An agent collaborative editing method and system for industry design

Abstract

The application relates to the technical field of information processing, and discloses an intelligent agent collaborative editing method and system for industry design, which comprises the following steps: obtaining post role information and design task description text, and generating a composite feature description; constructing a semantic query vector based on the composite feature description, and determining a relevant design material set; performing permission verification based on an intelligent agent role on the relevant design material set, and generating a safe design material subset; obtaining version change records and solving conflicts, constructing an ownership link from a contributor to a user, and generating a traceable design material group; performing safe transmission by using a hierarchical encryption protocol, and establishing a protective transmission channel; transmitting the traceable design material group to a design creation auxiliary module, intelligently processing the traceable design material group in combination with a semantic analysis result, and generating a final output material. The application significantly improves the intelligent level, safety and output quality of multi-intelligent agent collaborative design.

Description

Technical Field

[0001] This application relates to the field of information processing technology, and in particular to an intelligent agent collaborative editing method and system designed for industry applications. Background Technology

[0002] Currently, in industries such as film and television production, advertising design, industrial design, and digital content creation, multi-position collaborative editing has become the mainstream work mode. Multiple intelligent agents such as planning, design, review, and publishing need to work closely together, using and processing massive amounts of design materials around the same design task. This collaborative method, which relies heavily on resource sharing and version iteration, directly determines the delivery cycle and quality of the design project. Therefore, the efficient, secure, and controllable flow of materials is particularly crucial in industry design scenarios. However, existing collaborative editing systems have revealed significant technical flaws when dealing with high-frequency collaboration among multiple agents. Most systems only set static permission lists for single roles or departments, resulting in either overly strict permissions causing repeated applications and waiting, severely slowing down the design process, or overly lenient permissions leading to the risk of leaking core design drafts or trade secrets when transferring design materials across roles. At the same time, due to the lack of a unified version management mechanism, modifications made by different agents on the same design material often conflict. For example, after one designer adjusts the composition of the image, another reviewer may make annotations based on the old version, resulting in basic errors such as mismatched visual elements with the design script in the final product. More seriously, there is a lack of clear contribution records throughout the entire process of obtaining design materials from acquisition to finalization. Once copyright disputes or content quality issues arise, it is difficult for the design team to accurately trace the specific person and stage of modification, leading to unclear attribution of responsibility and high rework costs.

[0003] These problems are not isolated; rather, they are intertwined by complex interdependencies among factors such as the suitability of materials for specific roles, the sensitivity of their content, their version evolution history, and the information contributed by users. For example, in an architectural design project, the planning team selects a set of architectural exterior images and adds design descriptions. The rendering team then replaces materials and optimizes lighting based on these images. The review team then extracts a portion of these images for use in a presentation document. If the regulatory authorities discover that one of the images contains unauthorized commercial elements, the entire design process needs to be traced back to investigate. However, due to the fragmented versions, scattered permissions, and missing contribution information, the team often has to compare each file individually and repeatedly ask for confirmation, which significantly extends the project cycle and may even affect client relationships due to the inability to quickly identify the responsible party. This problem of uncontrolled permissions, version chaos, and difficulty in tracing the source caused by the intertwining of the multidimensional attributes of materials is amplified in industry design scenarios with high-frequency iterations involving multiple agents.

[0004] Therefore, how to achieve accurate material matching based on job responsibilities and task scenarios among multiple agents, consistent version management across positions, and complete contribution tracking from creators to end users, while ensuring the safety and controllability of design materials, has become a key technical problem that urgently needs to be solved in the current field of collaborative design editing. Summary of the Invention

[0005] To address the aforementioned technical issues, this application provides an intelligent agent collaborative editing method and system for industry-specific designs, which aims to improve the efficiency of material reuse, the strength of intellectual property protection, and the quality of creative output.

[0006] Firstly, this application provides an intelligent agent collaborative editing method for industry-specific designs, the method comprising: Step S1: Obtain the job role information and design task description text of the collaborative editing agent in the industry design scenario, and perform multimodal feature fusion to generate a composite feature description for the collaborative editing agent and the design task; Step S2: Construct a semantic query vector based on the composite feature description, perform semantic matching in the industry design material library, and determine the set of relevant design materials associated with the design task; Step S3: Perform permission verification based on the agent role on the relevant design material set, and dynamically filter based on the job responsibilities of the collaborative editing agent and the sensitivity of the design materials to generate a safe subset of design materials that the collaborative editing agent can safely access; Step S4: Obtain the version change records of the design materials in the secure design material subset during the cross-agent collaborative editing process, identify and resolve version conflicts, generate a unified version of design materials that is consistent across agents, and construct a complete ownership link from design material contributors to design material users to generate a traceable design material group; Step S5: Securely transmit the traceable design material group using a hierarchical encryption protocol related to the intelligent agent role path, and establish a protected transmission channel; Step S6: Transmit the traceable design material group to the design creation assistance module through the protective transmission channel, and perform context-aware intelligent processing based on the semantic parsing results of the design task description text to generate the final output design material that meets the design task requirements.

[0007] Secondly, this application provides an intelligent agent collaborative editing system designed for industry applications, the system comprising: The feature integration unit is used to acquire the job role information and design task description text of the collaborative editing agent in the industry design scenario, and perform multimodal feature fusion to generate a composite feature description for the collaborative editing agent and the design task. The material retrieval unit is used to construct a semantic query vector based on the composite feature description, perform semantic matching in the industry design material library, and determine the set of relevant design materials associated with the design task. The permission filtering unit is used to perform permission verification based on the role of the intelligent agent on the relevant design material set, and to dynamically filter based on the job responsibilities of the collaborative editing intelligent agent and the sensitivity of the design materials to generate a safe subset of design materials that the collaborative editing intelligent agent can safely access. The version processing unit is used to obtain the version change records of the design materials in the secure design material subset during the cross-agent collaborative editing process, identify and resolve version conflicts, generate a unified version of design materials that is consistent across agents, and construct a complete ownership link from design material contributors to design material users to generate a traceable design material group. The link tracing unit is used to securely transmit the traceable design material group using a hierarchical encryption protocol related to the intelligent agent role path, and to establish a protected transmission channel. The intelligent generation unit is used to transmit the traceable design material group to the design creation assistance module through the protective transmission channel, and perform context-aware intelligent processing based on the semantic parsing results of the design task description text to generate the final output design material that meets the design task requirements.

[0008] Compared with the prior art, the beneficial effects of the present invention are at least as follows: 1. This application generates a composite feature description that accurately represents the responsibilities and task intentions of the agent by deeply fusing job role information with design task description text using multimodal features. Based on this composite feature description, a semantic query vector is constructed for material matching, thereby significantly improving the accuracy and recall rate of design material retrieval. At the same time, by introducing a real-time online sharing mechanism and multi-dimensional value assessment, the source of materials is effectively expanded and the quality of supplementary materials is ensured. This solves the problem of collaboration interruption caused by insufficient internal material library resources, and provides efficient and accurate material acquisition capabilities for multi-agent collaborative design.

[0009] 2. This application implements dynamic permission filtering based on agent roles on the relevant design material set, adopts hierarchical job role binding rules to identify and exclude materials containing sensitive elements, and combines a version tracking model to automatically detect and process version conflicts in the cross-agent collaborative editing process. For complex binary materials, a manual intervention guidance mechanism is introduced, which effectively solves the technical problems of crude permission control and chaotic version management. At the same time, by constructing a complete ownership link from contributors to users through an ownership mapping function, it ensures that every version evolution is traceable and that the rights and responsibilities of every design asset are clear, which significantly improves the data security and accountability of the design collaboration process.

[0010] 3. This application also employs a hierarchical encryption protocol related to the agent role path to securely transmit traceable design material groups. By constructing a protective transmission channel through segmented identity verification operations, it effectively prevents data leakage and identity impersonation risks during cross-position transmission. Based on this, combined with the semantic parsing results of the design task description text, it performs context style consistency rendering and quantitative verification on the material group. Through a collaborative mechanism of automatic comparison, intelligent adjustment, and manual guidance, it generates final output materials that conform to the design intent, thereby significantly improving the overall efficiency and product quality of multi-agent collaborative design while ensuring security. Attached Figure Description

[0011] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0012] Figure 1 This is a flowchart illustrating the steps of an industry-oriented intelligent agent collaborative editing method as described in this application embodiment; Figure 2 This is a schematic diagram showing the attention weight distribution of job characteristics under different design tasks in the embodiments of this application; Figure 3 This is a comparison chart of material value assessments in the embodiments of this application; Figure 4 This is a structural diagram of an industry-oriented intelligent agent collaborative editing system as described in this application embodiment. Detailed Implementation

[0013] This application provides an intelligent agent collaborative editing method and system designed for industry applications. The terms "first," "second," "third," "fourth," etc. (if present) in the specification, claims, and accompanying drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in a sequence other than that illustrated or described herein. Furthermore, the terms "comprising" or "having" and any variations thereof are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or device that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or devices.

[0014] Example 1: For ease of understanding, the specific process of the embodiments of this application is described below, such as... Figure 1 The embodiment of this application shows an industry-oriented intelligent agent collaborative editing method, which includes: Step S1: Obtain the job role information and design task description text of the collaborative editing agent in the industry design scenario, and perform multimodal feature fusion to generate a composite feature description for the collaborative editing agent and the design task.

[0015] The process of generating a composite feature description for the collaborative editing agent and the design task includes: extracting job responsibility feature vectors based on the job role information of the collaborative editing agent; performing context-dependent semantic parsing on the design task description text to generate semantic features of the design task scenario; and using an attention fusion mechanism to interactively fuse the job responsibility feature vectors and the semantic features of the design task scenario to generate a composite feature description.

[0016] Specifically, this technical solution aims to address the complex technical problems in existing collaborative editing methods, such as low material retrieval accuracy, coarse access control, chaotic version management, and broken contribution links, due to the inability to deeply integrate the role characteristics of designers with the semantics of dynamically changing design tasks. To solve this technical problem, the above technical solution will be described in detail below.

[0017] In industry design scenarios, after obtaining the job role information of the collaborative editing agent and the description text of the design task to be processed, the first step is to perform deep semantic understanding and fusion on these two types of heterogeneous data. Specifically, for job role information, such as "interior designer" or "structural engineer," this method does not simply treat it as a text tag, but rather maps it to a pre-defined job responsibility knowledge graph to extract quantitative features that can characterize the job's material usage preferences, permission levels, and scope of responsibilities. This forms a job responsibility feature vector, which is essentially a multi-dimensional numerical representation whose dimensional space covers things such as: The model incorporates implicit knowledge such as frequently used material types, material libraries, component libraries, historical editing habits, and default creation templates. Meanwhile, for design task description texts, such as designing a minimalist living room TV background wall, a deep neural network model based on the Transformer architecture is trained using supervised learning. Its input data consists of a large number of labeled industry design task description text samples, and its output data is the corresponding structured design task scene semantic feature vector. During training, the model's generated semantic features are compared with manually labeled real scene features, and the backpropagation algorithm is used to continuously optimize the model parameters. Leveraging its built-in self-attention mechanism, the model can accurately capture long-distance dependencies between keywords in the text, identifying core design objects such as "background wall," spatial constraints such as "living room," style requirements such as "minimalist style," and potential functional needs. Ultimately, it generates structured design task scene semantic features that contain both explicit textual information and deep task intent. These features are stored and used for subsequent semantic parsing results.

[0018] After obtaining the job responsibility feature vector and the design task scenario semantic features respectively, the key lies in how to effectively integrate the two to generate a unified composite feature description. This solution abandons simple vector concatenation or weighted summation, and instead introduces an attention fusion mechanism to enhance the interaction between the two. Specifically, this attention fusion mechanism uses the design task scenario semantic features as the query and the job responsibility feature vector as the key and value. It calculates an attention score to measure which dimensions of the job features should be given priority under the current specific design task. For example... Figure 2The attention weight distribution of job features under different design tasks is shown. When the task description is "rendering an image," the attention mechanism amplifies the weights of "color management" and "output resolution" in the job features. When the task description is "structural calculation," it focuses on features related to "material mechanical properties" and "load-bearing case examples." Conversely, the mechanism can also use job features as a benchmark to filter out the parts of the task semantics that are highly relevant to the job responsibilities and remove noise information. Through this interactive weighted fusion, the final composite feature description can accurately express the core requirement of "what kind of materials a specific design role needs in the current specific design task." This composite feature description is then used to construct a semantic query vector, which is then used for efficient matching in a pre-built structured material index.

[0019] Step S2: Construct a semantic query vector based on the composite feature description, perform semantic matching in the industry design material library, and determine the set of relevant design materials associated with the design task.

[0020] Before determining the set of relevant design materials associated with the design task, the process also includes: using a semantic tagging multi-dimensional automatic annotation method to sort out the design materials in the industry design material library and generate a structured design material index.

[0021] Specifically, this technical solution aims to address the technical problem in existing collaborative editing methods where the lack of a semantically structured organization of the design material library leads to low efficiency and insufficient recall accuracy in subsequent semantic matching based on composite feature descriptions. To solve this problem, the following section elaborates on the structured design material index construction process required before performing semantic matching in the industry design material library to determine the relevant design material set.

[0022] Before implementing semantic matching, it is necessary to systematically organize the tags of all design materials in the industry design material library to generate a structured design material index, laying the foundation for subsequent rapid retrieval and accurate matching. Specifically, this solution uses a multi-dimensional automatic semantic tagging method to process each material. The core of this method is to use a pre-trained word embedding model or visual feature extraction model to transform unstructured material content into a high-dimensional vector representation. For text-based materials such as design specifications, semantic vectors can be obtained by directly inputting a pre-trained language model based on the Transformer architecture. For image or graphic materials such as design sketches, visual feature vectors are extracted using a convolutional neural network pre-trained on a large dataset. These vectors are all mapped into a high-dimensional semantic space. Subsequently, by calculating the cosine similarity between this vector and each candidate tag vector in the preset tag vector library, tags with similarity exceeding a preset threshold are automatically assigned to the material. The preset tag vector library is predefined by domain experts, covering various semantic categories required by industry design, and converts each tag text into a corresponding tag vector through a word embedding model, thereby achieving automated mapping from raw materials to multi-dimensional semantic tags.

[0023] After automatic labeling, a tree-like index structure is constructed based on the labeled tags. A multi-level classification tree is built according to the semantic hierarchy of the tags. Each material is attached to one or more leaf nodes in the classification tree according to its labeled tags, realizing multi-path rapid location of materials. During the index construction process, semantically repetitive tags are merged according to a preset thesaurus to ensure the compactness of the index and the recall rate during retrieval. When it is necessary to determine the relevant design material set based on the composite feature description, a semantic query vector is extracted from the composite feature description. Using the tree structure of the index, the tree is pruned layer by layer according to the similarity between the query vector and the tag vectors at each level to quickly locate the most relevant candidate nodes and extract the corresponding material list from these nodes, ultimately forming a highly accurate set of relevant design materials. This process reduces the computational complexity of semantic matching from linear to logarithmic level, significantly improving the retrieval efficiency of large-scale material databases.

[0024] The process of determining the relevant design material set associated with the design task includes: extracting semantic query vectors from composite feature descriptions, calculating the matching degree between the semantic query vectors and the design materials in the structured design material index, and directly determining the matching design materials as the relevant design material set if the matching degree is greater than or equal to a preset threshold. If the matching degree is less than the preset threshold, a real-time online sharing mechanism is triggered to obtain new design materials from external contributors, perform value assessment on the new design materials, and add qualified new design materials to the relevant design material set.

[0025] Specifically, this technical solution aims to address the problems in existing collaborative editing methods, such as limited recall, insufficient responsiveness to scarce resources, and inconsistent quality of externally introduced materials in the semantic matching process based on composite feature descriptions. To solve this technical problem, the specific implementation method for determining the set of relevant design materials associated with the design task is described in detail below.

[0026] After generating the composite feature description, the first step is to extract semantic query vectors that can be used for retrieval. This extraction process is essentially a process of quantifying and normalizing the high-dimensional semantic information contained in the composite feature description. Specifically, the job responsibility feature vector in the composite feature description is jointly represented with the semantic features of the design task scenario, and then mapped to a unified semantic space using word embedding technology to form an initial vector representation. Word embedding technology is essentially a method of converting discrete semantic units such as job attributes and keywords into continuous numerical vectors, which can ensure that semantically similar content is also close in distance in the vector space. After obtaining the initial vector representation, it is normalized to eliminate the influence of dimensions and generate the final semantic query vector that can be used for similarity calculation.

[0027] Subsequently, the core matching process begins, calculating the matching degree between the semantic query vector and the dynamically updated related vectors of each material in the pre-built structured design material index. This matching degree is usually measured by calculating the cosine value of the angle between the two vectors. The closer the cosine value is to 1, the more consistent the two vectors are in the semantic direction, that is, the higher the relevance of the material to the current task. A matching degree threshold is preset as the dividing line for determining whether the material is directly usable. When the calculated matching degree is greater than or equal to this preset matching degree threshold, it indicates that there are already highly matching existing resources in the material library. These materials are then directly retrieved from the index and initially sorted according to the matching degree from high to low to form a set of related design materials for subsequent processing.

[0028] When the matching degree between the semantic query vector and existing materials is lower than a preset matching degree threshold, it indicates that the internal material library cannot fully meet the unique needs of the current design task. At this time, a real-time online sharing mechanism is automatically triggered to expand the source of resources. This real-time online sharing mechanism first activates the preset online sharing interface, generates a notification message containing a summary of the current semantic query vector, and pushes it to the external contributors related to the design field. After receiving the notification, the external contributors can determine whether their resources meet the requirements based on the summary information and choose to upload new design materials. After receiving the uploaded materials, a preliminary verification is first performed to check whether the file format is compatible with the current design platform and whether the data integrity verification is passed. Only materials that pass the preliminary verification will be temporarily stored in the buffer for further processing.

[0029] For newly added materials entering the buffer, a multi-dimensional value assessment needs to be performed to determine whether they should be ultimately included in the relevant design material set. This value assessment system unfolds from three core dimensions: First, the relevance score, which recalculates the cosine similarity between the semantic vector of the new material and the original semantic query vector to ensure its high relevance to the task theme; second, the originality score, which calculates the hash value of the new material and compares it with the hash values ​​of existing materials in the material library. A hash value is a fingerprint that maps data of any size to a fixed-length string using a specific algorithm, which can efficiently detect whether content is duplicated or similar. If the hash values ​​are highly similar, it indicates low originality. It should be noted that the description here in the document is based on the principle of perceptual hashing technology. Perceptual hashing generates similar hash values ​​for visually or content-similar materials. Therefore, by comparing the similarity of perceptual hash values... The three scores are: 1) Relevance score, Originality score, and 2) Suitability score. The Relevance score is determined by the authorization permissions bound to the current collaborative editing agent's role. For example, some materials may involve specific copyrights or usage restrictions; if the content exceeds the current agent's authorization scope, the Suitability score will decrease. All three scores are dimensionless values ​​ranging from 0 to 1, and each dimension has a pre-set weighting coefficient. For example, the Relevance score can be set to 0.4, the Originality score to 0.3, and the Suitability score to 0.3. These weighting coefficients can be dynamically configured based on different design domain characteristics or specific enterprise management strategies. Finally, the three scores are weighted and combined to obtain a total value score, and a passing threshold is preset. Only when the total value score exceeds the passing threshold is the new material officially added to the relevant design material set. For example... Figure 3 The comparison chart of material value assessments shows that a newly added animated character model material has a relevance score of 0.75, an originality score of 0.9, and an applicability score of 0.85. The weighted total score is 0.75×0.4+0.9×0.3+0.85×0.3=0.825, exceeding the acceptable threshold of 0.7, and is therefore adopted. Another poster element material, although having a high relevance score of 0.9, was found to be duplicated with a material in the library after hash comparison, resulting in an originality score of only 0.2. Its relevance score is 0.7, and its total score is 0.9×0.4+0.2×0.3+0.7×0.3=0.36+0.06+0.27=0.63, below the acceptable threshold of 0.7, and is therefore rejected. This multi-dimensional assessment mechanism effectively ensures the overall quality of supplementary materials.

[0030] Step S3: Perform permission verification based on agent role on the relevant design material set, dynamically filter based on the job responsibilities of the collaborative editing agent and the sensitivity of the design materials, and generate a safe subset of design materials that the collaborative editing agent can safely access.

[0031] The process of generating a secure design material subset that can be safely accessed by the collaborative editing agent includes: for the relevant design material set, using a hierarchical job role binding rule with custom authorization permissions for permission matching; if the authorization permissions of the target design material in the relevant design material set match the job responsibilities of the collaborative editing agent, the target design material is retained in the authorized access list; if the authorization permissions of the target design material do not match the job responsibilities of the collaborative editing agent, design materials containing sensitive elements are identified and excluded from the target design material set; and a secure design material subset is generated based on the design materials retained in the authorized access list.

[0032] Specifically, this technical solution aims to address the technical problems in existing collaborative editing methods, such as the high risk of sensitive material leakage and low efficiency of acquiring compliant materials during multi-position collaboration due to the crude and unresponsive access control mechanism. To solve this problem, the following section elaborates on the specific implementation method for generating a subset of secure design materials that can be safely accessed by the collaborative editing agent.

[0033] After obtaining the relevant design material set through semantic matching, strict permission verification needs to be performed on each material in the set to ensure that only materials that conform to the current collaborative editing agent's job responsibilities can be used in subsequent processes. The core of this verification process lies in a set of custom authorization permission hierarchical job role binding rules. These binding rules are not static access control lists, but rather a mechanism that dynamically associates the job responsibility level with the authorization tags of the materials. Specifically, firstly, based on the currently logged-in job role of the collaborative editing agent, the corresponding permission level is extracted from a preset job responsibility database. The job responsibility database pre-stores the scope of responsibilities, accessible material types, and operation permissions such as read-only, edit, and publish for each job, such as junior designer, senior architect, and project manager. At the same time, each material in the relevant design material set carries an authorization tag corresponding to its content sensitivity, copyright status, and scope of use, such as "public material," "shared within the project," "confidential," and "editable only by the design team."

[0034] Subsequently, the permission matching engine is activated to compare the extracted job permission levels with the authorization tags of each material one by one. The matching rules follow the hierarchical binding principle, that is, the inheritance and inclusion relationship of permissions is defined according to the job level. For example, the permissions of a senior project manager automatically include the permissions of intermediate project team members, and the permissions of intermediate members include the permissions of junior members. The role inheritance mechanism can effectively reduce the redundancy of permission definitions and ensure that higher-level positions can access all compliant resources under their subordinates. When the authorization tag of the target material falls within the scope allowed by the current job permission level, the material is marked as accessible and retained in the authorized access list.

[0035] If the authorization tag of the target material does not match the current job's permission level, this solution will not simply discard the material directly. Instead, it will activate the sensitive element identification module for in-depth investigation. This module scans the material's metadata and content characteristics to identify whether it contains sensitive information unrelated to the job's responsibilities, such as financial data related to other projects, unpublished design drafts, or client privacy information. During the scanning process, it will use a preset sensitive word library, such as keywords like "confidential," "internal use," and "draft," as well as regular expressions to match specific encoding formats such as project numbers and contract numbers. If sensitive elements are detected and the permissions do not match, the material is determined to pose a risk of leakage and will be removed from the relevant design material collection. If no sensitive elements are detected, but the permissions still do not match, it may be a rule configuration issue. In this case, manual intervention will be triggered, and such anomalies will be pushed to the administrator for review. The administrator will then determine whether to adjust the permission rules or add the material to the whitelist, thereby avoiding the accidental blocking of usable materials due to rigid rules. For example, a cross-departmental shared design specification document may only allow access to the design department. However, if the marketing agent retrieves the document while performing a promotional task, it cannot directly decide whether to grant access because of the permission mismatch and the fact that the document does not contain sensitive elements. In this case, a manual intervention request will be generated, and the marketing department head and the design department head will consult to decide whether to grant temporary access.

[0036] After the above permission matching and sensitive element exclusion operations, all materials retained in the authorized access list are integrated into a secure design material subset. Each material in this subset undergoes dual verification: first, its authorized permissions match the current agent's job responsibilities; second, its content does not contain sensitive information beyond the scope of the job's knowledge. This dynamic filtering mechanism not only ensures the security of core assets during the design collaboration process but also avoids the rigidity and misjudgment that may result from fully automated decision-making by introducing human intervention to guide the handling of boundary situations, thus achieving a balance between security and availability. The generated subset will be passed to the subsequent version processing module for cross-agent consistent version management.

[0037] Step S4: Obtain the version change records of the design materials in the secure design material subset during the cross-agent collaborative editing process, identify and resolve version conflicts, generate a unified version of design materials that is consistent across agents, and build a complete ownership chain from design material contributors to design material users to generate a traceable design material group.

[0038] The process of generating a unified version of design materials that are consistent across agents includes: for design materials in the subset of safe design materials, obtaining the version change records of the same design material during the cross-agent collaborative editing process, comparing the historical change differences in the version change records using a version tracking model, identifying conflicting change nodes, and if conflicting change nodes are identified, performing difference merging processing on the conflicting change content. The difference merging processing includes a three-way merging algorithm based on the base version; after the difference merging processing, a unified version of design materials is generated.

[0039] Specifically, this technical solution aims to address the technical problems in existing collaborative editing methods, such as coarse-grained version conflict detection, a single merging strategy, and a lack of effective human intervention mechanisms, which lead to chaotic design material versions, difficulty in ensuring consistency, and difficulty in tracing responsibility during cross-position collaboration. To solve this technical problem, the specific implementation method of generating a unified version of design materials with cross-agent consistency and constructing a complete ownership chain in step S4 above will be described in detail below.

[0040] After obtaining the subset of security design materials, version consistency processing needs to be performed on each design material in the subset. First, by querying the collaboration log of the material library, the complete version change record of the same design material in the cross-agent collaborative editing process is obtained. This record is stored in time series and includes the timestamp of each change, the job information of the agent that performed the change, the type of change operation, and the content difference data before and after the change. These records form the basis for all subsequent version analysis.

[0041] Subsequently, a version tracking model is activated to compare historical changes in the records to identify differences and potential conflicts. The version tracking model is not a single algorithm, but a composite engine that dynamically selects difference detection strategies based on the type of material. For text-based design materials, such as design specifications and annotation documents, the version tracking model uses an edit distance-based difference comparison method. It quantifies the degree of textual changes between two versions by calculating the Levenshtein distance and marks operations such as insertion, deletion, and modification as difference nodes, generating a text-dimensional difference map. For image-based design materials, such as renderings and sketches, the model extracts the pixel matrix for each version and calculates the mean square error of pixel values ​​in the corresponding region. When the mean square error exceeds a preset threshold—for example, for an 8-bit depth image, a pixel value difference exceeding 30 is considered a significant change—the region is marked as a pixel-level change node. For 3D models or audio / video materials, feature vectors or keyframe sequences are extracted and compared to form a corresponding multi-dimensional change representation.

[0042] Based on the aforementioned difference map, conflict change nodes are further identified. When two or more agents are detected to have performed mutually exclusive modification operations in the same area of ​​the same material, such as text paragraphs, image areas, and timeline intervals, that area is marked as a conflict node. For example, when one agent modifies the text of a design specification, and another agent replaces the content in the same paragraph, these two changes constitute a conflict.

[0043] After identifying the conflict, the process enters the difference merging stage. This stage employs a hierarchical merging strategy, selecting different merging paths based on the material type and the nature of the conflict. For text-based materials, a three-way merging algorithm based on the base version is prioritized: first, the common base version of the two conflicting versions is located; then, the two changes are compared with the base version respectively to generate candidate content for merging. In the automatic decision-making stage, selection can be made according to preset rules, such as prioritizing the retention of changes made by agents with higher authority, or selecting the version that better matches the semantic parsing results of the current design task scenario through semantic similarity analysis. If the automatic decision is successful and the merged result has no obvious errors after grammatical checking, the merged result is adopted.

[0044] However, automatic merging often lacks semantic feasibility for binary design materials such as images and 3D models. For example, when two agents make conflicting modifications on different layers of the same rendering, pixels cannot be automatically merged like text. To address this, a conflict marking and manual intervention strategy is adopted: conflict areas are highlighted in the visual interface, and a conflict report is generated and pushed to the relevant agents or administrators with adjudication authority through the collaborative editing channel. The report includes version previews of the conflicting parties, information on the modifiers, and system-suggested solutions such as retaining a version or creating a branch version. The final adjudication is made by human intervention to ensure that the accuracy of key design assets is not affected by misjudgments by automatic algorithms.

[0045] For audio and video materials, conflicts may manifest as audio track overlays or misaligned editing points. This solution also uses a timeline visualization marker to guide manual selection or merging of audio and video clips by dragging and dropping on the timeline. This mechanism respects the complexity of professional creative fields and reduces the burden of manual investigation through automated conflict detection.

[0046] After merging the differences, the merged results are applied to the latest base version to generate unified version design assets. Subsequently, a link tracing process is initiated for this unified version: using an ownership mapping function, all version change records of the asset from its initial creation to the current unified version are traversed, extracting the identification information and operation time of each contributor and user, constructing a complete ownership link from source to end; the integrity of this link is verified to ensure that all nodes from the initial contributor to the end user are connected without breaks, thus determining clear ownership of the asset. Finally, the unified version design assets with clear ownership are integrated into a traceable design asset group and delivered to the subsequent secure transmission and intelligent generation module. This process ensures that in a high-frequency collaborative environment involving multiple agents, every version evolution is traceable, and the ownership and responsibility of each design asset are clearly defined.

[0047] The process of generating a traceable design material group includes: for a unified version of design materials, using an ownership mapping function to construct a link tracing relationship between contributors and users among various collaborative editing agents, and verifying the integrity of the link tracing relationship. If all nodes in the link tracing relationship from the initial contributor to the final user are connected completely without breaks, then the ownership of the unified version of design materials is determined to be clear, and the unified version of design materials with clear ownership are integrated to generate a traceable design material group.

[0048] Specifically, this technical solution aims to address the technical problem in existing collaborative editing methods where the lack of a record of the entire lifecycle of design materials from creation to final use makes it impossible to quickly and accurately trace the responsible party when copyright disputes or quality issues arise, resulting in ambiguous ownership of materials and affecting trust and overall efficiency in multi-agent collaboration. To solve this technical problem, the specific implementation method for generating traceable design material groups is described in detail below.

[0049] After obtaining a unified version of design materials, a complete ownership chain is constructed from material contributors to material users to generate a traceable design material group. The core of this process lies in using an ownership mapping function to achieve efficient chain tracing. The ownership mapping function is essentially a hash table-based data structure used to establish a fast mapping relationship between contributor identifiers and user sequences. Specifically, the metadata accompanying the unified version of design materials is first parsed. This metadata is usually organized in JSON format, which structurally records every change event that occurs during the cross-agent collaborative editing process of the materials, including the timestamp of each change, the identifier of the agent that performed the change, and the specific operation type such as creation, modification, review, and publication. By parsing this metadata, a complete list of contributor identifier information and all users' access logs to the materials can be extracted.

[0050] For each design resource file, a hash value is calculated for each contributor's identifier, preferably using cryptographic hash algorithms such as SHA-256 to ensure the uniqueness and collision resistance of the identifier. Then, using this hash value as the key, an ordered list of all user identifiers that have subsequent usage relationships with that contributor, arranged in chronological order of operation, is stored in the hash table, thus forming directional links from a specific contributor to its subsequent users. When a resource file has multiple contributors, a multi-key hash mapping mechanism is used to establish a dedicated link branch for each contributor, ensuring that all agents involved in the resource evolution are covered by the links, avoiding tracking omissions due to the large number of contributors.

[0051] To address the storage and query performance challenges posed by the large number of contributors in large-scale collaborative scenarios, the attribution mapping function can be optimized using a Bloom filter. A Bloom filter is a highly space-efficient probabilistic data structure. Its core principle is to use multiple independent hash functions to map elements to a long bit array. By checking whether each bit in the bit array corresponding to the hash value of the target element is 1, it can be quickly determined whether the element might exist in the set. In the early stages of link building, all known contributor identifiers are added to the Bloom filter. When a link needs to be established for a new contributor, the Bloom filter is queried first. If it returns no result, subsequent hash table lookups can be skipped, thereby significantly reducing unnecessary computational overhead and memory accesses and improving overall processing efficiency.

[0052] After the link is constructed, the integrity verification phase begins. The integrity verification phase involves traversing all the directed links stored in the hash table to check whether the connections between each node, starting from each initial contributor, passing through intermediate users, and ending at the final user, are complete and unbroken. The verification process must ensure that there are no missing intermediate links in the link. For example, after a contributor's modification record, the operation logs of subsequent users can be continuously connected. If all links meet the condition of continuous traceability, the ownership of the unified version design material is determined to be clear.

[0053] For scenarios involving external contributors, a digital signature verification mechanism can be introduced to further enhance the credibility and tamper-proof capability of the link. When submitting materials, external contributors can use their private key to sign their own identity and the hash value of the contributed content. The signature is then verified using the contributor's public key to ensure the authenticity of the contributor's identity and the integrity of the associated content. The RSA algorithm is one of the typical asymmetric encryption algorithms that can be used to implement this type of signature verification. The verified digital signature can effectively prevent the external contributor's identity from being impersonated or the link information from being maliciously tampered with.

[0054] After all the above processing steps, the unified version of the design materials with clear ownership, along with their accompanying complete contribution relationship metadata, are integrated and packaged to generate a traceable design material group. This traceable design material group not only includes the final result of collaborative editing, but also carries the entire contribution map from the initial idea to the final delivery, providing a solid data foundation for subsequent encrypted transmission, permission auditing, and tracing back potential intellectual property disputes, significantly improving the reliability and transparency of resource management in multi-agent collaborative design scenarios.

[0055] It should be further explained that in the version change handling stage, this method fully considers the complexity of engineering practice. For materials in the security design material subset, it not only obtains their version change records during cross-agent collaborative editing, but also introduces a version tracking model to compare historical change differences. If conflicting changes are identified, automatic merging is not blindly executed. Instead, the materials are first classified according to their file type: for text or code materials, a difference-based automatic merging algorithm can be used to generate merging suggestions; for binary files such as images and 3D models, a conflict marking mechanism is triggered, highlighting the conflict area in the visual interface, and sending a manual intervention guidance request to the relevant agents through the collaborative editing channel. The agent with higher authority or originality rights then manually decides the merging scheme. This layered processing strategy, combining automatic detection and manual guidance, ensures both the efficiency of version control and the accuracy of key design assets, making the technical solution more versatile and stable. The aforementioned 3D model plays two core roles in the design process: First, as a digital twin containing multi-dimensional engineering data such as geometry, weight, stress analysis parameters, accuracy parameters, and material parameters, it enables precise fitting of parts, interference detection, and stress analysis, replacing physical prototypes for testing and evaluation. Second, it is transforming from a tool specific to the design department into a core communication language across roles and processes, allowing process engineers, production personnel, quality inspectors, and other staff to intuitively understand assembly requirements and structural relationships through a lightweight model, thereby facilitating the collaborative link from R&D to mass production. Because the 3D model simultaneously carries precise engineering logic and complex process semantics, when two agents make conflicting modifications on different layers, they cannot be easily and automatically merged like merging text—automatic merging may damage the model's mechanical properties or assembly relationships. Manual adjudication by an agent with higher authority or originality is necessary to ensure the engineering accuracy and production feasibility of the design asset.

[0056] Step S5: Securely transmit the traceable design material group using a hierarchical encryption protocol related to the intelligent agent role path, and establish a protected transmission channel.

[0057] The establishment of a protected transmission channel includes: implementing a hierarchical encryption transmission protocol based on the sensitivity level of the design materials and the role hierarchy of the collaborative editing agents in the transmission process of traceable design material groups in multi-role intelligent agent collaborative editing scenarios; identifying the transmission path of the traceable design material group, and if the transmission path involves multiple collaborative editing agents at different role levels, performing segmented identity verification operations on each node in the transmission path, with each transmission segment independently verifying the identity of the receiving intelligent agent; and establishing a protected transmission channel through hierarchical encryption transmission protocols and segmented identity verification operations.

[0058] Specifically, this technical solution aims to address the lack of dynamic and fine-grained security protection mechanisms in existing collaborative editing methods when design materials are transmitted across different roles. These methods often employ a "one-size-fits-all" encryption strategy, which makes it difficult to balance transmission efficiency and data security. Furthermore, they are unable to effectively address the technical problems of identity theft or man-in-the-middle attacks that may occur when multiple roles are involved. To solve this technical problem, the specific implementation method for establishing a protected transmission channel will be described in detail below.

[0059] After obtaining the traceable design material set, it needs to be securely transmitted to the target agent. To this end, a hierarchical encryption transmission protocol is first implemented for the material set. The core of this protocol is to dynamically select the appropriate encryption algorithm and key strength based on the sensitivity level of the design material and the role level of the receiving collaborative editing agent. Specifically, a sensitivity assessment is performed on each design material in the traceable design material set. This assessment process can be based on the material's metadata tags, such as whether it is marked as "trade secret" or "unpublished draft," or by identifying the density of sensitive information it contains through a content analysis model, thereby assigning each material a sensitivity level, such as high, medium, and low. At the same time, the role level information of the target agent is obtained from the job responsibility database. This level reflects its scope of authority and the risk level of the transmission environment. For example, senior management positions and core design positions usually have higher security requirements. Based on a combination of material sensitivity and job level, a pre-defined encryption strategy mapping table is established: For highly sensitive materials, regardless of the recipient's level, AES-256 encryption is mandatory; for moderately sensitive materials, if the recipient is in a senior position, AES-256 is also used to ensure security, while if the recipient is in a regular position, AES-128 can be used to reduce encryption / decryption overhead and improve transmission efficiency; if the material sensitivity is low and the transmission network is determined to be an internal trusted network, plaintext transmission can be selected; a hybrid mode combining symmetric and asymmetric encryption is adopted: first, a temporarily generated session key is encrypted using the recipient's RSA public key, and then this session key is used in conjunction with the AES algorithm to encrypt the material content, forming the final encrypted packet; this method utilizes the efficiency of symmetric encryption while ensuring the security of session key distribution through asymmetric encryption.

[0060] During transmission, the path changes of the data stream are continuously monitored. When it is detected that the traceable design material group needs to pass through multiple agents of different role levels to reach its final destination, i.e., the transmission path involves multiple job nodes, a segmented authentication operation will be automatically triggered. This operation first parses out the complete transmission path graph, taking each job agent as a node and the transmission direction as an edge; then, the entire path is divided into several independent transmission segments, each corresponding to the direct transmission between two adjacent nodes; for each transmission segment, independent authentication is performed: the sending node generates a one-time authentication token based on the HMAC algorithm. The token is calculated using a pre-shared key between the sender and receiver, along with the current timestamp, ensuring its uniqueness and timeliness. Upon receiving a data packet, the receiving node first verifies the token's legitimacy. Only after successful verification does it receive the data and allow it to flow to the next node. If authentication fails in any transmission segment, transmission is immediately interrupted, and detailed log information is recorded, including the failed node, timestamp, number of attempts, etc. Manual intervention can be triggered according to preset policies, such as sending an alarm notification to the security administrator. The administrator can then investigate whether there is a malicious attack or configuration error. Transmission can only be resumed or the transmission path adjusted after manual confirmation.

[0061] After all segmented verifications are successful, the encrypted packet and metadata generated during the verification process, such as the encryption algorithm used, key identifier, verification results of each segment, and timestamp, are integrated to form a structured channel descriptor. This descriptor serves as the credential for this secure transmission and is delivered to the final recipient along with the materials for subsequent auditing and traceability. Through the synergistic effect of the above-mentioned hierarchical encryption and segmented verification, an efficient and secure protective transmission channel is built for the traceable design material group. While ensuring the confidentiality and integrity of design assets, it significantly improves the transmission reliability and anti-attack capability in multi-position collaborative scenarios.

[0062] Step S6: Transmit the traceable design material group to the design creation assistance module through the protected transmission channel. Combine the semantic parsing results of the design task description text to perform context-aware intelligent processing and generate the final output design materials that meet the design task requirements.

[0063] The process of generating final output design materials that meet the design task requirements includes: transmitting the traceable design material group to the industry design intelligent creation assistance module through a protected transmission channel; obtaining the semantic parsing results of the design task description text and extracting the context style consistency rendering requirements contained in the semantic parsing results of the design task description text; performing graphic processing on the traceable design material group according to the context style consistency rendering requirements, verifying whether the rendered design materials after graphic processing meet the semantic parsing results of the design task scene, and if they do, determining the rendered design materials as the final output design materials.

[0064] Specifically, this technical solution aims to address the technical problem in existing collaborative editing methods where the final output materials often deviate from the semantic requirements of the design task description, lack automated verification and adjustment mechanisms, and result in excessive reliance on repeated manual modifications and low efficiency in multi-position collaboration. To solve this technical problem, the specific implementation method for generating final output design materials that meet the design task requirements is described in detail below.

[0065] After the traceable design material group is securely delivered through the protected transmission channel, it is handed over to the industry design intelligent creation assistance module. This module first obtains the semantic parsing results of the design task description text associated with the current design task. These parsing results were generated in step S1 through context-dependent semantic parsing, which structurally contains the core design elements of the task, such as style orientation, color tone, and composition. The module further extracts rendering requirements related to the consistency of the context style from the semantic parsing results. These requirements are specifically manifested as constraints on the visual attributes of the final output material, including but not limited to color matching schemes such as main color tone, contrasting colors, shape proportions such as the golden ratio, symmetry, line style such as consistency of thickness, rounded corner radius, and compositional balance.

[0066] Based on the extracted style consistency rendering requirements, the intelligent creation assistance module begins to perform graphic processing on each design material in the traceable design material group. The processing first enters the comparison and matching stage: the current visual attributes of all graphic elements in the material group are compared with the target style requirements item by item. Taking color requirements as an example, the color difference value of the main color area in the material is calculated using the CIEDE2000 color difference formula recommended by the International Commission on Illumination. This formula can simulate the human eye's perception of different color differences. When the calculated color difference value is less than the preset threshold, it can be determined that the color of the area initially meets the requirements. For shape proportion requirements, the aspect ratio of the graphic elements, the distance between key points, etc., are measured and compared with the preset ideal proportion to calculate the percentage deviation. After the comparison and matching is completed, if any non-compliance is found, the graphic adjustment stage begins.

[0067] The graphic adjustment stage employs a layered processing strategy: For quantifiable attributes, such as line width, fill area size, and element alignment, automated adjustments are directly executed based on the parameters specified in the style requirements; for example, if the style requirement specifies that all outline widths be uniformly 2 pixels, all vector graphics in the source material will be scanned, and any non-compliant line widths will be automatically corrected to the target value; if center alignment of graphic elements is required, the offset between the canvas center and the element's bounding box center will be calculated and translation correction will be performed; for complex adjustments involving subjective aesthetic judgment, such as visual balance between multiple elements and optimization of color harmony, a manual intervention guidance mechanism is introduced on top of automated adjustments; specifically, the first First, the system attempts to automatically adjust the position and size of elements based on preset compositional aesthetic rules such as the rule of thirds and the golden ratio, generating an adjustment preview. The preview results are then scored to determine their compliance with style requirements. If the score is below a preset threshold (e.g., below 80 out of 100), a manual intervention process is automatically triggered. The current version, adjustment suggestions, and semantic requirements are packaged and sent to the lead designer or an AI agent with final review authority. The human designer then fine-tunes the element attributes through an interactive interface until they are subjectively satisfied. This mechanism leverages the efficiency advantage of machines in repetitive adjustments while preserving the aesthetic judgment of humans in artistic creation, avoiding the potentially rigid results that complete automation might lead to.

[0068] After completing the graphic processing, the rendered design materials undergo a final consistency verification process based on multi-dimensional quantitative indicators. For example, the layout balance is assessed by calculating the centroid distribution uniformity of all elements in the image; color harmony is assessed by analyzing the overlap area between the color histogram and the target color scheme; and symmetry ratio is calculated by measuring the symmetry similarity of elements on both sides of the key axis. Taking symmetry requirements as an example, the ratio of the mirror overlap area of ​​the rendered graphic relative to the central axis to the total area is calculated to obtain the symmetry ratio value. If this ratio value reaches a preset qualified threshold, it is determined to meet the semantic parsing results of the task scenario. Only when the scores of all key indicators meet the preset requirements are the currently rendered design materials determined as the final output design materials and submitted to the delivery process or the next collaborative stage. Through the complete closed loop of automatic comparison, intelligent adjustment, manual guidance, and quantitative verification, this method ensures that the final output design materials can accurately match the semantic intent of the original design task, significantly reducing the number of reworks caused by misunderstandings and improving the overall efficiency and quality of multi-position collaborative creation.

[0069] Through the coordination of the above steps, this application significantly improves the level of intelligence in material processing and the quality of the final design results.

[0070] Example 2: The above describes an industry-oriented intelligent agent collaborative editing method according to embodiments of this application. The following describes an industry-oriented intelligent agent collaborative editing system according to embodiments of this application, such as... Figure 4 As shown in the figure, an industry-oriented intelligent agent collaborative editing system according to an embodiment of this application includes: The feature integration unit is used to acquire the job role information and design task description text of the collaborative editing agent in the industry design scenario, and perform multimodal feature fusion to generate a composite feature description for the collaborative editing agent and the design task.

[0071] The material retrieval unit is used to construct semantic query vectors based on composite feature descriptions, perform semantic matching in the industry design material library, and determine the set of relevant design materials associated with the design task.

[0072] The permission filtering unit is used to perform permission verification based on the role of the intelligent agent on the relevant design material set. It dynamically filters based on the job responsibilities of the collaborative editing intelligent agent and the sensitivity of the design material, and generates a safe subset of design materials that the collaborative editing intelligent agent can safely access.

[0073] The version processing unit is used to obtain the version change records of design materials in the safe design material subset during the cross-agent collaborative editing process, identify and resolve version conflicts, generate a unified version of design materials that is consistent across agents, and build a complete ownership link from design material contributors to design material users to generate a traceable design material group.

[0074] The link tracing unit is used to securely transmit traceable design material groups using a hierarchical encryption protocol related to the agent role path, and to establish a protected transmission channel.

[0075] The intelligent generation unit is used to transmit traceable design material sets to the design creation assistance module through a protected transmission channel. It combines the semantic analysis results of the design task description text to perform context-aware intelligent processing and generate the final output design materials that meet the design task requirements.

[0076] Through the collaborative efforts of the aforementioned components, this application further enhances the level of intelligence in material processing and the quality of the final design outcome.

[0077] In summary, this application provides an intelligent agent collaborative editing method and system for industry design. This method fuses job role information with design task description text using multimodal features to generate composite feature descriptions and construct semantic query vectors, achieving accurate retrieval of design materials and dynamic permission filtering. Based on this, version change records are obtained, conflicts are resolved, and a complete ownership chain from contributor to user is constructed to generate traceable material groups. A protected transmission channel is then established through a hierarchical encryption protocol and segmented authentication. Finally, contextual style consistency rendering and quantitative verification are performed based on semantic parsing results to generate output materials that meet task requirements. Through the above technical solutions, this application effectively solves the complex technical problems existing in current collaborative editing methods, such as low material retrieval accuracy, coarse permission control, chaotic version management, difficulty in tracing responsibility, and transmission security risks. It significantly improves the efficiency of material reuse, the strength of intellectual property protection, and the quality of final output in multi-agent collaborative design scenarios, demonstrating outstanding substantive features and significant technological advancements.

[0078] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the system and unit described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0079] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0080] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A collaborative editing method for intelligent agents designed for specific industries, characterized in that, The method includes: Step S1: Obtain the job role information and design task description text of the collaborative editing agent in the industry design scenario, and perform multimodal feature fusion to generate a composite feature description for the collaborative editing agent and the design task; Step S2: Construct a semantic query vector based on the composite feature description, perform semantic matching in the industry design material library, and determine the set of relevant design materials associated with the design task; Step S3: Perform permission verification based on the agent role on the relevant design material set, and dynamically filter based on the job responsibilities of the collaborative editing agent and the sensitivity of the design materials to generate a safe subset of design materials that the collaborative editing agent can safely access; Step S4: Obtain the version change records of the design materials in the secure design material subset during the cross-agent collaborative editing process, identify and resolve version conflicts, generate a unified version of design materials that is consistent across agents, and construct a complete ownership link from design material contributors to design material users to generate a traceable design material group; Step S5: Securely transmit the traceable design material group using a hierarchical encryption protocol related to the intelligent agent role path, and establish a protected transmission channel; Step S6: Transmit the traceable design material group to the design creation assistance module through the protective transmission channel, and perform context-aware intelligent processing based on the semantic parsing results of the design task description text to generate the final output design material that meets the design task requirements.

2. The intelligent agent collaborative editing method for industry-oriented design according to claim 1, characterized in that, The step S1 of generating a composite feature description for the collaborative editing agent and the design task includes: Based on the job role information of the collaborative editing agent, extract the job responsibility feature vector; perform context-dependent semantic parsing on the design task description text to generate design task scenario semantic features; and use an attention fusion mechanism to interactively fuse the job responsibility feature vector and the design task scenario semantic features to generate a composite feature description.

3. The intelligent agent collaborative editing method for industry-oriented design according to claim 1, characterized in that, Before determining the set of relevant design materials associated with the design task in step S2, the following steps are also included: A semantic tagging multi-dimensional automatic annotation method is used to organize the design materials in the industry design material library with tags and generate a structured design material index.

4. The intelligent agent collaborative editing method for industry-oriented design according to claim 3, characterized in that, The set of relevant design materials associated with the design task determined in step S2 includes: The semantic query vector is extracted from the composite feature description, and the matching degree between the semantic query vector and the design materials in the structured design material index is calculated. If the matching degree is greater than or equal to a preset threshold, the matching design material is directly determined as the relevant design material set. If the matching degree is less than the preset threshold, a real-time online sharing mechanism is triggered to obtain new design materials from external contributors, and the value of the new design materials is evaluated. The qualified new design materials are added to the relevant design material set.

5. The intelligent agent collaborative editing method for industry-oriented design according to claim 1, characterized in that, The subset of secure design materials that the collaborative editing agent can securely access in step S3 includes: For the aforementioned set of design materials, a hierarchical job role binding rule with custom authorization permissions is used for permission matching. If the authorization permission of the target design material in the set of design materials matches the job responsibilities of the collaborative editing agent, the target design material is retained in the authorized access list. If the authorization permission of the target design material does not match the job responsibilities of the collaborative editing agent, design materials containing sensitive elements are identified and excluded. Based on the design materials retained in the authorized access list, a subset of safe design materials is generated.

6. The intelligent agent collaborative editing method for industry-oriented design according to claim 1, characterized in that, The unified version design material generated in step S4, which ensures cross-agent consistency, includes: For the design materials in the aforementioned subset of security design materials, version change records of the same design material during cross-agent collaborative editing are obtained. A version tracking model is used to compare the historical change differences in the version change records and identify conflicting change nodes. If a conflicting change node is identified, a difference merging process is performed on the conflicting change content. The difference merging process includes a three-way merging algorithm based on the base version. After the difference merging process, a unified version of the design material is generated.

7. The intelligent agent collaborative editing method for industry-oriented design according to claim 6, characterized in that, The step S4 of generating a traceable design material group includes: For the unified version design materials, an ownership mapping function is used to construct the link tracing relationship between contributors and users among the collaborative editing agents, and the integrity of the link tracing relationship is verified. If all nodes from the initial contributor to the final user in the link tracing relationship are connected completely without breaks, it is determined that the ownership of the unified version design materials is clear, and the unified version design materials with clear ownership are integrated to generate a traceable design material group.

8. The intelligent agent collaborative editing method for industry-oriented design according to claim 1, characterized in that, Establishing the protected transmission channel in step S5 includes: For the transmission process of the traceable design material group in a multi-role intelligent agent collaborative editing scenario, a hierarchical encryption transmission protocol is implemented based on the sensitivity level of the design material and the role level of the collaborative editing intelligent agent; the transmission path of the traceable design material group is identified, and if the transmission path involves multiple collaborative editing intelligent agents at different role levels, segmented identity verification is performed on each node in the transmission path, with each transmission segment independently verifying the identity of the receiving intelligent agent; the protected transmission channel is established through the hierarchical encryption transmission protocol and the segmented identity verification operation.

9. The intelligent agent collaborative editing method for industry-oriented design according to claim 8, characterized in that, The final output design materials that meet the design task requirements generated in step S6 include: The traceable design material group is transmitted to the industry design intelligent creation assistance module through the protective transmission channel; the semantic parsing result of the design task description text is obtained, and the context style consistency rendering requirements contained in the semantic parsing result of the design task description text are extracted; according to the context style consistency rendering requirements, the traceable design material group is subjected to graphic processing, and the rendered design material after graphic processing is verified to conform to the semantic parsing result of the design task scene. If it conforms, the rendered design material is determined as the final output design material.

10. An industry-oriented intelligent agent collaborative editing system, used to implement the industry-oriented intelligent agent collaborative editing method as described in any one of claims 1-9, characterized in that, The system includes: The feature integration unit is used to acquire the job role information and design task description text of the collaborative editing agent in the industry design scenario, and perform multimodal feature fusion to generate a composite feature description for the collaborative editing agent and the design task. The material retrieval unit is used to construct a semantic query vector based on the composite feature description, perform semantic matching in the industry design material library, and determine the set of relevant design materials associated with the design task. The permission filtering unit is used to perform permission verification based on the role of the intelligent agent on the relevant design material set, and to dynamically filter based on the job responsibilities of the collaborative editing intelligent agent and the sensitivity of the design materials to generate a safe subset of design materials that the collaborative editing intelligent agent can safely access. The version processing unit is used to obtain the version change records of the design materials in the secure design material subset during the cross-agent collaborative editing process, identify and resolve version conflicts, generate a unified version of design materials that is consistent across agents, and construct a complete ownership link from design material contributors to design material users to generate a traceable design material group. The link tracing unit is used to securely transmit the traceable design material group using a hierarchical encryption protocol related to the intelligent agent role path, and to establish a protected transmission channel. The intelligent generation unit is used to transmit the traceable design material group to the design creation assistance module through the protective transmission channel, and perform context-aware intelligent processing based on the semantic parsing results of the design task description text to generate the final output design material that meets the design task requirements.

Images