, composite specifications, and organizational-level design assets. The edge relationships of the design cognitive ontology include explicit inheritance, implicit imitation, conflict overriding, and team adoption. The initial design cognitive graph is then... The input is fed into a contrastive graph neural network and causal reasoning network model to identify implicit style references and design intent, and outputs a high-order reference graph with intent labels and confidence levels. The contrastive graph neural network and causal reasoning network model includes a semantic encoder composed of Tree-LSTM network layers, an instance visual encoder composed of CNN and GNN, a causal intervention decoupler composed of counterfactual intervention layers and causal attention layers, and a multi-view contrast alignment head. When a designer selects any design element, the origin node and evolution tree of that design element, its usage context, and team consensus strength score are dynamically presented based on the high-order reference graph. When the main component or design token of that design element changes, the impact propagation simulation is performed based on the high-order reference graph to predict the visual consistency risk of downstream projects and generate a multi-level remediation strategy package. This invention achieves a systematic representation and organization of design style knowledge, significantly improving the accuracy and comprehensiveness of design style reference relationship identification, and enabling dynamic analysis and prediction of design style evolution trends, providing support for design decisions. Furthermore, by utilizing knowledge graphs and causal reasoning networks, it achieves implicit reference relationship mining across projects and documents and intelligent reasoning for global style conflicts.

[0107] Figure 4 A schematic diagram of the framework of a high-fidelity style reference relationship recognition enhancement device for cooperative intelligent agents according to an embodiment of the present invention is shown. The high-fidelity style reference relationship recognition enhancement device for cooperative intelligent agents includes:

[0108] The design context vector generation module 410 is used to collect the fine-grained interaction behavior flow of the designer during the canvas operation process in real time, and generate a design context vector based on the fine-grained interaction behavior flow and the structural metadata of the current design document; wherein, the fine-grained interaction behavior flow includes component instantiation, style overriding, main component synchronization, variant derivation and cross-project template reuse; the structural metadata includes layer topology, style token, version snapshot and collaborator identity;

[0109] The initial design cognitive graph construction module 420 is used to construct a dynamically evolving design cognitive ontology based on the design context vector, and to construct an initial design cognitive graph based on the design cognitive ontology and cross-document reference links. The nodes of the design cognitive ontology include atomic styles, composite specifications and organizational-level design assets, and the edge relationships of the design cognitive ontology include explicit inheritance, implicit imitation, conflict coverage and team adoption.

[0110] The high-order citation graph generation module 430 is used to input the initial design cognitive graph into the comparative graph neural network and causal reasoning network model to identify implicit style citations and design intentions, and output a high-order citation graph with intention labels and confidence scores.

[0111] The design element recognition and presentation module 440 is used to dynamically present the origin node and evolution tree, usage context and team consensus strength score of any design element when the designer selects it, based on the higher-order reference relationship graph. When the main component or design token of the design element changes, the module simulates the impact propagation based on the higher-order reference relationship graph, predicts the visual consistency risk of downstream projects and generates a multi-level repair strategy package.

[0112] Figure 5 The diagram shows a structural schematic of an embodiment of the computing device of the present invention. The specific embodiments of the present invention do not limit the specific implementation of the computing device.

[0113] like Figure 5 As shown, the computing device may include: a processor 502, a communications interface 504, a memory 506, and a communications bus 508.

[0114] The processor 502, communication interface 504, and memory 506 communicate with each other via communication bus 508. Communication interface 504 is used to communicate with other network elements, such as clients or other servers. Processor 502 executes program 510, specifically performing the relevant steps in the above-described embodiment of the high-fidelity style reference relationship recognition enhancement method for cooperative intelligent agents.

[0115] Specifically, program 510 may include program code that includes computer operation instructions.

[0116] Processor 502 may be a central processing unit (CPU), an application-specific integrated circuit (ASIC), or one or more integrated circuits configured to implement embodiments of the present invention. The computing device includes one or more processors, which may be processors of the same type, such as one or more CPUs; or processors of different types, such as one or more CPUs and one or more ASICs.

[0117] Memory 506 is used to store program 510. Memory 506 may include high-speed RAM memory, and may also include non-volatile memory, such as at least one disk storage device.

[0118] According to the solution provided by this invention, fine-grained interactive behavior flow of designers during canvas page operations is collected in real time. A design context vector is generated based on the fine-grained interactive behavior flow and the structural metadata of the current design document. The fine-grained interactive behavior flow includes component instantiation, style overriding, main component synchronization, variant derivation, and cross-project template reuse. The structural metadata includes layer topology, style tokens, version snapshots, and collaborator identities. A dynamically evolving design cognitive ontology is constructed based on the design context vector. An initial design cognitive graph is constructed based on the design cognitive ontology and cross-document reference links. The nodes of the design cognitive ontology include atomic styles, composite specifications, and organizational-level design assets. The edge relationships of the design cognitive ontology include explicit inheritance, implicit imitation, conflict overriding, and team adoption. The initial design cognitive graph is then... The input is fed into a contrastive graph neural network and causal reasoning network model to identify implicit style references and design intent, and outputs a high-order reference graph with intent labels and confidence levels. The contrastive graph neural network and causal reasoning network model includes a semantic encoder composed of Tree-LSTM network layers, an instance visual encoder composed of CNN and GNN, a causal intervention decoupler composed of counterfactual intervention layers and causal attention layers, and a multi-view contrast alignment head. When a designer selects any design element, the origin node and evolution tree of that design element, its usage context, and team consensus strength score are dynamically presented based on the high-order reference graph. When the main component or design token of that design element changes, the impact propagation simulation is performed based on the high-order reference graph to predict the visual consistency risk of downstream projects and generate a multi-level remediation strategy package. This invention achieves a systematic representation and organization of design style knowledge, significantly improving the accuracy and comprehensiveness of design style reference relationship identification, and enabling dynamic analysis and prediction of design style evolution trends, providing support for design decisions. Furthermore, by utilizing knowledge graphs and causal reasoning networks, it achieves implicit reference relationship mining across projects and documents and intelligent reasoning for global style conflicts.

[0119] Those skilled in the art will understand that modules in the device of the embodiments can be adaptively changed and placed in one or more devices different from that embodiment. Modules, units, or components in the embodiments can be combined into a single module, unit, or component, and further, they can be divided into multiple sub-modules, sub-units, or sub-components. Except where at least some of such features and / or processes or units are mutually exclusive, any combination of all features disclosed in this specification (including the accompanying claims, abstract, and drawings) and all processes or units of any method or device so disclosed can be employed. Unless expressly stated otherwise, each feature disclosed in this specification (including the accompanying claims, abstract, and drawings) may be replaced by an alternative feature that serves the same, equivalent, or similar purpose. Furthermore, those skilled in the art will understand that although some embodiments herein include certain features included in other embodiments but not others, combinations of features from different embodiments are meant to be within the scope of the invention and form different embodiments. For example, in the following claims, any of the claimed embodiments can be used in any combination. The invention can be implemented by means of hardware comprising several different elements and by means of a suitably programmed computer. In the unit claims listing several devices, several of these devices may be embodied by the same hardware item. Unless otherwise specified, the steps in the above embodiments should not be construed as limiting the order of execution.

Claims

1. A high-fidelity style reference relationship recognition enhancement method for collaborative intelligent agents, characterized in that, include: The system collects fine-grained interaction behavior flows from designers during canvas operations in real time, and generates a design context vector based on these fine-grained interaction behavior flows and the structural metadata of the current design document. The fine-grained interaction behavior flows include component instantiation, style overriding, main component synchronization, variant derivation, and cross-project template reuse. The structural metadata includes layer topology, style tokens, version snapshots, and collaborator identities. A dynamically evolving design cognitive ontology is constructed based on the design context vector. An initial design cognitive graph is constructed based on the design cognitive ontology and cross-document reference links. The nodes of the design cognitive ontology include atomic styles, composite specifications, and organizational-level design assets. The edge relationships of the design cognitive ontology include explicit inheritance, implicit imitation, conflict coverage, and team adoption. The initial design cognitive map is input into a contrastive graph neural network and causal reasoning network model to identify implicit style references and design intentions, and outputs a high-order reference relationship graph with intention labels and confidence scores; wherein, the contrastive graph neural network and causal reasoning network model includes a semantic encoder composed of Tree-LSTM network layers, an instance visual encoder composed of CNN and GNN, a causal intervention decoupler composed of counterfactual intervention layers and causal attention layers, and a multi-view contrast alignment head; When a designer selects any design element, the origin node and evolution tree of the design element, usage context, and team consensus strength score are dynamically presented according to the high-order reference relationship graph. When the main component or design token of the design element changes, the impact propagation simulation is performed according to the high-order reference relationship graph to predict the visual consistency risk of downstream projects and generate a multi-level remediation strategy package.

2. The high-fidelity style reference relationship recognition enhancement method for cooperative intelligent agents according to claim 1, characterized in that, Generating a design context vector based on the fine-grained interaction behavior flow and the structural metadata of the current design document further includes: Feature extraction is performed on the component instantiation operations in the fine-grained interactive behavior flow to identify the main component ID it references and the initial attribute snapshot at the time of instantiation; Perform a difference analysis on the style overlay to calculate the style difference vector between the attribute values ​​of the current component instance and the referenced main component or atomic style; Version tracking is performed on the main component, and style token change logs are recorded before and after synchronization. Perform topology analysis on the variant derivation to construct a variant inheritance chain between the derived component and the source component; The cross-project template reuse is matched with the organizational asset library to identify the reused template ID and its adaptation modification records in the target project; The extracted main component ID, initial attribute snapshot, style difference vector, style token change log, variant inheritance chain, template ID, adaptation modification record, and structural metadata are fused to generate the design context vector of the current design state and historical operation sequence.

3. The high-fidelity style reference relationship recognition enhancement method for cooperative intelligent agents according to claim 1, characterized in that, Constructing a dynamically evolving design cognitive ontology based on the design context vector further includes: The design context vector is subjected to temporal clustering to identify design patterns that frequently co-occur on the time axis and are semantically stable, and these patterns are extracted into atomic style nodes; multiple atomic style nodes with hierarchical dependencies or compositional relationships are aggregated into composite canonical nodes; Elevate frequently reused atomic patterns or composite specification nodes that have been confirmed by the team to organizational-level design asset nodes; identify the edge relationships between nodes by analyzing the operation sequences contained in the design context vector; Specifically, if a component is detected referencing a parent style via the main component ID during instantiation, an explicit inheritance edge is established; if two nodes with no direct reference relationship are detected to be highly similar in visual attributes via style difference vectors, an implicit imitation edge is established; if a style manually modified by the designer is detected to conflict with the inherited parent style, a conflict overriding edge is established; if an atomic style or composite specification is detected to be repeatedly used by multiple collaborators in different projects, a team adoption edge is established, thereby generating a design cognitive ontology containing nodes, edges, and their evolution sequence.

4. The high-fidelity style reference relationship recognition enhancement method for cooperative intelligent agents according to claim 1, characterized in that, Constructing an initial design cognitive graph based on the aforementioned design cognitive ontology and cross-document reference links further includes: The first type of node is defined as atomic patterns, composite specifications, and organizational-level design assets in the design cognitive ontology. Traverse the current project and its referenced external project libraries to extract cross-document reference links between all design elements. These cross-document reference links include cross-file main component reference relationships and cross-project style library link relationships. The extracted cross-document reference links are added to the design cognition ontology as second-type nodes and edges to form the initial design cognition graph; wherein, the node attributes of the initial design cognition graph include node type, unique identifier, visual attribute vector and creation timestamp, and the edge attributes include relation type, initial confidence value and last synchronization timestamp.

5. The high-fidelity style reference relationship recognition enhancement method for cooperative intelligent agents according to claim 4, characterized in that, The method further includes: When identifying implicit style references, the contrastive graph neural network and causal reasoning network model uses the canonical semantics of the first type of cognitive nodes and the visual features of the second type of instance nodes to perform comparative learning, thereby obtaining the potential implicit reference relationships between instances and canonical references or between instances. When conducting impact propagation simulations, the contrastive graph neural network and causal reasoning network model starts with the first type of cognitive node corresponding to the main component or design token that has been changed, and propagates along the instantiation reference edge to all second type of instance nodes. It predicts the visual impact of the change on the design instance, thereby providing downstream projects with visual consistency risk assessment and multi-level remediation strategy packages.

6. The high-fidelity style reference relationship recognition enhancement method for cooperative intelligent agents according to claim 1, characterized in that, The design element's origin node and evolution tree, usage context, and team consensus strength score are dynamically presented based on the aforementioned high-order reference graph, further including: Using the design element as an index, query the corresponding target node in the higher-order reference graph; By traversing back through the explicit inheritance edges and implicit reference edges with confidence exceeding a preset threshold pointing to the target node in the higher-order reference graph, all upstream nodes are traced to construct the origin node and evolution tree of the design element, so as to show its evolution path from the basic style to the current form. By forward traversing all outgoing edges from the target node, all direct and indirect downstream nodes are aggregated, and a list of usage contexts for the design element is generated by grouping and statistically analyzing them by project, page, or team. The team consensus strength score is calculated based on the number of team adoption edges converged to the target node, the diversity of their sources, and the authority of the relevant collaborator identities.

7. The high-fidelity style reference relationship recognition enhancement method for cooperative intelligent agents according to claim 1, characterized in that, Based on the aforementioned high-order reference graph, the impact propagation simulation is performed to predict the visual consistency risks of downstream projects and generate a multi-level remediation strategy package, which further includes: When a designer initiates a change operation on the main component or design token of the source design element, the source design element is locked. Starting from the node of the source design element in the higher-order reference graph, a breadth-first traversal is performed along all outgoing edges to identify all downstream elements that are directly or indirectly affected by the change and obtain the influence propagation graph. Based on the project to which the downstream element belongs, the type of relationship with the source design element, and the conflict coverage of its edges, assess the visual consistency risk of each affected project to generate a risk report, wherein the risks include style conflicts, component deviations, and brand specification violations. Based on the visual consistency risk of each downstream element in the impact propagation diagram, a multi-level remediation strategy package is automatically generated. The multi-level remediation strategy package includes an automatic batch synchronization strategy for downstream elements without conflict coverage, a manual review and recommendation strategy for downstream elements with conflict coverage, and a version rollback or new variant creation strategy for cross-project references.

8. The high-fidelity style reference relationship recognition enhancement method for cooperative intelligent agents according to claim 7, characterized in that, After generating the multi-level remediation strategy package, the method further includes: Automatically traverse all downstream elements that meet the preset conditions, synchronize the changed attributes of the source design elements to the downstream elements, and record this batch synchronization operation in the version history; In response to the designer clicking the manual review recommendation strategy button, highlight all downstream elements with conflicting coverage and display their current style difference vector and the latest value of the parent style. In response to designers clicking the version rollback or new variant strategy button, an isolated branch containing the current stable version is created for the affected downstream projects or component libraries, or the affected downstream elements are automatically derived into new component variants, breaking their inheritance relationship with the source design elements to maintain the visual stability of the original project.

9. A high-fidelity style reference relationship recognition enhancement device for collaborative intelligent agents, characterized in that, include: The design context vector generation module is used to collect fine-grained interaction behavior flows of designers during canvas page operations in real time, and generate design context vectors based on the fine-grained interaction behavior flows and the structural metadata of the current design document; wherein, the fine-grained interaction behavior flows include component instantiation, style overriding, main component synchronization, variant derivation, and cross-project template reuse; the structural metadata includes layer topology, style token, version snapshot, and collaborator identity; The initial design cognitive graph construction module is used to construct a dynamically evolving design cognitive ontology based on the design context vector, and to construct an initial design cognitive graph based on the design cognitive ontology and cross-document reference links. The nodes of the design cognitive ontology include atomic styles, composite specifications and organizational-level design assets, and the edge relationships of the design cognitive ontology include explicit inheritance, implicit imitation, conflict coverage and team adoption. The high-order citation graph generation module is used to input the initial design cognitive graph into a comparative graph neural network and causal reasoning network model to identify implicit style citations and design intentions, and output a high-order citation graph with intention labels and confidence scores. The design element recognition and presentation module is used to dynamically present the origin node and evolution tree of any design element, its usage context, and the team consensus strength score based on the high-order reference relationship graph when a designer selects any design element. When the main component or design token of the design element changes, the module simulates the impact propagation based on the high-order reference relationship graph, predicts the visual consistency risk of downstream projects, and generates a multi-level remediation strategy package.

10. A computing device, comprising: The processor, memory, communication interface, and communication bus are provided, wherein the processor, memory, and communication interface communicate with each other via the communication bus. The memory is used to store at least one executable instruction, which causes the processor to perform the operation corresponding to the high-fidelity style reference relationship recognition enhancement method for cooperative agents described above.

Images